Non-Parametric Methods

Presented by Hasnat Israq

Islamic University, Bangladesh

1
 Contents
 Classification of Hypothesis Test
 Basic Concept of Non-Parametric Test
 Assumptions
 Difference between Parametric & Non-parametric Test
 Why do we use Non-parametric Test
 Goodness of Fit Test
 Types
 Chi-square Goodness of Fit Test
 Kolmogorov-Smirnov Test
 Empirical Distribution Function
 Properties of Empirical Distribution Function
 Tests Based on Run
 Run & Length of the Run
 Different Types of Test Based on Runs
 Sign Test
 Rank Order Statistics
 Linear Rank Statistics
 Difference between Rank Order Statistics & Linear Rank Statistics
 Classification of
Hypothesis Test

Parametric Test Non-Parametric Test

t-Test, f-Test, z-test Kolmogorov-Smirnov

etc test, Rank Sum Test
 Basic Concept of Non-Parametric Test
 Non-parametric Test:
Don’t make any assumption about the form of the frequency function of the
parent population.
Population parameters are unknown.

 Assumptions:
The assumptions for Non-parametric tests are given below-
– Sample observations are independent.
– Lower order moments exist.
– Population is symmetrical.
.
Parametric Test Vs Non-parametric Test
SL. NO. Parametric Test Non-parametric Test
1. Information about population is No information about the population is
completely known. available.
2. Basic assumptions are made from the No assumptions are made regarding the
parent population is normal. population.
3. Null hypothesis is made on parameters The null hypothesis is free from
of the population distribution. parameters.
4. They can be used when the data are They can be used when the data are
Interval and ratio. nominal or ordinal.
Why do we use Non-parametric test ?
– Readily comprehensible, very simple and easy to apply.
– Used to test hypothesis that don’t involve population parameters.
– Results are needed in a hurry and calculations must be done in hand.
– Researchers with minimum preparation in mathematics and statistics
usually find the concepts and methods of Non-parametric procedures easy
to understand.
 Goodness of Fit Test
 Goodness of Fit Test:
– Goodness of fit test is a testing procedure of nonparametric test.
– To check the compatibility of a set of observed sample values with a normal
distribution or any other distribution.
– These tests are designed for a null hypothesis about the form of the cumulative
distribution function or probability function of the parent population from which
the sample is drawn.
Types of
Goodness of Fit Test

Chi-square Kolmogorov-Smirnov
Goodness of Fit Test Test
 Chi-square Goodness of Fit Test
 Chi-square Goodness of Fit Test:
– Designed for the null hypothesis concerning the discrete distribution
– Compares the observed frequencies with the frequencies expected under the
null hypothesis.

 Hypothesis:
A single random sample size n is drawn from a population with unknown
F ₓ function . Now we wish to test the null hypothesis
cumulative distribution Hₒ:
for allF, where
ₒ is completely specified, against alternativeH₁:
for all x.

Decision Rule:
The test statistics is less than null hypothesis then we accept it otherwise reject.
 Kolmogorov-Smirnov Test
 Kolmogorov-Smirnov Test:
The Kolmogorov-Smirnov one sample statistic is based on the differences between
the hypothesized cumulative distribution function and the empirical distribution
function of the sampleS ₙ(x )for all x. The test statistic is

 Hypothesis:
Assume we have the random sample x₁, x₂, x₃,…..,xₙ we want to test the hypothesis
Hₒ: for all x where is completely specified continuous distribution,
against alternative forH₁:
all x.
 Empirical Distribution Function
 Empirical Distribution Function:
The cumulative relative distribution function of a random sample is called the
empirical distribution function, may be considered an estimate of the population
cdf for the given observed values.

{ }
0
if x<X ₍₁ ₎
k
S ₙ (x )= if X ₍ ₖ ₎ < x < X ₍ ₖ ₊ ₁ ₎ ; k = 1 , 2 , … , n − 1
n
if x≥ X ₍ ₙ ₎
1

 Properties of Empirical Distribution Function:

– Sₙ(x) is sometimes called the statistical image of the population.
– It is a random variable.
– It is a consistent estimator of Fₓ(x).
 Tests Based on Run
 Run:
– A run is defined to be a succession of one or more identical symbols.
– The number of elements in a run is referred to as the length of the run.
– The maximum number of elements is known as the longest run.

 Example:
Suppose we observe the arrangement of five males and five females in the
line to be
M FF MMM F M FF
Here the number of run is 6, the longest run is MMM and length of the
longest run is 3.

Continued…
 to be Continued…

 Different types of Tests Based on Runs:

Tests of randomness of an ordered sequence can be tested by the theory of runs.
Types of test based on runs are
– Test based on the total number of runs.
– Test based on the length of the longest run.
– Test based on runs up and down.
– Test based on ranks.

 Where we use run test?

Run analysis useful in time series analysis and quality control studies.
 Sign Test

 Sign test:
– Based on the sign (+ or -) of observed difference.
– Used to test the probability of a (+) sign equal to the
probability of sign (-).
– Simplest nonparametric test.

 Assumptions:
– Observations are independent.
– Observation come from symmetrical distribution.
 Rank Order Statistics
 Rank Order Statistics:
The rank order statistics for a random sample are any set of constants which
indicate the order of the observations.

If Xᵢ(i=1,2,3,……..N) be a random sample then the rank order statistics for these
random samples are r(xᵢ)

 A functional form of the rank order statistics,

𝑁 𝑁
r(xᵢ)=∑ 𝑆 ( x ᵢ − =1+
x ⱼ) ∑ 𝑆 ( x ᵢ− xwhere,
ⱼ)
𝑗=1 𝑗 ≠𝑖

Continued…
 to be Continued…

 Rank order statistics follows discrete uniform distribution and it’s distribution is
defined as,
P[r(xᵢ)=j]= ; j=1,2,3,…N

 It is distribution free.

 It is usually useful in Non-parametric inference.

 Linear Rank Statistics
 Linear Rank Statistics:
Many commonly used two sample rank tests can be classified as linear
combinations of certain indicator variables for the combined order samples. Such
functions are often called linear rank statistics.

A linear function of this indicator variable is called a linear rank statistics and thus
the linear rank statistics can be written as,
N

∑ a ᵢ z ᵢ T (z)=
N

j=1
Rank Order Statistics Vs Linear Rank Statistics

SL. NO. Rank Order Statistics Linear Rank Statistics

1. Rank order statistics is used in single It is used for two samples problem of
sample problem. combined order sample.
Rank order statistics can’t be expressed It can be expressed in terms of linear
2. in terms of indicator variable. combination of an indicator variable for
combined sample.
3. This test provides the information of the This test provides the information about
single population. difference between two populations.
 Thank you
For More
Email : [email protected]
LinkedIn : hasnatrbm