1.

Exploratory Data Analysis

1. Exploratory Data Analysis

This chapter presents the assumptions, principles, and techniques necessary to gain
insight into data via EDA--exploratory data analysis.

1. EDA Introduction 2. EDA Assumptions

1. What is EDA? 1. Underlying Assumptions
2. EDA vs Classical & Bayesian 2. Importance
3. EDA vs Summary 3. Techniques for Testing
4. EDA Goals Assumptions
5. The Role of Graphics 4. Interpretation of 4-Plot
6. An EDA/Graphics Example 5. Consequences
7. General Problem Categories

3. EDA Techniques 4. EDA Case Studies

1. Introduction 1. Introduction
2. Analysis Questions 2. By Problem Category
3. Graphical Techniques: Alphabetical
4. Graphical Techniques: By Problem
Category
5. Quantitative Techniques
6. Probability Distributions

Detailed Chapter Table of Contents

References
Dataplot Commands for EDA Techniques

http://www.itl.nist.gov/div898/handbook/eda/eda.htm [7/1/2003 5:20:57 PM]

1. Exploratory Data Analysis

1. Exploratory Data Analysis - Detailed Table of

Contents [1.]

This chapter presents the assumptions, principles, and techniques necessary to gain insight into
data via EDA--exploratory data analysis.
1. EDA Introduction [1.1.]
1. What is EDA? [1.1.1.]
2. How Does Exploratory Data Analysis differ from Classical Data Analysis? [1.1.2.]
1. Model [1.1.2.1.]
2. Focus [1.1.2.2.]
3. Techniques [1.1.2.3.]
4. Rigor [1.1.2.4.]
5. Data Treatment [1.1.2.5.]
6. Assumptions [1.1.2.6.]
3. How Does Exploratory Data Analysis Differ from Summary Analysis? [1.1.3.]
4. What are the EDA Goals? [1.1.4.]
5. The Role of Graphics [1.1.5.]
6. An EDA/Graphics Example [1.1.6.]
7. General Problem Categories [1.1.7.]

2. EDA Assumptions [1.2.]

1. Underlying Assumptions [1.2.1.]
2. Importance [1.2.2.]
3. Techniques for Testing Assumptions [1.2.3.]
4. Interpretation of 4-Plot [1.2.4.]
5. Consequences [1.2.5.]
1. Consequences of Non-Randomness [1.2.5.1.]
2. Consequences of Non-Fixed Location Parameter [1.2.5.2.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (1 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

3. Consequences of Non-Fixed Variation Parameter [1.2.5.3.]

4. Consequences Related to Distributional Assumptions [1.2.5.4.]

3. EDA Techniques [1.3.]

1. Introduction [1.3.1.]
2. Analysis Questions [1.3.2.]
3. Graphical Techniques: Alphabetic [1.3.3.]
1. Autocorrelation Plot [1.3.3.1.]
1. Autocorrelation Plot: Random Data [1.3.3.1.1.]
2. Autocorrelation Plot: Moderate Autocorrelation [1.3.3.1.2.]
3. Autocorrelation Plot: Strong Autocorrelation and Autoregressive
Model [1.3.3.1.3.]
4. Autocorrelation Plot: Sinusoidal Model [1.3.3.1.4.]
2. Bihistogram [1.3.3.2.]
3. Block Plot [1.3.3.3.]
4. Bootstrap Plot [1.3.3.4.]
5. Box-Cox Linearity Plot [1.3.3.5.]
6. Box-Cox Normality Plot [1.3.3.6.]
7. Box Plot [1.3.3.7.]
8. Complex Demodulation Amplitude Plot [1.3.3.8.]
9. Complex Demodulation Phase Plot [1.3.3.9.]
10. Contour Plot [1.3.3.10.]
1. DEX Contour Plot [1.3.3.10.1.]
11. DEX Scatter Plot [1.3.3.11.]
12. DEX Mean Plot [1.3.3.12.]
13. DEX Standard Deviation Plot [1.3.3.13.]
14. Histogram [1.3.3.14.]
1. Histogram Interpretation: Normal [1.3.3.14.1.]
2. Histogram Interpretation: Symmetric, Non-Normal,
Short-Tailed [1.3.3.14.2.]
3. Histogram Interpretation: Symmetric, Non-Normal,
Long-Tailed [1.3.3.14.3.]
4. Histogram Interpretation: Symmetric and Bimodal [1.3.3.14.4.]
5. Histogram Interpretation: Bimodal Mixture of 2 Normals [1.3.3.14.5.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (2 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

6. Histogram Interpretation: Skewed (Non-Normal) Right [1.3.3.14.6.]

7. Histogram Interpretation: Skewed (Non-Symmetric) Left [1.3.3.14.7.]
8. Histogram Interpretation: Symmetric with Outlier [1.3.3.14.8.]
15. Lag Plot [1.3.3.15.]
1. Lag Plot: Random Data [1.3.3.15.1.]
2. Lag Plot: Moderate Autocorrelation [1.3.3.15.2.]
3. Lag Plot: Strong Autocorrelation and Autoregressive
Model [1.3.3.15.3.]
4. Lag Plot: Sinusoidal Models and Outliers [1.3.3.15.4.]
16. Linear Correlation Plot [1.3.3.16.]
17. Linear Intercept Plot [1.3.3.17.]
18. Linear Slope Plot [1.3.3.18.]
19. Linear Residual Standard Deviation Plot [1.3.3.19.]
20. Mean Plot [1.3.3.20.]
21. Normal Probability Plot [1.3.3.21.]
1. Normal Probability Plot: Normally Distributed Data [1.3.3.21.1.]
2. Normal Probability Plot: Data Have Short Tails [1.3.3.21.2.]
3. Normal Probability Plot: Data Have Long Tails [1.3.3.21.3.]
4. Normal Probability Plot: Data are Skewed Right [1.3.3.21.4.]
22. Probability Plot [1.3.3.22.]
23. Probability Plot Correlation Coefficient Plot [1.3.3.23.]
24. Quantile-Quantile Plot [1.3.3.24.]
25. Run-Sequence Plot [1.3.3.25.]
26. Scatter Plot [1.3.3.26.]
1. Scatter Plot: No Relationship [1.3.3.26.1.]
2. Scatter Plot: Strong Linear (positive correlation)
Relationship [1.3.3.26.2.]
3. Scatter Plot: Strong Linear (negative correlation)
Relationship [1.3.3.26.3.]
4. Scatter Plot: Exact Linear (positive correlation)
Relationship [1.3.3.26.4.]
5. Scatter Plot: Quadratic Relationship [1.3.3.26.5.]
6. Scatter Plot: Exponential Relationship [1.3.3.26.6.]
7. Scatter Plot: Sinusoidal Relationship (damped) [1.3.3.26.7.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (3 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

8. Scatter Plot: Variation of Y Does Not Depend on X

(homoscedastic) [1.3.3.26.8.]
9. Scatter Plot: Variation of Y Does Depend on X
(heteroscedastic) [1.3.3.26.9.]
10. Scatter Plot: Outlier [1.3.3.26.10.]
11. Scatterplot Matrix [1.3.3.26.11.]
12. Conditioning Plot [1.3.3.26.12.]
27. Spectral Plot [1.3.3.27.]
1. Spectral Plot: Random Data [1.3.3.27.1.]
2. Spectral Plot: Strong Autocorrelation and Autoregressive
Model [1.3.3.27.2.]
3. Spectral Plot: Sinusoidal Model [1.3.3.27.3.]
28. Standard Deviation Plot [1.3.3.28.]
29. Star Plot [1.3.3.29.]
30. Weibull Plot [1.3.3.30.]
31. Youden Plot [1.3.3.31.]
1. DEX Youden Plot [1.3.3.31.1.]
32. 4-Plot [1.3.3.32.]
33. 6-Plot [1.3.3.33.]
4. Graphical Techniques: By Problem Category [1.3.4.]
5. Quantitative Techniques [1.3.5.]
1. Measures of Location [1.3.5.1.]
2. Confidence Limits for the Mean [1.3.5.2.]
3. Two-Sample t-Test for Equal Means [1.3.5.3.]
1. Data Used for Two-Sample t-Test [1.3.5.3.1.]
4. One-Factor ANOVA [1.3.5.4.]
5. Multi-factor Analysis of Variance [1.3.5.5.]
6. Measures of Scale [1.3.5.6.]
7. Bartlett's Test [1.3.5.7.]
8. Chi-Square Test for the Standard Deviation [1.3.5.8.]
1. Data Used for Chi-Square Test for the Standard Deviation [1.3.5.8.1.]
9. F-Test for Equality of Two Standard Deviations [1.3.5.9.]
10. Levene Test for Equality of Variances [1.3.5.10.]
11. Measures of Skewness and Kurtosis [1.3.5.11.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (4 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

12. Autocorrelation [1.3.5.12.]

13. Runs Test for Detecting Non-randomness [1.3.5.13.]
14. Anderson-Darling Test [1.3.5.14.]
15. Chi-Square Goodness-of-Fit Test [1.3.5.15.]
16. Kolmogorov-Smirnov Goodness-of-Fit Test [1.3.5.16.]
17. Grubbs' Test for Outliers [1.3.5.17.]
18. Yates Analysis [1.3.5.18.]
1. Defining Models and Prediction Equations [1.3.5.18.1.]
2. Important Factors [1.3.5.18.2.]
6. Probability Distributions [1.3.6.]
1. What is a Probability Distribution [1.3.6.1.]
2. Related Distributions [1.3.6.2.]
3. Families of Distributions [1.3.6.3.]
4. Location and Scale Parameters [1.3.6.4.]
5. Estimating the Parameters of a Distribution [1.3.6.5.]
1. Method of Moments [1.3.6.5.1.]
2. Maximum Likelihood [1.3.6.5.2.]
3. Least Squares [1.3.6.5.3.]
4. PPCC and Probability Plots [1.3.6.5.4.]
6. Gallery of Distributions [1.3.6.6.]
1. Normal Distribution [1.3.6.6.1.]
2. Uniform Distribution [1.3.6.6.2.]
3. Cauchy Distribution [1.3.6.6.3.]
4. t Distribution [1.3.6.6.4.]
5. F Distribution [1.3.6.6.5.]
6. Chi-Square Distribution [1.3.6.6.6.]
7. Exponential Distribution [1.3.6.6.7.]
8. Weibull Distribution [1.3.6.6.8.]
9. Lognormal Distribution [1.3.6.6.9.]
10. Fatigue Life Distribution [1.3.6.6.10.]
11. Gamma Distribution [1.3.6.6.11.]
12. Double Exponential Distribution [1.3.6.6.12.]
13. Power Normal Distribution [1.3.6.6.13.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (5 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

14. Power Lognormal Distribution [1.3.6.6.14.]

15. Tukey-Lambda Distribution [1.3.6.6.15.]
16. Extreme Value Type I Distribution [1.3.6.6.16.]
17. Beta Distribution [1.3.6.6.17.]
18. Binomial Distribution [1.3.6.6.18.]
19. Poisson Distribution [1.3.6.6.19.]
7. Tables for Probability Distributions [1.3.6.7.]
1. Cumulative Distribution Function of the Standard Normal
Distribution [1.3.6.7.1.]
2. Upper Critical Values of the Student's-t Distribution [1.3.6.7.2.]
3. Upper Critical Values of the F Distribution [1.3.6.7.3.]
4. Critical Values of the Chi-Square Distribution [1.3.6.7.4.]
5. Critical Values of the t* Distribution [1.3.6.7.5.]
6. Critical Values of the Normal PPCC Distribution [1.3.6.7.6.]

4. EDA Case Studies [1.4.]

1. Case Studies Introduction [1.4.1.]
2. Case Studies [1.4.2.]
1. Normal Random Numbers [1.4.2.1.]
1. Background and Data [1.4.2.1.1.]
2. Graphical Output and Interpretation [1.4.2.1.2.]
3. Quantitative Output and Interpretation [1.4.2.1.3.]
4. Work This Example Yourself [1.4.2.1.4.]
2. Uniform Random Numbers [1.4.2.2.]
1. Background and Data [1.4.2.2.1.]
2. Graphical Output and Interpretation [1.4.2.2.2.]
3. Quantitative Output and Interpretation [1.4.2.2.3.]
4. Work This Example Yourself [1.4.2.2.4.]
3. Random Walk [1.4.2.3.]
1. Background and Data [1.4.2.3.1.]
2. Test Underlying Assumptions [1.4.2.3.2.]
3. Develop A Better Model [1.4.2.3.3.]
4. Validate New Model [1.4.2.3.4.]
5. Work This Example Yourself [1.4.2.3.5.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (6 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

4. Josephson Junction Cryothermometry [1.4.2.4.]

1. Background and Data [1.4.2.4.1.]
2. Graphical Output and Interpretation [1.4.2.4.2.]
3. Quantitative Output and Interpretation [1.4.2.4.3.]
4. Work This Example Yourself [1.4.2.4.4.]
5. Beam Deflections [1.4.2.5.]
1. Background and Data [1.4.2.5.1.]
2. Test Underlying Assumptions [1.4.2.5.2.]
3. Develop a Better Model [1.4.2.5.3.]
4. Validate New Model [1.4.2.5.4.]
5. Work This Example Yourself [1.4.2.5.5.]
6. Filter Transmittance [1.4.2.6.]
1. Background and Data [1.4.2.6.1.]
2. Graphical Output and Interpretation [1.4.2.6.2.]
3. Quantitative Output and Interpretation [1.4.2.6.3.]
4. Work This Example Yourself [1.4.2.6.4.]
7. Standard Resistor [1.4.2.7.]
1. Background and Data [1.4.2.7.1.]
2. Graphical Output and Interpretation [1.4.2.7.2.]
3. Quantitative Output and Interpretation [1.4.2.7.3.]
4. Work This Example Yourself [1.4.2.7.4.]
8. Heat Flow Meter 1 [1.4.2.8.]
1. Background and Data [1.4.2.8.1.]
2. Graphical Output and Interpretation [1.4.2.8.2.]
3. Quantitative Output and Interpretation [1.4.2.8.3.]
4. Work This Example Yourself [1.4.2.8.4.]
9. Airplane Glass Failure Time [1.4.2.9.]
1. Background and Data [1.4.2.9.1.]
2. Graphical Output and Interpretation [1.4.2.9.2.]
3. Weibull Analysis [1.4.2.9.3.]
4. Lognormal Analysis [1.4.2.9.4.]
5. Gamma Analysis [1.4.2.9.5.]
6. Power Normal Analysis [1.4.2.9.6.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (7 of 8) [7/1/2003 5:20:49 PM]

1. Exploratory Data Analysis

7. Power Lognormal Analysis [1.4.2.9.7.]

8. Work This Example Yourself [1.4.2.9.8.]
10. Ceramic Strength [1.4.2.10.]
1. Background and Data [1.4.2.10.1.]
2. Analysis of the Response Variable [1.4.2.10.2.]
3. Analysis of the Batch Effect [1.4.2.10.3.]
4. Analysis of the Lab Effect [1.4.2.10.4.]
5. Analysis of Primary Factors [1.4.2.10.5.]
6. Work This Example Yourself [1.4.2.10.6.]
3. References For Chapter 1: Exploratory Data Analysis [1.4.3.]

http://www.itl.nist.gov/div898/handbook/eda/eda_d.htm (8 of 8) [7/1/2003 5:20:49 PM]

1.1. EDA Introduction

1. Exploratory Data Analysis

1.1. EDA Introduction

Summary What is exploratory data analysis? How did it begin? How and where
did it originate? How is it differentiated from other data analysis
approaches, such as classical and Bayesian? Is EDA the same as
statistical graphics? What role does statistical graphics play in EDA? Is
statistical graphics identical to EDA?
These questions and related questions are dealt with in this section. This
section answers these questions and provides the necessary frame of
reference for EDA assumptions, principles, and techniques.

Table of 1. What is EDA?

Contents for 2. EDA versus Classical and Bayesian
Section 1
1. Models
2. Focus
3. Techniques
4. Rigor
5. Data Treatment
6. Assumptions
3. EDA vs Summary
4. EDA Goals
5. The Role of Graphics
6. An EDA/Graphics Example
7. General Problem Categories

http://www.itl.nist.gov/div898/handbook/eda/section1/eda1.htm [7/1/2003 5:20:57 PM]

1.1.1. What is EDA?

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.1. What is EDA?

Approach Exploratory Data Analysis (EDA) is an approach/philosophy for data
analysis that employs a variety of techniques (mostly graphical) to
1. maximize insight into a data set;
2. uncover underlying structure;
3. extract important variables;
4. detect outliers and anomalies;
5. test underlying assumptions;
6. develop parsimonious models; and
7. determine optimal factor settings.

Focus The EDA approach is precisely that--an approach--not a set of

techniques, but an attitude/philosophy about how a data analysis should
be carried out.

Philosophy EDA is not identical to statistical graphics although the two terms are
used almost interchangeably. Statistical graphics is a collection of
techniques--all graphically based and all focusing on one data
characterization aspect. EDA encompasses a larger venue; EDA is an
approach to data analysis that postpones the usual assumptions about
what kind of model the data follow with the more direct approach of
allowing the data itself to reveal its underlying structure and model.
EDA is not a mere collection of techniques; EDA is a philosophy as to
how we dissect a data set; what we look for; how we look; and how we
interpret. It is true that EDA heavily uses the collection of techniques
that we call "statistical graphics", but it is not identical to statistical
graphics per se.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda11.htm (1 of 2) [7/1/2003 5:20:57 PM]

1.1.1. What is EDA?

History The seminal work in EDA is Exploratory Data Analysis, Tukey, (1977).
Over the years it has benefitted from other noteworthy publications such
as Data Analysis and Regression, Mosteller and Tukey (1977),
Interactive Data Analysis, Hoaglin (1977), The ABC's of EDA,
Velleman and Hoaglin (1981) and has gained a large following as "the"
way to analyze a data set.

Techniques Most EDA techniques are graphical in nature with a few quantitative
techniques. The reason for the heavy reliance on graphics is that by its
very nature the main role of EDA is to open-mindedly explore, and
graphics gives the analysts unparalleled power to do so, enticing the
data to reveal its structural secrets, and being always ready to gain some
new, often unsuspected, insight into the data. In combination with the
natural pattern-recognition capabilities that we all possess, graphics
provides, of course, unparalleled power to carry this out.
The particular graphical techniques employed in EDA are often quite
simple, consisting of various techniques of:
1. Plotting the raw data (such as data traces, histograms,
bihistograms, probability plots, lag plots, block plots, and Youden
plots.
2. Plotting simple statistics such as mean plots, standard deviation
plots, box plots, and main effects plots of the raw data.
3. Positioning such plots so as to maximize our natural
pattern-recognition abilities, such as using multiple plots per
page.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda11.htm (2 of 2) [7/1/2003 5:20:57 PM]

1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.2. How Does Exploratory Data Analysis

differ from Classical Data Analysis?
Data EDA is a data analysis approach. What other data analysis approaches
Analysis exist and how does EDA differ from these other approaches? Three
Approaches popular data analysis approaches are:
1. Classical
2. Exploratory (EDA)
3. Bayesian

Paradigms These three approaches are similar in that they all start with a general
for Analysis science/engineering problem and all yield science/engineering
Techniques conclusions. The difference is the sequence and focus of the
intermediate steps.
For classical analysis, the sequence is
Problem => Data => Model => Analysis => Conclusions
For EDA, the sequence is
Problem => Data => Analysis => Model => Conclusions
For Bayesian, the sequence is
Problem => Data => Model => Prior Distribution => Analysis =>
Conclusions

http://www.itl.nist.gov/div898/handbook/eda/section1/eda12.htm (1 of 2) [7/1/2003 5:20:57 PM]

1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

Method of Thus for classical analysis, the data collection is followed by the
dealing with imposition of a model (normality, linearity, etc.) and the analysis,
underlying estimation, and testing that follows are focused on the parameters of
model for that model. For EDA, the data collection is not followed by a model
the data imposition; rather it is followed immediately by analysis with a goal of
distinguishes inferring what model would be appropriate. Finally, for a Bayesian
the 3 analysis, the analyst attempts to incorporate scientific/engineering
approaches knowledge/expertise into the analysis by imposing a data-independent
distribution on the parameters of the selected model; the analysis thus
consists of formally combining both the prior distribution on the
parameters and the collected data to jointly make inferences and/or test
assumptions about the model parameters.
In the real world, data analysts freely mix elements of all of the above
three approaches (and other approaches). The above distinctions were
made to emphasize the major differences among the three approaches.

Further Focusing on EDA versus classical, these two approaches differ as

discussion of follows:
the 1. Models
distinction
between the 2. Focus
classical and 3. Techniques
EDA 4. Rigor
approaches
5. Data Treatment
6. Assumptions

http://www.itl.nist.gov/div898/handbook/eda/section1/eda12.htm (2 of 2) [7/1/2003 5:20:57 PM]

1.1.2.1. Model

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.1. Model
Classical The classical approach imposes models (both deterministic and
probabilistic) on the data. Deterministic models include, for example,
regression models and analysis of variance (ANOVA) models. The most
common probabilistic model assumes that the errors about the
deterministic model are normally distributed--this assumption affects the
validity of the ANOVA F tests.

Exploratory The Exploratory Data Analysis approach does not impose deterministic
or probabilistic models on the data. On the contrary, the EDA approach
allows the data to suggest admissible models that best fit the data.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda121.htm [7/1/2003 5:20:58 PM]

1.1.2.2. Focus

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.2. Focus
Classical The two approaches differ substantially in focus. For classical analysis,
the focus is on the model--estimating parameters of the model and
generating predicted values from the model.

Exploratory For exploratory data analysis, the focus is on the data--its structure,
outliers, and models suggested by the data.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda122.htm [7/1/2003 5:20:58 PM]

1.1.2.3. Techniques

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.3. Techniques
Classical Classical techniques are generally quantitative in nature. They include
ANOVA, t tests, chi-squared tests, and F tests.

Exploratory EDA techniques are generally graphical. They include scatter plots,
character plots, box plots, histograms, bihistograms, probability plots,
residual plots, and mean plots.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda123.htm [7/1/2003 5:20:58 PM]

1.1.2.4. Rigor

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.4. Rigor
Classical Classical techniques serve as the probabilistic foundation of science and
engineering; the most important characteristic of classical techniques is
that they are rigorous, formal, and "objective".

Exploratory EDA techniques do not share in that rigor or formality. EDA techniques
make up for that lack of rigor by being very suggestive, indicative, and
insightful about what the appropriate model should be.
EDA techniques are subjective and depend on interpretation which may
differ from analyst to analyst, although experienced analysts commonly
arrive at identical conclusions.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda124.htm [7/1/2003 5:20:58 PM]

1.1.2.5. Data Treatment

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.5. Data Treatment

Classical Classical estimation techniques have the characteristic of taking all of
the data and mapping the data into a few numbers ("estimates"). This is
both a virtue and a vice. The virtue is that these few numbers focus on
important characteristics (location, variation, etc.) of the population. The
vice is that concentrating on these few characteristics can filter out other
characteristics (skewness, tail length, autocorrelation, etc.) of the same
population. In this sense there is a loss of information due to this
"filtering" process.

Exploratory The EDA approach, on the other hand, often makes use of (and shows)
all of the available data. In this sense there is no corresponding loss of
information.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda125.htm [7/1/2003 5:20:58 PM]

1.1.2.6. Assumptions

1. Exploratory Data Analysis

1.1. EDA Introduction
1.1.2. How Does Exploratory Data Analysis differ from Classical Data Analysis?

1.1.2.6. Assumptions
Classical The "good news" of the classical approach is that tests based on
classical techniques are usually very sensitive--that is, if a true shift in
location, say, has occurred, such tests frequently have the power to
detect such a shift and to conclude that such a shift is "statistically
significant". The "bad news" is that classical tests depend on underlying
assumptions (e.g., normality), and hence the validity of the test
conclusions becomes dependent on the validity of the underlying
assumptions. Worse yet, the exact underlying assumptions may be
unknown to the analyst, or if known, untested. Thus the validity of the
scientific conclusions becomes intrinsically linked to the validity of the
underlying assumptions. In practice, if such assumptions are unknown
or untested, the validity of the scientific conclusions becomes suspect.

Exploratory Many EDA techniques make little or no assumptions--they present and

show the data--all of the data--as is, with fewer encumbering
assumptions.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda126.htm [7/1/2003 5:20:58 PM]

1.1.3. How Does Exploratory Data Analysis Differ from Summary Analysis?

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.3. How Does Exploratory Data Analysis

Differ from Summary Analysis?
Summary A summary analysis is simply a numeric reduction of a historical data
set. It is quite passive. Its focus is in the past. Quite commonly, its
purpose is to simply arrive at a few key statistics (for example, mean
and standard deviation) which may then either replace the data set or be
added to the data set in the form of a summary table.

Exploratory In contrast, EDA has as its broadest goal the desire to gain insight into
the engineering/scientific process behind the data. Whereas summary
statistics are passive and historical, EDA is active and futuristic. In an
attempt to "understand" the process and improve it in the future, EDA
uses the data as a "window" to peer into the heart of the process that
generated the data. There is an archival role in the research and
manufacturing world for summary statistics, but there is an enormously
larger role for the EDA approach.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda13.htm [7/1/2003 5:20:58 PM]

1.1.4. What are the EDA Goals?

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.4. What are the EDA Goals?

Primary and The primary goal of EDA is to maximize the analyst's insight into a data
Secondary set and into the underlying structure of a data set, while providing all of
Goals the specific items that an analyst would want to extract from a data set,
such as:
1. a good-fitting, parsimonious model
2. a list of outliers
3. a sense of robustness of conclusions
4. estimates for parameters
5. uncertainties for those estimates
6. a ranked list of important factors
7. conclusions as to whether individual factors are statistically
significant
8. optimal settings

Insight into Insight implies detecting and uncovering underlying structure in the
the Data data. Such underlying structure may not be encapsulated in the list of
items above; such items serve as the specific targets of an analysis, but
the real insight and "feel" for a data set comes as the analyst judiciously
probes and explores the various subtleties of the data. The "feel" for the
data comes almost exclusively from the application of various graphical
techniques, the collection of which serves as the window into the
essence of the data. Graphics are irreplaceable--there are no quantitative
analogues that will give the same insight as well-chosen graphics.
To get a "feel" for the data, it is not enough for the analyst to know what
is in the data; the analyst also must know what is not in the data, and the
only way to do that is to draw on our own human pattern-recognition
and comparative abilities in the context of a series of judicious graphical
techniques applied to the data.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda14.htm [7/1/2003 5:20:58 PM]

1.1.5. The Role of Graphics

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.5. The Role of Graphics

Quantitative/ Statistics and data analysis procedures can broadly be split into two
Graphical parts:
● quantitative

● graphical

Quantitative Quantitative techniques are the set of statistical procedures that yield
numeric or tabular output. Examples of quantitative techniques include:
● hypothesis testing

● analysis of variance
● point estimates and confidence intervals
● least squares regression
These and similar techniques are all valuable and are mainstream in
terms of classical analysis.

Graphical On the other hand, there is a large collection of statistical tools that we
generally refer to as graphical techniques. These include:
● scatter plots

● histograms
● probability plots
● residual plots
● box plots
● block plots

http://www.itl.nist.gov/div898/handbook/eda/section1/eda15.htm (1 of 2) [7/1/2003 5:20:59 PM]

1.1.5. The Role of Graphics

EDA The EDA approach relies heavily on these and similar graphical
Approach techniques. Graphical procedures are not just tools that we could use in
Relies an EDA context, they are tools that we must use. Such graphical tools
Heavily on are the shortest path to gaining insight into a data set in terms of
Graphical ● testing assumptions
Techniques
● model selection

● model validation

● estimator selection

● relationship identification

● factor effect determination

● outlier detection

If one is not using statistical graphics, then one is forfeiting insight into
one or more aspects of the underlying structure of the data.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda15.htm (2 of 2) [7/1/2003 5:20:59 PM]

1.1.6. An EDA/Graphics Example

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.6. An EDA/Graphics Example

Anscombe A simple, classic (Anscombe) example of the central role that graphics
Example play in terms of providing insight into a data set starts with the
following data set:

Data
X Y
10.00 8.04
8.00 6.95
13.00 7.58
9.00 8.81
11.00 8.33
14.00 9.96
6.00 7.24
4.00 4.26
12.00 10.84
7.00 4.82
5.00 5.68

Summary If the goal of the analysis is to compute summary statistics plus

Statistics determine the best linear fit for Y as a function of X, the results might
be given as:
N = 11
Mean of X = 9.0
Mean of Y = 7.5
Intercept = 3
Slope = 0.5
Residual standard deviation = 1.237
Correlation = 0.816
The above quantitative analysis, although valuable, gives us only
limited insight into the data.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda16.htm (1 of 5) [7/1/2003 5:20:59 PM]

1.1.6. An EDA/Graphics Example

Scatter Plot In contrast, the following simple scatter plot of the data

suggests the following:

1. The data set "behaves like" a linear curve with some scatter;
2. there is no justification for a more complicated model (e.g.,
quadratic);
3. there are no outliers;
4. the vertical spread of the data appears to be of equal height
irrespective of the X-value; this indicates that the data are
equally-precise throughout and so a "regular" (that is,
equi-weighted) fit is appropriate.

