Chapter 36a

advertisement
INFERENTIAL STATISTICS
© LOUIS COHEN, LAWRENCE
MANION & KEITH MORRISON
STRUCTURE OF THE CHAPTER
• Measures of difference between groups
• The t-test (a difference test for parametric data)
• Analysis of variance (a difference test for
parametric data)
• The chi-square test (a difference test and a test of
goodness of fit for non-parametric data)
• Degrees of freedom (a statistic that is used in
calculating statistical significance in considering
difference tests)
• The Mann-Whitney and Wilcoxon tests (difference
tests for non-parametric data)
STRUCTURE OF THE CHAPTER
• The Kruskal-Wallis and Friedman tests (difference
tests for non-parametric data)
• Regression analysis (prediction tests for parametric
data)
• Simple linear regression (predicting the value of one
variable from the known value of another variable)
• Multiple regression (calculating the different
weightings of independent variables on a dependent
variable)
• Standardized scores (used in calculating
regressions and comparing sets of data with
different means and standard deviations)
MEASURES OF DIFFERENCE
BETWEEN GROUPS
• Are there differences between two or more
groups of sub-samples, e.g.:
– Is there a significant difference between
the amount of homework done by boys and
girls?
– Is there a significant difference between
test scores from four similarly mixed-ability
classes studying the same syllabus?
– Does school A differ significantly from
school B in the stress level of its sixth form
students?
MEASURES OF DIFFERENCE
BETWEEN GROUPS
• The t-test (for two groups): parametric data
• Analysis of Variance (ANOVA) (for three or
more groups: parametric data
• The chi-square test: for categorical data
• The Mann-Whitney and Wilcoxon tests (for
two groups): non-parametric data
• The Kruskal-Wallis and the Friedman tests
(for three or more groups): non-parametric
data
t-TEST
• Devised by William Gossett in 1908;
• Used when we have 2 conditions; the t-test
assesses whether there is a statistically
significant difference between the means of the
two conditions;
• The independent t-test is used when the
participants perform in only one of two
conditions;
• The related or paired t-test is used when the
participants perform in both conditions.
t-TEST FOR PARAMETRIC DATA
• t-tests (parametric, interval and ratio data)
– To find if there are differences between two
groups
– Decide whether they are are independent
or related samples
Independent sample: two different groups on
one occasion
Related sample: one group on two occasions
t-TEST FOR PARAMETRIC DATA
Formula for computing the t-test
Sample one mean – sample two mean
t = 
Standard error of the difference in means
Formula for calculating t
t
M1  M 2
  d12   d 22  1

1








 N1  N 2  2  N1 N 2 
M = Mean
d = difference between the means
N = Number of cases
t-TEST FOR INDEPENDENT SAMPLES
The t-test computes a ratio between a measure of
the between-groups variance and the within group
variance.
The larger the variance between the groups
(columns), compared with the variance within the
groups (rows), the larger the t-value.
INDEPENDENT AND RELATED
SAMPLES IN A t-TEST: EXAMPLES
1. Independent sample (two groups):
• A group of scientists wants to study the effects of a
new drug for insomnia. They have applied this drug to
a random group of people (control group) and to a
group of people suffering from insomnia (experimental
group);
2. Related sample (same group in two conditions):
• A group of therapists wants to study whether there is
any difference in doing relaxation techniques on the
beach or in an apartment. A group of people is asked
to first do relaxation on the beach and later in an
apartment;
INDEPENDENT AND RELATED
SAMPLES IN A t-Test: AN EXAMPLE
24 people were involved in an experiment to
determine whether background noise affects
short-term memory (recall of words);
– If half of the sample were allocated to the
NOISE condition and the other half to the
NO NOISE condition (independent
sample) – we use independent t-test;
– If everyone in the sample has performed at
both conditions (related sample) – we use
paired or related t-test.
AN EXAMPLE OF A t-TEST
Participants were asked to memorize a list of 20
words in two minutes.
