Unit -3 Non parametric test:

Parametric Vs. Non parametric test:

➢ All the parametric test (like t – test, Z – test, F – test) are based upon the
assumption that the random samples are drawn from a normal population.
These tests are concerned with statistical hypothesis about the parameters of
the population (distribution of the random variables) from which the samples
are drawn.
➢ But in practice, there are many situations in which the samples are selected
from the non-normal population. In such cases, the hypothesis tests are not
concerned about the parameters of the population. These tests are called non
parametric tests and also called distribution free tests.
➢ Generally, non parametric tests are performed when the assumptions about
the underlying population parameters are in a weak state, data are not
normally distributed, data are measured in nominal and ordinal scale, and the
hypothesis is concerned about the qualitative characteristics.
 Difference between parametric and Non
parametric test
 Parametric test: ➢ Non parametric test:
1. Parametric test is concerned with the 1. Non parametric test is not concerned
hypothesis about the parameters of the with the hypothesis about the
population from which the samples are parameters of the population from
drawn. which the samples are drawn.
2. It is used in testing of hypothesis and 2. It is used in testing of hypothesis
estimation of parameters of the only.
population. 3. This test is used if data are measured
3. This test is used if data are measured in in nominal and ordinal scale.
interval and ratio scale. 4. It is less powerful test as compared
4. It is more powerful test as compared to to parametric test.
non parametric test. 5. It does not require complicated
sampling techniques.
5. It requires complicated sampling
techniques. 6. The commonly used Non parametric
tests are Run test, Binomial test,
6. The commonly used parametric tests
Mann Whitney test, Chi-square test,
are t test, Z test, and F test.
Kolmogorov – Smirnov test, etc.
 Advantages of Non parametric test:
i. Non parametric test is simple and easy to apply and do not require complicated
theory.
ii. It is less time consuming.
iii. It needs no assumptions about the population parameters form which the samples
are drawn.
iv. It can be applied to qualitative as well as quantitative data.
v. There is no restriction for the minimum size of sample for valid and reliable
results.
➢ Disadvantages of Non parametric test:
i. All Non parametric tests are not simple and easy to apply.
ii. Non parametric tests can not be used to estimate the unknown parameters of the
population.
iii. These tests are less reliable and less powerful than parametric tests.
iv. Lot of tables are needed for tests.
v. These tests are not suitable for the data measured in interval and ratio scale.
When to apply?
i. When data are not normally distributed.
ii. When assumptions of parametric procedure are not satisfied.
iii. When data are measured in nominal and ordinal scales.
iv. When data size is small.
v. When basic question of interest is distribution free in nature.

❑ Assumptions of Non parametric test:

i. The sample observations are independent.
ii. The variable under study is continuous.
iii. Sample probability density function is continuous.
iv. Lower order moments exists.
Commonly used Non Parametric Tests:
❑ One sample test:
1. Run test.
2. Binomial test.
3. Kolmogorov – Smirnov test
❑ Two independent samples test:
1. Median test.
2. Kolmogorov – Smirnov test.
3. Mann Whitney test.
4. Chi-square test
❑ Paired-sample test:
1. Wilcoxon signed rank test
❑ K samples test:
1. Cochran Q test.
2. Kruskal Wallish H test.
3. Friedman F test.
Run test:
 A run is a sequence of letters of one kind bounded by letters of the other
kind.
 For example:
 A sequence {HHHTTHTHTTTH} of symbols H and T has 7 runs i.e. R = 7
 Run test is a simple non-parametric test used to test the randomness of a
given set of samples.
 Testing procedure:
 Setting of hypothesis:
 Null hypothesis, H0:The sample observations are in random order.
 Alternative hypothesis, H1:The sample observations are not in random order.
 Level of significance:
Fix the level of significance 5 % or 1 % unless or otherwise stated.
 For small sample size: n1 or n2 ≤ 20
Test statistics:
 Under H0
 r = Number of runs.
 Calculation:
a. If letters or symbols are given, then the test statistic is easily calculated.
b. If sample observations (data) are given, then the test statistics is calculated
as follows.
➢ At first, find median from the arranged data.
➢ Assign a symbol A to the observation xi if xi > Md
➢ Assign a symbol B to the observation xi if xi < Md
➢ And omit if xi = Md
➢ Obtain a sequence of As and Bs and then find the
➢ r = no. of runs
➢ n1 = no. of A
➢ n2 = no. of B
 Critical value:
 The critical values (𝑟 and 𝑟)ҧ for the given level of significance (α), d.f. (n1,n2)
are obtained from the critical value table of the run test.

 Decision:
 If the observed value of ‘r’ lies within the lower critical value (𝑟) and upper
critical value (𝑟),
ҧ then the null hypothesis is accepted and rejected
otherwise.