Three This kind of characterization for the data serves as the core for getting
Additional insight/feel for the data. Such insight/feel does not come from the
Data Sets quantitative statistics; on the contrary, calculations of quantitative
statistics such as intercept and slope should be subsequent to the
characterization and will make sense only if the characterization is
true. To illustrate the loss of information that results when the graphics
insight step is skipped, consider the following three data sets
[Anscombe data sets 2, 3, and 4]:

X2 Y2 X3 Y3 X4 Y4
10.00 9.14 10.00 7.46 8.00 6.58
8.00 8.14 8.00 6.77 8.00 5.76
13.00 8.74 13.00 12.74 8.00 7.71

http://www.itl.nist.gov/div898/handbook/eda/section1/eda16.htm (2 of 5) [7/1/2003 5:20:59 PM]

1.1.6. An EDA/Graphics Example

9.00 8.77 9.00 7.11 8.00 8.84

11.00 9.26 11.00 7.81 8.00 8.47
14.00 8.10 14.00 8.84 8.00 7.04
6.00 6.13 6.00 6.08 8.00 5.25
4.00 3.10 4.00 5.39 19.00 12.50
12.00 9.13 12.00 8.15 8.00 5.56
7.00 7.26 7.00 6.42 8.00 7.91
5.00 4.74 5.00 5.73 8.00 6.89

Quantitative A quantitative analysis on data set 2 yields

Statistics for N = 11
Data Set 2 Mean of X = 9.0
Mean of Y = 7.5
Intercept = 3
Slope = 0.5
Residual standard deviation = 1.237
Correlation = 0.816
which is identical to the analysis for data set 1. One might naively
assume that the two data sets are "equivalent" since that is what the
statistics tell us; but what do the statistics not tell us?

Quantitative Remarkably, a quantitative analysis on data sets 3 and 4 also yields

Statistics for N = 11
Data Sets 3 Mean of X = 9.0
and 4 Mean of Y = 7.5
Intercept = 3
Slope = 0.5
Residual standard deviation = 1.236
Correlation = 0.816 (0.817 for data set 4)
which implies that in some quantitative sense, all four of the data sets
are "equivalent". In fact, the four data sets are far from "equivalent"
and a scatter plot of each data set, which would be step 1 of any EDA
approach, would tell us that immediately.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda16.htm (3 of 5) [7/1/2003 5:20:59 PM]

1.1.6. An EDA/Graphics Example

Scatter Plots

Interpretation Conclusions from the scatter plots are:

of Scatter 1. data set 1 is clearly linear with some scatter.
Plots
2. data set 2 is clearly quadratic.
3. data set 3 clearly has an outlier.
4. data set 4 is obviously the victim of a poor experimental design
with a single point far removed from the bulk of the data
"wagging the dog".

Importance These points are exactly the substance that provide and define "insight"
of and "feel" for a data set. They are the goals and the fruits of an open
Exploratory exploratory data analysis (EDA) approach to the data. Quantitative
Analysis statistics are not wrong per se, but they are incomplete. They are
incomplete because they are numeric summaries which in the
summarization operation do a good job of focusing on a particular
aspect of the data (e.g., location, intercept, slope, degree of relatedness,
etc.) by judiciously reducing the data to a few numbers. Doing so also
filters the data, necessarily omitting and screening out other sometimes
crucial information in the focusing operation. Quantitative statistics
focus but also filter; and filtering is exactly what makes the
quantitative approach incomplete at best and misleading at worst.
The estimated intercepts (= 3) and slopes (= 0.5) for data sets 2, 3, and
4 are misleading because the estimation is done in the context of an
assumed linear model and that linearity assumption is the fatal flaw in
this analysis.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda16.htm (4 of 5) [7/1/2003 5:20:59 PM]

1.1.6. An EDA/Graphics Example

The EDA approach of deliberately postponing the model selection until

further along in the analysis has many rewards, not the least of which is
the ultimate convergence to a much-improved model and the
formulation of valid and supportable scientific and engineering
conclusions.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda16.htm (5 of 5) [7/1/2003 5:20:59 PM]

1.1.7. General Problem Categories

1. Exploratory Data Analysis

1.1. EDA Introduction

1.1.7. General Problem Categories

Problem The following table is a convenient way to classify EDA problems.
Classification

Univariate
UNIVARIATE CONTROL
and Control
Data: Data:
A single column of A single column of
numbers, Y. numbers, Y.
Model: Model:
y = constant + error y = constant + error
Output: Output:
1. A number (the estimated A "yes" or "no" to the
constant in the model). question "Is the system
2. An estimate of uncertainty out of control?".
for the constant. Techniques:
3. An estimate of the ● Control Charts
distribution for the error.
Techniques:
● 4-Plot

● Probability Plot
● PPCC Plot

http://www.itl.nist.gov/div898/handbook/eda/section1/eda17.htm (1 of 4) [7/1/2003 5:20:59 PM]

1.1.7. General Problem Categories

Comparative
COMPARATIVE SCREENING
and
Screening Data: Data:
A single response variable A single response variable
and k independent and k independent
variables (Y, X1, X2, ... , variables (Y, X1, X2, ... ,
Xk), primary focus is on Xk).
one (the primary factor) of Model:
these independent
y = f(x1, x2, ..., xk) + error
variables.
Model: Output:
y = f(x1, x2, ..., xk) + error 1. A ranked list (from most
important to least
Output: important) of factors.
A "yes" or "no" to the 2. Best settings for the
question "Is the primary factors.
factor significant?".
3. A good model/prediction
Techniques: equation relating Y to the
● Block Plot factors.
● Scatter Plot Techniques:
● Box Plot ● Block Plot

● Probability Plot
● Bihistogram

Optimization
OPTIMIZATION REGRESSION
and
Regression Data: Data:
A single response variable A single response variable
and k independent and k independent
variables (Y, X1, X2, ... , variables (Y, X1, X2, ... ,
Xk). Xk). The independent
Model: variables can be
continuous.
y = f(x1, x2, ..., xk) + error
Model:
Output:
y = f(x1, x2, ..., xk) + error
Best settings for the factor
variables. Output:
Techniques: A good model/prediction
equation relating Y to the
● Block Plot
factors.

http://www.itl.nist.gov/div898/handbook/eda/section1/eda17.htm (2 of 4) [7/1/2003 5:20:59 PM]

1.1.7. General Problem Categories

● Least Squares Fitting Techniques:

● Contour Plot ● Least Squares Fitting

● Scatter Plot
● 6-Plot

Time Series
TIME SERIES MULTIVARIATE
and
Multivariate Data: Data:
A column of time k factor variables (X1, X2, ... ,
dependent numbers, Y. Xk).
In addition, time is an
indpendent variable. Model:
The time variable can The model is not explicit.
be either explicit or Output:
implied. If the data are Identify underlying
not equi-spaced, the correlation structure in the
time variable should be data.
explicitly provided.
Techniques:
Model:
● Star Plot
yt = f(t) + error
● Scatter Plot Matrix
The model can be either
a time domain based or ● Conditioning Plot
frequency domain ● Profile Plot
based.
● Principal Components
Output:
Clustering
●
A good
● Discrimination/Classification
model/prediction
equation relating Y to Note that multivarate analysis is
previous values of Y. only covered lightly in this
Techniques: Handbook.
● Autocorrelation Plot

● Spectrum
● Complex Demodulation
Amplitude Plot
● Complex Demodulation
Phase Plot
● ARIMA Models

http://www.itl.nist.gov/div898/handbook/eda/section1/eda17.htm (3 of 4) [7/1/2003 5:20:59 PM]

1.1.7. General Problem Categories

http://www.itl.nist.gov/div898/handbook/eda/section1/eda17.htm (4 of 4) [7/1/2003 5:20:59 PM]

1.2. EDA Assumptions

1. Exploratory Data Analysis

1.2. EDA Assumptions

Summary The gamut of scientific and engineering experimentation is virtually
limitless. In this sea of diversity is there any common basis that allows
the analyst to systematically and validly arrive at supportable, repeatable
research conclusions?
Fortunately, there is such a basis and it is rooted in the fact that every
measurement process, however complicated, has certain underlying
assumptions. This section deals with what those assumptions are, why
they are important, how to go about testing them, and what the
consequences are if the assumptions do not hold.

Table of 1. Underlying Assumptions

Contents for 2. Importance
Section 2
3. Testing Assumptions
4. Importance of Plots
5. Consequences

http://www.itl.nist.gov/div898/handbook/eda/section2/eda2.htm [7/1/2003 5:20:59 PM]

1.2.1. Underlying Assumptions

1. Exploratory Data Analysis

1.2. EDA Assumptions

1.2.1. Underlying Assumptions

Assumptions There are four assumptions that typically underlie all measurement
Underlying a processes; namely, that the data from the process at hand "behave
Measurement like":
Process 1. random drawings;
2. from a fixed distribution;
3. with the distribution having fixed location; and
4. with the distribution having fixed variation.

Univariate or The "fixed location" referred to in item 3 above differs for different
Single problem types. The simplest problem type is univariate; that is, a
Response single variable. For the univariate problem, the general model
Variable response = deterministic component + random component
becomes
response = constant + error

Assumptions For this case, the "fixed location" is simply the unknown constant. We
for Univariate can thus imagine the process at hand to be operating under constant
Model conditions that produce a single column of data with the properties
that
● the data are uncorrelated with one another;

● the random component has a fixed distribution;

● the deterministic component consists of only a constant; and

● the random component has fixed variation.

Extrapolation The universal power and importance of the univariate model is that it
to a Function can easily be extended to the more general case where the
of Many deterministic component is not just a constant, but is in fact a function
Variables of many variables, and the engineering objective is to characterize and
model the function.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda21.htm (1 of 2) [7/1/2003 5:20:59 PM]

1.2.1. Underlying Assumptions

Residuals Will The key point is that regardless of how many factors there are, and
Behave regardless of how complicated the function is, if the engineer succeeds
According to in choosing a good model, then the differences (residuals) between the
Univariate raw response data and the predicted values from the fitted model
Assumptions should themselves behave like a univariate process. Furthermore, the
residuals from this univariate process fit will behave like:
● random drawings;

● from a fixed distribution;

● with fixed location (namely, 0 in this case); and

● with fixed variation.

Validation of Thus if the residuals from the fitted model do in fact behave like the
Model ideal, then testing of underlying assumptions becomes a tool for the
validation and quality of fit of the chosen model. On the other hand, if
the residuals from the chosen fitted model violate one or more of the
above univariate assumptions, then the chosen fitted model is
inadequate and an opportunity exists for arriving at an improved
model.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda21.htm (2 of 2) [7/1/2003 5:20:59 PM]

1.2.2. Importance

1. Exploratory Data Analysis

1.2. EDA Assumptions

1.2.2. Importance
Predictability Predictability is an all-important goal in science and engineering. If the
and four underlying assumptions hold, then we have achieved probabilistic
Statistical predictability--the ability to make probability statements not only
Control about the process in the past, but also about the process in the future.
In short, such processes are said to be "in statistical control".

Validity of Moreover, if the four assumptions are valid, then the process is
Engineering amenable to the generation of valid scientific and engineering
Conclusions conclusions. If the four assumptions are not valid, then the process is
drifting (with respect to location, variation, or distribution),
unpredictable, and out of control. A simple characterization of such
processes by a location estimate, a variation estimate, or a distribution
"estimate" inevitably leads to engineering conclusions that are not
valid, are not supportable (scientifically or legally), and which are not
repeatable in the laboratory.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda22.htm [7/1/2003 5:21:00 PM]

1.2.3. Techniques for Testing Assumptions

1. Exploratory Data Analysis

1.2. EDA Assumptions

1.2.3. Techniques for Testing Assumptions

Testing Because the validity of the final scientific/engineering conclusions
Underlying is inextricably linked to the validity of the underlying univariate
Assumptions assumptions, it naturally follows that there is a real necessity that
Helps Assure the each and every one of the above four assumptions be routinely
Validity of tested.
Scientific and
Engineering
Conclusions

Four Techniques The following EDA techniques are simple, efficient, and powerful
to Test for the routine testing of underlying assumptions:
Underlying 1. run sequence plot (Yi versus i)
Assumptions
2. lag plot (Yi versus Yi-1)
3. histogram (counts versus subgroups of Y)
4. normal probability plot (ordered Y versus theoretical ordered
Y)

Plot on a Single The four EDA plots can be juxtaposed for a quick look at the
Page for a characteristics of the data. The plots below are ordered as follows:
Quick 1. Run sequence plot - upper left
Characterization
2. Lag plot - upper right
of the Data
3. Histogram - lower left
4. Normal probability plot - lower right

http://www.itl.nist.gov/div898/handbook/eda/section2/eda23.htm (1 of 3) [7/1/2003 5:21:00 PM]

1.2.3. Techniques for Testing Assumptions

Sample Plot:
Assumptions
Hold

This 4-plot reveals a process that has fixed location, fixed variation,
is random, apparently has a fixed approximately normal
distribution, and has no outliers.

Sample Plot: If one or more of the four underlying assumptions do not hold, then
Assumptions Do it will show up in the various plots as demonstrated in the following
Not Hold example.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda23.htm (2 of 3) [7/1/2003 5:21:00 PM]

1.2.3. Techniques for Testing Assumptions

This 4-plot reveals a process that has fixed location, fixed variation,
is non-random (oscillatory), has a non-normal, U-shaped
distribution, and has several outliers.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda23.htm (3 of 3) [7/1/2003 5:21:00 PM]

1.2.4. Interpretation of 4-Plot

1. Exploratory Data Analysis

1.2. EDA Assumptions

1.2.4. Interpretation of 4-Plot

Interpretation The four EDA plots discussed on the previous page are used to test the
of EDA Plots: underlying assumptions:
Flat and 1. Fixed Location:
Equi-Banded, If the fixed location assumption holds, then the run sequence
Random, plot will be flat and non-drifting.
Bell-Shaped,
and Linear 2. Fixed Variation:
If the fixed variation assumption holds, then the vertical spread
in the run sequence plot will be the approximately the same over
the entire horizontal axis.
3. Randomness:
If the randomness assumption holds, then the lag plot will be
structureless and random.
4. Fixed Distribution:
If the fixed distribution assumption holds, in particular if the
fixed normal distribution holds, then
1. the histogram will be bell-shaped, and
2. the normal probability plot will be linear.

Plots Utilized Conversely, the underlying assumptions are tested using the EDA
to Test the plots:
Assumptions ● Run Sequence Plot:
If the run sequence plot is flat and non-drifting, the
fixed-location assumption holds. If the run sequence plot has a
vertical spread that is about the same over the entire plot, then
the fixed-variation assumption holds.
● Lag Plot:
If the lag plot is structureless, then the randomness assumption
holds.
● Histogram:
If the histogram is bell-shaped, the underlying distribution is
symmetric and perhaps approximately normal.
● Normal Probability Plot:

http://www.itl.nist.gov/div898/handbook/eda/section2/eda24.htm (1 of 2) [7/1/2003 5:21:00 PM]

1.2.4. Interpretation of 4-Plot

If the normal probability plot is linear, the underlying

distribution is approximately normal.
If all four of the assumptions hold, then the process is said
definitionally to be "in statistical control".

http://www.itl.nist.gov/div898/handbook/eda/section2/eda24.htm (2 of 2) [7/1/2003 5:21:00 PM]

1.2.5. Consequences

1. Exploratory Data Analysis

1.2. EDA Assumptions

1.2.5. Consequences
What If If some of the underlying assumptions do not hold, what can be done
Assumptions about it? What corrective actions can be taken? The positive way of
Do Not Hold? approaching this is to view the testing of underlying assumptions as a
framework for learning about the process. Assumption-testing
promotes insight into important aspects of the process that may not
have surfaced otherwise.

Primary Goal The primary goal is to have correct, validated, and complete
is Correct and scientific/engineering conclusions flowing from the analysis. This
Valid usually includes intermediate goals such as the derivation of a
Scientific good-fitting model and the computation of realistic parameter
Conclusions estimates. It should always include the ultimate goal of an
understanding and a "feel" for "what makes the process tick". There is
no more powerful catalyst for discovery than the bringing together of
an experienced/expert scientist/engineer and a data set ripe with
intriguing "anomalies" and characteristics.

Consequences The following sections discuss in more detail the consequences of

of Invalid invalid assumptions:
Assumptions 1. Consequences of non-randomness
2. Consequences of non-fixed location parameter
3. Consequences of non-fixed variation
4. Consequences related to distributional assumptions

http://www.itl.nist.gov/div898/handbook/eda/section2/eda25.htm [7/1/2003 5:21:00 PM]

1.2.5.1. Consequences of Non-Randomness

1. Exploratory Data Analysis

1.2. EDA Assumptions
1.2.5. Consequences

1.2.5.1. Consequences of Non-Randomness

Randomness There are four underlying assumptions:
Assumption 1. randomness;
2. fixed location;
3. fixed variation; and
4. fixed distribution.
The randomness assumption is the most critical but the least tested.

Consequeces of If the randomness assumption does not hold, then

Non-Randomness 1. All of the usual statistical tests are invalid.
2. The calculated uncertainties for commonly used statistics
become meaningless.
3. The calculated minimal sample size required for a
pre-specified tolerance becomes meaningless.
4. The simple model: y = constant + error becomes invalid.
5. The parameter estimates become suspect and
non-supportable.

Non-Randomness One specific and common type of non-randomness is

Due to autocorrelation. Autocorrelation is the correlation between Yt and
Autocorrelation Yt-k, where k is an integer that defines the lag for the
autocorrelation. That is, autocorrelation is a time dependent
non-randomness. This means that the value of the current point is
highly dependent on the previous point if k = 1 (or k points ago if k
is not 1). Autocorrelation is typically detected via an
autocorrelation plot or a lag plot.

If the data are not random due to autocorrelation, then

1. Adjacent data values may be related.
2. There may not be n independent snapshots of the
phenomenon under study.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda251.htm (1 of 2) [7/1/2003 5:21:00 PM]

1.2.5.1. Consequences of Non-Randomness

3. There may be undetected "junk"-outliers.

4. There may be undetected "information-rich"-outliers.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda251.htm (2 of 2) [7/1/2003 5:21:00 PM]

1.2.5.2. Consequences of Non-Fixed Location Parameter

1. Exploratory Data Analysis

1.2. EDA Assumptions
1.2.5. Consequences

1.2.5.2. Consequences of Non-Fixed

Location Parameter
Location The usual estimate of location is the mean
Estimate

from N measurements Y1, Y2, ... , YN.

Consequences If the run sequence plot does not support the assumption of fixed
of Non-Fixed location, then
Location 1. The location may be drifting.
2. The single location estimate may be meaningless (if the process
is drifting).
3. The choice of location estimator (e.g., the sample mean) may be
sub-optimal.
4. The usual formula for the uncertainty of the mean:

may be invalid and the numerical value optimistically small.

5. The location estimate may be poor.
6. The location estimate may be biased.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda252.htm [7/1/2003 5:21:01 PM]

1.2.5.3. Consequences of Non-Fixed Variation Parameter

1. Exploratory Data Analysis

1.2. EDA Assumptions
1.2.5. Consequences

1.2.5.3. Consequences of Non-Fixed

Variation Parameter
Variation The usual estimate of variation is the standard deviation
Estimate

from N measurements Y1, Y2, ... , YN.

Consequences If the run sequence plot does not support the assumption of fixed
of Non-Fixed variation, then
Variation 1. The variation may be drifting.
2. The single variation estimate may be meaningless (if the process
variation is drifting).
3. The variation estimate may be poor.
4. The variation estimate may be biased.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda253.htm [7/1/2003 5:21:06 PM]

1.2.5.4. Consequences Related to Distributional Assumptions

1. Exploratory Data Analysis

1.2. EDA Assumptions
1.2.5. Consequences

1.2.5.4. Consequences Related to

Distributional Assumptions
Distributional Scientists and engineers routinely use the mean (average) to estimate
Analysis the "middle" of a distribution. It is not so well known that the
variability and the noisiness of the mean as a location estimator are
intrinsically linked with the underlying distribution of the data. For
certain distributions, the mean is a poor choice. For any given
distribution, there exists an optimal choice-- that is, the estimator
with minimum variability/noisiness. This optimal choice may be, for
example, the median, the midrange, the midmean, the mean, or
something else. The implication of this is to "estimate" the
distribution first, and then--based on the distribution--choose the
optimal estimator. The resulting engineering parameter estimators
will have less variability than if this approach is not followed.

Case Studies The airplane glass failure case study gives an example of determining
an appropriate distribution and estimating the parameters of that
distribution. The uniform random numbers case study gives an
example of determining a more appropriate centrality parameter for a
non-normal distribution.

Other consequences that flow from problems with distributional

assumptions are:

Distribution 1. The distribution may be changing.

2. The single distribution estimate may be meaningless (if the
process distribution is changing).
3. The distribution may be markedly non-normal.
4. The distribution may be unknown.
5. The true probability distribution for the error may remain
unknown.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda254.htm (1 of 2) [7/1/2003 5:21:06 PM]

1.2.5.4. Consequences Related to Distributional Assumptions

Model 1. The model may be changing.

2. The single model estimate may be meaningless.
3. The default model
Y = constant + error
may be invalid.
4. If the default model is insufficient, information about a better
model may remain undetected.
5. A poor deterministic model may be fit.
6. Information about an improved model may go undetected.

Process 1. The process may be out-of-control.

2. The process may be unpredictable.
3. The process may be un-modelable.

http://www.itl.nist.gov/div898/handbook/eda/section2/eda254.htm (2 of 2) [7/1/2003 5:21:06 PM]

1.3. EDA Techniques

1. Exploratory Data Analysis

1.3. EDA Techniques

Summary After you have collected a set of data, how do you do an exploratory
data analysis? What techniques do you employ? What do the various
techniques focus on? What conclusions can you expect to reach?
This section provides answers to these kinds of questions via a gallery
of EDA techniques and a detailed description of each technique. The
techniques are divided into graphical and quantitative techniques. For
exploratory data analysis, the emphasis is primarily on the graphical
techniques.

Table of 1. Introduction
Contents for 2. Analysis Questions
Section 3
3. Graphical Techniques: Alphabetical
4. Graphical Techniques: By Problem Category
5. Quantitative Techniques: Alphabetical
6. Probability Distributions

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3.htm [7/1/2003 5:21:06 PM]

1.3.1. Introduction

1. Exploratory Data Analysis

1.3. EDA Techniques

1.3.1. Introduction
Graphical This section describes many techniques that are commonly used in
and exploratory and classical data analysis. This list is by no means meant
Quantitative to be exhaustive. Additional techniques (both graphical and
Techniques quantitative) are discussed in the other chapters. Specifically, the
product comparisons chapter has a much more detailed description of
many classical statistical techniques.
EDA emphasizes graphical techniques while classical techniques
emphasize quantitative techniques. In practice, an analyst typically
uses a mixture of graphical and quantitative techniques. In this section,
we have divided the descriptions into graphical and quantitative
techniques. This is for organizational clarity and is not meant to
discourage the use of both graphical and quantitiative techniques when
analyzing data.

Use of This section emphasizes the techniques themselves; how the graph or
Techniques test is defined, published references, and sample output. The use of the
Shown in techniques to answer engineering questions is demonstrated in the case
Case Studies studies section. The case studies do not demonstrate all of the
techniques.

Availability The sample plots and output in this section were generated with the
in Software Dataplot software program. Other general purpose statistical data
analysis programs can generate most of the plots, intervals, and tests
discussed here, or macros can be written to acheive the same result.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda31.htm [7/1/2003 5:21:06 PM]

1.3.2. Analysis Questions

1. Exploratory Data Analysis

1.3. EDA Techniques

1.3.2. Analysis Questions

EDA Some common questions that exploratory data analysis is used to
Questions answer are:
1. What is a typical value?
2. What is the uncertainty for a typical value?
3. What is a good distributional fit for a set of numbers?
4. What is a percentile?
5. Does an engineering modification have an effect?
6. Does a factor have an effect?
7. What are the most important factors?
8. Are measurements coming from different laboratories equivalent?
9. What is the best function for relating a response variable to a set
of factor variables?
10. What are the best settings for factors?
11. Can we separate signal from noise in time dependent data?
12. Can we extract any structure from multivariate data?
13. Does the data have outliers?

Analyst A critical early step in any analysis is to identify (for the engineering
Should problem at hand) which of the above questions are relevant. That is, we
Identify need to identify which questions we want answered and which questions
Relevant have no bearing on the problem at hand. After collecting such a set of
Questions questions, an equally important step, which is invaluable for maintaining
for his focus, is to prioritize those questions in decreasing order of importance.
Engineering EDA techniques are tied in with each of the questions. There are some
Problem EDA techniques (e.g., the scatter plot) that are broad-brushed and apply
almost universally. On the other hand, there are a large number of EDA
techniques that are specific and whose specificity is tied in with one of
the above questions. Clearly if one chooses not to explicitly identify
relevant questions, then one cannot take advantage of these
question-specific EDA technqiues.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda32.htm (1 of 2) [7/1/2003 5:21:06 PM]

1.3.2. Analysis Questions

EDA Most of these questions can be addressed by techniques discussed in this

Approach chapter. The process modeling and process improvement chapters also
Emphasizes address many of the questions above. These questions are also relevant
Graphics for the classical approach to statistics. What distinguishes the EDA
approach is an emphasis on graphical techniques to gain insight as
opposed to the classical approach of quantitative tests. Most data
analysts will use a mix of graphical and classical quantitative techniques
to address these problems.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda32.htm (2 of 2) [7/1/2003 5:21:06 PM]

1.3.3. Graphical Techniques: Alphabetic

1. Exploratory Data Analysis

1.3. EDA Techniques

1.3.3. Graphical Techniques: Alphabetic

This section provides a gallery of some useful graphical techniques. The
techniques are ordered alphabetically, so this section is not intended to
be read in a sequential fashion. The use of most of these graphical
techniques is demonstrated in the case studies in this chapter. A few of
these graphical techniques are demonstrated in later chapters.

Autocorrelation Bihistogram: Block Plot: 1.3.3.3 Bootstrap Plot:

Plot: 1.3.3.1 1.3.3.2 1.3.3.4

Box-Cox Linearity Box-Cox Box Plot: 1.3.3.7 Complex

Plot: 1.3.3.5 Normality Plot: Demodulation
1.3.3.6 Amplitude Plot:
1.3.3.8

Complex Contour Plot: DEX Scatter Plot: DEX Mean Plot:

Demodulation 1.3.3.10 1.3.3.11 1.3.3.12
Phase Plot: 1.3.3.9

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33.htm (1 of 3) [7/1/2003 5:21:08 PM]

1.3.3. Graphical Techniques: Alphabetic

DEX Standard Histogram: Lag Plot: 1.3.3.15 Linear Correlation

Deviation Plot: 1.3.3.14 Plot: 1.3.3.16
1.3.3.13

Linear Intercept Linear Slope Plot: Linear Residual Mean Plot: 1.3.3.20
Plot: 1.3.3.17 1.3.3.18 Standard Deviation
Plot: 1.3.3.19

Normal Probability Probability Plot: Probability Plot Quantile-Quantile

Plot: 1.3.3.21 1.3.3.22 Correlation Plot: 1.3.3.24
Coefficient Plot:
1.3.3.23

Run Sequence Scatter Plot: Spectrum: 1.3.3.27 Standard Deviation

Plot: 1.3.3.25 1.3.3.26 Plot: 1.3.3.28

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33.htm (2 of 3) [7/1/2003 5:21:08 PM]

1.3.3. Graphical Techniques: Alphabetic

Star Plot: 1.3.3.29 Weibull Plot: Youden Plot: 4-Plot: 1.3.3.32

1.3.3.30 1.3.3.31

6-Plot: 1.3.3.33

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33.htm (3 of 3) [7/1/2003 5:21:08 PM]

1.3.3.1. Autocorrelation Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.1. Autocorrelation Plot

Purpose: Autocorrelation plots (Box and Jenkins, pp. 28-32) are a
Check commonly-used tool for checking randomness in a data set. This
Randomness randomness is ascertained by computing autocorrelations for data
values at varying time lags. If random, such autocorrelations should
be near zero for any and all time-lag separations. If non-random,
then one or more of the autocorrelations will be significantly
non-zero.
In addition, autocorrelation plots are used in the model identification
stage for Box-Jenkins autoregressive, moving average time series
models.

