Scribd
Upload
3 views19 pages

Non Parametric Methods

Non-parametric tests are distribution-free methods that do not assume a specific probability distribution, making them suitable for ordinal data and skewed variables. They are less powerful than parametric tests but are advantageous when assumptions for parametric methods are not met, and they can handle outliers effectively. Common non-parametric tests include the Mann Whitney U test, Wilcoxon signed-rank test, and Kruskal Wallis ANOVA, each with specific procedures for ranking and hypothesis testing.

Uploaded by

Ankita Ahirwar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
3 views19 pages

Non Parametric Methods

Non-parametric tests are distribution-free methods that do not assume a specific probability distribution, making them suitable for ordinal data and skewed variables. They are less powerful than parametric tests but are advantageous when assumptions for parametric methods are not met, and they can handle outliers effectively. Common non-parametric tests include the Mann Whitney U test, Wilcoxon signed-rank test, and Kruskal Wallis ANOVA, each with specific procedures for ranking and hypothesis testing.

Uploaded by

Ankita Ahirwar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19

Non Parametric Tests of

significance
Introduction
• Non-parametric methods are often referred to as
distribution free methods as they do not rely on
assumptions that the data are drawn from a given
probability distribution.
• Non-parametric methods are widely used for
studying populations that take on a ranked order
(ordinal scale) and skewed outcome variables
Difference between Parametric and Non
parametric
• Parametric test Appropriate measures
of central tendency
Assumes normality of • Non Parametric
population It is called distribution
Requires data in free test
interval Based on ranks
More powerful test Less powerful
Parametric test Non parametric Test

Two independent group Independent sample T test U test

Two paired group Paired T test Wilcoxon’s signed-ranked


test

Three/ more independent One way AVOVA Kruskal Wallis test


test

Three /more repeated Repeated measure ANOVA


measure

Quantify association Pearson’s correlation Spearman’s correlation


between 2 variables

Two/more independent Chi Square


sample (nominal data)
Advantages
• Nonparametric methods require no or very limited
assumptions to be made about the format of the
data, and they may therefore be preferable when
the assumptions required for parametric methods
are not valid.
• Nonparametric methods can be useful for dealing
with unexpected, outlying observations that might
be problematic with a parametric approach.
• Nonparametric methods are intuitive and are simple
to carry out by hand, for small samples at least.
Disadvantages
• Nonparametric methods may lack power as compared
parametric test of significance. This is a particular concern if
the sample size is small or if the assumptions for the
corresponding parametric method (e.g. Normality of the
data) hold.
• Nonparametric methods are geared toward hypothesis
testing rather than estimation of effects. It is often possible
to obtain nonparametric estimates and associated confidence
intervals, but this is not generally straight forward.
• Tied values can be problematic when these are common,
adjustments to the test statistic may be necessary.
Different non parametric methods
• Mann Whitney U test
• Willcoxon test
• Kruskal Walis Anova
Report

Baseline CD4 Count


Tuberculosis Mean N Std. Median Minimum Maximu
Deviation m

No 191.81 1213 140.135 176.00 40 908

Yes 159.06 763 96.710 143.00 57 684

Total 179.16 1976 126.151 163.00 40 908


Mann Whitney U test
• It is an non parametric analogue of
independent sample t test.
• Outcome of interest should be ordinal or
continuous(non normal)
• Objective is to compare medians of two
independent groups.
• observations in both groups are random
samples and have unknown distribution.
procedure
• Rank all the observations in increasing order
of magnitude, ignoring which group they
come from, if two observations have same
magnitude, regardless of group, then they are
given average ranking.
• b) Add up the ranks in the smaller of the
two groups (S). This is our test statistic.
• c) One can calculate an asymptotic or an
exact P value based on the test statistic.
Exact Inference(permutation test)
• Combine m+n observations into one group and rank the
observations from smallest to largest. Find the observed
rank sum S for the smallest of one of the groups.
• b) Find all possible permutations of the ranks in which
m ranks are assigned to group 1 and n ranks to group 2.
There can be (m+n)!/(n!m!) number of possible
combinations.
• c) For each permutation of the ranks, find the sum of
ranks for smallest size of the group I (or) II.
• Find the exact P value as
number of ranksum observed ranksum W
Puppertail 
 
n
m n
Example
comparison of change in cd4 cell count after
initialization of ART
Willcoxon sign Rank Test
It is a non parametric test used to compare
ordinal/ non normal variable collected at two
different time points. Hypothesis of interest is
to test median of difference of ordinal or non
normal variable collected at two different time
points is zero (or) not.
Procedure

• Calculate the difference of outcome variable collected at two


different time points
• Rank the observation from minimum to maximum ignoring
sign. Smaller rank is given for smaller number.
• Compute SR+, Sum of ranks of positive differences
• Compute SR+ for all 2n possible assignment of plus and minus
signs to the ranks of the absolute value of differences.
• The upper tail p value is the fraction of the SR+ that are greater
(or) equal to SR+ obs.
• The two tail p value is double the one tail p value
Example
• Change in CD4 cell counts from baseline to
one year
Kruskal Wallis ANOVA
• It is non parametric analogue of one way
Anova used when basic assumptions of one
way ANOVA is violated (or) out come variable
is ordinal in nature.
• This test is based on ranks.
Data layout

Treatment Ranks Sample size Mean


x11 - - - x1n1
1 N1 R1
R11 - - - - R1n1

2 x21 - - - x2n2 N2 R2
|
|
|
xk1 - - - xknk
K Nk
Rk1 - - - Rknk Rk
Procedure
• Combine data from all groups and give rank from
smaller to largest value give rank 1 for smaller
observations.
R1 , R2      Rk
• Let be the mean ranks of groups
1 ---- k. Then Kruskal Wallisk test statistic2 is given by
12  N 1 
KW   N
N N  1 i 1 
i  Ri 
2 

2
k
 N 1
 ni  Ri  
With i 1  is the treatment sum of squares with
2 

(N+1) /2 is the mean of all ranks.


• The test statistic follows chi-square distribution with
k – 1 degree of freedom.
Thank you

You might also like