0% found this document useful (0 votes)
70 views22 pages

Chapter 17

This document discusses various nonparametric statistical tests that do not rely on assumptions of normality. It describes the Mann-Whitney rank-sum test, sign test, Wilcoxon matched-pairs signed-ranks test, runs test, binomial test, Kolmogorov-Smirnov one-sample test, one-sample chi-square test, Friedman one-way ANOVA, and K-sample median test. Examples are provided to demonstrate how to conduct and interpret each test using the grade.sav dataset.

Uploaded by

Newte
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
70 views22 pages

Chapter 17

This document discusses various nonparametric statistical tests that do not rely on assumptions of normality. It describes the Mann-Whitney rank-sum test, sign test, Wilcoxon matched-pairs signed-ranks test, runs test, binomial test, Kolmogorov-Smirnov one-sample test, one-sample chi-square test, Friedman one-way ANOVA, and K-sample median test. Examples are provided to demonstrate how to conduct and interpret each test using the grade.sav dataset.

Uploaded by

Newte
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 22

Nonparametric

Chapter 17 Procedure

Instructor: Fei Tian


No:2

Content
• Nonparametric tests
• Step by step
• Mann-Whitney Rank-Sum Test
• The sign test
• Wilcoxon Matched-Pairs Signed-Ranks test
• The Runs test
• The Binomial test
• The Kolmogorov-smirnov one-sample test
• The One-sample chi-square test
• The Fridman one-way ANOVA
• The K-Sample median test
Department of Sociology, AU Fei Tian
No:3

Nonparametric tests
• What is parametric test?
• It is one that is based on certain parameters.
• The critical parameter that most of the procedures describ
ed in this book are based on is that data from samples ( an
d the populations from which they are drawn) are normall
y distributed (or something close to normal).
• Although some operations are based on other assumptions
or parameters (e.g., binomial or Possion distributions of d
ata), the Nonparametric Tests procedure deals with primar
ily with populations that are not normally distributed and
considered how to conduct statistical tests if the assumptio
n of normality is violated.
Department of Sociology, AU Fei Tian
• For data that are not normally distributed the
No:4

nonparametric tests use other statistical techniques


to test hypotheses.
• These techniques may include, among others,
analyses based on:
– Ranked values
– summation of how many values in one distribution are larger (or
smaller )than values in another distribution.
– Use of weighted comparisons.
– Tests of determine whether a distribution of values deviates from
randomness or is binomially distributed.
– Single-group tests of deviation from normality.
– Comparisons of frequencies.
– Calculation of the frequency of values above or below a grand
median to compare
Department groups.
of Sociology, AU Fei Tian
No:5

Nonparametric tests
• Despite this seeming complexity, most nonparame
tric tests are quite understandable and easy to cond
uct.
• The old data set grade.sav will be used again.
• The output and step by step section will be combin
ed into one for each test.
• Nine different nonparametric tests will be showed
one by one:

Department of Sociology, AU Fei Tian


No:6

Nonparametric tests
1. Mann-Whitney Rank-Sum Test:
a test of whether two groups differ from each other based o
n ranked scores. (independent t test)
2. The sign test:
Tests whether two distributions differ based on a comparis
on of paired scores. (paired t test)
3. Wilcoxon Matched-Pairs Signed-Ranks test:
The same as the sign test except the positive and negative s
ign are weighted by the mean rank of positive versus ne
gative comparison.
Department of Sociology, AU Fei Tian
No:7

Nonparametric tests
4. The runs test:
Tests whether the elements of a single dichotomous grou
p differ from a random distribution.
5. The binomial test
Tests whether the elements of a single dichotomous grou
p differ from a binomial distribution (each outcome eq
ually likely)
6. The kolmogorov-smirnov one-sample test
Tests whether the distribution of then members of a singl
e group differ significantly from a normal (or uniform
, or Possion) distribution.
Department of Sociology, AU Fei Tian
No:8

Nonparametric tests
7. The ons-sample chi-square test
Tests whether observed score differ significantly from ex
pected scores for level of a single variable.
8. The fridman one-way ANOVA
Tests whether three or more groups differ significantly fr
om each other, based on the average rank of groups ra
ther than the distribution of values.
9. The K-Sample median test:
Tests whether two or more groups differ on the number o
f instances (within each group) greater than the media
n value or less than the median value.
Department of Sociology, AU Fei Tian
No:9

Steps to Nonparametric Tests

Department of Sociology, AU Fei Tian


No:10

Mann-Whitney Rank-Sum Test


• If the distributions do not differ significantly from normal,
then the t test should be used because it has greater power.
• In our example we consider whether females and males
differ significantly on their scores on the final exam.
• Mann-Whitney procedure rank all 105 scores, determines
the rank of each subject, and then computes the average
rank for the two groups.
• Clearly the group with the higher average rank scored
higher on the test.
• The U test determines whether that difference is
significant.
Department of Sociology, AU Fei Tian
No:11

