ANSWERS TO EXERCISES AND REVIEW

QUESTIONS

PART FIVE: STATISTICAL TECHNIQUES TO COMPARE GROUPS

Before attempting these questions read through the introduction to Part Five and Chapters 16-
21 of the SPSS Survival Manual.

T-tests

5.1 Using the data file survey.sav follow the instructions in Chapter 16 of the SPSS Survival
Manual to find out if there is a statistically significant difference in the mean score for males
and females on the Total Life Satisfaction Scale (tlifesat). Present this information in a brief
report.

T-Test
Group Statistics

sex sex N Mean Std. Deviation Std. Error Mean

tlifesat total life satisfaction MALES 185 21.67 6.525 .480
FEMALES 251 22.90 6.911 .436

Independent Samples Test

Levene's Test
for Equality of
Variances t-test for Equality of Means
95% Confidence
Interval of the
Sig. Mean Std. Error Difference
F Sig. t df (2-tailed) Difference Difference Lower Upper
tlifesat total Equal variances
.706 .401 -1.881 434 .061 -1.230 .654 -2.516 .055
life satisfaction assumed
Equal variances not
-1.897 408.528 .059 -1.230 .648 -2.505 .044
assumed

An independent-samples t-test was conducted to compare total life satisfaction scores for males
and females. There was no statistically significant difference between the two groups [t(434)
=-1.88, p=.06].

1
5.2 Using the data file experim.sav apply whichever of the t-test procedures covered in
Chapter 16 of the SPSS Survival Manual that you think are appropriate to answer the
following questions.

(a) Who has the greatest fear of statistics at time 1, males or females?

Group Statistics

sex N Mean Std. Deviation Std. Error Mean

fost1 fear of stats time1 male 15 41.20 5.685 1.468
female 15 39.13 4.533 1.171

Independent Samples Test

Levene's Test for

Equality of
Variances t-test for Equality of Means
95% Confidence
Interval of the
Sig. Mean Std. Error Difference
F Sig. t df (2-tailed) Difference Difference Lower Upper
fost1 fear of Equal variances
2.087 .160 1.101 28 .280 2.067 1.877 -1.779 5.912
stats time1 assumed
Equal variances
1.101 26.679 .281 2.067 1.877 -1.788 5.921
not assumed

An independent-samples t-test was conducted to compare fear of statistics scores for males and
females. There was no statistically significant difference between the two groups [t(28) =1.10,
p=.28].

(b) Was the intervention effective in increasing students’ confidence in their ability to cope
with statistics? You will need to use the variables, confidence time1 (conf1) and confidence
time2 (conf2). Write your results up in a report.

Paired Samples Statistics

Std. Std. Error

Mean N Deviation Mean
Pair 1 confid1
confidence 19.00 30 5.369 .980
time1
confid2
confidence 21.87 30 5.594 1.021
time2

2
 Paired Samples Test

Paired Differences
95% Confidence
Interval of the
Std. Std. Error Difference Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)
Pair 1 confid1
confidence
time1 - confid2 -2.867 4.754 .868 -4.642 -1.091 -3.303 29 .003
confidence
time2

A paired-samples t-test was conducted to assess whether there was a change in students’
confidence scores from time 1 (pre-intervention) to time 2 (post-intervention). There was a
statistically significant difference between the two sets of scores [t(29) =-3.30, p=.003]. Mean
scores increased from 19.0 (SD=5.37) at Time 1 to 21.87(SD=5.59) at Time 2.

(c) What impact did the intervention have on students’ levels of depression?

Paired Samples Statistics

Mean N Std. Deviation Std. Error Mean

Pair 1 depress1 depression time1 42.53 30 4.592 .838
depress2 depression time2 40.73 30 5.521 1.008

Paired Samples Test

Paired Differences

Std. 95% Confidence Interval of

Std. Error the Difference Sig.
Mean Deviation Mean Lower Upper t df (2-tailed)
Pair 1 depress1 depression
time1 - depress2 1.800 2.497 .456 .868 2.732 3.949 29 .000
depression time2

A paired-samples t-test was conducted to assess whether there was a change in students’
depression scores from time 1 (pre-intervention) to time 2 (post-intervention). There was a
statistically significant difference between the two sets of scores [t(29) =-3.95, p<.001]. Mean
scores decreased from 42.53 (SD=4.59) at Time 1 to 40.73(SD=5.52) at Time 2.

3
One-way analysis of variance

For exercises 5.3 and 5.4 you will need to open the data file survey.sav.

5.3 Perform a one-way between-groups ANOVA to compare the levels of perceived stress
(tpstress) for the five different age groups (agegp5), 18-24yrs, 25-32yrs, 33-40yrs, 41-49yrs
and 50+yrs.

Descriptives

tpstress total perceived stress

95% Confidence
Interval for Mean Between-
Std. Lower Upper Componen
N Mean Deviation Std. Error Bound Bound Minimum Maximum t Variance
18-24 93 28.60 6.094 .632 27.35 29.86 12 46
25-32 86 25.65 4.920 .531 24.60 26.71 14 39
33-40 82 26.77 5.918 .654 25.47 28.07 13 40
41-49 95 26.62 5.706 .585 25.46 27.78 12 42
50+ 77 25.75 6.178 .704 24.35 27.16 13 42
Total 433 26.73 5.848 .281 26.18 27.28 12 46
Model Fixed
5.774 .277 26.18 27.27
Effects
Random
.539 25.23 28.22 1.062
Effects

Test of Homogeneity of Variances

tpstress total perceived stress

Levene Statistic df1 df2 Sig.
1.340 4 428 .254

ANOVA

tpstress total perceived stress

Sum of Squares df Mean Square F Sig.
Between Groups 500.761 4 125.190 3.755 .005
Within Groups 14271.082 428 33.344
Total 14771.843 432

Robust Tests of Equality of Means

tpstress total perceived stress

a
Statistic df1 df2 Sig.
Welch 3.651 4 211.303 .007
Brown-Forsythe 3.744 4 411.700 .005
a. Asymptotically F distributed.