Half of the sample performs in a noisy environment
and the other half in a quiet environment;
Independent variable - two types of environment:
Quiet environment (NO NOISE condition)
Noisy environment (NOISE condition)
Dependent variable – the number of words each
participant can recall.
NOISE
5
10
6
6
7
3
6
9
5
10
11
9
 = 87
X = 7.3
SD = 2.5
NO NOISE
15
9
16
15
16
18
17
13
11
12
13
11
 = 166
X = 13.8
SD = 2.8
NOTE: participants vary within
conditions: in the NOISE condition,
the scores range from 3 to 11, and in
the NO NOISE condition. They range
from 9 to 18;
The participants differ between the
conditions too: the scores of the NO
NOISE condition, in general, are
higher than those in the NOISE
condition – the means confirm it;
Are the differences between the
means of our groups large enough for
us to conclude that the differences are
due to our independent variable:
NOISE/NO NOISE manipulation?
t-TEST FOR INDEPENDENT SAMPLES
Group statistics
In which condition
are you?
N
Mean
How many words
can you recall?
Std. Deviation
Std. Error
Mean
NOISE
12
7.2500
.71906
NO NOISE
12
13.8333
.79614
This shows:
the name of 2 conditions;
the number of cases in each condition;
the mean of each condition;
the standard deviation and standard error of
the mean, of the two conditions.
t-TEST FOR INDEPENDENT SAMPLES (SPSS)
Independent Samples Test
Levene’s Test
for Equality of
Variances
F
How many Equal variances
words can assumed
you recall? Equal variances
Sig
t-test for Equality of Means
t
.177 .676 -6.137
df
Sig.
(2-tailed)
Mean
Std. Error
Differences Differences
95% Confidence
Interval of the
Difference
Lower
Upper
22
.000
-6.5833
1.07279
-8.808
-4.359
-6.137 21.78
.000
-6.5833
1.07279
-8.809
-4.357
not assumed
The Levene test is for ‘homogeneity of variance’,
and the t-test here indicates whether you should
use the upper or lower row.
Mean Difference means the difference between the
means of the two groups.
REPORTING FINDINGS FROM THE
EXAMPLE
Participants in the NOISE condition recalled fewer
words (t (22) = 7.25, SD = 2.49) than in the NO
NOISE condition (t (22) = 13.83, SD = 2.76). The
mean difference between conditions was 6.58; the
95% confidence interval for the estimated population
mean difference is between 4.36 and 8.81. An
independent t-test revealed that, if the null
hypothesis is true, such a result would be highly
unlikely to have arisen (t (22) = 6.14; p<0.001). It is
therefore concluded that listening to noise affects
short-term memory, at least in respect of word recall.
t-TEST FOR INDEPENDENT
SAMPLES WITH SPSS
Group Statistics
Which group are you
Mathematics post-test Control group
score
Experimental Group One
N
Mean
Std.
Std. Error
Deviation Mean
166
8.69
1.220
.095
166
9.45
.891
.069
Independent Samples Test
Levene's Test for
Equality of
Variances
t-test for Equality of Means
95% Confidence
Interval of the
Difference
Mathematics
post-test score
Equal variances
assumed
Equal variances
not assumed
F
28.856
Sig.
t
.000 -6.523
Sig. (2Mean
Std. Error
tailed) Difference Difference Lower
df
Upper
330
.000
-.765
.117
-.996
-.534
-6.523 302.064
.000
-.765
.117
-.996
-.534
Read the line ‘Levene’s Test for Equality of Variances’. If the
probability value is statistically significant then your variances
are unequal; otherwise they are regarded as equal. If the
Levene’s probability value is not statistically significant then
you need the row ‘equal variances assumed’; if the Levene’s
probability value is statistically significant then you need the
row ‘equal variances not assumed’. Look to the column ‘Sig.
(2-tailed)’ and the appropriate row, and see if the results are
statistically significant.