Department of Sociology, AU Fei Tian


No:12

Department of Sociology, AU Fei Tian


No:13

Ranks

gender N Mean Rank Sum of Ranks


final Female 64 55.81 3572.00
Male 41 48.61 1993.00
Total 105

Test Statisticsa

final
Mann-Whitney U 1132.000
Wilcoxon W 1993.000
Z -1.184
Asymp. Sig. (2-tailed) .237
Women didn’t score
a. Grouping Variable: gender significantly higher
than men

Department of Sociology, AU Fei Tian


No:14

The sign test


• The sign tests utilizes pairwise comparison of two differen
t distributions to identify which is larger than which, and t
hen from this information it determines if the two distribut
ions differ significantly from each other.
• By default, the sign test compares the second distribution
with the first distribution.
• For the first subject, quiz1 was 9 and quiz2 was 7, so this
would rate as a negative difference.
• The sign test sums all the positives, negatives, and ties an
d then computes a z score and a p value associated with th
e frequency of the positive and negatives.

Department of Sociology, AU Fei Tian


Wilcoxon Matched-Pairs Signed-
No:15

Ranks test
• The difficulty with the sign test is that a difference
between paired quizzes of 10 (10 on one, 0 on the other)
and a difference of 1 (1.g., 6 on one, 5on the other) will be
coded identically (as a negative).
• This test incorporates information about the magnitudes
of the differences between paired values.
• To compute this values, first the magnitude of the
differences (ignoring the signs) are ranked from high to
low. Then the ranks for the negative signs (quiz2<quiz1)
are summed and averaged, and the ranks for the positive
signs (when quiz2>quiz1) are summed and average.
Finally, significance values are calculated based on z
score.
Department of Sociology, AU Fei Tian
No:16

The runs test


• The test is used to see if the elements of a particular
data set are randomly distributed.
• If the sequence HHTHTTHTTHTTTHHHHTH
resulted from flipping a coin, does this sequence
differ significantly from randomness? In other
words, are we flipping a biased coin?
• Unfortunately this procedure works only with
dichotomous data (exactly two possible outcomes).
• We will test if the males and females in our file are
distributed randomly throughout our data set.
Department of Sociology, AU Fei Tian
No:17

The Binomial Test


• The test measures whether a distribution of values is
binomially distributed.
• binomial distribution assumes that any outcome is equally
likely.
• If you tossed an unbiased coin 100 times you would
expect approximately 50 heads and 50 tails.
• We will apply the test to the distribution of males and
females in our data set.
• In fact, we have known that there are 41 men and 64
women, so the use of binomial test is mainly to
demonstrate how the procedure work.
• It is of legitimate interest, however, to see if this
distribution differs significantly from 52.5 men and 52.5
women.
Department of Sociology, AU Fei Tian
The kolmogorov-smirnov one-sa No:18

mple test
• The test is designed to measure whether a particular distri
bution differs significantly from a normal distribution (ske
wness and kurtosis of the distribution =0)
• A uniform distribution ( values are distributed evenly, suc
h as the numbers 1-100 consecutively), a poisson distribut
ion (the values λ equals the mean and the variance of the d
istribution; as λ becomes large, the distribution approxima
tes normality) or an exponential distribution.
• The procedure is based on a comparison of the sample cu
mulative distribution to the hypothesized (normal, unifor
m, or Poisson) cumulative distribution.

Department of Sociology, AU Fei Tian


No:19

The One-Sample chi-square test


• The procedure conducts a noe-sample chi-square test rathe
r than the more traditional chi-square test of crosstabulate
d data.
• The expected values are the simple the total number of ca
ses divided by the number of levels of a variable.
• In grade.sav data, with five level of ethnicity and an N=10
5 the expected value for each cell will be 105/5=21.
• If you are drawing from a population that is 10% Native,
20% Asian, 20% Black, 40% White, and 10% Hispanic, y
ou may test against that distribution by type1, click Add, t
ype 2, click add, and so on.

Department of Sociology, AU Fei Tian


No:20

The Fridman one-way ANOVA


• It is similar to traditional analysis of variance with
two notable exceptions:
– comparisons in the Fridman procedure are based on mean rank o
f variables rather than on means and standard deviations of raw s
cores
– rather than calculating an F ratio, Fridman compares ranked valu
es with expected values in a chi-square analysis.
• The power of Fridman is, of cause, less than ANO
VA.
• The bad news is that no post hoc test and no plann
ed contrasts are allowed.
Department of Sociology, AU Fei Tian
No:21

The K-Sample median test


• It computes the median of two or more distributions and
then comparing whether the number of values below the
grand median (median for all groups) differs from the
number of values above the ground median for each group
compared.
• To demonstrate we will use final score in each of the
section.
• The procedure is to rank order all scores from all three
sections combined to determine the grand median.
• Then for each section the number of score above this
median and the number of scores below this median are
calculated.
• If any section deviates from approximately equal number
of scores above and below the grand median, this would
indicate that some biasing factor may be present for that
section.
Department of Sociology, AU Fei Tian
No:22

Assignment
• Review: chapter 17
• Preview: chapter 18
• Lab exercise: each example presented in the
chapter

Department of Sociology, AU Fei Tian

You might also like