Sample Plot:
Autocorrelations
should be
near-zero for
randomness.
Such is not the
case in this
example and
thus the
randomness
assumption fails

This sample autocorrelation plot shows that the time series is not
random, but rather has a high degree of autocorrelation between
adjacent and near-adjacent observations.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda331.htm (1 of 4) [7/1/2003 5:21:08 PM]

1.3.3.1. Autocorrelation Plot

Definition: Autocorrelation plots are formed by

r(h) versus h ● Vertical axis: Autocorrelation coefficient

where Ch is the autocovariance function

and C0 is the variance function

Note--Rh is between -1 and +1.

● Horizontal axis: Time lag h (h = 1, 2, 3, ...)
● The above line also contains several horizontal reference
lines. The middle line is at zero. The other four lines are 95%
and 99% confidence bands. Note that there are two distinct
formulas for generating the confidence bands.
1. If the autocorrelation plot is being used to test for
randomness (i.e., there is no time dependence in the
data), the following formula is recommended:

where N is the sample size, z is the percent point

function of the standard normal distribution and is
the. significance level. In this case, the confidence
bands have fixed width that depends on the sample
size. This is the formula that was used to generate the
confidence bands in the above plot.
2. Autocorrelation plots are also used in the model
identification stage for fitting ARIMA models. In this
case, a moving average model is assumed for the data
and the following confidence bands should be
generated:

where k is the lag, N is the sample size, z is the percent

http://www.itl.nist.gov/div898/handbook/eda/section3/eda331.htm (2 of 4) [7/1/2003 5:21:08 PM]

1.3.3.1. Autocorrelation Plot

point function of the standard normal distribution and

is. the significance level. In this case, the confidence
bands increase as the lag increases.

Questions The autocorrelation plot can provide answers to the following

questions:
1. Are the data random?
2. Is an observation related to an adjacent observation?
3. Is an observation related to an observation twice-removed?
(etc.)
4. Is the observed time series white noise?
5. Is the observed time series sinusoidal?
6. Is the observed time series autoregressive?
7. What is an appropriate model for the observed time series?
8. Is the model
Y = constant + error
valid and sufficient?
9. Is the formula valid?

Importance: Randomness (along with fixed model, fixed variation, and fixed
Ensure validity distribution) is one of the four assumptions that typically underlie all
of engineering measurement processes. The randomness assumption is critically
conclusions important for the following three reasons:
1. Most standard statistical tests depend on randomness. The
validity of the test conclusions is directly linked to the
validity of the randomness assumption.
2. Many commonly-used statistical formulae depend on the
randomness assumption, the most common formula being the
formula for determining the standard deviation of the sample
mean:

where is the standard deviation of the data. Although

heavily used, the results from using this formula are of no
value unless the randomness assumption holds.
3. For univariate data, the default model is
Y = constant + error
If the data are not random, this model is incorrect and invalid,
and the estimates for the parameters (such as the constant)
become nonsensical and invalid.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda331.htm (3 of 4) [7/1/2003 5:21:08 PM]

1.3.3.1. Autocorrelation Plot

In short, if the analyst does not check for randomness, then the
validity of many of the statistical conclusions becomes suspect. The
autocorrelation plot is an excellent way of checking for such
randomness.

Examples Examples of the autocorrelation plot for several common situations

are given in the following pages.
1. Random (= White Noise)
2. Weak autocorrelation
3. Strong autocorrelation and autoregressive model
4. Sinusoidal model

Related Partial Autocorrelation Plot

Techniques Lag Plot
Spectral Plot
Seasonal Subseries Plot

Case Study The autocorrelation plot is demonstrated in the beam deflection data
case study.

Software Autocorrelation plots are available in most general purpose

statistical software programs including Dataplot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda331.htm (4 of 4) [7/1/2003 5:21:08 PM]

1.3.3.1.1. Autocorrelation Plot: Random Data

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.1. Autocorrelation Plot

1.3.3.1.1. Autocorrelation Plot: Random

Data
Autocorrelation The following is a sample autocorrelation plot.
Plot

Conclusions We can make the following conclusions from this plot.

1. There are no significant autocorrelations.
2. The data are random.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3311.htm (1 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.1. Autocorrelation Plot: Random Data

Discussion Note that with the exception of lag 0, which is always 1 by

definition, almost all of the autocorrelations fall within the 95%
confidence limits. In addition, there is no apparent pattern (such as
the first twenty-five being positive and the second twenty-five being
negative). This is the abscence of a pattern we expect to see if the
data are in fact random.
A few lags slightly outside the 95% and 99% confidence limits do
not neccessarily indicate non-randomness. For a 95% confidence
interval, we might expect about one out of twenty lags to be
statistically significant due to random fluctuations.
There is no associative ability to infer from a current value Yi as to
what the next value Yi+1 will be. Such non-association is the essense
of randomness. In short, adjacent observations do not "co-relate", so
we call this the "no autocorrelation" case.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3311.htm (2 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.2. Autocorrelation Plot: Moderate Autocorrelation

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.1. Autocorrelation Plot

1.3.3.1.2. Autocorrelation Plot: Moderate

Autocorrelation
Autocorrelation The following is a sample autocorrelation plot.
Plot

Conclusions We can make the following conclusions from this plot.

1. The data come from an underlying autoregressive model with
moderate positive autocorrelation.

Discussion The plot starts with a moderately high autocorrelation at lag 1

(approximately 0.75) that gradually decreases. The decreasing
autocorrelation is generally linear, but with significant noise. Such a
pattern is the autocorrelation plot signature of "moderate
autocorrelation", which in turn provides moderate predictability if
modeled properly.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3312.htm (1 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.2. Autocorrelation Plot: Moderate Autocorrelation

Recommended The next step would be to estimate the parameters for the
Next Step autoregressive model:

Such estimation can be performed by using least squares linear

regression or by fitting a Box-Jenkins autoregressive (AR) model.

The randomness assumption for least squares fitting applies to the

residuals of the model. That is, even though the original data exhibit
randomness, the residuals after fitting Yi against Yi-1 should result in
random residuals. Assessing whether or not the proposed model in
fact sufficiently removed the randomness is discussed in detail in the
Process Modeling chapter.

The residual standard deviation for this autoregressive model will be

much smaller than the residual standard deviation for the default
model

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3312.htm (2 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.3. Autocorrelation Plot: Strong Autocorrelation and Autoregressive Model

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.1. Autocorrelation Plot

1.3.3.1.3. Autocorrelation Plot: Strong

Autocorrelation and
Autoregressive Model
Autocorrelation The following is a sample autocorrelation plot.
Plot for Strong
Autocorrelation

Conclusions We can make the following conclusions from the above plot.
1. The data come from an underlying autoregressive model with
strong positive autocorrelation.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3313.htm (1 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.3. Autocorrelation Plot: Strong Autocorrelation and Autoregressive Model

Discussion The plot starts with a high autocorrelation at lag 1 (only slightly less
than 1) that slowly declines. It continues decreasing until it becomes
negative and starts showing an incresing negative autocorrelation.
The decreasing autocorrelation is generally linear with little noise.
Such a pattern is the autocorrelation plot signature of "strong
autocorrelation", which in turn provides high predictability if
modeled properly.

Recommended The next step would be to estimate the parameters for the
Next Step autoregressive model:

Such estimation can be performed by using least squares linear

regression or by fitting a Box-Jenkins autoregressive (AR) model.

The randomness assumption for least squares fitting applies to the

residuals of the model. That is, even though the original data exhibit
randomness, the residuals after fitting Yi against Yi-1 should result in
random residuals. Assessing whether or not the proposed model in
fact sufficiently removed the randomness is discussed in detail in the
Process Modeling chapter.

The residual standard deviation for this autoregressive model will be

much smaller than the residual standard deviation for the default
model

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3313.htm (2 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.4. Autocorrelation Plot: Sinusoidal Model

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.1. Autocorrelation Plot

1.3.3.1.4. Autocorrelation Plot: Sinusoidal

Model
Autocorrelation The following is a sample autocorrelation plot.
Plot for
Sinusoidal
Model

Conclusions We can make the following conclusions from the above plot.
1. The data come from an underlying sinusoidal model.

Discussion The plot exhibits an alternating sequence of positive and negative

spikes. These spikes are not decaying to zero. Such a pattern is the
autocorrelation plot signature of a sinusoidal model.

Recommended The beam deflection case study gives an example of modeling a

Next Step sinusoidal model.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3314.htm (1 of 2) [7/1/2003 5:21:09 PM]

1.3.3.1.4. Autocorrelation Plot: Sinusoidal Model

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3314.htm (2 of 2) [7/1/2003 5:21:09 PM]

1.3.3.2. Bihistogram

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.2. Bihistogram
Purpose: The bihistogram is an EDA tool for assessing whether a
Check for a before-versus-after engineering modification has caused a change in
change in ● location;
location,
● variation; or
variation, or
distribution ● distribution.

It is a graphical alternative to the two-sample t-test. The bihistogram

can be more powerful than the t-test in that all of the distributional
features (location, scale, skewness, outliers) are evident on a single plot.
It is also based on the common and well-understood histogram.

Sample Plot:
This
bihistogram
reveals that
there is a
significant
difference in
ceramic
breaking
strength
between
batch 1
(above) and
batch 2
(below)

From the above bihistogram, we can see that batch 1 is centered at a

ceramic strength value of approximately 725 while batch 2 is centered
at a ceramic strength value of approximately 625. That indicates that
these batches are displaced by about 100 strength units. Thus the batch

http://www.itl.nist.gov/div898/handbook/eda/section3/eda332.htm (1 of 3) [7/1/2003 5:21:10 PM]

1.3.3.2. Bihistogram

factor has a significant effect on the location (typical value) for strength
and hence batch is said to be "significant" or to "have an effect". We
thus see graphically and convincingly what a t-test or analysis of
variance would indicate quantitatively.

With respect to variation, note that the spread (variation) of the

above-axis batch 1 histogram does not appear to be that much different
from the below-axis batch 2 histogram. With respect to distributional
shape, note that the batch 1 histogram is skewed left while the batch 2
histogram is more symmetric with even a hint of a slight skewness to
the right.
Thus the bihistogram reveals that there is a clear difference between the
batches with respect to location and distribution, but not in regard to
variation. Comparing batch 1 and batch 2, we also note that batch 1 is
the "better batch" due to its 100-unit higher average strength (around
725).

Definition: Bihistograms are formed by vertically juxtaposing two histograms:

Two ● Above the axis: Histogram of the response variable for condition
adjoined 1
histograms
● Below the axis: Histogram of the response variable for condition
2

Questions The bihistogram can provide answers to the following questions:

1. Is a (2-level) factor significant?
2. Does a (2-level) factor have an effect?
3. Does the location change between the 2 subgroups?
4. Does the variation change between the 2 subgroups?
5. Does the distributional shape change between subgroups?
6. Are there any outliers?

Importance: The bihistogram is an important EDA tool for determining if a factor

Checks 3 out "has an effect". Since the bihistogram provides insight into the validity
of the 4 of three (location, variation, and distribution) out of the four (missing
underlying only randomness) underlying assumptions in a measurement process, it
assumptions is an especially valuable tool. Because of the dual (above/below) nature
of a of the plot, the bihistogram is restricted to assessing factors that have
measurement only two levels. However, this is very common in the
process before-versus-after character of many scientific and engineering
experiments.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda332.htm (2 of 3) [7/1/2003 5:21:10 PM]

1.3.3.2. Bihistogram

Related t test (for shift in location)

Techniques F test (for shift in variation)
Kolmogorov-Smirnov test (for shift in distribution)
Quantile-quantile plot (for shift in location and distribution)

Case Study The bihistogram is demonstrated in the ceramic strength data case
study.

Software The bihistogram is not widely available in general purpose statistical

software programs. Bihistograms can be generated using Dataplot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda332.htm (3 of 3) [7/1/2003 5:21:10 PM]

1.3.3.3. Block Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.3. Block Plot

Purpose: The block plot (Filliben 1993) is an EDA tool for assessing whether the
Check to factor of interest (the primary factor) has a statistically significant effect
determine if on the response, and whether that conclusion about the primary factor
a factor of effect is valid robustly over all other nuisance or secondary factors in
interest has the experiment.
an effect
robust over It replaces the analysis of variance test with a less
all other assumption-dependent binomial test and should be routinely used
factors whenever we are trying to robustly decide whether a primary factor has
an effect.

Sample
Plot:
Weld
method 2 is
lower
(better) than
weld method
1 in 10 of 12
cases

This block plot reveals that in 10 of the 12 cases (bars), weld method 2
is lower (better) than weld method 1. From a binomial point of view,
weld method is statistically significant.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda333.htm (1 of 4) [7/1/2003 5:21:10 PM]

1.3.3.3. Block Plot

Definition Block Plots are formed as follows:

● Vertical axis: Response variable Y

● Horizontal axis: All combinations of all levels of all nuisance

(secondary) factors X1, X2, ...
● Plot Character: Levels of the primary factor XP

Discussion: Average number of defective lead wires per hour from a study with four
Primary factors,
factor is 1. weld strength (2 levels)
denoted by
2. plant (2 levels)
plot
character: 3. speed (2 levels)
within-bar 4. shift (3 levels)
plot are shown in the plot above. Weld strength is the primary factor and the
character. other three factors are nuisance factors. The 12 distinct positions along
the horizontal axis correspond to all possible combinations of the three
nuisance factors, i.e., 12 = 2 plants x 2 speeds x 3 shifts. These 12
conditions provide the framework for assessing whether any conclusions
about the 2 levels of the primary factor (weld method) can truly be
called "general conclusions". If we find that one weld method setting
does better (smaller average defects per hour) than the other weld
method setting for all or most of these 12 nuisance factor combinations,
then the conclusion is in fact general and robust.

Ordering In the above chart, the ordering along the horizontal axis is as follows:
along the ● The left 6 bars are from plant 1 and the right 6 bars are from plant
horizontal 2.
axis
● The first 3 bars are from speed 1, the next 3 bars are from speed
2, the next 3 bars are from speed 1, and the last 3 bars are from
speed 2.
● Bars 1, 4, 7, and 10 are from the first shift, bars 2, 5, 8, and 11 are
from the second shift, and bars 3, 6, 9, and 12 are from the third
shift.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda333.htm (2 of 4) [7/1/2003 5:21:10 PM]

1.3.3.3. Block Plot

Setting 2 is In the block plot for the first bar (plant 1, speed 1, shift 1), weld method
better than 1 yields about 28 defects per hour while weld method 2 yields about 22
setting 1 in defects per hour--hence the difference for this combination is about 6
10 out of 12 defects per hour and weld method 2 is seen to be better (smaller number
cases of defects per hour).
Is "weld method 2 is better than weld method 1" a general conclusion?
For the second bar (plant 1, speed 1, shift 2), weld method 1 is about 37
while weld method 2 is only about 18. Thus weld method 2 is again seen
to be better than weld method 1. Similarly for bar 3 (plant 1, speed 1,
shift 3), we see weld method 2 is smaller than weld method 1. Scanning
over all of the 12 bars, we see that weld method 2 is smaller than weld
method 1 in 10 of the 12 cases, which is highly suggestive of a robust
weld method effect.

An event What is the chance of 10 out of 12 happening by chance? This is

with chance probabilistically equivalent to testing whether a coin is fair by flipping it
probability and getting 10 heads in 12 tosses. The chance (from the binomial
of only 2% distribution) of getting 10 (or more extreme: 11, 12) heads in 12 flips of
a fair coin is about 2%. Such low-probability events are usually rejected
as untenable and in practice we would conclude that there is a difference
in weld methods.

Advantage: The advantages of the block plot are as follows:

Graphical ● A quantitative procedure (analysis of variance) is replaced by a
and graphical procedure.
binomial
● An F-test (analysis of variance) is replaced with a binomial test,
which requires fewer assumptions.

Questions The block plot can provide answers to the following questions:
1. Is the factor of interest significant?
2. Does the factor of interest have an effect?
3. Does the location change between levels of the primary factor?
4. Has the process improved?
5. What is the best setting (= level) of the primary factor?
6. How much of an average improvement can we expect with this
best setting of the primary factor?
7. Is there an interaction between the primary factor and one or more
nuisance factors?
8. Does the effect of the primary factor change depending on the
setting of some nuisance factor?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda333.htm (3 of 4) [7/1/2003 5:21:10 PM]

1.3.3.3. Block Plot

9. Are there any outliers?

Importance: The block plot is a graphical technique that pointedly focuses on

Robustly whether or not the primary factor conclusions are in fact robustly
checks the general. This question is fundamentally different from the generic
significance multi-factor experiment question where the analyst asks, "What factors
of the factor are important and what factors are not" (a screening problem)? Global
of interest data analysis techniques, such as analysis of variance, can potentially be
improved by local, focused data analysis techniques that take advantage
of this difference.

Related t test (for shift in location for exactly 2 levels)

Techniques ANOVA (for shift in location for 2 or more levels)
Bihistogram (for shift in location, variation, and distribution for exactly
2 levels).

Case Study The block plot is demonstrated in the ceramic strength data case study.

Software Block plots can be generated with the Dataplot software program. They
are not currently available in other statistical software programs.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda333.htm (4 of 4) [7/1/2003 5:21:10 PM]

1.3.3.4. Bootstrap Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.4. Bootstrap Plot

Purpose: The bootstrap (Efron and Gong) plot is used to estimate the uncertainty
Estimate of a statistic.
uncertainty

Generate To generate a bootstrap uncertainty estimate for a given statistic from a

subsamples set of data, a subsample of a size less than or equal to the size of the data
with set is generated from the data, and the statistic is calculated. This
replacement subsample is generated with replacement so that any data point can be
sampled multiple times or not sampled at all. This process is repeated
for many subsamples, typically between 500 and 1000. The computed
values for the statistic form an estimate of the sampling distribution of
the statistic.
For example, to estimate the uncertainty of the median from a dataset
with 50 elements, we generate a subsample of 50 elements and calculate
the median. This is repeated at least 500 times so that we have at least
500 values for the median. Although the number of bootstrap samples to
use is somewhat arbitrary, 500 subsamples is usually sufficient. To
calculate a 90% confidence interval for the median, the sample medians
are sorted into ascending order and the value of the 25th median
(assuming exactly 500 subsamples were taken) is the lower confidence
limit while the value of the 475th median (assuming exactly 500
subsamples were taken) is the upper confidence limit.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda334.htm (1 of 3) [7/1/2003 5:21:11 PM]

1.3.3.4. Bootstrap Plot

Sample
Plot:

This bootstrap plot was generated from 500 uniform random numbers.
Bootstrap plots and corresponding histograms were generated for the
mean, median, and mid-range. The histograms for the corresponding
statistics clearly show that for uniform random numbers the mid-range
has the smallest variance and is, therefore, a superior location estimator
to the mean or the median.

Definition The bootstrap plot is formed by:

● Vertical axis: Computed value of the desired statistic for a given
subsample.
● Horizontal axis: Subsample number.

The bootstrap plot is simply the computed value of the statistic versus
the subsample number. That is, the bootstrap plot generates the values
for the desired statistic. This is usually immediately followed by a
histogram or some other distributional plot to show the location and
variation of the sampling distribution of the statistic.

Questions The bootstrap plot is used to answer the following questions:

● What does the sampling distribution for the statistic look like?

● What is a 95% confidence interval for the statistic?

● Which statistic has a sampling distribution with the smallest

variance? That is, which statistic generates the narrowest
confidence interval?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda334.htm (2 of 3) [7/1/2003 5:21:11 PM]

1.3.3.4. Bootstrap Plot

Importance The most common uncertainty calculation is generating a confidence

interval for the mean. In this case, the uncertainty formula can be
derived mathematically. However, there are many situations in which
the uncertainty formulas are mathematically intractable. The bootstrap
provides a method for calculating the uncertainty in these cases.

Cautuion on The bootstrap is not appropriate for all distributions and statistics (Efron
use of the and Tibrashani). For example, because of the shape of the uniform
bootstrap distribution, the bootstrap is not appropriate for estimating the
distribution of statistics that are heavily dependent on the tails, such as
the range.

Related Histogram
Techniques Jackknife
The jacknife is a technique that is closely related to the bootstrap. The
jackknife is beyond the scope of this handbook. See the Efron and Gong
article for a discussion of the jackknife.

Case Study The bootstrap plot is demonstrated in the uniform random numbers case
study.

Software The bootstrap is becoming more common in general purpose statistical

software programs. However, it is still not supported in many of these
programs. Dataplot supports a bootstrap capability.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda334.htm (3 of 3) [7/1/2003 5:21:11 PM]

1.3.3.5. Box-Cox Linearity Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.5. Box-Cox Linearity Plot

Purpose: When performing a linear fit of Y against X, an appropriate
Find the transformation of X can often significantly improve the fit. The
transformation Box-Cox transformation (Box and Cox, 1964) is a particularly useful
of the X family of transformations. It is defined as:
variable that
maximizes the
correlation where X is the variable being transformed and is the transformation
between a Y parameter. For = 0, the natural log of the data is taken instead of
and an X using the above formula.
variable
The Box-Cox linearity plot is a plot of the correlation between Y and
the transformed X for given values of . That is, is the coordinate
for the horizontal axis variable and the value of the correlation
between Y and the transformed X is the coordinate for the vertical
axis of the plot. The value of corresponding to the maximum
correlation (or minimum for negative correlation) on the plot is then
the optimal choice for .
Transforming X is used to improve the fit. The Box-Cox
transformation applied to Y can be used as the basis for meeting the
error assumptions. That case is not covered here. See page 225 of
(Draper and Smith, 1981) or page 77 of (Ryan, 1997) for a discussion
of this case.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda335.htm (1 of 3) [7/1/2003 5:21:12 PM]

1.3.3.5. Box-Cox Linearity Plot

Sample Plot

The plot of the original data with the predicted values from a linear fit
indicate that a quadratic fit might be preferable. The Box-Cox
linearity plot shows a value of = 2.0. The plot of the transformed
data with the predicted values from a linear fit with the transformed
data shows a better fit (verified by the significant reduction in the
residual standard deviation).

Definition Box-Cox linearity plots are formed by

● Vertical axis: Correlation coefficient from the transformed X
and Y
● Horizontal axis: Value for

Questions The Box-Cox linearity plot can provide answers to the following
questions:
1. Would a suitable transformation improve my fit?
2. What is the optimal value of the transformation parameter?

Importance: Transformations can often significantly improve a fit. The Box-Cox

Find a linearity plot provides a convenient way to find a suitable
suitable transformation without engaging in a lot of trial and error fitting.
transformation

Related Linear Regression

Techniques Box-Cox Normality Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda335.htm (2 of 3) [7/1/2003 5:21:12 PM]

1.3.3.5. Box-Cox Linearity Plot

Case Study The Box-Cox linearity plot is demonstrated in the Alaska pipeline
data case study.

Software Box-Cox linearity plots are not a standard part of most general
purpose statistical software programs. However, the underlying
technique is based on a transformation and computing a correlation
coefficient. So if a statistical program supports these capabilities,
writing a macro for a Box-Cox linearity plot should be feasible.
Dataplot supports a Box-Cox linearity plot directly.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda335.htm (3 of 3) [7/1/2003 5:21:12 PM]

1.3.3.6. Box-Cox Normality Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.6. Box-Cox Normality Plot

Purpose: Many statistical tests and intervals are based on the assumption of
Find normality. The assumption of normality often leads to tests that are
transformation simple, mathematically tractable, and powerful compared to tests that
to normalize do not make the normality assumption. Unfortunately, many real data
data sets are in fact not approximately normal. However, an appropriate
transformation of a data set can often yield a data set that does follow
approximately a normal distribution. This increases the applicability
and usefulness of statistical techniques based on the normality
assumption.
The Box-Cox transformation is a particulary useful family of
transformations. It is defined as:

where Y is the response variable and is the transformation

parameter. For = 0, the natural log of the data is taken instead of
using the above formula.
Given a particular transformation such as the Box-Cox transformation
defined above, it is helpful to define a measure of the normality of the
resulting transformation. One measure is to compute the correlation
coefficient of a normal probability plot. The correlation is computed
between the vertical and horizontal axis variables of the probability
plot and is a convenient measure of the linearity of the probability plot
(the more linear the probability plot, the better a normal distribution
fits the data).
The Box-Cox normality plot is a plot of these correlation coefficients
for various values of the parameter. The value of corresponding
to the maximum correlation on the plot is then the optimal choice for
.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda336.htm (1 of 3) [7/1/2003 5:21:12 PM]

1.3.3.6. Box-Cox Normality Plot

Sample Plot

The histogram in the upper left-hand corner shows a data set that has
significant right skewness (and so does not follow a normal
distribution). The Box-Cox normality plot shows that the maximum
value of the correlation coefficient is at = -0.3. The histogram of the
data after applying the Box-Cox transformation with = -0.3 shows a
data set for which the normality assumption is reasonable. This is
verified with a normal probability plot of the transformed data.

Definition Box-Cox normality plots are formed by:

● Vertical axis: Correlation coefficient from the normal
probability plot after applying Box-Cox transformation
● Horizontal axis: Value for

Questions The Box-Cox normality plot can provide answers to the following
questions:
1. Is there a transformation that will normalize my data?
2. What is the optimal value of the transformation parameter?

Importance: Normality assumptions are critical for many univariate intervals and
Normalization hypothesis tests. It is important to test the normality assumption. If the
Improves data are in fact clearly not normal, the Box-Cox normality plot can
Validity of often be used to find a transformation that will approximately
Tests normalize the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda336.htm (2 of 3) [7/1/2003 5:21:12 PM]

1.3.3.6. Box-Cox Normality Plot

Related Normal Probability Plot

Techniques Box-Cox Linearity Plot

Software Box-Cox normality plots are not a standard part of most general
purpose statistical software programs. However, the underlying
technique is based on a normal probability plot and computing a
correlation coefficient. So if a statistical program supports these
capabilities, writing a macro for a Box-Cox normality plot should be
feasible. Dataplot supports a Box-Cox normality plot directly.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda336.htm (3 of 3) [7/1/2003 5:21:12 PM]

1.3.3.7. Box Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.7. Box Plot

Purpose: Box plots (Chambers 1983) are an excellent tool for conveying location
Check and variation information in data sets, particularly for detecting and
location and illustrating location and variation changes between different groups of
variation data.
shifts

Sample
Plot:
This box
plot reveals
that
machine has
a significant
effect on
energy with
respect to
location and
possibly
variation

This box plot, comparing four machines for energy output, shows that
machine has a significant effect on energy with respect to both location
and variation. Machine 3 has the highest energy response (about 72.5);
machine 4 has the least variable energy response with about 50% of its
readings being within 1 energy unit.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda337.htm (1 of 3) [7/1/2003 5:21:12 PM]

1.3.3.7. Box Plot

Definition Box plots are formed by

Vertical axis: Response variable
Horizontal axis: The factor of interest
More specifically, we
1. Calculate the median and the quartiles (the lower quartile is the
25th percentile and the upper quartile is the 75th percentile).
2. Plot a symbol at the median (or draw a line) and draw a box
(hence the name--box plot) between the lower and upper
quartiles; this box represents the middle 50% of the data--the
"body" of the data.
3. Draw a line from the lower quartile to the minimum point and
another line from the upper quartile to the maximum point.
Typically a symbol is drawn at these minimum and maximum
points, although this is optional.
Thus the box plot identifies the middle 50% of the data, the median, and
the extreme points.

Single or A single box plot can be drawn for one batch of data with no distinct
multiple box groups. Alternatively, multiple box plots can be drawn together to
plots can be compare multiple data sets or to compare groups in a single data set. For
drawn a single box plot, the width of the box is arbitrary. For multiple box
plots, the width of the box plot can be set proportional to the number of
points in the given group or sample (some software implementations of
the box plot simply set all the boxes to the same width).