PAIRED SAMPLE t-TEST (SAME GROUP
UNDER TWO CONDITIONS) WITH SPSS
Paired Samples Statistics
Std.
Std. Error
Mean N Deviation
Mean
Pair 1 Mathematics pre-test
score
Mathematics post-test
score
6.95 252
1.066
.067
8.94 252
1.169
.074
This indicates:
1.The two conditions;
2.The mean of each condition;
3.The number of cases in each condition;
4.The standard deviation and standard error of the
mean, for the two conditions.
PAIRED SAMPLE t-TEST (SAME GROUP
UNDER TWO CONDITIONS) WITH SPSS
Paired Samples Correlations
N
Pair 1 Mathematics pre-test
score & Mathematics
post-test score
Correlation
252
.020
Sig.
.749
This shows that there is no association between the
scores on the pre-test and the scores on the post
test for the group in question (r = .02 and  = .749).
PAIRED SAMPLE t-TEST (SAME GROUP UNDER
TWO CONDITIONS) WITH SPSS
Paired Samples Test
Paired Differences
95%
Confidence
Interval of the
Difference
Std.
Error
Std.
Mean Deviation Mean Lower Upper
Pair Mathematics
1
pre-test score - -1.992
Mathematics
post-test score
1.567
t
Sig.
df (2-tailed)
.099 -2.186 -1.798 -20.186 251
This shows that :
1.The difference between the mean of each condition (6.95 and
8.94) is 1.992.
2.The confidence intervals shows that we are 95% certain that the
population difference lies somewhere between -2.186 and -1.798.
3.There is a statistically significant difference between the two sets
of scores.
.000
RESULT
It can be seen from the paired t-test that the
hypothesis is not supported (t (251) = 20.186;
=.000).
DEGREES OF FREEDOM
The number of individual scores that can vary
without changing the sample mean.
The number of scores one needs to know before
one can calculate the others.
E.g.: If you are asked to choose 2 numbers that
must add up to 100, and the first is 89, then the
other has to be 11; there is 1 degree of freedom
(89 + x = 100).
If you are asked to choose 3 numbers that must
add to 100, and the first of these is 20, then you
have 2 degrees of freedom (20 + x + y = 100).
DEGREES OF FREEDOM (WITH SPSS)
Which group are you * Who are you Crosstabulation
Chinese or non-Chinese
Chinese Non-Chinese Total
Which group Control group
are you
Experimental
Group One
Experimental
Group Two
Total
156
94.0%
166
100.0%
143
85.1%
465
93.0%
10
166
6.0% 100.0%
0
166
.0% 100.0%
25
168
14.9% 100.0%
35
500
7.0% 100.0%
Degrees of freedom = 2 (1 degree of freedom in each of 2
rows, which fixes what must be in the third row)
ANALYSIS OF VARIANCE (ANOVA)
• Analysis of variance
– Parametric, interval and ratio data
– To see if there are any statistically significant
differences between the means of two or more
groups;
– It calculates the grand mean (i.e. the mean of
the means of each condition) and sees how
different each of the individual means is from
the grand mean.
– Premised on the same assumptions as t-tests
(random sampling, a normal distribution of
scores, independent variable(s) is/are
categorical (e.g. teachers, students,) and one
is a continuous variable (e.g. marks on a test).
ANOVA AND MANOVA
• One way analysis of variance (one categorical
independent variable and one continuous
dependent variable)
• Two-way analysis of variance (two categorical
independent variables and one continuous
dependent variable)
• Multiple analysis of variance (MANOVA) (one
categorical independent variable and two or
more continuous variables)
• Post-hoc tests (e.g. Tukey hsd test, Sheffe
test) to locate where differences between
means lie (in which group(s))
FORMULA FOR ANOVA
between - groups variance
F ratio 
within - groups variance

d X N

Between variance 
2
mean
df  groups1
Within variance 
d
2
df  N  groups
A1
9
9
9
9
9
X =9
A2
15
15
16
15
16
X = 15.4
A3
21
25
17
22
26
X = 22.2
Between-groups and within-groups variance:
Variation between the groups (9 to 22.2);
Variation within the first group (no variation since all
participants scored the same);
Variation within the second group (from 15 to 16);
Variation within the third group (from 17 to 26).
