Chapter 6 Test For K Independent Samples

TESTS FOR k
INDEPENDENT SAMPLES
STAT 323
Nonparametric Statistics
Department of Mathematics & Statistics

College of Arts & Sciences
University of Southeastern Philippines
k INDEPENDENT SAMPLES
- Tests included in this chapter are used if we
are interested to determine whether there is a
significant difference among three or more
independent groups or samples.
- These tests are used if assumptions of

parametric tests are not satisfied; or if the
measurement of data is weaker than interval; or if
the researcher wishes to avoid making restrictive
assumptions of the F test and thus to increase the
generality of his findings.
Tests for k Independent

Samples Page 2
Extension of Median Test
Median Test
- is deigned to examine whether several samples
came from populations having the same median.
Note: The notation c is simply the number of samples thus c = k.

Samples Page 3
Let a equal the total number of observations above

the grand median in all samples and let b equal the
total number of values less than or equal to the grand
median. The a + b = N, the total number of
observations.

Samples Page 4

Samples Page 5
Ho: All c (or k) populations have the same median.
Ha: At least two of the populations have different

medians

Samples Page 6

Samples Page 7

Samples Page 8
Null Distribution
The exact distribution of T is difficult to

tabulate, so large sample approximation is used to
approximate the distribution of T. The approximate
null distribution is the chi-squared distribution with
c – 1 degrees of freedom.

Samples Page 9

Samples Page 10

Samples Page 11

Samples Page 12
Power-Efficiency
Mood (1954) has shown that when the median

test is applied to data measured in at least an
interval scale from normal distributions with
common variance (i.e., data that might properly be
analyzed by the parametric t test), it has the same
power-efficiency as the sign test. That is, its power-
efficiency is about 95% per cent for n1 + n2 as low as
6. This power-efficiency decreases as the sample
size increases, reaching an asymptotic efficiency of
2/𝜋 = 63 per cent.
Samples Page 13
Example.
Four different methods of growing corn were

randomly assigned to a large number of different
plots of land and the yield per acre was computed
for each plot. In order to determine whether there is
a difference in yields as a result of the method used,
use median test procedure.

Samples Page 14

Samples Page 15
Solution.
1) Ho: All methods have the same median yield per

acre.
Ha: At least two of the methods differ with respect to

the median yield per acre.
2) ∝ = 0.05

Samples Page 16
3) Median Test for c (or k) independent samples
4) Computations (using SPSS with computations from

the book)

Samples Page 17
Computation (Con’t.)

Samples Page 18
5) Critical Region: Reject Ho if 𝒳 2 > 𝒳 2 0.05,3 = 7.815.

Samples Page 19
6) Decision/Conclusion: Since 17.6 > 7.815 or the p-

value associated to the 17.6 is less than 0.01, we
reject Ho and conclude that there is a difference in the
yield of the four methods.
Thus a multiple comparison test is needed.

Samples Page 20
Comparison of Method 1 versus Method 2

(Not significant)

Samples Page 21

(Significant)

Samples Page 22

(Significant)

Samples Page 23

(Significant)

Samples Page 24

(Significant)

Samples Page 25

(Significant)

Samples Page 26
It can be observed that only in Method 1 and

Method 2 that the p-value is less than 0.05, the rest
of the comparisons have p-values less than 0.05.
This result is the same if tabulated value is
used 𝒳 2 0.05,1 = 3.841.

Samples Page 27
Note: Such repeated testing procedures are for

one’s satisfaction or for use as an objective
“yardstick” for separating the various population but
they cannot receive the same interpretation as may
legitimate to the first test, overall test since multiple
comparison distorts the true level of significance.
Refer to the work of Gabriel (1966) and Knoke (1976).
Note: The implementation of the extension of Median

Test is similar with the process in median test with two
independent samples.
Samples Page 28
Multiple Comparison:

Samples Page 29

Samples Page 30
Procedure in SPSS:

Samples Page 31

Samples Page 32
Kruskal-Wallis Test
Kruskal-Wallis Test
- It is a rank-based nonparametric test that

can be used to determine if there are statistically
significant differences between two or more groups
of an independent variable on a continuous or
ordinal dependent variable.
- It is considered the nonparametric

alternative to the one-way ANOVA (sometimes also
called the "one-way ANOVA on ranks"), and an
extension of the Mann-Whitney U test to allow the
comparison of more than two independent groups.
Samples Page 33
Kruskal-Wallis Test

Samples Page 34
Kruskal-Wallis Test

Samples Page 35
Kruskal-Wallis Test

Samples Page 36
Kruskal-Wallis Test

Samples Page 37
Kruskal-Wallis Test

Samples Page 38
Kruskal-Wallis Test

Samples Page 39
Kruskal-Wallis Test

Samples Page 40
Kruskal-Wallis Test

Samples Page 41
Kruskal-Wallis Test
Example.
Four different methods of growing corn were

randomly assigned to a large number of different
plots of land and the yield per acre was computed
for each plot. In order to determine whether there is
a difference in yields as a result of the method used,
use Kruskal-Wallis test procedure.