Box plots There is a useful variation of the box plot that more specifically
with fences identifies outliers. To create this variation:
1. Calculate the median and the lower and upper quartiles.
2. Plot a symbol at the median and draw a box between the lower
and upper quartiles.
3. Calculate the interquartile range (the difference between the upper
and lower quartile) and call it IQ.
4. Calculate the following points:
L1 = lower quartile - 1.5*IQ
L2 = lower quartile - 3.0*IQ
U1 = upper quartile + 1.5*IQ
U2 = upper quartile + 3.0*IQ
5. The line from the lower quartile to the minimum is now drawn
from the lower quartile to the smallest point that is greater than
L1. Likewise, the line from the upper quartile to the maximum is
now drawn to the largest point smaller than U1.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda337.htm (2 of 3) [7/1/2003 5:21:12 PM]

1.3.3.7. Box Plot

6. Points between L1 and L2 or between U1 and U2 are drawn as

small circles. Points less than L2 or greater than U2 are drawn as
large circles.

Questions The box plot can provide answers to the following questions:
1. Is a factor significant?
2. Does the location differ between subgroups?
3. Does the variation differ between subgroups?
4. Are there any outliers?

Importance: The box plot is an important EDA tool for determining if a factor has a
Check the significant effect on the response with respect to either location or
significance variation.
of a factor
The box plot is also an effective tool for summarizing large quantities of
information.

Related Mean Plot

Techniques Analysis of Variance

Case Study The box plot is demonstrated in the ceramic strength data case study.

Software Box plots are available in most general purpose statistical software
programs, including Dataplot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda337.htm (3 of 3) [7/1/2003 5:21:12 PM]

1.3.3.8. Complex Demodulation Amplitude Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.8. Complex Demodulation Amplitude

Plot
Purpose: In the frequency analysis of time series models, a common model is the
Detect sinusoidal model:
Changing
Amplitude in
Sinusoidal In this equation, is the amplitude, is the phase shift, and is the
Models dominant frequency. In the above model, and are constant, that is
they do not vary with time, ti.

The complex demodulation amplitude plot (Granger, 1964) is used to

determine if the assumption of constant amplitude is justifiable. If the
slope of the complex demodulation amplitude plot is zero, then the
above model is typically replaced with the model:

where is some type of linear model fit with standard least squares.
The most common case is a linear fit, that is the model becomes

Quadratic models are sometimes used. Higher order models are

relatively rare.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda338.htm (1 of 3) [7/1/2003 5:21:14 PM]

1.3.3.8. Complex Demodulation Amplitude Plot

Sample
Plot:

This complex demodulation amplitude plot shows that:

● the amplitude is fixed at approximately 390;

● there is a start-up effect; and

● there is a change in amplitude at around x = 160 that should be

investigated for an outlier.

Definition: The complex demodulation amplitude plot is formed by:

● Vertical axis: Amplitude

● Horizontal axis: Time

The mathematical computations for determining the amplitude are

beyond the scope of the Handbook. Consult Granger (Granger, 1964)
for details.

Questions The complex demodulation amplitude plot answers the following

questions:
1. Does the amplitude change over time?
2. Are there any outliers that need to be investigated?
3. Is the amplitude different at the beginning of the series (i.e., is
there a start-up effect)?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda338.htm (2 of 3) [7/1/2003 5:21:14 PM]

1.3.3.8. Complex Demodulation Amplitude Plot

Importance: As stated previously, in the frequency analysis of time series models, a

Assumption common model is the sinusoidal model:
Checking

In this equation, is assumed to be constant, that is it does not vary

with time. It is important to check whether or not this assumption is
reasonable.
The complex demodulation amplitude plot can be used to verify this
assumption. If the slope of this plot is essentially zero, then the
assumption of constant amplitude is justified. If it is not, should be
replaced with some type of time-varying model. The most common
cases are linear (B0 + B1*t) and quadratic (B0 + B1*t + B2*t2).

Related Spectral Plot

Techniques Complex Demodulation Phase Plot
Non-Linear Fitting

Case Study The complex demodulation amplitude plot is demonstrated in the beam
deflection data case study.

Software Complex demodulation amplitude plots are available in some, but not
most, general purpose statistical software programs. Dataplot supports
complex demodulation amplitude plots.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda338.htm (3 of 3) [7/1/2003 5:21:14 PM]

1.3.3.9. Complex Demodulation Phase Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.9. Complex Demodulation Phase Plot

Purpose: As stated previously, in the frequency analysis of time series models, a
Improve the common model is the sinusoidal model:
estimate of
frequency in
sinusoidal In this equation, is the amplitude, is the phase shift, and is the
time series dominant frequency. In the above model, and are constant, that is
models they do not vary with time ti.

The complex demodulation phase plot (Granger, 1964) is used to

improve the estimate of the frequency (i.e., ) in this model.
If the complex demodulation phase plot shows lines sloping from left to
right, then the estimate of the frequency should be increased. If it shows
lines sloping right to left, then the frequency should be decreased. If
there is essentially zero slope, then the frequency estimate does not need
to be modified.

Sample
Plot:

http://www.itl.nist.gov/div898/handbook/eda/section3/eda339.htm (1 of 3) [7/1/2003 5:21:14 PM]

1.3.3.9. Complex Demodulation Phase Plot

This complex demodulation phase plot shows that:

● the specified demodulation frequency is incorrect;

● the demodulation frequency should be increased.

Definition The complex demodulation phase plot is formed by:

● Vertical axis: Phase

● Horizontal axis: Time

The mathematical computations for the phase plot are beyond the scope
of the Handbook. Consult Granger (Granger, 1964) for details.

Questions The complex demodulation phase plot answers the following question:
Is the specified demodulation frequency correct?

Importance The non-linear fitting for the sinusoidal model:

of a Good
Initial
Estimate for is usually quite sensitive to the choice of good starting values. The
the initial estimate of the frequency, , is obtained from a spectral plot. The
Frequency complex demodulation phase plot is used to assess whether this estimate
is adequate, and if it is not, whether it should be increased or decreased.
Using the complex demodulation phase plot with the spectral plot can
significantly improve the quality of the non-linear fits obtained.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda339.htm (2 of 3) [7/1/2003 5:21:14 PM]

1.3.3.9. Complex Demodulation Phase Plot

Related Spectral Plot

Techniques Complex Demodulation Phase Plot
Non-Linear Fitting

Case Study The complex demodulation amplitude plot is demonstrated in the beam
deflection data case study.

Software Complex demodulation phase plots are available in some, but not most,
general purpose statistical software programs. Dataplot supports
complex demodulation phase plots.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda339.htm (3 of 3) [7/1/2003 5:21:14 PM]

1.3.3.10. Contour Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.10. Contour Plot

Purpose: A contour plot is a graphical technique for representing a
Display 3-d 3-dimensional surface by plotting constant z slices, called contours, on
surface on a 2-dimensional format. That is, given a value for z, lines are drawn for
2-d plot connecting the (x,y) coordinates where that z value occurs.
The contour plot is an alternative to a 3-D surface plot.

Sample Plot:

This contour plot shows that the surface is symmetric and peaks in the
center.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a.htm (1 of 3) [7/1/2003 5:21:15 PM]

1.3.3.10. Contour Plot

Definition The contour plot is formed by:

● Vertical axis: Independent variable 2

● Horizontal axis: Independent variable 1

● Lines: iso-response values

The independent variables are usually restricted to a regular grid. The

actual techniques for determining the correct iso-response values are
rather complex and are almost always computer generated.
An additional variable may be required to specify the Z values for
drawing the iso-lines. Some software packages require explicit values.
Other software packages will determine them automatically.
If the data (or function) do not form a regular grid, you typically need
to perform a 2-D interpolation to form a regular grid.

Questions The contour plot is used to answer the question

How does Z change as a function of X and Y?

Importance: For univariate data, a run sequence plot and a histogram are considered
Visualizing necessary first steps in understanding the data. For 2-dimensional data,
3-dimensional a scatter plot is a necessary first step in understanding the data.
data
In a similar manner, 3-dimensional data should be plotted. Small data
sets, such as result from designed experiments, can typically be
represented by block plots, dex mean plots, and the like (here, "DEX"
stands for "Design of Experiments"). For large data sets, a contour plot
or a 3-D surface plot should be considered a necessary first step in
understanding the data.

DEX Contour The dex contour plot is a specialized contour plot used in the design of
Plot experiments. In particular, it is useful for full and fractional designs.

Related 3-D Plot

Techniques

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a.htm (2 of 3) [7/1/2003 5:21:15 PM]

1.3.3.10. Contour Plot

Software Contour plots are available in most general purpose statistical software
programs. They are also available in many general purpose graphics
and mathematics programs. These programs vary widely in the
capabilities for the contour plots they generate. Many provide just a
basic contour plot over a rectangular grid while others permit color
filled or shaded contours. Dataplot supports a fairly basic contour plot.

Most statistical software programs that support design of experiments

will provide a dex contour plot capability.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a.htm (3 of 3) [7/1/2003 5:21:15 PM]

1.3.3.10.1. DEX Contour Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.10. Contour Plot

1.3.3.10.1. DEX Contour Plot

DEX Contour The dex contour plot is a specialized contour plot used in the analysis of
Plot: full and fractional experimental designs. These designs often have a low
Introduction level, coded as "-1" or "-", and a high level, coded as "+1" or "+" for each
factor. In addition, there can optionally be one or more center points.
Center points are at the mid-point between the low and high level for each
factor and are coded as "0".
The dex contour plot is generated for two factors. Typically, this would be
the two most important factors as determined by previous analyses (e.g.,
through the use of the dex mean plots and a Yates analysis). If more than
two factors are important, you may want to generate a series of dex
contour plots, each of which is drawn for two of these factors. You can
also generate a matrix of all pairwise dex contour plots for a number of
important factors (similar to the scatter plot matrix for scatter plots).

The typical application of the dex contour plot is in determining settings

that will maximize (or minimize) the response variable. It can also be
helpful in determining settings that result in the response variable hitting a
pre-determined target value. The dex contour plot plays a useful role in
determining the settings for the next iteration of the experiment. That is,
the initial experiment is typically a fractional factorial design with a fairly
large number of factors. After the most important factors are determined,
the dex contour plot can be used to help define settings for a full factorial
or response surface design based on a smaller number of factors.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a1.htm (1 of 4) [7/1/2003 5:21:15 PM]

1.3.3.10.1. DEX Contour Plot

Construction The following are the primary steps in the construction of the dex contour
of DEX plot.
Contour Plot 1. The x and y axes of the plot represent the values of the first and
second factor (independent) variables.
2. The four vertex points are drawn. The vertex points are (-1,-1),
(-1,1), (1,1), (1,-1). At each vertex point, the average of all the
response values at that vertex point is printed.
3. Similarly, if there are center points, a point is drawn at (0,0) and the
average of the response values at the center points is printed.
4. The linear dex contour plot assumes the model:

where is the overall mean of the response variable. The values of

, , , and are estimated from the vertex points using a
Yates analysis (the Yates analysis utilizes the special structure of the
2-level full and fractional factorial designs to simplify the
computation of these parameter estimates). Note that for the dex
contour plot, a full Yates analysis does not need to performed,
simply the calculations for generating the parameter estimates.
In order to generate a single contour line, we need a value for Y, say
Y0. Next, we solve for U2 in terms of U1 and, after doing the
algebra, we have the equation:

We generate a sequence of points for U1 in the range -2 to 2 and

compute the corresponding values of U2. These points constitute a
single contour line corresponding to Y = Y0.

The user specifies the target values for which contour lines will be
generated.
The above algorithm assumes a linear model for the design. Dex contour
plots can also be generated for the case in which we assume a quadratic
model for the design. The algebra for solving for U2 in terms of U1
becomes more complicated, but the fundamental idea is the same.
Quadratic models are needed for the case when the average for the center
points does not fall in the range defined by the vertex point (i.e., there is
curvature).

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a1.htm (2 of 4) [7/1/2003 5:21:15 PM]

1.3.3.10.1. DEX Contour Plot

Sample DEX The following is a dex contour plot for the data used in the Eddy current
Contour Plot case study. The analysis in that case study demonstrated that X1 and X2
were the most important factors.

Interpretation From the above dex contour plot we can derive the following information.
of the Sample 1. Interaction significance;
DEX Contour
2. Best (data) setting for these 2 dominant factors;
Plot

Interaction Note the appearance of the contour plot. If the contour curves are linear,
Significance then that implies that the interaction term is not significant; if the contour
curves have considerable curvature, then that implies that the interaction
term is large and important. In our case, the contour curves do not have
considerable curvature, and so we conclude that the X1*X2 term is not
significant.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a1.htm (3 of 4) [7/1/2003 5:21:15 PM]

1.3.3.10.1. DEX Contour Plot

Best Settings To determine the best factor settings for the already-run experiment, we
first must define what "best" means. For the Eddy current data set used to
generate this dex contour plot, "best" means to maximize (rather than
minimize or hit a target) the response. Hence from the contour plot we
determine the best settings for the two dominant factors by simply
scanning the four vertices and choosing the vertex with the largest value
(= average response). In this case, it is (X1 = +1, X2 = +1).
As for factor X3, the contour plot provides no best setting information, and
so we would resort to other tools: the main effects plot, the interaction
effects matrix, or the ordered data to determine optimal X3 settings.

Case Study The Eddy current case study demonstrates the use of the dex contour plot
in the context of the analysis of a full factorial design.

Software DEX contour plots are available in many statistical software programs that
analyze data from designed experiments. Dataplot supports a linear dex
contour plot and it provides a macro for generating a quadratic dex contour
plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33a1.htm (4 of 4) [7/1/2003 5:21:15 PM]

1.3.3.11. DEX Scatter Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.11. DEX Scatter Plot

Purpose: The dex scatter plot shows the response values for each level of each
Determine factor (i.e., independent) variable. This graphically shows how the
Important location and scale vary for both within a factor variable and between
Factors with different factor variables. This graphically shows which are the
Respect to important factors and can help provide a ranked list of important
Location and factors from a designed experiment. The dex scatter plot is a
Scale complement to the traditional analyis of variance of designed
experiments.
Dex scatter plots are typically used in conjunction with the dex mean
plot and the dex standard deviation plot. The dex mean plot replaces
the raw response values with mean response values while the dex
standard deviation plot replaces the raw response values with the
standard deviation of the response values. There is value in generating
all 3 of these plots. The dex mean and standard deviation plots are
useful in that the summary measures of location and spread stand out
(they can sometimes get lost with the raw plot). However, the raw data
points can reveal subtleties, such as the presence of outliers, that might
get lost with the summary statistics.

Sample Plot:
Factors 4, 2,
3, and 7 are
the Important
Factors.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33b.htm (1 of 5) [7/1/2003 5:21:15 PM]

1.3.3.11. DEX Scatter Plot

Description For this sample plot, there are seven factors and each factor has two
of the Plot levels. For each factor, we define a distinct x coordinate for each level
of the factor. For example, for factor 1, level 1 is coded as 0.8 and level
2 is coded as 1.2. The y coordinate is simply the value of the response
variable. The solid horizontal line is drawn at the overall mean of the
response variable. The vertical dotted lines are added for clarity.
Although the plot can be drawn with an arbitrary number of levels for a
factor, it is really only useful when there are two or three levels for a
factor.

Conclusions This sample dex scatter plot shows that:

1. there does not appear to be any outliers;
2. the levels of factors 2 and 4 show distinct location differences;
and
3. the levels of factor 1 show distinct scale differences.

Definition: Dex scatter plots are formed by:

Response ● Vertical axis: Value of the response variable
Values
● Horizontal axis: Factor variable (with each level of the factor
Versus
coded with a slightly offset x coordinate)
Factor
Variables

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33b.htm (2 of 5) [7/1/2003 5:21:15 PM]

1.3.3.11. DEX Scatter Plot

Questions The dex scatter plot can be used to answer the following questions:
1. Which factors are important with respect to location and scale?
2. Are there outliers?

Importance: The goal of many designed experiments is to determine which factors

Identify are important with respect to location and scale. A ranked list of the
Important important factors is also often of interest. Dex scatter, mean, and
Factors with standard deviation plots show this graphically. The dex scatter plot
Respect to additionally shows if outliers may potentially be distorting the results.
Location and
Scale Dex scatter plots were designed primarily for analyzing designed
experiments. However, they are useful for any type of multi-factor data
(i.e., a response variable with 2 or more factor variables having a small
number of distinct levels) whether or not the data were generated from
a designed experiment.

Extension for
Interaction
Effects

Using the concept of the scatterplot matrix, the dex scatter plot can be
extended to display first order interaction effects.
Specifically, if there are k factors, we create a matrix of plots with k
rows and k columns. On the diagonal, the plot is simply a dex scatter
plot with a single factor. For the off-diagonal plots, we multiply the
values of Xi and Xj. For the common 2-level designs (i.e., each factor
has two levels) the values are typically coded as -1 and 1, so the
multiplied values are also -1 and 1. We then generate a dex scatter plot
for this interaction variable. This plot is called a dex interaction effects
plot and an example is shown below.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33b.htm (3 of 5) [7/1/2003 5:21:15 PM]

1.3.3.11. DEX Scatter Plot

Interpretation We can first examine the diagonal elements for the main effects. These
of the Dex diagonal plots show a great deal of overlap between the levels for all
Interaction three factors. This indicates that location and scale effects will be
Effects Plot relatively small.
We can then examine the off-diagonal plots for the first order
interaction effects. For example, the plot in the first row and second
column is the interaction between factors X1 and X2. As with the main
effect plots, no clear patterns are evident.

Related Dex mean plot

Techniques Dex standard deviation plot
Block plot
Box plot
Analysis of variance

Case Study The dex scatter plot is demonstrated in the ceramic strength data case
study.

Software Dex scatter plots are available in some general purpose statistical
software programs, although the format may vary somewhat between
these programs. They are essentially just scatter plots with the X
variable defined in a particular way, so it should be feasible to write
macros for dex scatter plots in most statistical software programs.
Dataplot supports a dex scatter plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33b.htm (4 of 5) [7/1/2003 5:21:15 PM]

1.3.3.11. DEX Scatter Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33b.htm (5 of 5) [7/1/2003 5:21:15 PM]

1.3.3.12. DEX Mean Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.12. DEX Mean Plot

Purpose: The dex mean plot is appropriate for analyzing data from a designed
Detect experiment, with respect to important factors, where the factors are at
Important two or more levels. The plot shows mean values for the two or more
Factors with levels of each factor plotted by factor. The means for a single factor are
Respect to connected by a straight line. The dex mean plot is a complement to the
Location traditional analysis of variance of designed experiments.

This plot is typically generated for the mean. However, it can be

generated for other location statistics such as the median.

Sample
Plot:
Factors 4, 2,
and 1 are
the Most
Important
Factors

This sample dex mean plot shows that:

1. factor 4 is the most important;
2. factor 2 is the second most important;
3. factor 1 is the third most important;

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33c.htm (1 of 3) [7/1/2003 5:21:16 PM]

1.3.3.12. DEX Mean Plot

4. factor 7 is the fourth most important;

5. factor 6 is the fifth most important;
6. factors 3 and 5 are relatively unimportant.
In summary, factors 4, 2, and 1 seem to be clearly important, factors 3
and 5 seem to be clearly unimportant, and factors 6 and 7 are borderline
factors whose inclusion in any subsequent models will be determined by
further analyses.

Definition: Dex mean plots are formed by:

Mean ● Vertical axis: Mean of the response variable for each level of the
Response factor
Versus
● Horizontal axis: Factor variable
Factor
Variables

Questions The dex mean plot can be used to answer the following questions:
1. Which factors are important? The dex mean plot does not provide
a definitive answer to this question, but it does help categorize
factors as "clearly important", "clearly not important", and
"borderline importance".
2. What is the ranking list of the important factors?

Importance: The goal of many designed experiments is to determine which factors

Determine are significant. A ranked order listing of the important factors is also
Significant often of interest. The dex mean plot is ideally suited for answering these
Factors types of questions and we recommend its routine use in analyzing
designed experiments.

Extension Using the concept of the scatter plot matrix, the dex mean plot can be
for extended to display first-order interaction effects.
Interaction
Effects Specifically, if there are k factors, we create a matrix of plots with k
rows and k columns. On the diagonal, the plot is simply a dex mean plot
with a single factor. For the off-diagonal plots, measurements at each
level of the interaction are plotted versus level, where level is Xi times
Xj and Xi is the code for the ith main effect level and Xj is the code for
the jth main effect. For the common 2-level designs (i.e., each factor has
two levels) the values are typically coded as -1 and 1, so the multiplied
values are also -1 and 1. We then generate a dex mean plot for this
interaction variable. This plot is called a dex interaction effects plot and
an example is shown below.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33c.htm (2 of 3) [7/1/2003 5:21:16 PM]

1.3.3.12. DEX Mean Plot

DEX
Interaction
Effects Plot

This plot shows that the most significant factor is X1 and the most
significant interaction is between X1 and X3.

Related Dex scatter plot

Techniques Dex standard deviation plot
Block plot
Box plot
Analysis of variance

Case Study The dex mean plot and the dex interaction effects plot are demonstrated
in the ceramic strength data case study.

Software Dex mean plots are available in some general purpose statistical
software programs, although the format may vary somewhat between
these programs. It may be feasible to write macros for dex mean plots in
some statistical software programs that do not support this plot directly.
Dataplot supports both a dex mean plot and a dex interaction effects
plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33c.htm (3 of 3) [7/1/2003 5:21:16 PM]

1.3.3.13. DEX Standard Deviation Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.13. DEX Standard Deviation Plot

Purpose: The dex standard deviation plot is appropriate for analyzing data from a
Detect designed experiment, with respect to important factors, where the
Important factors are at two or more levels and there are repeated values at each
Factors with level. The plot shows standard deviation values for the two or more
Respect to levels of each factor plotted by factor. The standard deviations for a
Scale single factor are connected by a straight line. The dex standard deviation
plot is a complement to the traditional analysis of variance of designed
experiments.
This plot is typically generated for the standard deviation. However, it
can also be generated for other scale statistics such as the range, the
median absolute deviation, or the average absolute deviation.

Sample Plot

This sample dex standard deviation plot shows that:

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33d.htm (1 of 3) [7/1/2003 5:21:16 PM]

1.3.3.13. DEX Standard Deviation Plot

1. factor 1 has the greatest difference in standard deviations between

factor levels;
2. factor 4 has a significantly lower average standard deviation than
the average standard deviations of other factors (but the level 1
standard deviation for factor 1 is about the same as the level 1
standard deviation for factor 4);
3. for all factors, the level 1 standard deviation is smaller than the
level 2 standard deviation.

Definition: Dex standard deviation plots are formed by:

Response ● Vertical axis: Standard deviation of the response variable for each
Standard level of the factor
Deviations
● Horizontal axis: Factor variable
Versus
Factor
Variables

Questions The dex standard deviation plot can be used to answer the following
questions:
1. How do the standard deviations vary across factors?
2. How do the standard deviations vary within a factor?
3. Which are the most important factors with respect to scale?
4. What is the ranked list of the important factors with respect to
scale?

Importance: The goal with many designed experiments is to determine which factors
Assess are significant. This is usually determined from the means of the factor
Variability levels (which can be conveniently shown with a dex mean plot). A
secondary goal is to assess the variability of the responses both within a
factor and between factors. The dex standard deviation plot is a
convenient way to do this.

Related Dex scatter plot

Techniques Dex mean plot
Block plot
Box plot
Analysis of variance

Case Study The dex standard deviation plot is demonstrated in the ceramic strength
data case study.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33d.htm (2 of 3) [7/1/2003 5:21:16 PM]

1.3.3.13. DEX Standard Deviation Plot

Software Dex standard deviation plots are not available in most general purpose
statistical software programs. It may be feasible to write macros for dex
standard deviation plots in some statistical software programs that do
not support them directly. Dataplot supports a dex standard deviation
plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33d.htm (3 of 3) [7/1/2003 5:21:16 PM]

1.3.3.14. Histogram

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.14. Histogram
Purpose: The purpose of a histogram (Chambers) is to graphically summarize the
Summarize distribution of a univariate data set.
a Univariate
Data Set The histogram graphically shows the following:
1. center (i.e., the location) of the data;
2. spread (i.e., the scale) of the data;
3. skewness of the data;
4. presence of outliers; and
5. presence of multiple modes in the data.
These features provide strong indications of the proper distributional
model for the data. The probability plot or a goodness-of-fit test can be
used to verify the distributional model.
The examples section shows the appearance of a number of common
features revealed by histograms.

Sample Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e.htm (1 of 4) [7/1/2003 5:21:17 PM]

1.3.3.14. Histogram

Definition The most common form of the histogram is obtained by splitting the
range of the data into equal-sized bins (called classes). Then for each
bin, the number of points from the data set that fall into each bin are
counted. That is
● Vertical axis: Frequency (i.e., counts for each bin)

● Horizontal axis: Response variable

The classes can either be defined arbitrarily by the user or via some
systematic rule. A number of theoretically derived rules have been
proposed by Scott (Scott 1992).

The cumulative histogram is a variation of the histogram in which the

vertical axis gives not just the counts for a single bin, but rather gives
the counts for that bin plus all bins for smaller values of the response
variable.
Both the histogram and cumulative histogram have an additional variant
whereby the counts are replaced by the normalized counts. The names
for these variants are the relative histogram and the relative cumulative
histogram.
There are two common ways to normalize the counts.
1. The normalized count is the count in a class divided by the total
number of observations. In this case the relative counts are
normalized to sum to one (or 100 if a percentage scale is used).
This is the intuitive case where the height of the histogram bar
represents the proportion of the data in each class.
2. The normalized count is the count in the class divided by the

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e.htm (2 of 4) [7/1/2003 5:21:17 PM]

1.3.3.14. Histogram

number of observations times the class width. For this

normalization, the area (or integral) under the histogram is equal
to one. From a probabilistic point of view, this normalization
results in a relative histogram that is most akin to the probability
density function and a relative cumulative histogram that is most
akin to the cumulative distribution function. If you want to
overlay a probability density or cumulative distribution function
on top of the histogram, use this normalization. Although this
normalization is less intuitive (relative frequencies greater than 1
are quite permissible), it is the appropriate normalization if you
are using the histogram to model a probability density function.

Questions The histogram can be used to answer the following questions:

1. What kind of population distribution do the data come from?
2. Where are the data located?
3. How spread out are the data?
4. Are the data symmetric or skewed?
5. Are there outliers in the data?

Examples 1. Normal
2. Symmetric, Non-Normal, Short-Tailed
3. Symmetric, Non-Normal, Long-Tailed
4. Symmetric and Bimodal
5. Bimodal Mixture of 2 Normals
6. Skewed (Non-Symmetric) Right
7. Skewed (Non-Symmetric) Left
8. Symmetric with Outlier

Related Box plot

Techniques Probability plot

The techniques below are not discussed in the Handbook. However,

they are similar in purpose to the histogram. Additional information on
them is contained in the Chambers and Scott references.

Frequency Plot
Stem and Leaf Plot
Density Trace

Case Study The histogram is demonstrated in the heat flow meter data case study.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e.htm (3 of 4) [7/1/2003 5:21:17 PM]

1.3.3.14. Histogram

Software Histograms are available in most general purpose statistical software

programs. They are also supported in most general purpose charting,
spreadsheet, and business graphics programs. Dataplot supports
histograms.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e.htm (4 of 4) [7/1/2003 5:21:17 PM]

1.3.3.14.1. Histogram Interpretation: Normal

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.1. Histogram Interpretation: Normal

Symmetric,
Moderate-
Tailed
Histogram

Note the classical bell-shaped, symmetric histogram with most of the

frequency counts bunched in the middle and with the counts dying off
out in the tails. From a physical science/engineering point of view, the
normal distribution is that distribution which occurs most often in
nature (due in part to the central limit theorem).