ANOVA
1. First, ANOVA calculates the mean for each of
the three groups;
2. Then it calculates the grand mean (the three
means added then divided by three);
3. For each group separately, the total deviation of
each individual’ s score from the mean of the
group is calculated (within-groups variation);
4. Then the deviation of each group mean from the
grand mean is calculated (between-groups
variation).
F RATIO
between - groups variance
F ratio 
within - groups variance
When we conduct our experiment, we hope that the
between-groups variance is very much larger than the
within-groups variance, in order to get a bigger F ratio;
This shows us that one (or more) of the individual
group means is significantly different from the grand
mean;
However, it does not tell us which means are
statistically significantly different.
De scri ptives
Records of student s' progress
95% Confidenc e Interval for
Mean
20-29
30-39
40-49
50+
Total
N
7
5
4
1
17
Mean
3.29
3.80
3.25
4.00
3.47
St d.
Deviation
.76
1.30
.96
.
.94
Between-groups
variation
St d.
Error
.29
.58
.48
.
.23
Lower Bound
2.59
2.18
1.73
.
2.99
Within-groups
variation
Upper Bound
3.98
5.42
4.77
.
3.96
Minimum
2
2
2
4
2
Maximum
4
5
4
4
5
F (3,13) = .420, =.742
ANOVA
Records of students ' progress
Between Groups
W ithin Groups
Total
Sum of
Squares
1.257
12.979
14.235
df
3
13
16
Mean Square
.419
.998
F
.420
Sig.
.742
RESULTS
An F ratio of .420 has been given, with a
probability of =.742. This tells us that there
is no statistically significant difference
between any of the groups.
EFFECT SIZE: PARTIAL ETA SQUARED
Partial eta squared (
2
partial
)
SS effect
SS effect  SS error
SSeffect = The sums of the squares for whatever
effect is of interest;
SSerror = the sums of the squares for whatever
error term is associated with that effect.
EFFECT SIZE: PARTIAL ETA SQUARED FOR
INDEPENDENT SAMPLES IN SPSS
Analyze  General Linear Model  Univariate 
Options  Estimates of effect size
EFFECT SIZE: PARTIAL ETA SQUARED
IN SPSS
Between-Subjects Factors
Value Label
Which group are you 1
N
Control group
166
2
Experimental Group One
166
3
Experimental Group Two
168
EFFECT SIZE: PARTIAL ETA
SQUARED IN SPSS
Tests of Between-Subjects Effects
Dependent Variable:Mathematics post-test score
Source
Type III
Sum of
Squares
Corrected Model
Intercept
48.583a
41113.093
2
1
group
48.583
2
Error
521.105
497
Total
41684.000
500
569.688
499
Corrected Total
df
Mean Square
F
24.291 23.168
41113.093 39211.30
1
24.291 23.168
1.049
a. R Squared = .085 (Adjusted R Squared = .082)
Sig.
Partial Eta
Squared
.000
.000
.085
.987
.000
.085
THE POST HOC TUKEY TEST
• The null hypothesis for the F-test ANOVA is
always that the samples come from populations
with the same Mean (i.e., no statistically
significant differences): H0 = μ1 = μ2 = μ3 = …
• If the p-value is so low that we reject the null
hypothesis, we have decided that, at least one of
these populations has a mean that is not equal to
the others;
• The F-test itself only tells us that there are
differences at least between one pair of means,
not where these differences lie.
