non parametric statistics

Non-parametric statistics
Anchal, BalRam, Kush
Environment Management 2016
USEM

Learning objectives
Compare and contrast parametric and nonparametric tests
Perform and interpret the Mann Whitney U Test
Perform and interpret the Sign test and Wilcoxon Signed Rank
Test
Compare and contrast the Sign test and Wilcoxon Signed Rank
Test
Perform and interpret the Kruskal Wallis test

INFERENTIAL STATISTICS
PARAMETRIC
STATISTICS
NON-
PARAMETRIC
STATISTICS

Parametric Non-parametric
Assumed distribution normal any
Typical data Ratio or interval Nominal or ordinal
Usual central measures mean Median
Benefits Can draw many
conclusions
Simplicity less affected by
outliers
Tests
Independent measures, 2
groups
Independent measure t
test
Mann- whitney test
Independent measures, >2
groups
One way independent
measures ANOVA
Kruskal wallis test
Repeated measures, 2
conditions
Matched pair t-test Wilcoxon test
Table: Distinguish between parametric and non parametric statistics

Nonparametric tests
• Also known as distribution-free tests because they are based on
fewer assumptions (e.g., they do not assume that the outcome is
approximately normally distributed).
• Parametric tests involve specific probability distributions (e.g., the
normal distribution) and the tests involve estimation of the key
parameters of that distribution (e.g., the mean or difference in
means) from the sample data.
• There are some situations when it is clear that the outcome does
not follow a normal distribution. These include:
when the outcome is an ordinal variable or a rank,
when there are definite outliers or
when the outcome has clear limits of detection.

Non-parametric Methods
• Sign Test
• Wilcoxon Signed-Rank Test
• Mann-Whitney-Wilcoxon Test
• Kruskal-Wallis Test

How to assign the ranks
Ordered observed data 0 2 3 5 7 9
ranks 1 2 3 4 5 6
ranks 1 2 3 4.5 4.5 6
ranks 1 2 3 5 5 5

1. Sign Test
• A common application of the sign test involves using a sample
of n potential customers to identify a preference for one of
two brands of a product.
• The objective is to determine whether there is a difference in
preference between the two items being compared.
• To record the preference data, we use a plus sign if the
individual prefers one brand and a minus sign if the individual
prefers the other brand.
• Because the data are recorded as plus and minus signs, this
test is called the sign test.

Example: Butter Taste Test
• Sign Test: Large-Sample Case
As part of a market research study, a sample of 36
consumers were asked to taste two brands of butter and
indicate a preference. Do the data shown below indicate a
significant difference in the consumer preferences for the
two brands?
18 preferred Amul Butter (+ sign recorded)
12 preferred Mother Dairy Butter (_ sign recorded)
6 had no preference
The analysis is based on a sample size of 18 + 12 = 30.

H0: No preference for one brand over the other exists
Ha: A preference for one brand over the other exists
• Sampling Distribution
2.74
Sampling distribution
of the number of “+”
values if there is no
brand preference
 =30/2 =15

• Rejection Rule
Using 0.05 level of significance,
Reject H0 if z < -1.96 or z > 1.96
• Test Statistic
z = (18 - 15)/2.74 = 3/2.74 = 1.095
• Conclusion
Do not reject H0. There is insufficient evidence in the
sample to conclude that a difference in preference exists for
the two brands of butter.

2. Wilcoxon Signed-Rank Test
• The Wilcoxon test is used when we are unwilling to make assumptions
about the form of the underlying population probability distributions,
• but we want compare paired samples.
• Analogous to the dependent t-test we are interested in the difference
• in two measurements taken from each person.
• The rank sum of the positive (T+) and negative (T−) differences are
calculated, the smallest of these is used as the test statistic to test the
hypothesis.
• Two assumptions underlie the use of this technique.
1. The paired data are selected randomly.
2. The underlying distributions are symmetrical.

Small-Sample Case (n ≤15)
• When sample size is small, a critical value against which to
compare T can be found by table, to determine whether the null
hypothesis should be rejected. The critical value is located by using
n and α.
• If the observed value of T is less than or equal to the critical value
of T, the decision is to reject the null hypothesis.

Problem: The survey by American Demographics estimated the
average annual household spending on healthcare. The U.S.
metropolitan average was $1,800. Suppose six families in
Pittsburgh, Pennsylvania, are matched demographically with six
families in Oakland, California, and their amounts of household
spending on healthcare for last year are obtained. The data follow
on the next page.
A healthcare analyst uses α = 0.05 to test to determine whether
there is a signiﬁcant difference in annual household healthcare
spending between these two cities.

STEP 1. The following hypotheses are being tested.
H0: Md= 0
Ha: Md≠ 0
STEP 2. Because the sample size of pairs is six, the small-
sample Wilcoxon matched pairs signed ranks test is
appropriate if the underlying distributions are assumed to
be symmetrical.

Step 3. calculated T = 3 is greater than critical T = 1 (at α=0.05,
the decision is to accept the null hypothesis.
Conclusion: there is no signiﬁcant difference in annual
household healthcare spending between Pittsburgh and
Oakland.