Recommended If the histogram indicates a symmetric, moderate tailed distribution,

Next Step then the recommended next step is to do a normal probability plot to
confirm approximate normality. If the normal probability plot is linear,
then the normal distribution is a good model for the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e1.htm (1 of 2) [7/1/2003 5:21:17 PM]

1.3.3.14.1. Histogram Interpretation: Normal

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e1.htm (2 of 2) [7/1/2003 5:21:17 PM]

1.3.3.14.2. Histogram Interpretation: Symmetric, Non-Normal, Short-Tailed

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.2. Histogram Interpretation:

Symmetric, Non-Normal,
Short-Tailed
Symmetric,
Short-Tailed
Histogram

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e2.htm (1 of 3) [7/1/2003 5:21:17 PM]

1.3.3.14.2. Histogram Interpretation: Symmetric, Non-Normal, Short-Tailed

Description of For a symmetric distribution, the "body" of a distribution refers to the

What "center" of the distribution--commonly that region of the distribution
Short-Tailed where most of the probability resides--the "fat" part of the distribution.
Means The "tail" of a distribution refers to the extreme regions of the
distribution--both left and right. The "tail length" of a distribution is a
term that indicates how fast these extremes approach zero.
For a short-tailed distribution, the tails approach zero very fast. Such
distributions commonly have a truncated ("sawed-off") look. The
classical short-tailed distribution is the uniform (rectangular)
distribution in which the probability is constant over a given range and
then drops to zero everywhere else--we would speak of this as having
no tails, or extremely short tails.
For a moderate-tailed distribution, the tails decline to zero in a
moderate fashion. The classical moderate-tailed distribution is the
normal (Gaussian) distribution.
For a long-tailed distribution, the tails decline to zero very slowly--and
hence one is apt to see probability a long way from the body of the
distribution. The classical long-tailed distribution is the Cauchy
distribution.
In terms of tail length, the histogram shown above would be
characteristic of a "short-tailed" distribution.
The optimal (unbiased and most precise) estimator for location for the
center of a distribution is heavily dependent on the tail length of the
distribution. The common choice of taking N observations and using
the calculated sample mean as the best estimate for the center of the
distribution is a good choice for the normal distribution (moderate
tailed), a poor choice for the uniform distribution (short tailed), and a
horrible choice for the Cauchy distribution (long tailed). Although for
the normal distribution the sample mean is as precise an estimator as
we can get, for the uniform and Cauchy distributions, the sample mean
is not the best estimator.
For the uniform distribution, the midrange
midrange = (smallest + largest) / 2
is the best estimator of location. For a Cauchy distribution, the median
is the best estimator of location.

Recommended If the histogram indicates a symmetric, short-tailed distribution, the

Next Step recommended next step is to generate a uniform probability plot. If the
uniform probability plot is linear, then the uniform distribution is an
appropriate model for the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e2.htm (2 of 3) [7/1/2003 5:21:17 PM]

1.3.3.14.2. Histogram Interpretation: Symmetric, Non-Normal, Short-Tailed

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e2.htm (3 of 3) [7/1/2003 5:21:17 PM]

1.3.3.14.3. Histogram Interpretation: Symmetric, Non-Normal, Long-Tailed

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.3. Histogram Interpretation:

Symmetric, Non-Normal,
Long-Tailed
Symmetric,
Long-Tailed
Histogram

Description of The previous example contains a discussion of the distinction between

Long-Tailed short-tailed, moderate-tailed, and long-tailed distributions.

In terms of tail length, the histogram shown above would be

characteristic of a "long-tailed" distribution.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e3.htm (1 of 2) [7/1/2003 5:21:17 PM]

1.3.3.14.3. Histogram Interpretation: Symmetric, Non-Normal, Long-Tailed

Recommended If the histogram indicates a symmetric, long tailed distribution, the

Next Step recommended next step is to do a Cauchy probability plot. If the
Cauchy probability plot is linear, then the Cauchy distribution is an
appropriate model for the data. Alternatively, a Tukey Lambda PPCC
plot may provide insight into a suitable distributional model for the
data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e3.htm (2 of 2) [7/1/2003 5:21:17 PM]

1.3.3.14.4. Histogram Interpretation: Symmetric and Bimodal

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.4. Histogram Interpretation:

Symmetric and Bimodal
Symmetric,
Bimodal
Histogram

Description of The mode of a distribution is that value which is most frequently

Bimodal occurring or has the largest probability of occurrence. The sample
mode occurs at the peak of the histogram.
For many phenomena, it is quite common for the distribution of the
response values to cluster around a single mode (unimodal) and then
distribute themselves with lesser frequency out into the tails. The
normal distribution is the classic example of a unimodal distribution.
The histogram shown above illustrates data from a bimodal (2 peak)
distribution. The histogram serves as a tool for diagnosing problems
such as bimodality. Questioning the underlying reason for
distributional non-unimodality frequently leads to greater insight and

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e4.htm (1 of 2) [7/1/2003 5:21:18 PM]

1.3.3.14.4. Histogram Interpretation: Symmetric and Bimodal

improved deterministic modeling of the phenomenon under study. For

example, for the data presented above, the bimodal histogram is
caused by sinusoidality in the data.

Recommended If the histogram indicates a symmetric, bimodal distribution, the

Next Step recommended next steps are to:
1. Do a run sequence plot or a scatter plot to check for
sinusoidality.
2. Do a lag plot to check for sinusoidality. If the lag plot is
elliptical, then the data are sinusoidal.
3. If the data are sinusoidal, then a spectral plot is used to
graphically estimate the underlying sinusoidal frequency.
4. If the data are not sinusoidal, then a Tukey Lambda PPCC plot
may determine the best-fit symmetric distribution for the data.
5. The data may be fit with a mixture of two distributions. A
common approach to this case is to fit a mixture of 2 normal or
lognormal distributions. Further discussion of fitting mixtures of
distributions is beyond the scope of this Handbook.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e4.htm (2 of 2) [7/1/2003 5:21:18 PM]

1.3.3.14.5. Histogram Interpretation: Bimodal Mixture of 2 Normals

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.5. Histogram Interpretation:

Bimodal Mixture of 2 Normals
Histogram
from Mixture
of 2 Normal
Distributions

Discussion of The histogram shown above illustrates data from a bimodal (2 peak)
Unimodal and distribution.
Bimodal
In contrast to the previous example, this example illustrates bimodality
due not to an underlying deterministic model, but bimodality due to a
mixture of probability models. In this case, each of the modes appears
to have a rough bell-shaped component. One could easily imagine the
above histogram being generated by a process consisting of two
normal distributions with the same standard deviation but with two
different locations (one centered at approximately 9.17 and the other
centered at approximately 9.26). If this is the case, then the research
challenge is to determine physically why there are two similar but
separate sub-processes.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e5.htm (1 of 2) [7/1/2003 5:21:19 PM]

1.3.3.14.5. Histogram Interpretation: Bimodal Mixture of 2 Normals

Recommended If the histogram indicates that the data might be appropriately fit with
Next Steps a mixture of two normal distributions, the recommended next step is:
Fit the normal mixture model using either least squares or maximum
likelihood. The general normal mixing model is

where p is the mixing proportion (between 0 and 1) and and are

normal probability density functions with location and scale
parameters , , , and , respectively. That is, there are 5
parameters to estimate in the fit.
Whether maximum likelihood or least squares is used, the quality of
the fit is sensitive to good starting values. For the mixture of two
normals, the histogram can be used to provide initial estimates for the
location and scale parameters of the two normal distributions.
Dataplot can generate a least squares fit of the mixture of two normals
with the following sequence of commands:
RELATIVE HISTOGRAM Y
LET Y2 = YPLOT
LET X2 = XPLOT
RETAIN Y2 X2 SUBSET TAGPLOT = 1
LET U1 = <estimated value from histogram>
LET SD1 = <estimated value from histogram>
LET U2 = <estimated value from histogram>
LET SD2 = <estimated value from histogram>
LET P = 0.5
FIT Y2 = NORMXPDF(X2,U1,S1,U2,S2,P)

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e5.htm (2 of 2) [7/1/2003 5:21:19 PM]

1.3.3.14.6. Histogram Interpretation: Skewed (Non-Normal) Right

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.6. Histogram Interpretation:

Skewed (Non-Normal) Right
Right-Skewed
Histogram

Discussion of A symmetric distribution is one in which the 2 "halves" of the

Skewness histogram appear as mirror-images of one another. A skewed
(non-symmetric) distribution is a distribution in which there is no such
mirror-imaging.
For skewed distributions, it is quite common to have one tail of the
distribution considerably longer or drawn out relative to the other tail.
A "skewed right" distribution is one in which the tail is on the right
side. A "skewed left" distribution is one in which the tail is on the left
side. The above histogram is for a distribution that is skewed right.
Skewed distributions bring a certain philosophical complexity to the
very process of estimating a "typical value" for the distribution. To be

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e6.htm (1 of 3) [7/1/2003 5:21:29 PM]

1.3.3.14.6. Histogram Interpretation: Skewed (Non-Normal) Right

specific, suppose that the analyst has a collection of 100 values

randomly drawn from a distribution, and wishes to summarize these
100 observations by a "typical value". What does typical value mean?
If the distribution is symmetric, the typical value is unambiguous-- it is
a well-defined center of the distribution. For example, for a
bell-shaped symmetric distribution, a center point is identical to that
value at the peak of the distribution.
For a skewed distribution, however, there is no "center" in the usual
sense of the word. Be that as it may, several "typical value" metrics are
often used for skewed distributions. The first metric is the mode of the
distribution. Unfortunately, for severely-skewed distributions, the
mode may be at or near the left or right tail of the data and so it seems
not to be a good representative of the center of the distribution. As a
second choice, one could conceptually argue that the mean (the point
on the horizontal axis where the distributiuon would balance) would
serve well as the typical value. As a third choice, others may argue
that the median (that value on the horizontal axis which has exactly
50% of the data to the left (and also to the right) would serve as a good
typical value.
For symmetric distributions, the conceptual problem disappears
because at the population level the mode, mean, and median are
identical. For skewed distributions, however, these 3 metrics are
markedly different. In practice, for skewed distributions the most
commonly reported typical value is the mean; the next most common
is the median; the least common is the mode. Because each of these 3
metrics reflects a different aspect of "centerness", it is recommended
that the analyst report at least 2 (mean and median), and preferably all
3 (mean, median, and mode) in summarizing and characterizing a data
set.

Some Causes Skewed data often occur due to lower or upper bounds on the data.
for Skewed That is, data that have a lower bound are often skewed right while data
Data that have an upper bound are often skewed left. Skewness can also
result from start-up effects. For example, in reliability applications
some processes may have a large number of initial failures that could
cause left skewness. On the other hand, a reliability process could
have a long start-up period where failures are rare resulting in
right-skewed data.
Data collected in scientific and engineering applications often have a
lower bound of zero. For example, failure data must be non-negative.
Many measurement processes generate only positive data. Time to
occurence and size are common measurements that cannot be less than
zero.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e6.htm (2 of 3) [7/1/2003 5:21:29 PM]

1.3.3.14.6. Histogram Interpretation: Skewed (Non-Normal) Right

Recommended If the histogram indicates a right-skewed data set, the recommended

Next Steps next steps are to:
1. Quantitatively summarize the data by computing and reporting
the sample mean, the sample median, and the sample mode.
2. Determine the best-fit distribution (skewed-right) from the
❍ Weibull family (for the maximum)

❍ Gamma family
❍ Chi-square family
❍ Lognormal family
❍ Power lognormal family
3. Consider a normalizing transformation such as the Box-Cox
transformation.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e6.htm (3 of 3) [7/1/2003 5:21:29 PM]

1.3.3.14.7. Histogram Interpretation: Skewed (Non-Symmetric) Left

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.7. Histogram Interpretation:

Skewed (Non-Symmetric) Left
Skewed Left
Histogram

The issues for skewed left data are similar to those for skewed right
data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e7.htm [7/1/2003 5:21:29 PM]

1.3.3.14.8. Histogram Interpretation: Symmetric with Outlier

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.14. Histogram

1.3.3.14.8. Histogram Interpretation:

Symmetric with Outlier
Symmetric
Histogram
with Outlier

Discussion of A symmetric distribution is one in which the 2 "halves" of the

Outliers histogram appear as mirror-images of one another. The above example
is symmetric with the exception of outlying data near Y = 4.5.
An outlier is a data point that comes from a distribution different (in
location, scale, or distributional form) from the bulk of the data. In the
real world, outliers have a range of causes, from as simple as
1. operator blunders
2. equipment failures
3. day-to-day effects
4. batch-to-batch differences
5. anomalous input conditions

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e8.htm (1 of 2) [7/1/2003 5:21:29 PM]

1.3.3.14.8. Histogram Interpretation: Symmetric with Outlier

6. warm-up effects
to more subtle causes such as
1. A change in settings of factors that (knowingly or unknowingly)
affect the response.
2. Nature is trying to tell us something.

Outliers All outliers should be taken seriously and should be investigated

Should be thoroughly for explanations. Automatic outlier-rejection schemes
Investigated (such as throw out all data beyond 4 sample standard deviations from
the sample mean) are particularly dangerous.
The classic case of automatic outlier rejection becoming automatic
information rejection was the South Pole ozone depletion problem.
Ozone depletion over the South Pole would have been detected years
earlier except for the fact that the satellite data recording the low
ozone readings had outlier-rejection code that automatically screened
out the "outliers" (that is, the low ozone readings) before the analysis
was conducted. Such inadvertent (and incorrect) purging went on for
years. It was not until ground-based South Pole readings started
detecting low ozone readings that someone decided to double-check as
to why the satellite had not picked up this fact--it had, but it had gotten
thrown out!
The best attitude is that outliers are our "friends", outliers are trying to
tell us something, and we should not stop until we are comfortable in
the explanation for each outlier.

Recommended If the histogram shows the presence of outliers, the recommended next
Next Steps steps are:
1. Graphically check for outliers (in the commonly encountered
normal case) by generating a box plot. In general, box plots are
a much better graphical tool for detecting outliers than are
histograms.
2. Quantitatively check for outliers (in the commonly encountered
normal case) by carrying out Grubbs test which indicates how
many sample standard deviations away from the sample mean
are the data in question. Large values indicate outliers.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33e8.htm (2 of 2) [7/1/2003 5:21:29 PM]

1.3.3.15. Lag Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.15. Lag Plot

Purpose: A lag plot checks whether a data set or time series is random or not.
Check for Random data should not exhibit any identifiable structure in the lag plot.
randomness Non-random structure in the lag plot indicates that the underlying data
are not random. Several common patterns for lag plots are shown in the
examples below.

Sample Plot

This sample lag plot exhibits a linear pattern. This shows that the data
are strongly non-random and further suggests that an autoregressive
model might be appropriate.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f.htm (1 of 2) [7/1/2003 5:21:29 PM]

1.3.3.15. Lag Plot

Definition A lag is a fixed time displacement. For example, given a data set Y1, Y2
..., Yn, Y2 and Y7 have lag 5 since 7 - 2 = 5. Lag plots can be generated
for any arbitrary lag, although the most commonly used lag is 1.
A plot of lag 1 is a plot of the values of Yi versus Yi-1
● Vertical axis: Yi for all i
● Horizontal axis: Yi-1 for all i

Questions Lag plots can provide answers to the following questions:

1. Are the data random?
2. Is there serial correlation in the data?
3. What is a suitable model for the data?
4. Are there outliers in the data?

Importance Inasmuch as randomness is an underlying assumption for most statistical

estimation and testing techniques, the lag plot should be a routine tool
for researchers.

Examples ● Random (White Noise)

● Weak autocorrelation
● Strong autocorrelation and autoregressive model
● Sinusoidal model and outliers

Related Autocorrelation Plot

Techniques Spectrum
Runs Test

Case Study The lag plot is demonstrated in the beam deflection data case study.

Software Lag plots are not directly available in most general purpose statistical
software programs. Since the lag plot is essentially a scatter plot with
the 2 variables properly lagged, it should be feasible to write a macro for
the lag plot in most statistical programs. Dataplot supports a lag plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f.htm (2 of 2) [7/1/2003 5:21:29 PM]

1.3.3.15.1. Lag Plot: Random Data

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.15. Lag Plot

1.3.3.15.1. Lag Plot: Random Data

Lag Plot

Conclusions We can make the following conclusions based on the above plot.
1. The data are random.
2. The data exhibit no autocorrelation.
3. The data contain no outliers.

Discussion The lag plot shown above is for lag = 1. Note the absence of structure.
One cannot infer, from a current value Yi-1, the next value Yi. Thus for a
known value Yi-1 on the horizontal axis (say, Yi-1 = +0.5), the Yi-th
value could be virtually anything (from Yi = -2.5 to Yi = +1.5). Such
non-association is the essence of randomness.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f1.htm (1 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.1. Lag Plot: Random Data

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f1.htm (2 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.2. Lag Plot: Moderate Autocorrelation

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.15. Lag Plot

1.3.3.15.2. Lag Plot: Moderate

Autocorrelation
Lag Plot

Conclusions We can make the conclusions based on the above plot.

1. The data are from an underlying autoregressive model with
moderate positive autocorrelation
2. The data contain no outliers.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f2.htm (1 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.2. Lag Plot: Moderate Autocorrelation

Discussion In the plot above for lag = 1, note how the points tend to cluster (albeit
noisily) along the diagonal. Such clustering is the lag plot signature of
moderate autocorrelation.
If the process were completely random, knowledge of a current
observation (say Yi-1 = 0) would yield virtually no knowledge about
the next observation Yi. If the process has moderate autocorrelation, as
above, and if Yi-1 = 0, then the range of possible values for Yi is seen
to be restricted to a smaller range (.01 to +.01). This suggests
prediction is possible using an autoregressive model.

Recommended Estimate the parameters for the autoregressive model:

Next Step
Since Yi and Yi-1 are precisely the axes of the lag plot, such estimation
is a linear regression straight from the lag plot.

The residual standard deviation for the autoregressive model will be

much smaller than the residual standard deviation for the default
model

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f2.htm (2 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.3. Lag Plot: Strong Autocorrelation and Autoregressive Model

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.15. Lag Plot

1.3.3.15.3. Lag Plot: Strong Autocorrelation

and Autoregressive Model
Lag Plot

Conclusions We can make the following conclusions based on the above plot.
1. The data come from an underlying autoregressive model with
strong positive autocorrelation
2. The data contain no outliers.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f3.htm (1 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.3. Lag Plot: Strong Autocorrelation and Autoregressive Model

Discussion Note the tight clustering of points along the diagonal. This is the lag
plot signature of a process with strong positive autocorrelation. Such
processes are highly non-random--there is strong association between
an observation and a succeeding observation. In short, if you know
Yi-1 you can make a strong guess as to what Yi will be.

If the above process were completely random, the plot would have a
shotgun pattern, and knowledge of a current observation (say Yi-1 = 3)
would yield virtually no knowledge about the next observation Yi (it
could here be anywhere from -2 to +8). On the other hand, if the
process had strong autocorrelation, as seen above, and if Yi-1 = 3, then
the range of possible values for Yi is seen to be restricted to a smaller
range (2 to 4)--still wide, but an improvement nonetheless (relative to
-2 to +8) in predictive power.

Recommended When the lag plot shows a strongly autoregressive pattern and only
Next Step successive observations appear to be correlated, the next steps are to:
1. Extimate the parameters for the autoregressive model:

Since Yi and Yi-1 are precisely the axes of the lag plot, such
estimation is a linear regression straight from the lag plot.

The residual standard deviation for this autoregressive model

will be much smaller than the residual standard deviation for the
default model

2. Reexamine the system to arrive at an explanation for the strong

autocorrelation. Is it due to the
1. phenomenon under study; or
2. drifting in the environment; or
3. contamination from the data acquisition system?
Sometimes the source of the problem is contamination and
carry-over from the data acquisition system where the system
does not have time to electronically recover before collecting
the next data point. If this is the case, then consider slowing
down the sampling rate to achieve randomness.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f3.htm (2 of 2) [7/1/2003 5:21:30 PM]

1.3.3.15.4. Lag Plot: Sinusoidal Models and Outliers

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.15. Lag Plot

1.3.3.15.4. Lag Plot: Sinusoidal Models and

Outliers
Lag Plot

Conclusions We can make the following conclusions based on the above plot.
1. The data come from an underlying single-cycle sinusoidal
model.
2. The data contain three outliers.

Discussion In the plot above for lag = 1, note the tight elliptical clustering of
points. Processes with a single-cycle sinusoidal model will have such
elliptical lag plots.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f4.htm (1 of 3) [7/1/2003 5:21:31 PM]

1.3.3.15.4. Lag Plot: Sinusoidal Models and Outliers

Consequences If one were to naively assume that the above process came from the
of Ignoring null model
Cyclical
Pattern
and then estimate the constant by the sample mean, then the analysis
would suffer because
1. the sample mean would be biased and meaningless;
2. the confidence limits would be meaningless and optimistically
small.
The proper model

(where is the amplitude, is the frequency--between 0 and .5

cycles per observation--, and is the phase) can be fit by standard
non-linear least squares, to estimate the coefficients and their
uncertainties.
The lag plot is also of value in outlier detection. Note in the above plot
that there appears to be 4 points lying off the ellipse. However, in a lag
plot, each point in the original data set Y shows up twice in the lag
plot--once as Yi and once as Yi-1. Hence the outlier in the upper left at
Yi = 300 is the same raw data value that appears on the far right at Yi-1
= 300. Thus (-500,300) and (300,200) are due to the same outlier,
namely the 158th data point: 300. The correct value for this 158th
point should be approximately -300 and so it appears that a sign got
dropped in the data collection. The other two points lying off the
ellipse, at roughly (100,100) and at (0,-50), are caused by two faulty
data values: the third data point of -15 should be about +125 and the
fourth data point of +141 should be about -50, respectively. Hence the
4 apparent lag plot outliers are traceable to 3 actual outliers in the
original run sequence: at points 4 (-15), 5 (141) and 158 (300). In
retrospect, only one of these (point 158 (= 300)) is an obvious outlier
in the run sequence plot.

Unexpected Frequently a technique (e.g., the lag plot) is constructed to check one
Value of EDA aspect (e.g., randomness) which it does well. Along the way, the
technique also highlights some other anomaly of the data (namely, that
there are 3 outliers). Such outlier identification and removal is
extremely important for detecting irregularities in the data collection
system, and also for arriving at a "purified" data set for modeling. The
lag plot plays an important role in such outlier identification.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f4.htm (2 of 3) [7/1/2003 5:21:31 PM]

1.3.3.15.4. Lag Plot: Sinusoidal Models and Outliers

Recommended When the lag plot indicates a sinusoidal model with possible outliers,
Next Step the recommended next steps are:
1. Do a spectral plot to obtain an initial estimate of the frequency
of the underlying cycle. This will be helpful as a starting value
for the subsequent non-linear fitting.
2. Omit the outliers.
3. Carry out a non-linear fit of the model to the 197 points.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33f4.htm (3 of 3) [7/1/2003 5:21:31 PM]

1.3.3.16. Linear Correlation Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.16. Linear Correlation Plot

Purpose: Linear correlation plots are used to assess whether or not correlations
Detect are consistent across groups. That is, if your data is in groups, you may
changes in want to know if a single correlation can be used across all the groups or
correlation whether separate correlations are required for each group.
between
groups Linear correlation plots are often used in conjunction with linear slope,
linear intercept, and linear residual standard deviation plots. A linear
correlation plot could be generated intially to see if linear fitting would
be a fruitful direction. If the correlations are high, this implies it is
worthwhile to continue with the linear slope, intercept, and residual
standard deviation plots. If the correlations are weak, a different model
needs to be pursued.
In some cases, you might not have groups. Instead you may have
different data sets and you want to know if the same correlation can be
adequately applied to each of the data sets. In this case, simply think of
each distinct data set as a group and apply the linear slope plot as for
groups.

Sample Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33g.htm (1 of 3) [7/1/2003 5:21:31 PM]

1.3.3.16. Linear Correlation Plot

This linear correlation plot shows that the correlations are high for all
groups. This implies that linear fits could provide a good model for
each of these groups.

Definition: Linear correlation plots are formed by:

Group ● Vertical axis: Group correlations
Correlations
● Horizontal axis: Group identifier
Versus
Group ID A reference line is plotted at the correlation between the full data sets.

Questions The linear correlation plot can be used to answer the following
questions.
1. Are there linear relationships across groups?
2. Are the strength of the linear relationships relatively constant
across the groups?

Importance: For grouped data, it may be important to know whether the different
Checking groups are homogeneous (i.e., similar) or heterogeneous (i.e., different).
Group Linear correlation plots help answer this question in the context of
Homogeneity linear fitting.

Related Linear Intercept Plot

Techniques Linear Slope Plot
Linear Residual Standard Deviation Plot
Linear Fitting

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33g.htm (2 of 3) [7/1/2003 5:21:31 PM]

1.3.3.16. Linear Correlation Plot

Case Study The linear correlation plot is demonstrated in the Alaska pipeline data
case study.

Software Most general purpose statistical software programs do not support a

linear correlation plot. However, if the statistical program can generate
correlations over a group, it should be feasible to write a macro to
generate this plot. Dataplot supports a linear correlation plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33g.htm (3 of 3) [7/1/2003 5:21:31 PM]

1.3.3.17. Linear Intercept Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.17. Linear Intercept Plot

Purpose: Linear intercept plots are used to graphically assess whether or not
Detect linear fits are consistent across groups. That is, if your data have
changes in groups, you may want to know if a single fit can be used across all the
linear groups or whether separate fits are required for each group.
intercepts
between Linear intercept plots are typically used in conjunction with linear slope
groups and linear residual standard deviation plots.

In some cases you might not have groups. Instead, you have different
data sets and you want to know if the same fit can be adequately applied
to each of the data sets. In this case, simply think of each distinct data
set as a group and apply the linear intercept plot as for groups.

Sample Plot

This linear intercept plot shows that there is a shift in intercepts.

Specifically, the first three intercepts are lower than the intercepts for

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33h.htm (1 of 2) [7/1/2003 5:21:31 PM]

1.3.3.17. Linear Intercept Plot

the other groups. Note that these are small differences in the intercepts.

Definition: Linear intercept plots are formed by:

Group ● Vertical axis: Group intercepts from linear fits
Intercepts
● Horizontal axis: Group identifier
Versus
Group ID A reference line is plotted at the intercept from a linear fit using all the
data.

Questions The linear intercept plot can be used to answer the following questions.
1. Is the intercept from linear fits relatively constant across groups?
2. If the intercepts vary across groups, is there a discernible pattern?

Importance: For grouped data, it may be important to know whether the different
Checking groups are homogeneous (i.e., similar) or heterogeneous (i.e., different).
Group Linear intercept plots help answer this question in the context of linear
Homogeneity fitting.

Related Linear Correlation Plot

Techniques Linear Slope Plot
Linear Residual Standard Deviation Plot
Linear Fitting

Case Study The linear intercept plot is demonstrated in the Alaska pipeline data
case study.

Software Most general purpose statistical software programs do not support a

linear intercept plot. However, if the statistical program can generate
linear fits over a group, it should be feasible to write a macro to
generate this plot. Dataplot supports a linear intercept plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33h.htm (2 of 2) [7/1/2003 5:21:31 PM]

1.3.3.18. Linear Slope Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.18. Linear Slope Plot

Purpose: Linear slope plots are used to graphically assess whether or not linear
Detect fits are consistent across groups. That is, if your data have groups, you
changes in may want to know if a single fit can be used across all the groups or
linear slopes whether separate fits are required for each group.
between
groups Linear slope plots are typically used in conjunction with linear intercept
and linear residual standard deviation plots.

In some cases you might not have groups. Instead, you have different
data sets and you want to know if the same fit can be adequately applied
to each of the data sets. In this case, simply think of each distinct data
set as a group and apply the linear slope plot as for groups.

Sample Plot

This linear slope plot shows that the slopes are about 0.174 (plus or
minus 0.002) for all groups. There does not appear to be a pattern in the
variation of the slopes. This implies that a single fit may be adequate.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33i.htm (1 of 2) [7/1/2003 5:21:32 PM]

1.3.3.18. Linear Slope Plot

Definition: Linear slope plots are formed by:

Group ● Vertical axis: Group slopes from linear fits
Slopes
● Horizontal axis: Group identifier
Versus
Group ID A reference line is plotted at the slope from a linear fit using all the
data.

Questions The linear slope plot can be used to answer the following questions.
1. Do you get the same slope across groups for linear fits?
2. If the slopes differ, is there a discernible pattern in the slopes?

Importance: For grouped data, it may be important to know whether the different
Checking groups are homogeneous (i.e., similar) or heterogeneous (i.e., different).
Group Linear slope plots help answer this question in the context of linear
Homogeneity fitting.

Related Linear Intercept Plot

Techniques Linear Correlation Plot
Linear Residual Standard Deviation Plot
Linear Fitting

Case Study The linear slope plot is demonstrated in the Alaska pipeline data case
study.

