TESTS OF
SIGNIFICANCE
DR DINESH WALIA
DEPT OF COMMUNITY MEDICINE
GMCH CHANDIGARH
" Statistics may be defined as "a
body of methods for making wise
decisions in the face of uncertainty."
~W.A. Wallis.
Meaning of Tests of Significance:
Statistical procedures to draw
inferences from samples about
population
Why required?
Whether difference between sample
estimate and population values is
significant or not?
Differences between different sample
estimates significant or not?
Meaning by Significant and Non
Significant Differences
Terminology
Hypothesis
Null Hypothesis
Normal Curve
Alternate Hypothesis
Acceptance & Rejection Regions
Right tail & left tail tests.
Level of significance
Type I error (α-error)
Type II error (β-error)
Power of the Test
Steps in Tests of Significance
State clearly Null Hypo (Ho)
Choose Level of Significance (α)
Decide test of Significance
Calculate value of test statistic
Obtain P-Value and Conclude Ho
More About Acceptance and
Rejection Regions
Acceptance Region :Region in which Ho is
accepted
Rejection Region:Region in which Ho is
rejected.
If z cal > Zα
Then P < α → Ho is rejected
If Zcal/ < Zα
Then P > α → Ho is accepted
Usually α = 0.05, May be 0.01.,0…..
Pictorial Representation of a Statistical
Test
Testing Distribution
-Zα s 0.0 +Zα s
(e.g. – 1.96s) (e.g. + 1.96s)
Criterion For Testing Based on
Normal Distribution
If │z cal│< 1.96 → P > 0.05 (Non Sig)
If │z cal│ > 1.96 → P < 0.05 (Sig)
If z > 2.58 →P < 0.01 (Highly Sig)
If z > 3 → P < 0.001 (Highly Sig)
Parametric Statistics
Z-test (for large samples) for testing
significance of
single population mean,
difference of two population means,
single population proportion,
difference of two population
proportions.
t-test (for small samples)
for testing significance of
single population mean,
difference of two population means,
paired data (Effectiveness of Drug)
correlation coefficient.
F-Test:
Testing equality of several means
Testing equality of two population
variances.
One-way, Two-way and Multi-way
ANOVA
Chi-square:
Testing Associations
Goodness of Fit
Significance of Risk
H0: RR=1 or OR =1
Normal Test (z-Test):(For large samples,
Quantitative & Qualitative Data))
(a) Test for Significance of Difference
Between Two Group Means
Ho :µ1= µ2
z= X1- X2
s√ +1
1
n1 n2
Z= X -X
1 2
__________
√s12 + s22
n1 n2
Ex. Normal Test of Means
Mid Arm circumference (MAC) of 625
American and 625 Indian Children.
MAC American Indian
Children Children
Mean= 15.5
SD = 5.0 5.4
Z= X1- X2
20.5
√s12 + s22
n1 n2
= 20.5 -15.5
√52 + (5.4)2= 16.98
625
625
Z cal >3 → P < 0.001 Highly Significant
(b) Tests for two proportions
Ho:------------- (Qualitative Data)
z = p1 – p2
√p q( 1 + 1 )
n1 n2
p1 = prop. In I Ip.} p= pooled
p2 = prop in II Ip.} q= prop. 1-p
Prob. From a popn, 40 females using
oral contraceptives and 60 females
using other contra, were randomly
selected and the number of
hypertensive cases from both groups
were recorded:
Type of Contra Total Women Hyp.
oral 40 8
others 60 15
Test the hypothesis that the prop. Of
patients with hyp. Is same for the
two groups.
Ex : Normal Test of Proportions
Gp I Gp II
Sample size n1=150 n2=160
Cp Abn 129 110
Ho: Prop. Of CP. Abn. Is same
p1= 129 = 0.86
150
p2= 110 = 0.68
160
p= overall prop. = (129+110)
(150+160)
(n1 p1 + n2 p2)
= (n1+n2)
= 0.77
q = 1-p = 0.23
Z= p1-p2
√pq (1 + 1) = 3.61
n1 n2 P < 0.001
(Highly Sig)
t-test ( for small samples Quantative
Data)
(a) Comparison of means of two
independent samples
student’s t-test : Ho:---------
t = X1- X2
s √ 1 + 1)
n1 n2
X = Mean of I group.
1
X = Mean of II group.
2
n1= no. of cases in I group.
n2= no. of cases in II group.
s = pooled s.d.
s2= (n1-1) s12 + (n2-1) s22
n1 + n2 – 2
(b) Comparison criterion:
→ 2f t cal < t
tab (n1 + n2 -2) d.f.
Then P > 0.05 → Ho is accepted
If t cal > t tab (.05), (n1 + n2 -2) d.f.
Then P < 0.05 → Ho is rejected
Prob. Cumulative weight losses during
insulin induced hypoglycemia for 12
patients treated with propanol
and 12 control patients. Test the
significance of difference between mean
cumulative wt losses.
Group Mean SD
Propanol 120 10
Control placebo 8
70
t = X -X
1 2
s √1 + 1
n1 n2
S2 = (n2-1) s12 + (n2-1) s22
n1 + n2 -2
n1 = 12 , n2 = 12
(b) Test for effectiveness of a
particular drug.