POST HOC TESTS
• To determine which samples are statistically
significantly different; after having performed the
F-test and rejected the null hypothesis, we turn
to post hoc comparisons;
• The purpose of a post hoc analysis is to find out
exactly where those differences are;
• Post hoc tests allow us to make multiple pair
wise comparisons and determine which pairs are
statistically significantly different from each other
and which are not.
THE POST HOC TUKEY TEST
Tukey’s Honestly Significant Difference (HSD)
Test is used to test the hypothesis that all
possible pairs of means are equal;
Tukey’s HSD test compares the mean
differences between each pair of means to a
critical value. If the mean difference from a pair
of means exceeds the critical value, we
conclude that there is a significant difference
between these pairs.
THE SCHEFFE TEST
The Scheffe test is very similar to the Tukey
hsd test, but it is more stringent that the Tukey
test in respect of reducing the risk of a Type I
error, though this comes with some loss of
power – one may be less likely to find a
difference between groups in the Sheffe test.
FINDING PARTIAL ETA SQUARED IN SPSS
Multivariate Testsb
Effect
scores
Pillai's Trace
Wilks' Lambda
Hotelling's Trace
Roy's Largest Root
scores * Pillai's Trace
group
Wilks' Lambda
Hotelling's Trace
Roy's Largest Root
a. Exact statistic
b. Design: Intercept + group
Within Subjects Design: scores
Value
.675
.325
2.079
2.079
.040
.960
.042
.042
Hypothesis
F
df
1033.477a
1.000
1033.477a
1.000
1033.477a
1.000
1033.477a
1.000
10.366a
2.000
10.366a
2.000
10.366a
2.000
10.366a
2.000
Error df
497.000
497.000
497.000
497.000
497.000
497.000
497.000
497.000
Sig.
.000
.000
.000
.000
.000
.000
.000
.000
Partial Eta
Squared
.675
.675
.675
.675
.040
.040
.040
.040
USING TUKEY TO LOCATE
DIFFERENCE IN SPSS
Multiple Comparisons
MEASURE_1
Tukey HSD
95% Confidence
Interval
Mean
Difference Std.
Lower
Error Sig. Bound
(I-J)
*
-.42
.084 .000
-.61
-.15 .084 .173
-.35
.42* .084 .000
.22
.27* .084 .005
.07
.15 .084 .173
-.05
-.27* .084 .005
-.46
(I) Which
group are you (J) Which group are you
Control group Experimental Group One
Experimental Group Two
Experimental Control group
Group One
Experimental Group Two
Experimental Control group
Group Two
Experimental Group One
Based on observed means.
The error term is Mean Square(Error) = .588.
*. The mean difference is significant at the .05 level.
Upper
Bound
-.22
.05
.61
.46
.35
-.07
USING TUKEY TO LOCATE DIFFERENCE IN SPSS
MEASURE_1
Tukey HSDa,,b,,c
Which group are you
N
Subset
1
Control group
166
7.86
Experimental Group Two
168
8.01
Experimental Group One
166
Sig.
2
8.27
.174
Means for groups in homogeneous subsets are displayed.
Based on observed means.
The error term is Mean Square(Error) = .588.
a. Uses Harmonic Mean Sample Size = 166.661.
b. The group sizes are unequal. The harmonic mean of the
group sizes is used. Type I error levels are not guaranteed.
c. Alpha = .05.
1.000
CHI-SQUARE
• A measure of a relationship or an association
developed by Karl Pearson in 1900;
• Measures the association between two
categorical variables;
• Compares the observed frequencies with the
expected frequencies;
• Determines whether two variables are
independent;
• Allows us to find out whether various subgroups are homogeneous.
TYPES OF CHI-SQUARE
• One-variable Chi-Square (goodness-of-fit test)
– used when we have one variable;
• Chi-Square test for independence: 2 x 2 – used
when we are looking for an association
between two variables, with two levels, e.g. the
association between (drinking alcohol/does not
drink alcohol) and (smoke/does not smoke);
• Chi-Square test for independence: r x c – used
when we are looking for an association
between two variables, where one has more
than two levels (heavy smoker, moderate
smoker, does not smoke) and (heavy drinker,
moderate drinker, does not drink).