Samples Page 42
Kruskal-Wallis Test

Samples Page 43
Kruskal-Wallis Test
Sol’n.
1. Ho: The four methods are equivalent

Ha: Some methods of growing corn tend to
furnish higher yields than other.
2. ∝= 0.05
3. Kruskal-Wallis Tesr

Samples Page 44
Kruskal-Wallis Test
4. Computation:

Samples Page 45
Kruskal-Wallis Test

Samples Page 46
Kruskal-Wallis Test
5) Critical Region: Reject Ho if 𝒳 2 0.05,3 = 7.815
6) Decision/Conclusion:
Since 𝒳 2 = 25.46 > 𝒳 2 0.05,1 = 7.815, we

reject Ho and conclude that some methods of
growing corn tend to furnish higher yields than
other.

Samples Page 47
Kruskal-Wallis Test
Multiple Comparison:

Samples Page 48
Kruskal-Wallis Test

Samples Page 49
Kruskal-Wallis Test
Using SPSS:

Samples Page 50
Kruskal-Wallis Test

Samples Page 51
Kruskal-Wallis Test
Procedure in SPSS:

Samples Page 52
Kruskal-Wallis Test

Samples Page 53
Jonckheere-Terpstra Test
- is
a rank-based nonparametric test that can be
used to determine if there is a statistically significant
trend between an ordinal independent variable and a
continuous or ordinal dependent variable.
- It tests for an ordered difference in medians

where you need to state the direction of this order.
- It is also known as the Jonckheere-Terpstra test

for ordered alternatives.
Samples Page 54
Note:
The Jonckheere-Terpstra test is similar to the Kruskal-

Wallis Test, which can be used to determine if there are
statistically significant differences between two or more
groups of an independent variable on a continuous or
ordinal dependent variable. However, unlike the
Jonckheere-Terpstra test, the Kruskal-Wallis H test does
not predict how the differences in the scores of the
dependent variable will depend on the ordinal nature of
the groups of the independent variable

Samples Page 55
Examples
Use a Jonckheere-Terpstra test to understand whether

test scores, measured on a continuous scale from 0-
100, differed based on time spent revising. We expect
that median test scores increases with increasing hours
spent revising
Use Jonckheere-Terpstra test to understand whether

job satisfaction, measured on an ordinal scale, differed
based on job position. We expect job satisfaction to
increase with higher job position.

Samples Page 56
Assumptions:
1. The dependent variable should be measured at

the ordinal or continuous level.
2. The independent variable should consist of two or

more ordinal, independent groups. Typically, a
Jonckheere-Terpstra test is used when there are three
or more ordinal, independent groups, but it can be used
for just two groups.
3. There should be independence of observations,

which means that there is no relationship between the
observations in each group or between the groups
themselves.
Samples Page 57
4. In order to know how to interpret the results from a
Jonckheere-Terpstra test, you have to determine
whether the distributions in each group (i.e., the
distribution of scores for each group of the independent
variable) have the same shape and the same
variability.
5. You need to predict, a priori, the order of the groups

of the independent variable when running the
Jonckheere-Terpstra test (i.e., the term, "a priori", simply
means that you have to predict this order "before" you
collect your data, typically when you set the research
questions/hypotheses for your study).

Samples Page 58
6. You need to predict, "a priori", the direction of the
alternative hypothesis. The term "a priori" simply means
that you need to predict the direction of the alternative
hypothesis without looking at your data (i.e., "before"
looking at your data)..