 Conclusion:
 If H0 is accepted, then statement of H0 is accepted and the statement of H0 is
written as a conclusion.
 If H0 is rejected, then statement of H1 is accepted and the statement of H1 is
written as a conclusion.
For large sample size: n1 or n2 > 20
 For large sample size, r is normally distributed with mean 𝜇𝑟 and variance 𝜎𝑟2 .
 Test statistic:
 Under H0
𝑟 − 𝜇𝑟
 Z= ~ N(0, 1)
𝜎𝑟
 Where,
2𝑛1 𝑛2
 Mean (𝜇𝑟 ) = +1
𝑛1 +𝑛2
2𝑛1 𝑛2 (2𝑛1 𝑛2 −𝑛1 −𝑛2 )
 Standard deviation (𝜎𝑟 ) =
𝑛1 +𝑛2 2 (𝑛1 +𝑛2 −1)

 Critical value:
 Same as Z test.

 Decision:
 Same as Z test.

 Conclusion:
 Same

 Remark: Even if the size of the sample is small, when the level of significance (α) is other than 5%, we will use the
large sample test (Z – test).
Binomial test:
 Binomial test is a non parametric test used to test whether the binomial
population has two distinct groups of two equal numbers of outcomes
(dichotomous) or not i.e. to test whether the total number of outcomes in first
group is equal to that in second group or not.
 This test is appropriate if the data are measured in nominal scale. This test is an
alternative test of Z test of single proportion.
 Testing procedure:
 Setting of hypothesis:
1
 Null hypothesis, H0: P = P0 (= 2) i.e. i.e. the sample has been drawn from the
1
binomial population with specified proportion P0 (= 2).
1
 Alternative hypothesis, H1: P ≠ P0 (= 2) i.e. the sample has not been drawn from
1
the binomial population with specified proportion P0 (= 2) (two tailed test).
1
 Alternative hypothesis, H1: P > P0 (= ) (one tailed test).
2
1
 Alternative hypothesis, H1: P < P0 (= ) (one tailed test).
2
Level of significance:
Fix the level of significance 5 % or 1 % unless or otherwise stated.
 For small sample size: n ≤ 25
 Test statistic:
 Under H0
 X0 = Min.(n1, n1)
 Where,
 n1 = No. of observations belonging to the first group.
 n1 = No. of observations belonging to the second group.
 p – value/Critical value:
 The p-value of the test is obtained from the binomial p-value table by using
the formula
𝑋0 1
 p = P(X ≤ X0) = σ𝑥=0 𝐶(𝑛, 𝑥)(2)𝑛
Decision:
1. For one tailed test:
If p-value is less than or equal to alpha, then H0 is rejected and accepted
otherwise.
That is, If p0 ≤ α, then Ho is rejected and if p0 > α, then Ho is accepted.
2. For two tailed test:
If 2p0 value is less than or equal to alpha, then H0 is rejected and accepted
otherwise.
That is, If 2p0 ≤ α, then H0 is rejected and if 2p0 > α then H0 is accepted.

Conclusion:
Same
For large sample size: n > 25
 For large sample size, X0 is normally distributed with mean np and variance npq.
 Test statistic:
 Under H0
𝑋0 − 𝜇 𝑋0 −𝑛𝑝
 Z= = ~ N(0, 1)
𝜎 𝑛𝑝𝑞

 Since 𝑋0 is discrete so that continuity correction is made as

(𝑋0 ±0.5) −𝑛𝑝
 Z= 𝑛𝑝𝑞
use +0.5 if 𝑋0 < np and use – 0.5 if if 𝑋0 > np

 Critical value:
 Same as Z test.
 Decision:
 Same as Z test.

 Conclusion:
 Same
Kolmogorov-Smirnov test:
 Kolmogorov-Smirnov one sample test is another non parametric test used to
test the goodness of fit between observed frequencies and expected
frequencies.
 This test is alternative to chi-square test for goodness of fit when sample size
is small.
 Testing procedure:

 Setting of hypothesis:
 Null hypothesis, H0: The random samples are drawn from the population with
specified distribution F0(x).
 Alternative hypothesis, H1: The random samples are not drawn from the
population with specified distribution F0(x).

 Level of significance:
 Fix the level of significance 5 % or 1 % unless or otherwise stated.
Test statistic:
 Under H0
 𝐷0 = 𝑀𝑎𝑥. 𝐹𝑒 − 𝐹0
 Where,
 𝐹𝑒 = expected relative frequency.
 𝐹0 = observed relative frequency.

 Critical value:
 The critical value (𝐷(𝑛, 𝛼) ) is obtained from the critical value table of
Kolmogorov-Smirnov test.
 Remark:
 For large sample n > 40, the critical value is obtained from the following
table. Level of significance, α 0.20 0.10 0.05 0.02 0.01
Approximate critical values, 𝐷(𝑛,𝛼) 1.07 1.22 1.36 1.52 1.63
𝑛 𝑛 𝑛 𝑛 𝑛
Decision:
 If 𝐷0 ≥ 𝐷(𝑛, 𝛼) then H0 is rejected and accepted otherwise.

 Conclusion:
 Same