Software Most general purpose statistical software programs do not support a

linear slope plot. However, if the statistical program can generate linear
fits over a group, it should be feasible to write a macro to generate this
plot. Dataplot supports a linear slope plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33i.htm (2 of 2) [7/1/2003 5:21:32 PM]

1.3.3.19. Linear Residual Standard Deviation Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.19. Linear Residual Standard

Deviation Plot
Purpose: Linear residual standard deviation (RESSD) plots are used to
Detect graphically assess whether or not linear fits are consistent across
Changes in groups. That is, if your data have groups, you may want to know if a
Linear single fit can be used across all the groups or whether separate fits are
Residual required for each group.
Standard
Deviation The residual standard deviation is a goodness-of-fit measure. That is,
Between the smaller the residual standard deviation, the closer is the fit to the
Groups data.
Linear RESSD plots are typically used in conjunction with linear
intercept and linear slope plots. The linear intercept and slope plots
convey whether or not the fits are consistent across groups while the
linear RESSD plot conveys whether the adequacy of the fit is consistent
across groups.
In some cases you might not have groups. Instead, you have different
data sets and you want to know if the same fit can be adequately applied
to each of the data sets. In this case, simply think of each distinct data
set as a group and apply the linear RESSD plot as for groups.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33j.htm (1 of 3) [7/1/2003 5:21:32 PM]

1.3.3.19. Linear Residual Standard Deviation Plot

Sample Plot

This linear RESSD plot shows that the residual standard deviations
from a linear fit are about 0.0025 for all the groups.

Definition: Linear RESSD plots are formed by:

Group ● Vertical axis: Group residual standard deviations from linear fits
Residual
● Horizontal axis: Group identifier
Standard
Deviation A reference line is plotted at the residual standard deviation from a
Versus linear fit using all the data. This reference line will typically be much
Group ID greater than any of the individual residual standard deviations.

Questions The linear RESSD plot can be used to answer the following questions.
1. Is the residual standard deviation from a linear fit constant across
groups?
2. If the residual standard deviations vary, is there a discernible
pattern across the groups?

Importance: For grouped data, it may be important to know whether the different
Checking groups are homogeneous (i.e., similar) or heterogeneous (i.e., different).
Group Linear RESSD plots help answer this question in the context of linear
Homogeneity fitting.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33j.htm (2 of 3) [7/1/2003 5:21:32 PM]

1.3.3.19. Linear Residual Standard Deviation Plot

Related Linear Intercept Plot

Techniques Linear Slope Plot
Linear Correlation Plot
Linear Fitting

Case Study The linear residual standard deviation plot is demonstrated in the
Alaska pipeline data case study.

Software Most general purpose statistical software programs do not support a

linear residual standard deviation plot. However, if the statistical
program can generate linear fits over a group, it should be feasible to
write a macro to generate this plot. Dataplot supports a linear residual
standard deviation plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33j.htm (3 of 3) [7/1/2003 5:21:32 PM]

1.3.3.20. Mean Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.20. Mean Plot

Purpose: Mean plots are used to see if the mean varies between different groups
Detect of the data. The grouping is determined by the analyst. In most cases,
changes in the data set contains a specific grouping variable. For example, the
location groups may be the levels of a factor variable. In the sample plot below,
between the months of the year provide the grouping.
groups
Mean plots can be used with ungrouped data to determine if the mean is
changing over time. In this case, the data are split into an arbitrary
number of equal-sized groups. For example, a data series with 400
points can be divided into 10 groups of 40 points each. A mean plot can
then be generated with these groups to see if the mean is increasing or
decreasing over time.
Although the mean is the most commonly used measure of location, the
same concept applies to other measures of location. For example,
instead of plotting the mean of each group, the median or the trimmed
mean might be plotted instead. This might be done if there were
significant outliers in the data and a more robust measure of location
than the mean was desired.
Mean plots are typically used in conjunction with standard deviation
plots. The mean plot checks for shifts in location while the standard
deviation plot checks for shifts in scale.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33k.htm (1 of 3) [7/1/2003 5:21:32 PM]

1.3.3.20. Mean Plot

Sample Plot

This sample mean plot shows a shift of location after the 6th month.

Definition: Mean plots are formed by:

Group ● Vertical axis: Group mean
Means
● Horizontal axis: Group identifier
Versus
Group ID A reference line is plotted at the overall mean.

Questions The mean plot can be used to answer the following questions.
1. Are there any shifts in location?
2. What is the magnitude of the shifts in location?
3. Is there a distinct pattern in the shifts in location?

Importance: A common assumption in 1-factor analyses is that of constant location.

Checking That is, the location is the same for different levels of the factor
Assumptions variable. The mean plot provides a graphical check for that assumption.
A common assumption for univariate data is that the location is
constant. By grouping the data into equal intervals, the mean plot can
provide a graphical test of this assumption.

Related Standard Deviation Plot

Techniques Dex Mean Plot
Box Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33k.htm (2 of 3) [7/1/2003 5:21:32 PM]

1.3.3.20. Mean Plot

Software Most general purpose statistical software programs do not support a

mean plot. However, if the statistical program can generate the mean
over a group, it should be feasible to write a macro to generate this plot.
Dataplot supports a mean plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33k.htm (3 of 3) [7/1/2003 5:21:32 PM]

1.3.3.21. Normal Probability Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.21. Normal Probability Plot

Purpose: The normal probability plot (Chambers 1983) is a graphical technique
Check If Data for assessing whether or not a data set is approximately normally
Are distributed.
Approximately
Normally The data are plotted against a theoretical normal distribution in such a
Distributed way that the points should form an approximate straight line.
Departures from this straight line indicate departures from normality.
The normal probability plot is a special case of the probability plot.
We cover the normal probability plot separately due to its importance
in many applications.

Sample Plot

The points on this plot form a nearly linear pattern, which indicates
that the normal distribution is a good model for this data set.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l.htm (1 of 3) [7/1/2003 5:21:33 PM]

1.3.3.21. Normal Probability Plot

Definition: The normal probability plot is formed by:

Ordered ● Vertical axis: Ordered response values
Response
● Horizontal axis: Normal order statistic medians
Values Versus
Normal Order The observations are plotted as a function of the corresponding normal
Statistic order statistic medians which are defined as:
Medians N(i) = G(U(i))
where U(i) are the uniform order statistic medians (defined below) and
G is the percent point function of the normal distribution. The percent
point function is the inverse of the cumulative distribution function
(probability that x is less than or equal to some value). That is, given a
probability, we want the corresponding x of the cumulative
distribution function.
The uniform order statistic medians are defined as:
m(i) = 1 - m(n) for i = 1
m(i) = (i - 0.3175)/(n + 0.365) for i = 2, 3, ..., n-1
m(i) = 0.5(1/n) for i = n
In addition, a straight line can be fit to the points and added as a
reference line. The further the points vary from this line, the greater
the indication of departures from normality.
Probability plots for distributions other than the normal are computed
in exactly the same way. The normal percent point function (the G) is
simply replaced by the percent point function of the desired
distribution. That is, a probability plot can easily be generated for any
distribution for which you have the percent point function.
One advantage of this method of computing probability plots is that
the intercept and slope estimates of the fitted line are in fact estimates
for the location and scale parameters of the distribution. Although this
is not too important for the normal distribution since the location and
scale are estimated by the mean and standard deviation, respectively, it
can be useful for many other distributions.
The correlation coefficient of the points on the normal probability plot
can be compared to a table of critical values to provide a formal test of
the hypothesis that the data come from a normal distribution.

Questions The normal probability plot is used to answer the following questions.
1. Are the data normally distributed?
2. What is the nature of the departure from normality (data
skewed, shorter than expected tails, longer than expected tails)?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l.htm (2 of 3) [7/1/2003 5:21:33 PM]

1.3.3.21. Normal Probability Plot

Importance: The underlying assumptions for a measurement process are that the
Check data should behave like:
Normality 1. random drawings;
Assumption
2. from a fixed distribution;
3. with fixed location;
4. with fixed scale.
Probability plots are used to assess the assumption of a fixed
distribution. In particular, most statistical models are of the form:
response = deterministic + random
where the deterministic part is the fit and the random part is error. This
error component in most common statistical models is specifically
assumed to be normally distributed with fixed location and scale. This
is the most frequent application of normal probability plots. That is, a
model is fit and a normal probability plot is generated for the residuals
from the fitted model. If the residuals from the fitted model are not
normally distributed, then one of the major assumptions of the model
has been violated.

Examples 1. Data are normally distributed

2. Data have fat tails
3. Data have short tails
4. Data are skewed right

Related Histogram
Techniques Probability plots for other distributions (e.g., Weibull)
Probability plot correlation coefficient plot (PPCC plot)
Anderson-Darling Goodness-of-Fit Test
Chi-Square Goodness-of-Fit Test
Kolmogorov-Smirnov Goodness-of-Fit Test

Case Study The normal probability plot is demonstrated in the heat flow meter
data case study.

Software Most general purpose statistical software programs can generate a

normal probability plot. Dataplot supports a normal probability plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l.htm (3 of 3) [7/1/2003 5:21:33 PM]

1.3.3.21.1. Normal Probability Plot: Normally Distributed Data

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.21. Normal Probability Plot

1.3.3.21.1. Normal Probability Plot:

Normally Distributed Data
Normal The following normal probability plot is from the heat flow meter data.
Probability
Plot

Conclusions We can make the following conclusions from the above plot.
1. The normal probability plot shows a strongly linear pattern. There
are only minor deviations from the line fit to the points on the
probability plot.
2. The normal distribution appears to be a good model for these
data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l1.htm (1 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.1. Normal Probability Plot: Normally Distributed Data

Discussion Visually, the probability plot shows a strongly linear pattern. This is
verified by the correlation coefficient of 0.9989 of the line fit to the
probability plot. The fact that the points in the lower and upper extremes
of the plot do not deviate significantly from the straight-line pattern
indicates that there are not any significant outliers (relative to a normal
distribution).
In this case, we can quite reasonably conclude that the normal
distribution provides an excellent model for the data. The intercept and
slope of the fitted line give estimates of 9.26 and 0.023 for the location
and scale parameters of the fitted normal distribution.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l1.htm (2 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.2. Normal Probability Plot: Data Have Short Tails

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.21. Normal Probability Plot

1.3.3.21.2. Normal Probability Plot: Data

Have Short Tails
Normal The following is a normal probability plot for 500 random numbers
Probability generated from a Tukey-Lambda distribution with the parameter equal
Plot for to 1.1.
Data with
Short Tails

Conclusions We can make the following conclusions from the above plot.
1. The normal probability plot shows a non-linear pattern.
2. The normal distribution is not a good model for these data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l2.htm (1 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.2. Normal Probability Plot: Data Have Short Tails

Discussion For data with short tails relative to the normal distribution, the
non-linearity of the normal probability plot shows up in two ways. First,
the middle of the data shows an S-like pattern. This is common for both
short and long tails. Second, the first few and the last few points show a
marked departure from the reference fitted line. In comparing this plot
to the long tail example in the next section, the important difference is
the direction of the departure from the fitted line for the first few and
last few points. For short tails, the first few points show increasing
departure from the fitted line above the line and last few points show
increasing departure from the fitted line below the line. For long tails,
this pattern is reversed.
In this case, we can reasonably conclude that the normal distribution
does not provide an adequate fit for this data set. For probability plots
that indicate short-tailed distributions, the next step might be to generate
a Tukey Lambda PPCC plot. The Tukey Lambda PPCC plot can often
be helpful in identifying an appropriate distributional family.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l2.htm (2 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.3. Normal Probability Plot: Data Have Long Tails

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.21. Normal Probability Plot

1.3.3.21.3. Normal Probability Plot: Data

Have Long Tails
Normal The following is a normal probability plot of 500 numbers generated
Probability from a double exponential distribution. The double exponential
Plot for distribution is symmetric, but relative to the normal it declines rapidly
Data with and has longer tails.
Long Tails

Conclusions We can make the following conclusions from the above plot.
1. The normal probability plot shows a reasonably linear pattern in
the center of the data. However, the tails, particularly the lower
tail, show departures from the fitted line.
2. A distribution other than the normal distribution would be a good
model for these data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l3.htm (1 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.3. Normal Probability Plot: Data Have Long Tails

Discussion For data with long tails relative to the normal distribution, the
non-linearity of the normal probability plot can show up in two ways.
First, the middle of the data may show an S-like pattern. This is
common for both short and long tails. In this particular case, the S
pattern in the middle is fairly mild. Second, the first few and the last few
points show marked departure from the reference fitted line. In the plot
above, this is most noticeable for the first few data points. In comparing
this plot to the short-tail example in the previous section, the important
difference is the direction of the departure from the fitted line for the
first few and the last few points. For long tails, the first few points show
increasing departure from the fitted line below the line and last few
points show increasing departure from the fitted line above the line. For
short tails, this pattern is reversed.
In this case we can reasonably conclude that the normal distribution can
be improved upon as a model for these data. For probability plots that
indicate long-tailed distributions, the next step might be to generate a
Tukey Lambda PPCC plot. The Tukey Lambda PPCC plot can often be
helpful in identifying an appropriate distributional family.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l3.htm (2 of 2) [7/1/2003 5:21:33 PM]

1.3.3.21.4. Normal Probability Plot: Data are Skewed Right

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.21. Normal Probability Plot

1.3.3.21.4. Normal Probability Plot: Data are

Skewed Right
Normal
Probability
Plot for
Data that
are Skewed
Right

Conclusions We can make the following conclusions from the above plot.
1. The normal probability plot shows a strongly non-linear pattern.
Specifically, it shows a quadratic pattern in which all the points
are below a reference line drawn between the first and last points.
2. The normal distribution is not a good model for these data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l4.htm (1 of 2) [7/1/2003 5:21:34 PM]

1.3.3.21.4. Normal Probability Plot: Data are Skewed Right

Discussion This quadratic pattern in the normal probability plot is the signature of a
significantly right-skewed data set. Similarly, if all the points on the
normal probability plot fell above the reference line connecting the first
and last points, that would be the signature pattern for a significantly
left-skewed data set.
In this case we can quite reasonably conclude that we need to model
these data with a right skewed distribution such as the Weibull or
lognormal.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33l4.htm (2 of 2) [7/1/2003 5:21:34 PM]

1.3.3.22. Probability Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.22. Probability Plot

Purpose: The probability plot (Chambers 1983) is a graphical technique for
Check If assessing whether or not a data set follows a given distribution such as
Data Follow the normal or Weibull.
a Given
Distribution The data are plotted against a theoretical distribution in such a way that
the points should form approximately a straight line. Departures from
this straight line indicate departures from the specified distribution.
The correlation coefficient associated with the linear fit to the data in
the probability plot is a measure of the goodness of the fit. Estimates of
the location and scale parameters of the distribution are given by the
intercept and slope. Probability plots can be generated for several
competing distributions to see which provides the best fit, and the
probability plot generating the highest correlation coefficient is the best
choice since it generates the straightest probability plot.
For distributions with shape parameters (not counting location and
scale parameters), the shape parameters must be known in order to
generate the probability plot. For distributions with a single shape
parameter, the probability plot correlation coefficient (PPCC) plot
provides an excellent method for estimating the shape parameter.
We cover the special case of the normal probability plot separately due
to its importance in many statistical applications.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33m.htm (1 of 4) [7/1/2003 5:21:34 PM]

1.3.3.22. Probability Plot

Sample Plot

This data is a set of 500 Weibull random numbers with a shape

parameter = 2, location parameter = 0, and scale parameter = 1. The
Weibull probability plot indicates that the Weibull distribution does in
fact fit these data well.

Definition: The probability plot is formed by:

Ordered ● Vertical axis: Ordered response values
Response
● Horizontal axis: Order statistic medians for the given distribution
Values
Versus Order The order statistic medians are defined as:
Statistic N(i) = G(U(i))
Medians for
the Given where the U(i) are the uniform order statistic medians (defined below)
Distribution and G is the percent point function for the desired distribution. The
percent point function is the inverse of the cumulative distribution
function (probability that x is less than or equal to some value). That is,
given a probability, we want the corresponding x of the cumulative
distribution function.
The uniform order statistic medians are defined as:
m(i) = 1 - m(n) for i = 1
m(i) = (i - 0.3175)/(n + 0.365) for i = 2, 3, ..., n-1
m(i) = 0.5**(1/n) for i = n
In addition, a straight line can be fit to the points and added as a
reference line. The further the points vary from this line, the greater the

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33m.htm (2 of 4) [7/1/2003 5:21:34 PM]

1.3.3.22. Probability Plot

indication of a departure from the specified distribution.

This definition implies that a probability plot can be easily generated
for any distribution for which the percent point function can be
computed.
One advantage of this method of computing proability plots is that the
intercept and slope estimates of the fitted line are in fact estimates for
the location and scale parameters of the distribution. Although this is
not too important for the normal distribution (the location and scale are
estimated by the mean and standard deviation, respectively), it can be
useful for many other distributions.

Questions The probability plot is used to answer the following questions:

● Does a given distribution, such as the Weibull, provide a good fit
to my data?
● What distribution best fits my data?

● What are good estimates for the location and scale parameters of
the chosen distribution?

Importance: The discussion for the normal probability plot covers the use of
Check probability plots for checking the fixed distribution assumption.
distributional
assumption Some statistical models assume data have come from a population with
a specific type of distribution. For example, in reliability applications,
the Weibull, lognormal, and exponential are commonly used
distributional models. Probability plots can be useful for checking this
distributional assumption.

Related Histogram
Techniques Probability Plot Correlation Coefficient (PPCC) Plot
Hazard Plot
Quantile-Quantile Plot
Anderson-Darling Goodness of Fit
Chi-Square Goodness of Fit
Kolmogorov-Smirnov Goodness of Fit

Case Study The probability plot is demonstrated in the airplane glass failure time
data case study.

Software Most general purpose statistical software programs support probability

plots for at least a few common distributions. Dataplot supports
probability plots for a large number of distributions.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33m.htm (3 of 4) [7/1/2003 5:21:34 PM]

1.3.3.22. Probability Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33m.htm (4 of 4) [7/1/2003 5:21:34 PM]

1.3.3.23. Probability Plot Correlation Coefficient Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.23. Probability Plot Correlation

Coefficient Plot
Purpose: The probability plot correlation coefficient (PPCC) plot (Filliben
Graphical 1975) is a graphical technique for identifying the shape parameter for
Technique for a distributional family that best describes the data set. This technique
Finding the is appropriate for families, such as the Weibull, that are defined by a
Shape single shape parameter and location and scale parameters, and it is not
Parameter of appropriate for distributions, such as the normal, that are defined only
a by location and scale parameters.
Distributional
Family that The PPCC plot is generated as follows. For a series of values for the
Best Fits a shape parameter, the correlation coefficient is computed for the
Data Set probability plot associated with a given value of the shape parameter.
These correlation coefficients are plotted against their corresponding
shape parameters. The maximum correlation coefficient corresponds
to the optimal value of the shape parameter. For better precision, two
iterations of the PPCC plot can be generated; the first is for finding
the right neighborhood and the second is for fine tuning the estimate.
The PPCC plot is used first to find a good value of the shape
parameter. The probability plot is then generated to find estimates of
the location and scale parameters and in addition to provide a
graphical assessment of the adequacy of the distributional fit.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33n.htm (1 of 4) [7/1/2003 5:21:34 PM]

1.3.3.23. Probability Plot Correlation Coefficient Plot

Compare In addition to finding a good choice for estimating the shape

Distributions parameter of a given distribution, the PPCC plot can be useful in
deciding which distributional family is most appropriate. For example,
given a set of reliabilty data, you might generate PPCC plots for a
Weibull, lognormal, gamma, and inverse Gaussian distributions, and
possibly others, on a single page. This one page would show the best
value for the shape parameter for several distributions and would
additionally indicate which of these distributional families provides
the best fit (as measured by the maximum probability plot correlation
coefficient). That is, if the maximum PPCC value for the Weibull is
0.99 and only 0.94 for the lognormal, then we could reasonably
conclude that the Weibull family is the better choice.

Tukey-Lambda The Tukey Lambda PPCC plot, with shape parameter , is

PPCC Plot for particularly useful for symmetric distributions. It indicates whether a
Symmetric distribution is short or long tailed and it can further indicate several
Distributions common distributions. Specifically,
1. = -1: distribution is approximately Cauchy
2. = 0: distribution is exactly logistic
3. = 0.14: distribution is approximately normal
4. = 0.5: distribution is U-shaped
5. = 1: distribution is exactly uniform
If the Tukey Lambda PPCC plot gives a maximum value near 0.14,
we can reasonably conclude that the normal distribution is a good
model for the data. If the maximum value is less than 0.14, a
long-tailed distribution such as the double exponential or logistic
would be a better choice. If the maximum value is near -1, this implies
the selection of very long-tailed distribution, such as the Cauchy. If
the maximum value is greater than 0.14, this implies a short-tailed
distribution such as the Beta or uniform.
The Tukey-Lambda PPCC plot is used to suggest an appropriate
distribution. You should follow-up with PPCC and probability plots of
the appropriate alternatives.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33n.htm (2 of 4) [7/1/2003 5:21:34 PM]

1.3.3.23. Probability Plot Correlation Coefficient Plot

Use When comparing distributional models, do not simply choose the one
Judgement with the maximum PPCC value. In many cases, several distributional
When fits provide comparable PPCC values. For example, a lognormal and
Selecting An Weibull may both fit a given set of reliability data quite well.
Appropriate Typically, we would consider the complexity of the distribution. That
Distributional is, a simpler distribution with a marginally smaller PPCC value may
Family be preferred over a more complex distribution. Likewise, there may be
theoretical justification in terms of the underlying scientific model for
preferring a distribution with a marginally smaller PPCC value in
some cases. In other cases, we may not need to know if the
distributional model is optimal, only that it is adequate for our
purposes. That is, we may be able to use techniques designed for
normally distributed data even if other distributions fit the data
somewhat better.

Sample Plot The following is a PPCC plot of 100 normal random numbers. The
maximum value of the correlation coefficient = 0.997 at = 0.099.

This PPCC plot shows that:

1. the best-fit symmetric distribution is nearly normal;
2. the data are not long tailed;
3. the sample mean would be an appropriate estimator of location.
We can follow-up this PPCC plot with a normal probability plot to
verify the normality model for the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33n.htm (3 of 4) [7/1/2003 5:21:34 PM]

1.3.3.23. Probability Plot Correlation Coefficient Plot

Definition: The PPCC plot is formed by:

● Vertical axis: Probability plot correlation coefficient;

● Horizontal axis: Value of shape parameter.

Questions The PPCC plot answers the following questions:

1. What is the best-fit member within a distributional family?
2. Does the best-fit member provide a good fit (in terms of
generating a probability plot with a high correlation
coefficient)?
3. Does this distributional family provide a good fit compared to
other distributions?
4. How sensitive is the choice of the shape parameter?

Importance Many statistical analyses are based on distributional assumptions

about the population from which the data have been obtained.
However, distributional families can have radically different shapes
depending on the value of the shape parameter. Therefore, finding a
reasonable choice for the shape parameter is a necessary step in the
analysis. In many analyses, finding a good distributional model for the
data is the primary focus of the analysis. In both of these cases, the
PPCC plot is a valuable tool.

Related Probability Plot

Techniques Maximum Likelihood Estimation
Least Squares Estimation
Method of Moments Estimation

Case Study The PPCC plot is demonstrated in the airplane glass failure data case
study.

Software PPCC plots are currently not available in most common general
purpose statistical software programs. However, the underlying
technique is based on probability plots and correlation coefficients, so
it should be possible to write macros for PPCC plots in statistical
programs that support these capabilities. Dataplot supports PPCC
plots.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33n.htm (4 of 4) [7/1/2003 5:21:34 PM]

1.3.3.24. Quantile-Quantile Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.24. Quantile-Quantile Plot

Purpose: The quantile-quantile (q-q) plot is a graphical technique for determining
Check If if two data sets come from populations with a common distribution.
Two Data
Sets Can Be A q-q plot is a plot of the quantiles of the first data set against the
Fit With the quantiles of the second data set. By a quantile, we mean the fraction (or
Same percent) of points below the given value. That is, the 0.3 (or 30%)
Distribution quantile is the point at which 30% percent of the data fall below and
70% fall above that value.
A 45-degree reference line is also plotted. If the two sets come from a
population with the same distribution, the points should fall
approximately along this reference line. The greater the departure from
this reference line, the greater the evidence for the conclusion that the
two data sets have come from populations with different distributions.
The advantages of the q-q plot are:
1. The sample sizes do not need to be equal.
2. Many distributional aspects can be simultaneously tested. For
example, shifts in location, shifts in scale, changes in symmetry,
and the presence of outliers can all be detected from this plot. For
example, if the two data sets come from populations whose
distributions differ only by a shift in location, the points should lie
along a straight line that is displaced either up or down from the
45-degree reference line.
The q-q plot is similar to a probability plot. For a probability plot, the
quantiles for one of the data samples are replaced with the quantiles of a
theoretical distribution.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33o.htm (1 of 3) [7/1/2003 5:21:35 PM]

1.3.3.24. Quantile-Quantile Plot

Sample Plot

This q-q plot shows that

1. These 2 batches do not appear to have come from populations
with a common distribution.
2. The batch 1 values are significantly higher than the corresponding
batch 2 values.
3. The differences are increasing from values 525 to 625. Then the
values for the 2 batches get closer again.

Definition: The q-q plot is formed by:

Quantiles ● Vertical axis: Estimated quantiles from data set 1
for Data Set
● Horizontal axis: Estimated quantiles from data set 2
1 Versus
Quantiles of Both axes are in units of their respective data sets. That is, the actual
Data Set 2 quantile level is not plotted. For a given point on the q-q plot, we know
that the quantile level is the same for both points, but not what that
quantile level actually is.
If the data sets have the same size, the q-q plot is essentially a plot of
sorted data set 1 against sorted data set 2. If the data sets are not of equal
size, the quantiles are usually picked to correspond to the sorted values
from the smaller data set and then the quantiles for the larger data set are
interpolated.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33o.htm (2 of 3) [7/1/2003 5:21:35 PM]

1.3.3.24. Quantile-Quantile Plot

Questions The q-q plot is used to answer the following questions:

● Do two data sets come from populations with a common
distribution?
● Do two data sets have common location and scale?

● Do two data sets have similar distributional shapes?

● Do two data sets have similar tail behavior?

Importance: When there are two data samples, it is often desirable to know if the
Check for assumption of a common distribution is justified. If so, then location and
Common scale estimators can pool both data sets to obtain estimates of the
Distribution common location and scale. If two samples do differ, it is also useful to
gain some understanding of the differences. The q-q plot can provide
more insight into the nature of the difference than analytical methods
such as the chi-square and Kolmogorov-Smirnov 2-sample tests.

Related Bihistogram
Techniques T Test
F Test
2-Sample Chi-Square Test
2-Sample Kolmogorov-Smirnov Test

Case Study The quantile-quantile plot is demonstrated in the ceramic strength data
case study.

Software Q-Q plots are available in some general purpose statistical software
programs, including Dataplot. If the number of data points in the two
samples are equal, it should be relatively easy to write a macro in
statistical programs that do not support the q-q plot. If the number of
points are not equal, writing a macro for a q-q plot may be difficult.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33o.htm (3 of 3) [7/1/2003 5:21:35 PM]

1.3.3.25. Run-Sequence Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.25. Run-Sequence Plot

Purpose: Run sequence plots (Chambers 1983) are an easy way to graphically
Check for summarize a univariate data set. A common assumption of univariate
Shifts in data sets is that they behave like:
Location 1. random drawings;
and Scale
and Outliers 2. from a fixed distribution;
3. with a common location; and
4. with a common scale.
With run sequence plots, shifts in location and scale are typically quite
evident. Also, outliers can easily be detected.

Sample
Plot:
Last Third
of Data
Shows a
Shift of
Location

This sample run sequence plot shows that the location shifts up for the
last third of the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33p.htm (1 of 2) [7/1/2003 5:21:35 PM]

1.3.3.25. Run-Sequence Plot

Definition: Run sequence plots are formed by:

y(i) Versus i ● Vertical axis: Response variable Y(i)

● Horizontal axis: Index i (i = 1, 2, 3, ... )

Questions The run sequence plot can be used to answer the following questions
1. Are there any shifts in location?
2. Are there any shifts in variation?
3. Are there any outliers?
The run sequence plot can also give the analyst an excellent feel for the
data.

Importance: For univariate data, the default model is

Check Y = constant + error
Univariate
where the error is assumed to be random, from a fixed distribution, and
Assumptions
with constant location and scale. The validity of this model depends on
the validity of these assumptions. The run sequence plot is useful for
checking for constant location and scale.
Even for more complex models, the assumptions on the error term are
still often the same. That is, a run sequence plot of the residuals (even
from very complex models) is still vital for checking for outliers and for
detecting shifts in location and scale.