FORMULA FOR CHI-SQUARE

2
=
(O  E )
 E
2
Where:
O = observed frequencies
E = expected frequencies
 = the sum of
ONE-VARIABLE CHI-SQUARE OR
GOODNESS-OF-FIT TEST
• Enables us to discover whether a set of
obtained frequencies differs from an expected
set of frequencies;
• One variable only;
• The numbers that we find in the various
categories are called the observed frequencies;
• The numbers that we expect to find in the
categories, if the null hypothesis is true, are the
expected frequencies;
• Chi-Square compares the observed and the
expected frequencies.
EXAMPLE: PREFERENCE FOR
CHOCOLATE BARS
A sample of 120 people were asked which of
four chocolate bars they preferred;
• We want to find out whether some brands (or
one brand) are preferred over others –
Research Hypothesis;
• If some brands are not preferred over others,
then all brands should be equally represented
– Null Hypothesis;
• If the Null Hypothesis is true, then we expect
30 (120/4) people in each category
ONE-VARIABLE CHI-SQUARE OR
GOODNESS-OF-FIT TEST
Frequencies Chocolate A
Chocolate B Chocolate C
Chocolate D
Observed
20
70
10
20
Expected
30
30
30
30
If all brands of chocolate are equally popular, the
observed frequencies will not differ much from the
expected frequencies;
If, however, the observed frequencies differ a lot
from the expected frequencies, then it is likely that
all brands are not equally popular;
ONE-VARIABLE CHI-SQUARE/GOODNESSOF-FIT TEST
Observed
N
Expected
N
Residual
(Difference between
observed and expected
frequencies)
Brand A
20
30
-10.0
Brand B
70
30
40.0
Brand C
10
30
-20.0
Brand D
20
30
-10.0
120
120
Total
Chocolate
Chisquare
df
Asymp.
Sig
73.333
3
.000
A chi-square value of 73.3, df = 3 was found to
have an associated probability level of 0.000. A
statistically significant difference was found
between the observed and the expected
frequencies, i.e. all brands of chocolate are not
equally popular. More people prefer chocolate B
(70) than the other bars of chocolate.
CHI-SQUARE TEST FOR
INDEPENDENCE (BIVARIATE): 2 X 2
Enables us to discover whether there is a
relationship or association between two
categorical variables of 2 levels;
If there is no association between the two
variables, then we conclude that the variables
are independent of each other.
A WORKED EXAMPLE
Imagine that we have asked 110 students the
following:
A. Do you smoke and drink?
B. Do you smoke but do not drink?
C. Do you not smoke but drink?
D. Do you abstain from both?
Each student can only fall into one group, and
thus we have 4 groups (they must be
mutually exclusive);
CHI-SQUARE TEST FOR
INDEPENDENCE: 2 X 2 (WITH SPSS)
Do you drink? * Do you smoke? Crosstabulation
Do you smoke?
Yes
No
Do you
drink?
Yes Count
50
15
Expected Count
41.4
23.6
No Count
20
25
Expected Count
28.6
16.4
Total
Count
70
40
Expected Count
70.0
40.0
Total
65
65.0
45
45.0
110
110.0
row total x column total
Expected value of a cell 
Overall total
CHI-SQUARE TEST FOR
INDEPENDENCE: 2 X 2 (WITH SPSS)
Value
df
Asymp. Sig.
Pearson Chi-Square
12.12
1
.000
Continuity Correction
10.759
1
.001
Likelihood Ratio
12.153
1
.001
Fisher’s Exact Test
Exact Sig.
(2-sided)
.001
Linear-by-Linear
Association
12.011
N of Valid Cases
110
1
Exact Sig.