Large-Sample Case (n >15)
For large samples, the T statistic is approximately normally
distributed and a z score can be used as the test statistic.

Problem: Suppose a company implemented a quality-
control program and has been operating under it for 2
years. The company’s president wants to determine
whether worker productivity signiﬁcantly increased since
installation of the program. Use a non parametric statistics
for the following data at α=0.01.

STEP 1. The following hypotheses are being tested.
H0: Md= 0
Ha: Md≠ 0
STEP 2. Wilcoxon matched-pairs signed rank test to be
applied on the data to test the difference in productivity from
before to after.
STEP 3. Computes the difference values and because zero
differences are to be eliminated, deletes worker 3 from the
study. This reduces n from 20 to 19, then ranks the
differences regardless of sign.

STEP 4. This test is one tailed. The critical value is z =-2.33. Here calculate
z is less than critical z, hence we reject the null hypothesis.
Conclusion: The productivity is signiﬁcantly greater after the
implementation of quality control at this company.

3. Mann-Whitney-Wilcoxon Test
(U test)
• Also known as Wilcoxon rank sum test.
• It is a nonparametric counterpart of the t test used to compare the
means of two independent populations.
• The following assumptions underlie the use of the Mann-Whitney
U test.
1. The samples are independent.
2. The level of data is at least ordinal.
• The two-tailed hypotheses being tested with the Mann-Whitney U
test are as follows.
H0: The two populations are identical.
Ha: The two populations are not identical.

Mann-Whitney-Wilcoxon Test
(U test)
• Computation of the U test begins by arbitrarily designating two
samples as group 1 and group 2. The data from the two groups are
combined into one group, with each data value retaining a group
identiﬁer of its original group. The pooled values are then ranked
from 1 to n.
• W 1 and W2 are the sum of the ranks of values from group 1 and
group 2 respectively.
• Small sample case: when both n1 and n2 ≤10
• Large sample case: when both n1 and n2 >10

Small sample case
• The test statistic is the smallest of these two U values.
• Determine the p-value for a U from the table.
• Note: For a two-tailed test, double the p-value shown in the table.

• Because U2 is the smaller value of U, we use U = 3
as the test statistic. Because it is the smallest size,
let n1= 7; n2= 8.
• STEP 7. from table find a p-value of 0.0011. Because
this test is two tailed, we double the table p-value,
producing a ﬁnal p-value of .0022. Because the p-
value is less than α = 0.05, the null hypothesis is
rejected.
• The statistical conclusion is that the populations are
not identical.

Large-Sample Case
• For large sample sizes, the value of U is approximately normally
distributed and hence compute a z score for the U value. A
decision is then made whether to reject the null hypothesis. A z
score can be calculated from U by the following formulas.

4. Kruskal Wallis Test
• Like the one-way analysis of variance, the Kruskal-Wallis test is
used to determine whether c ≥3 samples come from the same or
different populations.
• The Kruskal-Wallis test is based on the assumption that the c
groups are independent and that individual items are selected
randomly. The hypotheses tested by the Kruskal-Wallis test follow.
H0 :The c populations are identical.
Ha: At least one of the c populations is different.

Kruskal Wallis Test
o This test determines whether all of the groups come from the
same or equal populations or whether at least one group comes
from a different population.
o The process of computing a Kruskal-Wallis K statistic begins with
ranking the data in all the groups together, as though they were
from one group.

Start
Are the samples
independent? Use Mann
whitney U test
Are the data
atleast interval ?
Use sign test
Use Wilcoxon
signed rank test
yes
yes
no
no
Deciding which test to use

Advantages of
Nonparametric Tests
• Used with all scales
• Easier to compute
— Developed originally before wide
computer use
• Make fewer assumptions
• Need not involve population
parameters
• Results may be as exact as
parametric procedures
.

Disadvantages of
Nonparametric Tests
• May waste information
— If data permit using parametric
procedures
— Example: converting data from
ratio to ordinal scale
• Difficult to compute by hand for
large samples
• Tables not widely available
.

• Jarkko Isotalo, Basics of Statistics (Available online at:
http://www.mv.helsinki.fi/home/jmisotal/BoS.pdf)
• Ken Black, 6th edition, Business Statistics For Contemporary Decision
Making
• Lisa Sullivan, Non parametric statistics, Boston University School of Public
Health (available online at:
http://sphweb.bumc.bu.edu/otlt/MPHModules/BS/BS704_Nonparametri
c/BS704_Nonparametric_print.html)
• Arora, P.N and Malhan P.K; Biostatistics, 2009 Edition
• http://blog.minitab.com/blog/adventures-in-statistics/choosing-
between-a-nonparametric-test-and-a-parametric-test

non parametric statistics

More Related Content

What's hot (20)

Viewers also liked (10)

Similar to non parametric statistics (20)

More from Anchal Garg (8)

Recently uploaded (20)

non parametric statistics