Related Scatter Plot

Techniques Histogram
Autocorrelation Plot
Lag Plot

Case Study The run sequence plot is demonstrated in the Filter transmittance data
case study.

Software Run sequence plots are available in most general purpose statistical
software programs, including Dataplot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33p.htm (2 of 2) [7/1/2003 5:21:35 PM]

1.3.3.26. Scatter Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.26. Scatter Plot

Purpose: A scatter plot (Chambers 1983) reveals relationships or association
Check for between two variables. Such relationships manifest themselves by any
Relationship non-random structure in the plot. Various common types of patterns are
demonstrated in the examples.

Sample
Plot:
Linear
Relationship
Between
Variables Y
and X

This sample plot reveals a linear relationship between the two variables
indicating that a linear regression model might be appropriate.

Definition: A scatter plot is a plot of the values of Y versus the corresponding

Y Versus X values of X:
● Vertical axis: variable Y--usually the response variable

● Horizontal axis: variable X--usually some variable we suspect

may ber related to the response

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q.htm (1 of 3) [7/1/2003 5:21:35 PM]

1.3.3.26. Scatter Plot

Questions Scatter plots can provide answers to the following questions:

1. Are variables X and Y related?
2. Are variables X and Y linearly related?
3. Are variables X and Y non-linearly related?
4. Does the variation in Y change depending on X?
5. Are there outliers?

Examples 1. No relationship
2. Strong linear (positive correlation)
3. Strong linear (negative correlation)
4. Exact linear (positive correlation)
5. Quadratic relationship
6. Exponential relationship
7. Sinusoidal relationship (damped)
8. Variation of Y doesn't depend on X (homoscedastic)
9. Variation of Y does depend on X (heteroscedastic)
10. Outlier

Combining Scatter plots can also be combined in multiple plots per page to help
Scatter Plots understand higher-level structure in data sets with more than two
variables.
The scatterplot matrix generates all pairwise scatter plots on a single
page. The conditioning plot, also called a co-plot or subset plot,
generates scatter plots of Y versus X dependent on the value of a third
variable.

Causality Is The scatter plot uncovers relationships in data. "Relationships" means

Not Proved that there is some structured association (linear, quadratic, etc.) between
By X and Y. Note, however, that even though
Association
causality implies association
association does NOT imply causality.
Scatter plots are a useful diagnostic tool for determining association, but
if such association exists, the plot may or may not suggest an underlying
cause-and-effect mechanism. A scatter plot can never "prove" cause and
effect--it is ultimately only the researcher (relying on the underlying
science/engineering) who can conclude that causality actually exists.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q.htm (2 of 3) [7/1/2003 5:21:35 PM]

1.3.3.26. Scatter Plot

Appearance The most popular rendition of a scatter plot is

1. some plot character (e.g., X) at the data points, and
2. no line connecting data points.
Other scatter plot format variants include
1. an optional plot character (e.g, X) at the data points, but
2. a solid line connecting data points.
In both cases, the resulting plot is referred to as a scatter plot, although
the former (discrete and disconnected) is the author's personal
preference since nothing makes it onto the screen except the data--there
are no interpolative artifacts to bias the interpretation.

Related Run Sequence Plot

Techniques Box Plot
Block Plot

Case Study The scatter plot is demonstrated in the load cell calibration data case
study.

Software Scatter plots are a fundamental technique that should be available in any
general purpose statistical software program, including Dataplot. Scatter
plots are also available in most graphics and spreadsheet programs as
well.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q.htm (3 of 3) [7/1/2003 5:21:35 PM]

1.3.3.26.1. Scatter Plot: No Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.1. Scatter Plot: No Relationship

Scatter Plot
with No
Relationship

Discussion Note in the plot above how for a given value of X (say X = 0.5), the
corresponding values of Y range all over the place from Y = -2 to Y = +2.
The same is true for other values of X. This lack of predictablility in
determining Y from a given value of X, and the associated amorphous,
non-structured appearance of the scatter plot leads to the summary
conclusion: no relationship.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q1.htm [7/1/2003 5:21:36 PM]

1.3.3.26.2. Scatter Plot: Strong Linear (positive correlation) Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.2. Scatter Plot: Strong Linear

(positive correlation)
Relationship
Scatter Plot
Showing
Strong
Positive
Linear
Correlation

Discussion Note in the plot above how a straight line comfortably fits through the
data; hence a linear relationship exists. The scatter about the line is quite
small, so there is a strong linear relationship. The slope of the line is
positive (small values of X correspond to small values of Y; large values
of X correspond to large values of Y), so there is a positive co-relation
(that is, a positive correlation) between X and Y.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q2.htm [7/1/2003 5:21:36 PM]

1.3.3.26.3. Scatter Plot: Strong Linear (negative correlation) Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.3. Scatter Plot: Strong Linear

(negative correlation)
Relationship
Scatter Plot
Showing a
Strong
Negative
Correlation

Discussion Note in the plot above how a straight line comfortably fits through the
data; hence there is a linear relationship. The scatter about the line is
quite small, so there is a strong linear relationship. The slope of the line
is negative (small values of X correspond to large values of Y; large
values of X correspond to small values of Y), so there is a negative
co-relation (that is, a negative correlation) between X and Y.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q3.htm [7/1/2003 5:21:36 PM]

1.3.3.26.4. Scatter Plot: Exact Linear (positive correlation) Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.4. Scatter Plot: Exact Linear

(positive correlation)
Relationship
Scatter Plot
Showing an
Exact
Linear
Relationship

Discussion Note in the plot above how a straight line comfortably fits through the
data; hence there is a linear relationship. The scatter about the line is
zero--there is perfect predictability between X and Y), so there is an
exact linear relationship. The slope of the line is positive (small values
of X correspond to small values of Y; large values of X correspond to
large values of Y), so there is a positive co-relation (that is, a positive
correlation) between X and Y.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q4.htm (1 of 2) [7/1/2003 5:21:36 PM]

1.3.3.26.4. Scatter Plot: Exact Linear (positive correlation) Relationship

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q4.htm (2 of 2) [7/1/2003 5:21:36 PM]

1.3.3.26.5. Scatter Plot: Quadratic Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.5. Scatter Plot: Quadratic

Relationship
Scatter Plot
Showing
Quadratic
Relationship

Discussion Note in the plot above how no imaginable simple straight line could
ever adequately describe the relationship between X and Y--a curved (or
curvilinear, or non-linear) function is needed. The simplest such
curvilinear function is a quadratic model

for some A, B, and C. Many other curvilinear functions are possible, but
the data analysis principle of parsimony suggests that we try fitting a
quadratic function first.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q5.htm (1 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.5. Scatter Plot: Quadratic Relationship

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q5.htm (2 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.6. Scatter Plot: Exponential Relationship

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.6. Scatter Plot: Exponential

Relationship
Scatter Plot
Showing
Exponential
Relationship

Discussion Note that a simple straight line is grossly inadequate in describing the
relationship between X and Y. A quadratic model would prove lacking,
especially for large values of X. In this example, the large values of X
correspond to nearly constant values of Y, and so a non-linear function
beyond the quadratic is needed. Among the many other non-linear
functions available, one of the simpler ones is the exponential model

for some A, B, and C. In this case, an exponential function would, in

fact, fit well, and so one is led to the summary conclusion of an
exponential relationship.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q6.htm (1 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.6. Scatter Plot: Exponential Relationship

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q6.htm (2 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.7. Scatter Plot: Sinusoidal Relationship (damped)

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.7. Scatter Plot: Sinusoidal

Relationship (damped)
Scatter Plot
Showing a
Sinusoidal
Relationship

Discussion The complex relationship between X and Y appears to be basically

oscillatory, and so one is naturally drawn to the trigonometric sinusoidal
model:

Closer inspection of the scatter plot reveals that the amount of swing
(the amplitude in the model) does not appear to be constant but rather
is decreasing (damping) as X gets large. We thus would be led to the
conclusion: damped sinusoidal relationship, with the simplest
corresponding model being

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q7.htm (1 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.7. Scatter Plot: Sinusoidal Relationship (damped)

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q7.htm (2 of 2) [7/1/2003 5:21:37 PM]

1.3.3.26.8. Scatter Plot: Variation of Y Does Not Depend on X (homoscedastic)

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.8. Scatter Plot: Variation of Y Does

Not Depend on X
(homoscedastic)
Scatter Plot
Showing
Homoscedastic
Variability

Discussion This scatter plot reveals a linear relationship between X and Y: for a
given value of X, the predicted value of Y will fall on a line. The plot
further reveals that the variation in Y about the predicted value is
about the same (+- 10 units), regardless of the value of X.
Statistically, this is referred to as homoscedasticity. Such
homoscedasticity is very important as it is an underlying assumption
for regression, and its violation leads to parameter estimates with
inflated variances. If the data are homoscedastic, then the usual
regression estimates can be used. If the data are not homoscedastic,
then the estimates can be improved using weighting procedures as
shown in the next example.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q8.htm (1 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.8. Scatter Plot: Variation of Y Does Not Depend on X (homoscedastic)

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q8.htm (2 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.9. Scatter Plot: Variation of Y Does Depend on X (heteroscedastic)

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.9. Scatter Plot: Variation of Y Does

Depend on X (heteroscedastic)
Scatter Plot
Showing
Heteroscedastic
Variability

Discussion This scatter plot reveals an approximate linear relationship between

X and Y, but more importantly, it reveals a statistical condition
referred to as heteroscedasticity (that is, nonconstant variation in Y
over the values of X). For a heteroscedastic data set, the variation in
Y differs depending on the value of X. In this example, small values
of X yield small scatter in Y while large values of X result in large
scatter in Y.
Heteroscedasticity complicates the analysis somewhat, but its effects
can be overcome by:
1. proper weighting of the data with noisier data being weighted
less, or by

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q9.htm (1 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.9. Scatter Plot: Variation of Y Does Depend on X (heteroscedastic)

2. performing a Y variable transformation to achieve

homoscedasticity. The Box-Cox normality plot can help
determine a suitable transformation.

Impact of Fortunately, unweighted regression analyses on heteroscedastic data

Ignoring produce estimates of the coefficients that are unbiased. However, the
Unequal coefficients will not be as precise as they would be with proper
Variability in weighting.
the Data
Note further that if heteroscedasticity does exist, it is frequently
useful to plot and model the local variation as a
function of X, as in . This modeling has
two advantages:
1. it provides additional insight and understanding as to how the
response Y relates to X; and
2. it provides a convenient means of forming weights for a
weighted regression by simply using

The topic of non-constant variation is discussed in some detail in the

process modeling chapter.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33q9.htm (2 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.10. Scatter Plot: Outlier

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.10. Scatter Plot: Outlier

Scatter Plot
Showing
Outliers

Discussion The scatter plot here reveals

1. a basic linear relationship between X and Y for most of the data,
and
2. a single outlier (at X = 375).
An outlier is defined as a data point that emanates from a different
model than do the rest of the data. The data here appear to come from a
linear model with a given slope and variation except for the outlier
which appears to have been generated from some other model.
Outlier detection is important for effective modeling. Outliers should be
excluded from such model fitting. If all the data here are included in a
linear regression, then the fitted model will be poor virtually
everywhere. If the outlier is omitted from the fitting process, then the
resulting fit will be excellent almost everywhere (for all points except

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qa.htm (1 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.10. Scatter Plot: Outlier

the outlying point).

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qa.htm (2 of 2) [7/1/2003 5:21:44 PM]

1.3.3.26.11. Scatterplot Matrix

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.11. Scatterplot Matrix

Purpose: Given a set of variables X1, X2, ... , Xk, the scatterplot matrix contains
Check all the pairwise scatter plots of the variables on a single page in a
Pairwise matrix format. That is, if there are k variables, the scatterplot matrix
Relationships will have k rows and k columns and the ith row and jth column of this
Between matrix is a plot of Xi versus Xj.
Variables
Although the basic concept of the scatterplot matrix is simple, there are
numerous alternatives in the details of the plots.
1. The diagonal plot is simply a 45-degree line since we are plotting
Xi versus Xi. Although this has some usefulness in terms of
showing the univariate distribution of the variable, other
alternatives are common. Some users prefer to use the diagonal
to print the variable label. Another alternative is to plot the
univariate histogram on the diagonal. Alternatively, we could
simply leave the diagonal blank.
2. Since Xi versus Xj is equivalent to Xj versus Xi with the axes
reversed, some prefer to omit the plots below the diagonal.
3. It can be helpful to overlay some type of fitted curve on the
scatter plot. Although a linear or quadratic fit can be used, the
most common alternative is to overlay a lowess curve.
4. Due to the potentially large number of plots, it can be somewhat
tricky to provide the axes labels in a way that is both informative
and visually pleasing. One alternative that seems to work well is
to provide axis labels on alternating rows and columns. That is,
row one will have tic marks and axis labels on the left vertical
axis for the first plot only while row two will have the tic marks
and axis labels for the right vertical axis for the last plot in the
row only. This alternating pattern continues for the remaining
rows. A similar pattern is used for the columns and the horizontal
axes labels. Another alternative is to put the minimum and
maximum scale value in the diagonal plot with the variable

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qb.htm (1 of 3) [7/1/2003 5:21:45 PM]

1.3.3.26.11. Scatterplot Matrix

name.
5. Some analysts prefer to connect the scatter plots. Others prefer to
leave a little gap between each plot.
6. Although this plot type is most commonly used for scatter plots,
the basic concept is both simple and powerful and extends easily
to other plot formats that involve pairwise plots such as the
quantile-quantile plot and the bihistogram.

Sample Plot

This sample plot was generated from pollution data collected by NIST
chemist Lloyd Currie.
There are a number of ways to view this plot. If we are primarily
interested in a particular variable, we can scan the row and column for
that variable. If we are interested in finding the strongest relationship,
we can scan all the plots and then determine which variables are
related.

Definition Given k variables, scatter plot matrices are formed by creating k rows
and k columns. Each row and column defines a single scatter plot
The individual plot for row i and column j is defined as
● Vertical axis: Variable Xi

● Horizontal axis: Variable Xj

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qb.htm (2 of 3) [7/1/2003 5:21:45 PM]

1.3.3.26.11. Scatterplot Matrix

Questions The scatterplot matrix can provide answers to the following questions:
1. Are there pairwise relationships between the variables?
2. If there are relationships, what is the nature of these
relationships?
3. Are there outliers in the data?
4. Is there clustering by groups in the data?

Linking and The scatterplot matrix serves as the foundation for the concepts of
Brushing linking and brushing.
By linking, we mean showing how a point, or set of points, behaves in
each of the plots. This is accomplished by highlighting these points in
some fashion. For example, the highlighted points could be drawn as a
filled circle while the remaining points could be drawn as unfilled
circles. A typical application of this would be to show how an outlier
shows up in each of the individual pairwise plots. Brushing extends this
concept a bit further. In brushing, the points to be highlighted are
interactively selected by a mouse and the scatterplot matrix is
dynamically updated (ideally in real time). That is, we can select a
rectangular region of points in one plot and see how those points are
reflected in the other plots. Brushing is discussed in detail by Becker,
Cleveland, and Wilks in the paper "Dynamic Graphics for Data
Analysis" (Cleveland and McGill, 1988).

Related Star plot

Techniques Scatter plot
Conditioning plot
Locally weighted least squares

Software Scatterplot matrices are becoming increasingly common in general

purpose statistical software programs, including Dataplot. If a software
program does not generate scatterplot matrices, but it does provide
multiple plots per page and scatter plots, it should be possible to write a
macro to generate a scatterplot matrix. Brushing is available in a few of
the general purpose statistical software programs that emphasize
graphical approaches.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qb.htm (3 of 3) [7/1/2003 5:21:45 PM]

1.3.3.26.12. Conditioning Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.26. Scatter Plot

1.3.3.26.12. Conditioning Plot

Purpose: A conditioning plot, also known as a coplot or subset plot, is a plot of
Check two variables conditional on the value of a third variable (called the
pairwise conditioning variable). The conditioning variable may be either a
relationship variable that takes on only a few discrete values or a continuous variable
between two that is divided into a limited number of subsets.
variables
conditional One limitation of the scatterplot matrix is that it cannot show interaction
on a third effects with another variable. This is the strength of the conditioning
variable plot. It is also useful for displaying scatter plots for groups in the data.
Although these groups can also be plotted on a single plot with different
plot symbols, it can often be visually easier to distinguish the groups
using the conditioning plot.
Although the basic concept of the conditioning plot matrix is simple,
there are numerous alternatives in the details of the plots.
1. It can be helpful to overlay some type of fitted curve on the
scatter plot. Although a linear or quadratic fit can be used, the
most common alternative is to overlay a lowess curve.
2. Due to the potentially large number of plots, it can be somewhat
tricky to provide the axis labels in a way that is both informative
and visually pleasing. One alternative that seems to work well is
to provide axis labels on alternating rows and columns. That is,
row one will have tic marks and axis labels on the left vertical
axis for the first plot only while row two will have the tic marks
and axis labels for the right vertical axis for the last plot in the
row only. This alternating pattern continues for the remaining
rows. A similar pattern is used for the columns and the horizontal
axis labels. Note that this approach only works if the axes limits
are fixed to common values for all of the plots.
3. Some analysts prefer to connect the scatter plots. Others prefer to
leave a little gap between each plot. Alternatively, each plot can
have its own labeling with the plots not connected.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qc.htm (1 of 3) [7/1/2003 5:21:45 PM]

1.3.3.26.12. Conditioning Plot

4. Although this plot type is most commonly used for scatter plots,
the basic concept is both simple and powerful and extends easily
to other plot formats.

Sample Plot

In this case, temperature has six distinct values. We plot torque versus
time for each of these temperatures. This example is discussed in more
detail in the process modeling chapter.

Definition Given the variables X, Y, and Z, the conditioning plot is formed by

dividing the values of Z into k groups. There are several ways that these
groups may be formed. There may be a natural grouping of the data, the
data may be divided into several equal sized groups, the grouping may
be determined by clusters in the data, and so on. The page will be
divided into n rows and c columns where . Each row and
column defines a single scatter plot.
The individual plot for row i and column j is defined as
● Vertical axis: Variable Y

● Horizontal axis: Variable X

where only the points in the group corresponding to the ith row and jth
column are used.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qc.htm (2 of 3) [7/1/2003 5:21:45 PM]

1.3.3.26.12. Conditioning Plot

Questions The conditioning plot can provide answers to the following questions:
1. Is there a relationship between two variables?
2. If there is a relationship, does the nature of the relationship
depend on the value of a third variable?
3. Are groups in the data similar?
4. Are there outliers in the data?

Related Scatter plot

Techniques Scatterplot matrix
Locally weighted least squares

Software Scatter plot matrices are becoming increasingly common in general

purpose statistical software programs, including Dataplot. If a software
program does not generate conditioning plots, but it does provide
multiple plots per page and scatter plots, it should be possible to write a
macro to generate a conditioning plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33qc.htm (3 of 3) [7/1/2003 5:21:45 PM]

1.3.3.27. Spectral Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.27. Spectral Plot

Purpose: A spectral plot ( Jenkins and Watts 1968 or Bloomfield 1976) is a
Examine graphical technique for examining cyclic structure in the frequency
Cyclic domain. It is a smoothed Fourier transform of the autocovariance
Structure function.
The frequency is measured in cycles per unit time where unit time is
defined to be the distance between 2 points. A frequency of 0
corresponds to an infinite cycle while a frequency of 0.5 corresponds to
a cycle of 2 data points. Equi-spaced time series are inherently limited to
detecting frequencies between 0 and 0.5.
Trends should typically be removed from the time series before
applying the spectral plot. Trends can be detected from a run sequence
plot. Trends are typically removed by differencing the series or by
fitting a straight line (or some other polynomial curve) and applying the
spectral analysis to the residuals.
Spectral plots are often used to find a starting value for the frequency,
, in the sinusoidal model

See the beam deflection case study for an example of this.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r.htm (1 of 3) [7/1/2003 5:21:45 PM]

1.3.3.27. Spectral Plot

Sample Plot

This spectral plot shows one dominant frequency of approximately 0.3

cycles per observation.

Definition: The spectral plot is formed by:

Variance ● Vertical axis: Smoothed variance (power)
Versus
● Horizontal axis: Frequency (cycles per observation)
Frequency
The computations for generating the smoothed variances can be
involved and are not discussed further here. The details can be found in
the Jenkins and Bloomfield references and in most texts that discuss the
frequency analysis of time series.

Questions The spectral plot can be used to answer the following questions:
1. How many cyclic components are there?
2. Is there a dominant cyclic frequency?
3. If there is a dominant cyclic frequency, what is it?

Importance The spectral plot is the primary technique for assessing the cyclic nature
Check of univariate time series in the frequency domain. It is almost always the
Cyclic second plot (after a run sequence plot) generated in a frequency domain
Behavior of analysis of a time series.
Time Series

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r.htm (2 of 3) [7/1/2003 5:21:45 PM]

1.3.3.27. Spectral Plot

Examples 1. Random (= White Noise)

2. Strong autocorrelation and autoregressive model
3. Sinusoidal model

Related Autocorrelation Plot

Techniques Complex Demodulation Amplitude Plot
Complex Demodulation Phase Plot

Case Study The spectral plot is demonstrated in the beam deflection data case study.

Software Spectral plots are a fundamental technique in the frequency analysis of

time series. They are available in many general purpose statistical
software programs, including Dataplot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r.htm (3 of 3) [7/1/2003 5:21:45 PM]

1.3.3.27.1. Spectral Plot: Random Data

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.27. Spectral Plot

1.3.3.27.1. Spectral Plot: Random Data

Spectral
Plot of 200
Normal
Random
Numbers

Conclusions We can make the following conclusions from the above plot.
1. There are no dominant peaks.
2. There is no identifiable pattern in the spectrum.
3. The data are random.

Discussion For random data, the spectral plot should show no dominant peaks or
distinct pattern in the spectrum. For the sample plot above, there are no
clearly dominant peaks and the peaks seem to fluctuate at random. This
type of appearance of the spectral plot indicates that there are no
significant cyclic patterns in the data.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r1.htm (1 of 2) [7/1/2003 5:21:46 PM]

1.3.3.27.1. Spectral Plot: Random Data

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r1.htm (2 of 2) [7/1/2003 5:21:46 PM]

1.3.3.27.2. Spectral Plot: Strong Autocorrelation and Autoregressive Model

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.27. Spectral Plot

1.3.3.27.2. Spectral Plot: Strong

Autocorrelation and
Autoregressive Model
Spectral Plot
for Random
Walk Data

Conclusions We can make the following conclusions from the above plot.
1. Strong dominant peak near zero.
2. Peak decays rapidly towards zero.
3. An autoregressive model is an appropriate model.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r2.htm (1 of 2) [7/1/2003 5:21:46 PM]

1.3.3.27.2. Spectral Plot: Strong Autocorrelation and Autoregressive Model

Discussion This spectral plot starts with a dominant peak near zero and rapidly
decays to zero. This is the spectral plot signature of a process with
strong positive autocorrelation. Such processes are highly non-random
in that there is high association between an observation and a
succeeding observation. In short, if you know Yi you can make a
strong guess as to what Yi+1 will be.

Recommended The next step would be to determine the parameters for the
Next Step autoregressive model:

Such estimation can be done by linear regression or by fitting a

Box-Jenkins autoregressive (AR) model.

The residual standard deviation for this autoregressive model will be

much smaller than the residual standard deviation for the default
model

Then the system should be reexamined to find an explanation for the

strong autocorrelation. Is it due to the
1. phenomenon under study; or
2. drifting in the environment; or
3. contamination from the data acquisition system (DAS)?
Oftentimes the source of the problem is item (3) above where
contamination and carry-over from the data acquisition system result
because the DAS does not have time to electronically recover before
collecting the next data point. If this is the case, then consider slowing
down the sampling rate to re-achieve randomness.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r2.htm (2 of 2) [7/1/2003 5:21:46 PM]

1.3.3.27.3. Spectral Plot: Sinusoidal Model

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.27. Spectral Plot

1.3.3.27.3. Spectral Plot: Sinusoidal Model

Spectral Plot
for Sinusoidal
Model

Conclusions We can make the following conclusions from the above plot.
1. There is a single dominant peak at approximately 0.3.
2. There is an underlying single-cycle sinusoidal model.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r3.htm (1 of 2) [7/1/2003 5:21:46 PM]

1.3.3.27.3. Spectral Plot: Sinusoidal Model

Discussion This spectral plot shows a single dominant frequency. This indicates
that a single-cycle sinusoidal model might be appropriate.
If one were to naively assume that the data represented by the graph
could be fit by the model

and then estimate the constant by the sample mean, the analysis would
be incorrect because
● the sample mean is biased;

● the confidence interval for the mean, which is valid only for
random data, is meaningless and too small.
On the other hand, the choice of the proper model

where is the amplitude, is the frequency (between 0 and .5 cycles

per observation), and is the phase can be fit by non-linear least
squares. The beam deflection data case study demonstrates fitting this
type of model.

Recommended The recommended next steps are to:

Next Steps 1. Estimate the frequency from the spectral plot. This will be
helpful as a starting value for the subsequent non-linear fitting.
A complex demodulation phase plot can be used to fine tune the
estimate of the frequency before performing the non-linear fit.
2. Do a complex demodulation amplitude plot to obtain an initial
estimate of the amplitude and to determine if a constant
amplitude is justified.
3. Carry out a non-linear fit of the model

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33r3.htm (2 of 2) [7/1/2003 5:21:46 PM]

1.3.3.28. Standard Deviation Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.28. Standard Deviation Plot

Purpose: Standard deviation plots are used to see if the standard deviation varies
Detect between different groups of the data. The grouping is determined by the
Changes in analyst. In most cases, the data provide a specific grouping variable. For
Scale example, the groups may be the levels of a factor variable. In the sample
Between plot below, the months of the year provide the grouping.
Groups
Standard deviation plots can be used with ungrouped data to determine
if the standard deviation is changing over time. In this case, the data are
broken into an arbitrary number of equal-sized groups. For example, a
data series with 400 points can be divided into 10 groups of 40 points
each. A standard deviation plot can then be generated with these groups
to see if the standard deviation is increasing or decreasing over time.
Although the standard deviation is the most commonly used measure of
scale, the same concept applies to other measures of scale. For example,
instead of plotting the standard deviation of each group, the median
absolute deviation or the average absolute deviation might be plotted
instead. This might be done if there were significant outliers in the data
and a more robust measure of scale than the standard deviation was
desired.
Standard deviation plots are typically used in conjunction with mean
plots. The mean plot would be used to check for shifts in location while
the standard deviation plot would be used to check for shifts in scale.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33s.htm (1 of 3) [7/1/2003 5:21:46 PM]

1.3.3.28. Standard Deviation Plot

Sample Plot

This sample standard deviation plot shows

1. there is a shift in variation;
2. greatest variation is during the summer months.

Definition: Standard deviation plots are formed by:

Group ● Vertical axis: Group standard deviations
Standard
● Horizontal axis: Group identifier
Deviations
Versus A reference line is plotted at the overall standard deviation.
Group ID

Questions The standard deviation plot can be used to answer the following
questions.
1. Are there any shifts in variation?
2. What is the magnitude of the shifts in variation?
3. Is there a distinct pattern in the shifts in variation?

Importance: A common assumption in 1-factor analyses is that of equal variances.

Checking That is, the variance is the same for different levels of the factor
Assumptions variable. The standard deviation plot provides a graphical check for that
assumption. A common assumption for univariate data is that the
variance is constant. By grouping the data into equi-sized intervals, the
standard deviation plot can provide a graphical test of this assumption.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33s.htm (2 of 3) [7/1/2003 5:21:46 PM]

1.3.3.28. Standard Deviation Plot

Related Mean Plot

Techniques Dex Standard Deviation Plot

Software Most general purpose statistical software programs do not support a

standard deviation plot. However, if the statistical program can generate
the standard deviation for a group, it should be feasible to write a macro
to generate this plot. Dataplot supports a standard deviation plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33s.htm (3 of 3) [7/1/2003 5:21:46 PM]

1.3.3.29. Star Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.29. Star Plot

Purpose: The star plot (Chambers 1983) is a method of displaying multivariate
Display data. Each star represents a single observation. Typically, star plots are
Multivariate generated in a multi-plot format with many stars on each page and each
Data star representing one observation.
Star plots are used to examine the relative values for a single data point
(e.g., point 3 is large for variables 2 and 4, small for variables 1, 3, 5,
and 6) and to locate similar points or dissimilar points.