(1-sided)
.001
.001
Chi-Square = 12.12
df (degrees of freedom) = (columns -1) x (rows -1)
= (2-1) x (2-1) = 1
RESULTS
A 2 x 2 Chi-square was carried out to discover
whether there was a significant relationship
between smoking and drinking. The Chisquare value of 12.12 has an associated
probability value of p<0.001, df = 1, showing
that such an association is extremely unlikely
to have arisen as a result of sampling error. It
can therefore be concluded that there is a
significant association between smoking and
drinking.
MANN-WHITNEY U-TEST FOR
INDEPENDENT SAMPLES
• Mann-Whitney (non-parametric, nominal and
ordinal data) for two groups under one condition
– Difference between two independent groups
(independent samples), based on ranks
• This is the non-parametric equivalent of the t-test
for independent samples.
• Find the significant differences and then run a
crosstabs to look at where the differences lie.
• Note where there are NO statistically significant
differences as well as where there are statistically
significant differences
MANN-WHITNEY U-TEST (SPSS)
Ranks
the contents
are interesting
form
Primary 3
Primary 4
Total
N
22
64
86
Mean Rank
43.52
43.49
Sum of Ranks
957.50
2783.50
MANN-WHITNEY U-TEST (SPSS)
Test Statisticsa
the
contents
are
interesting
703.500
2783.500
-.006
Mann-Whitney U
Wilcoxon W
Z
Asymp. Sig.
.996
(2-tailed)
a. Grouping Variable: form
THE WILCOXON TEST FOR RELATED
SAMPLES
This is the non-parametric equivalent of the t-test
for related samples.
For paired (related) samples in a non-parametric
test, e.g. the same group under two conditions.
For example, here is the result for one group of
females who have rated (a) their own ability in
mathematics and (b) their enjoyment of
mathematics, both variables using a 5-point scale
(‘not at all’ to ‘a very great deal’).
THE WILCOXON TEST FOR RELATED SAMPLES (SPSS)
Ranks
How good at
mathematics
do you think
you are? How much do
you enjoy
mathematics?
Negative Ranks
Positive Ranks
Ties
N Mean Rank Sum of Ranks
11
94.08
11101.00
73
99.11
7235.00
57
Total
248
Test Statisticsb
How good at mathematics do you
think you are? - How much do you
enjoy mathematics?
Z
Asymp. Sig. (2-tailed)
a. Based on positive ranks.
b. Wilcoxon Signed Ranks Test
-2.631a
.009
KRUSKAL-WALLIS TEST FOR
INDEPENDENT SAMPLES
• Kruskal-Wallis (non-parametric, nominal and
ordinal data) for three or more independent groups
under one condition
– Difference between more than two independent
groups (independent samples), based on ranks
• This is the non-parametric equivalent of ANOVA
for independent samples.
• Find the statistically significant differences and
then run a crosstabs to look at where the
differences lie.
• Note where there are NO statistically significant
differences as well as where there are statistically
significant differences.
KRUSKAL-WALLIS TEST (SPSS)
Ranks
own made-up tests
Age
20-29
30-39
40-49
Total
N
7
5
5
17
Mean Rank
6.57
10.70
10.70
KRUSKAL-WALLIS TEST (SPSS)
Test Statisticsa, b
Chi-Square
df
Asymp. Sig.
own made-up
tests
4.319
2
.115
a. Kruskal Wallis Test
b. Grouping Variable: Age
THE FRIEDMAN TEST FOR 3 OR MORE
RELATED GROUPS
This is the non-parametric equivalent of ANOVA for
related samples.
For three or more related samples in a non-parametric
test, e.g. the same groups under two conditions. For
example, the result for 4 groups of students, grouped
according to their IQ (Group 1= IQ up to 90; Group 2 =
IQ from 91-110; Group 3 = IQ from 111-125; Group 4 =
IQ over 125) who have rated (a) their own ability in
mathematics and (b) their enjoyment of mathematics,
both variables using a 5-point scale (‘not at all’ to ‘a
very great deal’).
Download