(case of 2- related/dependant
samples)
t = đ
s/√n
d = AT – BT
đ=Σd = Sum of all diff
n No. of paired observations
s2 = variance of differences
(s= s.d.)
s2 = 1 [Σ d2 – (Σ d)2]
(n-1) n
Conclusions
If t cal < t tab 5%, (n-1)d.f.
Then P > 0.05, Ho is accepted.
If t cal > t tab 5%, (n-1) d.f.
Then P < 0.05, Ho is rejected
Ex. Paired t-test
Sr.No. Serum Cholesterol
d2= (x2-x1)2
Before (x1) d=AT-BT
After (x 2) x2-x1
1 201 200
2 231 236 5
3 221 216 -5
4 260 223 -27
5 228 224 -1 -4
6 237 216 -21
7 326 296 -30
8 235 195 -40
9 240
207
10 267 247 -20
11 284 210 -74
12 201 209 8
n=12 Σ d= -242
Σ d2= 10766
đ=Σ d= -20.17
n
S2d= [Σ d2 – (Σ d)2]
1(n-1) n
= 1 [10766-(-242)2 ]
11 12
Sd = 6.67
t = 3.11
tab, II df
1% level = 2.20
tab, II df
5% level
t= đ
s/√=n 3.02 < t tab
x2- Chi Square Test
(For Qualitative Data)
Ho: There is no significant association
between two attributes.
Ho: Ro=1 or Ho: R R=1
(Case control) (cohort)
Formula:
x2 = Σ (O-E)2
E
O = observed freq.
E = Expected freq.
E =R*C
N
D+ D-
E+ a b R1
c d R2
E-
C1 C2
d.f. = (r-1) (c-1)
(2-1)*(2-1)
1*1=1
NOTE :
(1) Not necessarily Disease & exposure
may be any two attributes A & B
(2) Not necessarily 2*2 table any
number of categories of A & B.
Short cut Method :
X2 = N (ad-bc)2
R 1 R 1 C1 C2
If X2 cal < X2 tab, 5%
Then P > 0.05 → Ho is accepted
If X2 cal > X2
Then P < 0.05,→ Ho is rejected
For 2 * 2 table (1 d.f.)
X2 tab, 5% = 3.84
Examples
Prob 1. Test whether there is any
association between death from
C.H.D and ECG Abnormalities:
ECG Abn Death due to CHD
Yes No Total
+ 10 40 50
- 15 135 150
Total 25 175 200
Ho : -------------
X2 = N (ad-bc)2
R1 R2 C1 C2
= 200(10 * 135 -15 * 40)2 =…
50 * 150 * 25 * 175
IInd Method: 2 = Σ (0-E)2
E
E (10) = .. X
E (40) = ..
E (15) = ..
E (135)=..
Prob 2 :
Test whether there is any
asso between smoking and
hypertension?
or
Test whether there is any evidence
that smoking is a risk factor for
hypertension.
+ HTN -
Smoking
Yes 120 280
No 30 570
Parametric Statistics: where to apply ?
The given observations are independent
Observations come from normal
distribution
The hypothesis involve population
parameters
Parametric procedures may be applied on
measurement data only
Computations are sometimes difficult
Parametric procedures are more robust
than Non-Parametric procedures.
Choice of statistical test for independent observations
Outcome variable
Nominal Cate Ordinal Quantitat Quantitativ Quantitative
goric ive e Non- Normal
al Discrete Normal
I Nominal χ² or χ² χ² or Mann- Mann-Whitney Student's t test
Whitney or log-rank (a
n Fisher's Mann-
Whitney
p Categorical (>2 χ² χ² Kruskal- Kruskal- Kruskal-Wallis ANOVA (c)
u categories) Wallis (b) Wallis (b) (b)
t Ordinal (Orde red χ² or (e) Spearman Spearman Spearman rank Spearman rank
categories) rank rank or linear
Mann-
regression(d)
Whitney
V
a Quantitative Logistic (e) (e) Spearman Spearman rank Spearman rank
Discrete regression rank or linear
r regression (d)
i Quantitative non- Logistic (e) (e) (e) Plot data and Plot data and
a Normal regression Pearson or Pearson or
Spearman rank Spearman rank
b and linear
regression
l
e
Quantitative Logistic (e) (e) (e) Linear Pearson and
Normal regression regression (d) linear regression
Non-Parametric Statistics: where to apply ?
The given observations are independent
Observations may not be from normal
distribution
Based on minimum assumptions
The hypothesis is not concerned about
population parameters
Non-Parametric procedures may be applied
on weak measurement scale i.e. on count
data or rank data
Computations are simple.
Non-parametric Statistics
Sign test: Testing median
Wilcoxon signed rank test: Testing
median
Run test: For randomness
Median test: For testing equality of two
medians
Mann-Whitney test: For testing equality
of two medians
Chi-square test: Testing “goodness of
fit”, testing independence, homogeneity
Kruskal Wallis: One way ANOVA
Friedman: Two-way ANOVA
Thank You