Sample Plot The plot below contains the star plots of 16 cars. The data file actually
contains 74 cars, but we restrict the plot to what can reasonably be
shown on one page. The variable list for the sample star plot is
1 Price
2 Mileage (MPG)
3 1978 Repair Record (1 = Worst, 5 = Best)
4 1977 Repair Record (1 = Worst, 5 = Best)
5 Headroom
6 Rear Seat Room
7 Trunk Space
8 Weight
9 Length

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33t.htm (1 of 3) [7/1/2003 5:21:47 PM]

1.3.3.29. Star Plot

We can look at these plots individually or we can use them to identify

clusters of cars with similar features. For example, we can look at the
star plot of the Cadillac Seville and see that it is one of the most
expensive cars, gets below average (but not among the worst) gas
mileage, has an average repair record, and has average-to-above-average
roominess and size. We can then compare the Cadillac models (the last
three plots) with the AMC models (the first three plots). This
comparison shows distinct patterns. The AMC models tend to be
inexpensive, have below average gas mileage, and are small in both
height and weight and in roominess. The Cadillac models are expensive,
have poor gas mileage, and are large in both size and roominess.

Definition The star plot consists of a sequence of equi-angular spokes, called radii,
with each spoke representing one of the variables. The data length of a
spoke is proportional to the magnitude of the variable for the data point
relative to the maximum magnitude of the variable across all data
points. A line is drawn connecting the data values for each spoke. This
gives the plot a star-like appearance and the origin of the name of this
plot.

Questions The star plot can be used to answer the following questions:
1. What variables are dominant for a given observation?
2. Which observations are most similar, i.e., are there clusters of
observations?
3. Are there outliers?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33t.htm (2 of 3) [7/1/2003 5:21:47 PM]

1.3.3.29. Star Plot

Weakness in Star plots are helpful for small-to-moderate-sized multivariate data sets.
Technique Their primary weakness is that their effectiveness is limited to data sets
with less than a few hundred points. After that, they tend to be
overwhelming.
Graphical techniques suited for large data sets are discussed by Scott.

Related Alternative ways to plot multivariate data are discussed in Chambers, du

Techniques Toit, and Everitt.

Software Star plots are available in some general purpose statistical software
progams, including Dataplot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33t.htm (3 of 3) [7/1/2003 5:21:47 PM]

1.3.3.30. Weibull Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.30. Weibull Plot

Purpose: The Weibull plot (Nelson 1982) is a graphical technique for
Graphical determining if a data set comes from a population that would logically
Check To See be fit by a 2-parameter Weibull distribution (the location is assumed to
If Data Come be zero).
From a
Population The Weibull plot has special scales that are designed so that if the data
That Would do in fact follow a Weibull distribution, the points will be linear (or
Be Fit by a nearly linear). The least squares fit of this line yields estimates for the
Weibull shape and scale parameters of the Weibull distribution. Weibull
Distribution distribution (the location is assumed to be zero).

Sample Plot

This Weibull plot shows that:

1. the assumption of a Weibull distribution is reasonable;
2. the shape parameter estimate is computed to be 33.32;
3. the scale parameter estimate is computed to be 5.28; and

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33u.htm (1 of 3) [7/1/2003 5:21:47 PM]

1.3.3.30. Weibull Plot

4. there are no outliers.

Definition: The Weibull plot is formed by:

Weibull ● Vertical axis: Weibull cumulative probability expressed as a
Cumulative percentage
Probability
● Horizontal axis: LN of ordered response
Versus
LN(Ordered The vertical scale is ln-ln(1-p) where p=(i-0.3)/(n+0.4) and i is the rank
Response) of the observation. This scale is chosen in order to linearize the
resulting plot for Weibull data.

Questions The Weibull plot can be used to answer the following questions:
1. Do the data follow a 2-parameter Weibull distribution?
2. What is the best estimate of the shape parameter for the
2-parameter Weibull distribution?
3. What is the best estimate of the scale (= variation) parameter for
the 2-parameter Weibull distribution?

Importance: Many statistical analyses, particularly in the field of reliability, are

Check based on the assumption that the data follow a Weibull distribution. If
Distributional the analysis assumes the data follow a Weibull distribution, it is
Assumptions important to verify this assumption and, if verified, find good estimates
of the Weibull parameters.

Related Weibull Probability Plot

Techniques Weibull PPCC Plot
Weibull Hazard Plot

The Weibull probability plot (in conjunction with the Weibull PPCC
plot), the Weibull hazard plot, and the Weibull plot are all similar
techniques that can be used for assessing the adequacy of the Weibull
distribution as a model for the data, and additionally providing
estimation for the shape, scale, or location parameters.
The Weibull hazard plot and Weibull plot are designed to handle
censored data (which the Weibull probability plot does not).

Case Study The Weibull plot is demonstrated in the airplane glass failure data case
study.

Software Weibull plots are generally available in statistical software programs

that are designed to analyze reliability data. Dataplot supports the
Weibull plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33u.htm (2 of 3) [7/1/2003 5:21:47 PM]

1.3.3.30. Weibull Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33u.htm (3 of 3) [7/1/2003 5:21:47 PM]

1.3.3.31. Youden Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.31. Youden Plot

Purpose: Youden plots are a graphical technique for analyzing interlab data when
Interlab each lab has made two runs on the same product or one run on two
Comparisons different products.
The Youden plot is a simple but effective method for comparing both
the within-laboratory variability and the between-laboratory variability.

Sample Plot

This plot shows:

1. Not all labs are equivalent.
2. Lab 4 is biased low.
3. Lab 3 has within-lab variability problems.
4. Lab 5 has an outlying run.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3331.htm (1 of 2) [7/1/2003 5:21:47 PM]

1.3.3.31. Youden Plot

Definition: Youden plots are formed by:

Response 1 1. Vertical axis: Response variable 1 (i.e., run 1 or product 1
Versus response value)
Response 2
2. Horizontal axis: Response variable 2 (i.e., run 2 or product 2
Coded by
response value)
Lab
In addition, the plot symbol is the lab id (typically an integer from 1 to k
where k is the number of labs). Sometimes a 45-degree reference line is
drawn. Ideally, a lab generating two runs of the same product should
produce reasonably similar results. Departures from this reference line
indicate inconsistency from the lab. If two different products are being
tested, then a 45-degree line may not be appropriate. However, if the
labs are consistent, the points should lie near some fitted straight line.

Questions The Youden plot can be used to answer the following questions:
1. Are all labs equivalent?
2. What labs have between-lab problems (reproducibility)?
3. What labs have within-lab problems (repeatability)?
4. What labs are outliers?

Importance In interlaboratory studies or in comparing two runs from the same lab, it
is useful to know if consistent results are generated. Youden plots
should be a routine plot for analyzing this type of data.

DEX Youden The dex Youden plot is a specialized Youden plot used in the design of
Plot experiments. In particular, it is useful for full and fractional designs.

Related Scatter Plot

Techniques

Software The Youden plot is essentially a scatter plot, so it should be feasible to

write a macro for a Youden plot in any general purpose statistical
program that supports scatter plots. Dataplot supports a Youden plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3331.htm (2 of 2) [7/1/2003 5:21:47 PM]

1.3.3.31.1. DEX Youden Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic
1.3.3.31. Youden Plot

1.3.3.31.1. DEX Youden Plot

DEX Youden The dex (Design of Experiments) Youden plot is a specialized Youden
Plot: plot used in the analysis of full and fractional experiment designs. In
Introduction particular, it is used in support of a Yates analysis. These designs may
have a low level, coded as "-1" or "-", and a high level, coded as "+1"
or "+", for each factor. In addition, there can optionally be one or more
center points. Center points are at the midpoint between the low and
high levels for each factor and are coded as "0".
The Yates analysis and the the dex Youden plot only use the "-1" and
"+1" points. The Yates analysis is used to estimate factor effects. The
dex Youden plot can be used to help determine the approriate model to
use from the Yates analysis.

Construction The following are the primary steps in the construction of the dex
of DEX Youden plot.
Youden Plot
1. For a given factor or interaction term, compute the mean of the
response variable for the low level of the factor and for the high
level of the factor. Any center points are omitted from the
computation.
2. Plot the point where the y-coordinate is the mean for the high
level of the factor and the x-coordinate is the mean for the low
level of the factor. The character used for the plot point should
identify the factor or interaction term (e.g., "1" for factor 1, "13"
for the interaction between factors 1 and 3).
3. Repeat steps 1 and 2 for each factor and interaction term of the
data.
The high and low values of the interaction terms are obtained by
multiplying the corresponding values of the main level factors. For
example, the interaction term X13 is obtained by multiplying the values
for X1 with the corresponding values of X3. Since the values for X1 and
X3 are either "-1" or "+1", the resulting values for X13 are also either

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33311.htm (1 of 3) [7/1/2003 5:21:48 PM]

1.3.3.31.1. DEX Youden Plot

"-1" or "+1".
In summary, the dex Youden plot is a plot of the mean of the response
variable for the high level of a factor or interaction term against the
mean of the response variable for the low level of that factor or
interaction term.
For unimportant factors and interaction terms, these mean values
should be nearly the same. For important factors and interaction terms,
these mean values should be quite different. So the interpretation of the
plot is that unimportant factors should be clustered together near the
grand mean. Points that stand apart from this cluster identify important
factors that should be included in the model.

Sample DEX The following is a dex Youden plot for the data used in the Eddy
Youden Plot current case study. The analysis in that case study demonstrated that
X1 and X2 were the most important factors.

Interpretation From the above dex Youden plot, we see that factors 1 and 2 stand out
of the Sample from the others. That is, the mean response values for the low and high
DEX Youden levels of factor 1 and factor 2 are quite different. For factor 3 and the 2
Plot and 3-term interactions, the mean response values for the low and high
levels are similar.
We would conclude from this plot that factors 1 and 2 are important
and should be included in our final model while the remaining factors
and interactions should be omitted from the final model.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33311.htm (2 of 3) [7/1/2003 5:21:48 PM]

1.3.3.31.1. DEX Youden Plot

Case Study The Eddy current case study demonstrates the use of the dex Youden
plot in the context of the analysis of a full factorial design.

Software DEX Youden plots are not typically available as built-in plots in
statistical software programs. However, it should be relatively
straightforward to write a macro to generate this plot in most general
purpose statistical software programs.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda33311.htm (3 of 3) [7/1/2003 5:21:48 PM]

1.3.3.32. 4-Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.32. 4-Plot
Purpose: The 4-plot is a collection of 4 specific EDA graphical techniques
Check whose purpose is to test the assumptions that underlie most
Underlying measurement processes. A 4-plot consists of a
Statistical 1. run sequence plot;
Assumptions
2. lag plot;
3. histogram;
4. normal probability plot.
If the 4 underlying assumptions of a typical measurement process
hold, then the above 4 plots will have a characteristic appearance (see
the normal random numbers case study below); if any of the
underlying assumptions fail to hold, then it will be revealed by an
anomalous appearance in one or more of the plots. Several commonly
encountered situations are demonstrated in the case studies below.
Although the 4-plot has an obvious use for univariate and time series
data, its usefulness extends far beyond that. Many statistical models of
the form

have the same underlying assumptions for the error term. That is, no
matter how complicated the functional fit, the assumptions on the
underlying error term are still the same. The 4-plot can and should be
routinely applied to the residuals when fitting models regardless of
whether the model is simple or complicated.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3332.htm (1 of 5) [7/1/2003 5:21:49 PM]

1.3.3.32. 4-Plot

Sample Plot:
Process Has
Fixed
Location,
Fixed
Variation,
Non-Random
(Oscillatory),
Non-Normal
U-Shaped
Distribution,
and Has 3
Outliers.

This 4-plot reveals the following:

1. the fixed location assumption is justified as shown by the run
sequence plot in the upper left corner.
2. the fixed variation assumption is justified as shown by the run
sequence plot in the upper left corner.
3. the randomness assumption is violated as shown by the
non-random (oscillatory) lag plot in the upper right corner.
4. the assumption of a common, normal distribution is violated as
shown by the histogram in the lower left corner and the normal
probability plot in the lower right corner. The distribution is
non-normal and is a U-shaped distribution.
5. there are several outliers apparent in the lag plot in the upper
right corner.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3332.htm (2 of 5) [7/1/2003 5:21:49 PM]

1.3.3.32. 4-Plot

Definition: The 4-plot consists of the following:

1. Run 1. Run sequence plot to test fixed location and variation.
Sequence
❍ Vertically: Yi
Plot;
2. Lag Plot; Horizontally: i
❍
3. Histogram; 2. Lag Plot to test randomness.
4. Normal ❍ Vertically: Yi
Probability
Plot ❍ Horizontally: Yi-1
3. Histogram to test (normal) distribution.
❍ Vertically: Counts

❍ Horizontally: Y

4. Normal probability plot to test normal distribution.

❍ Vertically: Ordered Yi

❍ Horizontally: Theoretical values from a normal N(0,1)

distribution for ordered Yi

Questions 4-plots can provide answers to many questions:

1. Is the process in-control, stable, and predictable?
2. Is the process drifting with respect to location?
3. Is the process drifting with respect to variation?
4. Are the data random?
5. Is an observation related to an adjacent observation?
6. If the data are a time series, is is white noise?
7. If the data are a time series and not white noise, is it sinusoidal,
autoregressive, etc.?
8. If the data are non-random, what is a better model?
9. Does the process follow a normal distribution?
10. If non-normal, what distribution does the process follow?
11. Is the model

valid and sufficient?

12. If the default model is insufficient, what is a better model?
13. Is the formula valid?
14. Is the sample mean a good estimator of the process location?
15. If not, what would be a better estimator?
16. Are there any outliers?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3332.htm (3 of 5) [7/1/2003 5:21:49 PM]

1.3.3.32. 4-Plot

Importance: There are 4 assumptions that typically underlie all measurement

Testing processes; namely, that the data from the process at hand "behave
Underlying like":
Assumptions 1. random drawings;
Helps Ensure
2. from a fixed distribution;
the Validity of
the Final 3. with that distribution having a fixed location; and
Scientific and 4. with that distribution having fixed variation.
Engineering Predictability is an all-important goal in science and engineering. If
Conclusions the above 4 assumptions hold, then we have achieved probabilistic
predictability--the ability to make probability statements not only
about the process in the past, but also about the process in the future.
In short, such processes are said to be "statistically in control". If the 4
assumptions do not hold, then we have a process that is drifting (with
respect to location, variation, or distribution), is unpredictable, and is
out of control. A simple characterization of such processes by a
location estimate, a variation estimate, or a distribution "estimate"
inevitably leads to optimistic and grossly invalid engineering
conclusions.
Inasmuch as the validity of the final scientific and engineering
conclusions is inextricably linked to the validity of these same 4
underlying assumptions, it naturally follows that there is a real
necessity for all 4 assumptions to be routinely tested. The 4-plot (run
sequence plot, lag plot, histogram, and normal probability plot) is seen
as a simple, efficient, and powerful way of carrying out this routine
checking.

Interpretation: Of the 4 underlying assumptions:

Flat, 1. If the fixed location assumption holds, then the run sequence
Equi-Banded, plot will be flat and non-drifting.
Random,
2. If the fixed variation assumption holds, then the vertical spread
Bell-Shaped,
in the run sequence plot will be approximately the same over
and Linear
the entire horizontal axis.
3. If the randomness assumption holds, then the lag plot will be
structureless and random.
4. If the fixed distribution assumption holds (in particular, if the
fixed normal distribution assumption holds), then the histogram
will be bell-shaped and the normal probability plot will be
approximatelylinear.
If all 4 of the assumptions hold, then the process is "statistically in
control". In practice, many processes fall short of achieving this ideal.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3332.htm (4 of 5) [7/1/2003 5:21:49 PM]

1.3.3.32. 4-Plot

Related Run Sequence Plot

Techniques Lag Plot
Histogram
Normal Probability Plot

Autocorrelation Plot
Spectral Plot
PPCC Plot

Case Studies The 4-plot is used in most of the case studies in this chapter:
1. Normal random numbers (the ideal)
2. Uniform random numbers
3. Random walk
4. Josephson junction cryothermometry
5. Beam deflections
6. Filter transmittance
7. Standard resistor
8. Heat flow meter 1

Software It should be feasible to write a macro for the 4-plot in any general
purpose statistical software program that supports the capability for
multiple plots per page and supports the underlying plot techniques.
Dataplot supports the 4-plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3332.htm (5 of 5) [7/1/2003 5:21:49 PM]

1.3.3.33. 6-Plot

1. Exploratory Data Analysis

1.3. EDA Techniques
1.3.3. Graphical Techniques: Alphabetic

1.3.3.33. 6-Plot
Purpose: The 6-plot is a collection of 6 specific graphical techniques whose
Graphical purpose is to assess the validity of a Y versus X fit. The fit can be a
Model linear fit, a non-linear fit, a LOWESS (locally weighted least squares)
Validation fit, a spline fit, or any other fit utilizing a single independent variable.
The 6 plots are:
1. Scatter plot of the response and predicted values versus the
independent variable;
2. Scatter plot of the residuals versus the independent variable;
3. Scatter plot of the residuals versus the predicted values;
4. Lag plot of the residuals;
5. Histogram of the residuals;
6. Normal probability plot of the residuals.

Sample Plot

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3333.htm (1 of 4) [7/1/2003 5:21:50 PM]

1.3.3.33. 6-Plot

This 6-plot, which followed a linear fit, shows that the linear model is
not adequate. It suggests that a quadratic model would be a better
model.

Definition: The 6-plot consists of the following:

6 1. Response and predicted values
Component
❍ Vertical axis: Response variable, predicted values
Plots
❍ Horizontal axis: Independent variable

2. Residuals versus independent variable

❍ Vertical axis: Residuals

❍ Horizontal axis: Independent variable

3. Residuals versus predicted values

❍ Vertical axis: Residuals

❍ Horizontal axis: Predicted values

4. Lag plot of residuals

❍ Vertical axis: RES(I)

❍ Horizontal axis: RES(I-1)

5. Histogram of residuals
❍ Vertical axis: Counts

❍ Horizontal axis: Residual values

6. Normal probability plot of residuals

❍ Vertical axis: Ordered residuals

❍ Horizontal axis: Theoretical values from a normal N(0,1)

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3333.htm (2 of 4) [7/1/2003 5:21:50 PM]

1.3.3.33. 6-Plot

distribution for ordered residuals

Questions The 6-plot can be used to answer the following questions:

1. Are the residuals approximately normally distributed with a fixed
location and scale?
2. Are there outliers?
3. Is the fit adequate?
4. Do the residuals suggest a better fit?

Importance: A model involving a response variable and a single independent variable

Validating has the form:
Model

where Y is the response variable, X is the independent variable, f is the

linear or non-linear fit function, and E is the random component. For a
good model, the error component should behave like:
1. random drawings (i.e., independent);
2. from a fixed distribution;
3. with fixed location; and
4. with fixed variation.
In addition, for fitting models it is usually further assumed that the fixed
distribution is normal and the fixed location is zero. For a good model
the fixed variation should be as small as possible. A necessary
component of fitting models is to verify these assumptions for the error
component and to assess whether the variation for the error component
is sufficiently small. The histogram, lag plot, and normal probability
plot are used to verify the fixed distribution, location, and variation
assumptions on the error component. The plot of the response variable
and the predicted values versus the independent variable is used to
assess whether the variation is sufficiently small. The plots of the
residuals versus the independent variable and the predicted values is
used to assess the independence assumption.
Assessing the validity and quality of the fit in terms of the above
assumptions is an absolutely vital part of the model-fitting process. No
fit should be considered complete without an adequate model validation
step.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3333.htm (3 of 4) [7/1/2003 5:21:50 PM]

1.3.3.33. 6-Plot

Related Linear Least Squares

Techniques Non-Linear Least Squares
Scatter Plot
Run Sequence Plot
Lag Plot
Normal Probability Plot
Histogram

Case Study The 6-plot is used in the Alaska pipeline data case study.

Software It should be feasible to write a macro for the 6-plot in any general
purpose statistical software program that supports the capability for
multiple plots per page and supports the underlying plot techniques.
Dataplot supports the 6-plot.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda3333.htm (4 of 4) [7/1/2003 5:21:50 PM]

1.3.4. Graphical Techniques: By Problem Category

1. Exploratory Data Analysis

1.3. EDA Techniques

1.3.4. Graphical
Techniques: By
Problem
Category
Univariate
y=c+e

Run Sequence Lag Plot: Histogram:

Plot: 1.3.3.25 1.3.3.15 1.3.3.14

Normal 4-Plot: 1.3.3.32 PPCC Plot:

Probability Plot: 1.3.3.23
1.3.3.21

Weibull Plot: Probability Plot: Box-Cox

1.3.3.30 1.3.3.22 Linearity Plot:
1.3.3.5

http://www.itl.nist.gov/div898/handbook/eda/section3/eda34.htm (1 of 4) [7/1/2003 5:21:50 PM]

1.3.4. Graphical Techniques: By Problem Category

Box-Cox Bootstrap Plot:

Normality Plot: 1.3.3.4
1.3.3.6

Time Series
y = f(t) + e

Run Sequence Spectral Plot: Autocorrelation

Plot: 1.3.3.25 1.3.3.27 Plot: 1.3.3.1

Complex Complex
Demodulation Demodulation
Amplitude Plot: Phase Plot:
1.3.3.8 1.3.3.9

1 Factor
y = f(x) + e

Scatter Plot: Box Plot: 1.3.3.7 Bihistogram:

1.3.3.26 1.3.3.2

http://www.itl.nist.gov/div898/handbook/eda/section3/eda34.htm (2 of 4) [7/1/2003 5:21:50 PM]

1.3.4. Graphical Techniques: By Problem Category

Quantile-Quantile Mean Plot: Standard

Plot: 1.3.3.24 1.3.3.20 Deviation Plot:
1.3.3.28

Multi-Factor/Comparative
y = f(xp, x1,x2,...,xk) + e

Block Plot:
1.3.3.3

Multi-Factor/Screening
y = f(x1,x2,x3,...,xk) + e

DEX Scatter DEX Mean Plot: DEX Standard

Plot: 1.3.3.11 1.3.3.12 Deviation Plot:
1.3.3.13

Contour Plot:
1.3.3.10

http://www.itl.nist.gov/div898/handbook/eda/section3/eda34.htm (3 of 4) [7/1/2003 5:21:50 PM]

1.3.4. Graphical Techniques: By Problem Category

Regression
y = f(x1,x2,x3,...,xk) + e

Scatter Plot: 6-Plot: 1.3.3.33 Linear

1.3.3.26 Correlation Plot:
1.3.3.16

Linear Intercept Linear Slope Linear Residual

Plot: 1.3.3.17 Plot: 1.3.3.18 Standard
Deviation
Plot:1.3.3.19

Interlab
(y1,y2) = f(x) + e

Youden Plot:
1.3.3.31

Multivariate
(y1,y2,...,yp)

Star Plot:
1.3.3.29

http://www.itl.nist.gov/div898/handbook/eda/section3/eda34.htm (4 of 4) [7/1/2003 5:21:50 PM]

1.3.5. Quantitative Techniques

1. Exploratory Data Analysis

1.3. EDA Techniques

1.3.5. Quantitative Techniques

Confirmatory The techniques discussed in this section are classical statistical methods
Statistics as opposed to EDA techniques. EDA and classical techniques are not
mutually exclusive and can be used in a complamentary fashion. For
example, the analysis can start with some simple graphical techniques
such as the 4-plot followed by the classical confirmatory methods
discussed herein to provide more rigorous statments about the
conclusions. If the classical methods yield different conclusions than
the graphical analysis, then some effort should be invested to explain
why. Often this is an indication that some of the assumptions of the
classical techniques are violated.
Many of the quantitative techniques fall into two broad categories:
1. Interval estimation
2. Hypothesis tests

Interval It is common in statistics to estimate a parameter from a sample of data.

Estimates The value of the parameter using all of the possible data, not just the
sample data, is called the population parameter or true value of the
parameter. An estimate of the true parameter value is made using the
sample data. This is called a point estimate or a sample estimate.
For example, the most commonly used measure of location is the mean.
The population, or true, mean is the sum of all the members of the
given population divided by the number of members in the population.
As it is typically impractical to measure every member of the
population, a random sample is drawn from the population. The sample
mean is calculated by summing the values in the sample and dividing
by the number of values in the sample. This sample mean is then used
as the point estimate of the population mean.
Interval estimates expand on point estimates by incorporating the
uncertainty of the point estimate. In the example for the mean above,
different samples from the same population will generate different
values for the sample mean. An interval estimate quantifies this
uncertainty in the sample estimate by computing lower and upper

http://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm (1 of 4) [7/1/2003 5:21:51 PM]

1.3.5. Quantitative Techniques

values of an interval which will, with a given level of confidence (i.e.,

probability), contain the population parameter.

Hypothesis Hypothesis tests also address the uncertainty of the sample estimate.
Tests However, instead of providing an interval, a hypothesis test attempts to
refute a specific claim about a population parameter based on the
sample data. For example, the hypothesis might be one of the
following:
● the population mean is equal to 10

● the population standard deviation is equal to 5

● the means from two populations are equal

● the standard deviations from 5 populations are equal

To reject a hypothesis is to conclude that it is false. However, to accept

a hypothesis does not mean that it is true, only that we do not have
evidence to believe otherwise. Thus hypothesis tests are usually stated
in terms of both a condition that is doubted (null hypothesis) and a
condition that is believed (alternative hypothesis).
A common format for a hypothesis test is:
H0: A statement of the null hypothesis, e.g., two
population means are equal.
Ha: A statement of the alternative hypothesis, e.g., two
population means are not equal.
Test Statistic: The test statistic is based on the specific
hypothesis test.
Significance Level: The significance level, , defines the sensitivity of
the test. A value of = 0.05 means that we
inadvertently reject the null hypothesis 5% of the
time when it is in fact true. This is also called the
type I error. The choice of is somewhat
arbitrary, although in practice values of 0.1, 0.05,
and 0.01 are commonly used.
The probability of rejecting the null hypothesis
when it is in fact false is called the power of the
test and is denoted by 1 - . Its complement, the
probability of accepting the null hypothesis when
the alternative hypothesis is, in fact, true (type II
error), is called and can only be computed for a
specific alternative hypothesis.

http://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm (2 of 4) [7/1/2003 5:21:51 PM]

1.3.5. Quantitative Techniques

Critical Region: The critical region encompasses those values of

the test statistic that lead to a rejection of the null
hypothesis. Based on the distribution of the test
statistic and the significance level, a cut-off value
for the test statistic is computed. Values either
above or below or both (depending on the
direction of the test) this cut-off define the critical
region.

Practical It is important to distinguish between statistical significance and

Versus practical significance. Statistical significance simply means that we
Statistical reject the null hypothesis. The ability of the test to detect differences
Significance that lead to rejection of the null hypothesis depends on the sample size.
For example, for a particularly large sample, the test may reject the null
hypothesis that two process means are equivalent. However, in practice
the difference between the two means may be relatively small to the
point of having no real engineering significance. Similarly, if the
sample size is small, a difference that is large in engineering terms may
not lead to rejection of the null hypothesis. The analyst should not just
blindly apply the tests, but should combine engineering judgement with
statistical analysis.

Bootstrap In some cases, it is possible to mathematically derive appropriate

Uncertainty uncertainty intervals. This is particularly true for intervals based on the
Estimates assumption of a normal distribution. However, there are many cases in
which it is not possible to mathematically derive the uncertainty. In
these cases, the bootstrap provides a method for empirically
determining an appropriate interval.

Table of Some of the more common classical quantitative techniques are listed
Contents below. This list of quantitative techniques is by no means meant to be
exhaustive. Additional discussions of classical statistical techniques are
contained in the product comparisons chapter.
● Location
1. Measures of Location
2. Confidence Limits for the Mean and One Sample t-Test
3. Two Sample t-Test for Equal Means
4. One Factor Analysis of Variance
5. Multi-Factor Analysis of Variance
● Scale (or variability or spread)
1. Measures of Scale
2. Bartlett's Test

http://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm (3 of 4) [7/1/2003 5:21:51 PM]

1.3.5. Quantitative Techniques

3. Chi-Square Test
4. F-Test
5. Levene Test
● Skewness and Kurtosis
1. Measures of Skewness and Kurtosis
● Randomness
1. Autocorrelation