IBM SPSS Statistics evolved from an original

product that was released in 1968. That product was called “Statistical Package for Social
Sciences,” or “SPSS.” IBM SPSS Statistics is a statistical and machine
learning software application and is widely used in academia, government agencies, and
large enterprises. It’s used to build predictive models, perform statistical analysis of data,
and conduct other analytic tasks. It has a visual interface, which enables users to leverage
statistical and data mining algorithms without programming, although the interface is very
different from Modeler. As you can see, the main section of the screen looks very much
like a spreadsheet; it displays data and allows manual editing. This particular small data
set, called “Employee Data”, was created some time ago and does not represent real
people. It is shipped with the product for use in demos and tutorials. At the bottom of the
screen, we can see two
tabs: Data View and Variable View. In the Variable View, we can see and edit the information
about all variables, including names, labels, data types, and measurement levels. We can
also specify labels for values of categorical variables, and missing values. At the top of the data
window is a menu. Under
File, if you select “Import Data,” you will see a list of a wide variety of data
formats that you can import. The product uses its own data file format with the extension
“.sav” that saves all the information about the variables we just saw in Variable
view. The menu enables importing from and exporting to many other formats. Under “Data,”
you’ll find an extensive
menu of possible data operations. Note that Data Validation can be performed using user-
defined
rules that specify the expected behavior of variable values. For example, if the date
and month are kept in separate columns, the date cannot exceed “31,” but for February,
the date can’t exceed “29.” A special rule can therefore be created and applied
during data validation. Additionally, you can enable some checks, such as percentage
of missing values in a record or in the field. When you click the “Transform” menu item,
you’ll find a variety of available data transformations.
Under “Compute Variable…” you can write a formula for a new variable based on existing
variables. You can use any of the many mathematical and statistical functions available in the
product. You also have the option to use automatic
data preparation, similar to Modeler. In the “Analyze” menu, you will see many
types of statistical and machine learning analysis. Under “Regression,” there are
a variety of regression-related models. There are other kinds of regressions that appear
separately on the Analyze menu, including General Linear Model, Generalized Linear Models,
Mixed Models, and Loglinear. Now let’s build a decision-tree model on
the data. For this exercise we’ll try to predict the "Employment category" field based
on other fields. In the “Analyze” menu, select “Classify” and then “Tree”.
<Click> In the Decision Tree window, we can specify the dependent variable “Employment
Category,” and use most other fields -- except id and bdate -- as predictors, or independent
variables. Usually the ID variable should not be used as a predictor, because it will
not help with new cases, and the birthdate does not seem to be a useful predictor in
this example either. We’ll select “Exhaustive CHAID” as our Growing Method, although there
are also three other options available. Data scientists often try many different models
to see which one works best for their data. Here we are just looking at one example model
in order to illustrate how the product works. Click the “Validation” button to open
the Decision Tree Validation window. Here, we select “Split-sample validation” to
make sure we test the model on new data. Click “OK” in the Decision Tree window, to <Click>
generate the output, including the tree diagram shown here. <Click> A Classification table
is also displayed that shows how well the model works on training and test data. In
this case, the accuracy is 91.2% on training data and only 85.6% on test data, which means
the model does not generalize to new data very well. It’s possible that by using different
models, we can get better results. Let’s move to the next menu item. When you
click “Graphs,” you’ll open a versatile Chart Builder, in addition to several other
options. The Chart Builder enables us to choose a style
from the gallery and to drag required fields onto the canvas, select colors, and choose
from other options. Here’s an example after we drag the “Previous
Experience,” “Current Salary,” and Gender variables to the corresponding slots to define
the axis and colors for the dots on the chart. The plot in the canvas is not based on real
data, this example simply gives you an idea of what to expect. Here is the real plot obtained
from the data
that we’ve been using. It shows different colored dots for gender, and regression lines
that show the relationship of the current salary to previous experience for each gender.
Throughout IBM SPSS Statistics, you’ll see
a “Paste” button. When you click the “Paste” button, instead of executing the task right
away the application will open another window, called the Syntax editor. Here, you can see
the code called “syntax” pasted for you. SPSS syntax is a special programming language. For
example, here is the code for the decision
tree we just built. Once we have the syntax, we can execute it, manually edit it, store
it for later use, or send it to other users of IBM SPSS Statistics. Experienced SPSS users
can write the code from scratch, while others might prefer to have it generated by the graphical
interface. Remember, the option to paste syntax is available in throughout the program.
If the syntax is generated by all the steps in a data analytics process -- opening the
data set, applying any data transformations, building models -- and then saved as a syntax
file with the extension “.sps”, it’s similar to saving a stream in IBM SPSS Modeler.
However, one important difference is that it does not allow for an easy way of scoring
new records with the model. We’ll talk about different ways to deploy models in the next
section. You’ve learned how IBM SPSS Statistics helps
data scientists to analyze their data using many statistical and machine learning techniques.
Using a graphical user interface, we can create complicated analysis that can be saved in
the form of syntax and reused later. Next, we will talk about predictive model
deployment, an important part of the overall data science lifecycle.