3. CONTENT
• HISTORY
• KRUSKAL-WALLIS TEST
• PRE-REQUISITES
• ASSUMPTIONS MADE
• LIMITATIONS
• USES
• ADVANTAGES
• DRAWBACK
• EXAMPLES
• REFERENCES
4. HISTORY
• Described for the first time by William Henry Kruskal
&
Wilson Allen Wallis
• They used this test for quantitative data & it tests the equality of population
medians among group
5. KRUSKAL-WALLIS TEST
WHAT
Non parametric Analysis Of Variance in one way
classification (data replaced by their ranks)
Extension of Mann-Whitney U test for three or more
groups
6. WHY
To Test Homogeneity of samples drawn from the same population
To test the equality of population medians among groups
7. WHEN
When there are three or more comparable groups
The number of observations in each comparable group is 5 or more
The distribution approximates the chi-square distribution
When the assumptions of one-way ANOVA are not met
When there is one nominal variable & one measurement/ ranked
variable
9. PRE REQUISITES
There must be three or more groups of observations
The observations should be measured on a ratio or interval scale
Number of observations in each group should be 5 or more
10. ASSUMPTIONS MADE
The groups are comparable but independent of each other
Within each group, the observations are independent of each
other but identically distributed
11. LIMITATION
When we substitute ranks for the original values, we lose
information which makes KW test somewhat less powerful than
one-way ANOVA
12. USES
For small sample size
Distribution of population from where the sample is drawn, is not
normal
13. ADVANTAGES
Simple to understand & easy to calculate
Based on observations
No information is deleted or omitted
Totally numerical based & quantitative test
15. EXAMPLE
• The serum protein levels in gm/dl of 15 cases with Grade I, II, & III under nutrition
are given below:
Test whether differences in serum protein levels are statistically significant at 5% l.o.s.
Grade I Grade II Grade III
6.8 6.2 5.6
7 6.3 5.5
7.2 6.5 5.4
7.1 6.7 5.7
7.4 7.3 6.9
16. SOLUTION
• H0 - There is no difference in the serum protein levels of the 3
groups
• H1- There is significant difference in the serum protein levels of
the 3 groups
17. Serum Protein (gm/dl) Rank Grade
5.4 1 III
5.5 2 III
5.6 3 III
5.7 4 III
6.2 5 II
6.3 6 II
6.5 7 II
6.7 8 II
6.8 9 I
6.9 10 III
7 11 I
7.1 12 I
7.2 13 I
7.3 14 II
7.4 15 I
19. REFERENCES
• J. Dixit, text book of Principles and Practice of Biostatistics, edition 7 pg. 257
• K. S. Negi, text book of methods in biostatistics, pg. 204 to 210
• S. Kartikeyan, comprehensive text book of biostatistics & research methodology, 1st
edition pg. 365 to 367
• W. Daniel, text book of Biostatistics basic concepts and methodology for the health
sciences, 10th edition pg.704 to 709
• http://www.statisticssolutions.com/kruskal-wallis-test/
• https://www.ncbi.nlm.nih.gov/pubmed/9413454
*William Henry Kruskal was an American mathematician and statistician. He is best known for having formulated the Kruskal–Wallis one-way analysis of variance, a widely used nonparametric statistical method
&
The Wilson Allen Wallis was an American economist and statistician who was the companion of Kruskal…..
Here I’m trying to explain this particular statistical test in a different way like ….. What is the test is???? Why we use this test???? When we use the test????
How it is Being used???
*It is a non parametric counterpart of one way anova…….
* It resembles one way anova , only that , the data is replaced by their ranks………
**e.g. for Ranked……Dominance hierarchies (in behavioral biology) and developmental stages are the ranked variables…….
Step 1: state the null hypothesis…………… which will be,, there is no significant difference between the medians of the groups……….
Step 2: state the altenative hypothesis……..
Step 3: assigning of ranks to all the observations…….. 1st arrange all observations in ascending order & then give the rank from smallest one……….xxx
Step 4: calculate the sum of ranks of each group……….
Step 5: calculate Kruskal-wallis statistic from the given formula………
n. = total no. of observations in all samples together……….
Ri = sum of ranks in the ith sample.
Step.6: compare the calculated value of KW with Tabulated value from chi-square table……. Where degree of freedom is equal to
(k – 1) where ‘k’ is the no. of groups…….
With the level of significance 0.05…..
Inference: if the calculated value of KW statistic (H) is less than the tabulated value from chi-square table accept the H0…….
& if it is more then reject the H0…………
There are some pre-requesites for application of this test
There are some assumptions like…….
Like all other tests this test also have some limitations………..
2. Being a non parametric test it is less powerfull than one way anova if population follows normal distribution…….
Used when the sample size is small & the sample is drawn from a population that does not have a normal distribution.
We can not perform this test for very large data……
Now Dr. Kalyani will explain the test with examples…..
Grades of under nutrition are comparable but independent of each other & the no. of observations in each group is 5 hence KW test can be used…………..