Samples Page 59
Note: Kruskal-Wallis test is sensitive against any differences in

means, while this usage of Spearman’s 𝜌 or Kendall’s 𝜏 is sensitive
against only the ordering specified in the H1 given above. When
Kendall’s 𝜏 is used, this test is equivalent to the Jonckheere-
Terpstra test.
Samples Page 60
Procedure:
1. If there are k samples, calculate the sum, U, of all

Mann-Whitney statistics Urs relevant to the rth
sample (r = 1, 2, … , k – 1) and any sample s for
which s > r.
2. Add all Urs to give a total of U.
3. Calculate the probability associated to the observed

sum Urs. Then compare the p-value to the level of
significance to determine whether to reject or accept
Ho.
Samples Page 61
4. For large N and the ni not too small the distribution of
𝑈−𝐸(𝑈)
Z=
𝑉𝑎𝑟(𝑈)
is approximately standard normal where
1
𝐸 𝑈 = 𝑁 2 − σ𝑖 𝑛𝑖2 and
4
1
𝑉𝑎𝑟 𝑈 = 72 𝑁 2 2𝑁 + 3 − σ𝑖 𝑛𝑖2 (2𝑛𝑖 + 3)
Note: The expression for Var(U) needs adjustment for ties

but the adjustment is trivial for few ties. For many ties,
Var (U) is given by Lehmann (1975, p. 235)
Samples
Page 62
Example. Hinkley (1989) gives breaking distances taken
by motorists to stop when travelling at various speeds. A of
subset of this data is:
Speed (mph) Breaking distances (feet)

20 48
25 33 59 48 56
30 60 101 67
35 85 107
Use the Jonckheere-Terpstra test to assess the evidence

for a tendency for braking distance to increase as speed
increases.
Samples Page 63
Sol’n.
1) Ho: F1(x) = F2(x) = F3(x) = F4(x)

Ha: F1(x) ≥ F2(x) ≥ F3(x) ≥ F4(x) with at least one
inequality
2) 𝛼 = 0.05
3) Jonckheere-Terpstra test

Samples Page 64
4) Computations:
Speed (mph) Breaking distances (feet)

20 48
25 33 59 48 56
30 60 101 67
35 85 107 This refers to the number of
data in sample 2 that is
U12 = 1 + 0.5 + 1 = 2.5 greater than or equal to the
data in sample 1. In this
U13 = 1 + 1 + 1 = 3 case U12 = 2.5 since there is
U14 = 1 + 1 = 2 one tie {48} and two data
{59, 56} are greater than 48.
U23 = 3 + 3 + 3 + 3 = 12
U24 = 2 + 2 + 2 + 2 = 8
U34 = 2 + 1 + 2 = 5
U = 32.5
Samples Page 65
5) Decision/Conclusion: Since p-value for both one-tailed
and two-tailed test is less than 0,05, we reject Ho. Thus,
we conclude that increasing speed increases braking
distance.

Samples Page 66
Procedure in SPSS:

Samples Page 67

Samples Page 68

Chapter 6 Test For K Independent Samples

Uploaded by

Copyright:

Available Formats

Chapter 6 Test For K Independent Samples

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 6 Test For K Independent Samples

Uploaded by

Copyright:

Available Formats

TESTS FOR k

Department of Mathematics & Statistics

- These tests are used if assumptions of

Tests for k Independent

Note: The notation c is simply the number of samples thus c = k.

Let a equal the total number of observations above

Tests for k Independent

Tests for k Independent

Ho: All c (or k) populations have the same median.

Ha: At least two of the populations have different

Tests for k Independent

Tests for k Independent

Tests for k Independent

The exact distribution of T is difficult to

Tests for k Independent

Tests for k Independent

Tests for k Independent

Tests for k Independent

Mood (1954) has shown that when the median

Four different methods of growing corn were

Tests for k Independent

Tests for k Independent

1) Ho: All methods have the same median yield per

Ha: At least two of the methods differ with respect to

Tests for k Independent

3) Median Test for c (or k) independent samples

4) Computations (using SPSS with computations from

Tests for k Independent

Tests for k Independent

5) Critical Region: Reject Ho if 𝒳 2 > 𝒳 2 0.05,3 = 7.815.

Tests for k Independent

6) Decision/Conclusion: Since 17.6 > 7.815 or the p-

Thus a multiple comparison test is needed.

Tests for k Independent

Comparison of Method 1 versus Method 2

Tests for k Independent

Comparison of Method 1 versus Method 3

Tests for k Independent

Comparison of Method 1 versus Method 4

Tests for k Independent

Comparison of Method 2 versus Method 3

Tests for k Independent

Comparison of Method 2 versus Method 4

Tests for k Independent

Comparison of Method 3 versus Method 4

Tests for k Independent

It can be observed that only in Method 1 and

Tests for k Independent

Note: Such repeated testing procedures are for

Refer to the work of Gabriel (1966) and Knoke (1976).

Note: The implementation of the extension of Median

Tests for k Independent

Tests for k Independent

Tests for k Independent

Tests for k Independent

- It is a rank-based nonparametric test that

- It is considered the nonparametric

Tests for k Independent

Tests for k Independent