4
 29

28
Mean of tpstress

27

26

18-24 25-32 33-40 41-49 50+

age 5 groups

The results of the one way ANOVA indicate that there is a difference in the perceived stress
levels amongst the age groups [F(4, 428)=3.76, p=.005]. Inspection of the means plot
suggests that the younger age group (18 to 24yrs) has higher stress levels than the other age
groups.

5.4 Perform post-hoc tests to compare the Self esteem scores for people across the three
different age groups (use the agegp3 variable).

Descriptives

tslfest total self esteem

95% Confidence
Interval for Mean Between-
Std. Std. Lower Upper Componen
N Mean Deviation Error Bound Bound Minimum Maximum t Variance
18-29 149 32.60 5.589 .458 31.69 33.50 18 40
30-44 152 33.59 5.288 .429 32.74 34.43 18 40
45+ 135 34.50 5.151 .443 33.63 35.38 20 40
Total 436 33.53 5.395 .258 33.02 34.04 18 40
Model Fixed Effects 5.352 .256 33.03 34.04
Random Effects .545 31.19 35.88 .692

5
 tslfest total self esteem

Subset for alpha = .05

agegp3 age 3 groups N 1 2
Tukey HSD a,b 18-29 149 32.60
30-44 152 33.59 33.59
45+ 135 34.50
Sig. .259 .311
Tukey B a,b 18-29 149 32.60
30-44 152 33.59 33.59
45+ 135 34.50
Sig.

Means for groups in homogeneous subsets are displayed.

a. Uses Harmonic Mean Sample Size = 144.943.

b. The group sizes are unequal. The harmonic mean of the group sizes is
used. Type I error levels are not guaranteed.

Post-hoc comparisons using the Tukey Honestly Significant Difference test indicated that the
mean score for Group 1 (M=32.6, SD=5.59) was significantly different from Group 3 (M=34.5,
SD=5.15). Group 2 (M=33.59, SD=5.29) did not differ significantly from either Group 1 or 3.

For the following exercise you will need to open the data file experim.sav.

5.5 Use one-way repeated measures ANOVA to compare the Fear of Statistics scores for the
three time periods (time1, time2 and time3). Inspect the means plots and describe the impact
of the intervention and the subsequent follow-up three months later.

General Linear Model

Within-Subjects Factors

Measure: MEASURE_1
Dependent
time Variable
1 fost1
2 fost2
3 fost3

Descriptive Statistics

Mean Std. Deviation N

fost1 fear of stats time1 40.17 5.160 30
fost2 fear of stats time2 37.50 5.151 30
fost3 fear of stats time3 35.23 6.015 30

6
 Multivariate Tests b

Partial Eta
Effect Value F Hypothesis df Error df Sig. Squared
time Pillai's Trace .635 24.356a 2.000 28.000 .000 .635
Wilks' Lambda .365 24.356a 2.000 28.000 .000 .635
Hotelling's Trace 1.740 24.356a 2.000 28.000 .000 .635
Roy's Largest Root 1.740 24.356a 2.000 28.000 .000 .635
a. Exact statistic
b.
Design: Intercept
Within Subjects Design: time

Mauchly's Test of Sphericity b

Measure: MEASURE_1

a
Epsilon
Approx. Greenhouse-
Within Subjects Effect Mauchly's W Chi-Square df Sig. Geisser Huynh-Feldt Lower-bound
time .342 30.071 2 .000 .603 .615 .500

Tests the null hypothesis that the error covariance matrix of the orthonormalized transformed dependent variables is proportional to an
identity matrix.
a. May be used to adjust the degrees of freedom for the averaged tests of significance. Corrected tests are displayed in the
Tests of Within-Subjects Effects table.
b.
Design: Intercept
Within Subjects Design: time

Tests of Within-Subjects Effects

Measure: MEASURE_1
Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
time Sphericity Assumed 365.867 2 182.933 41.424 .000 .588
Greenhouse-Geisser 365.867 1.206 303.368 41.424 .000 .588
Huynh-Feldt 365.867 1.230 297.506 41.424 .000 .588
Lower-bound 365.867 1.000 365.867 41.424 .000 .588
Error(time) Sphericity Assumed 256.133 58 4.416
Greenhouse-Geisser 256.133 34.974 7.323
Huynh-Feldt 256.133 35.664 7.182
Lower-bound 256.133 29.000 8.832

Tests of Within-Subjects Contrasts

Measure: MEASURE_1
Type III Sum Partial Eta
Source time of Squares df Mean Square F Sig. Squared
time Linear 365.067 1 365.067 46.652 .000 .617
Quadratic .800 1 .800 .795 .380 .027
Error(time) Linear 226.933 29 7.825
Quadratic 29.200 29 1.007

7
 Tests of Between-Subjects Effects

Measure: MEASURE_1
Transformed Variable: Average
Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
Intercept 127464.100 1 127464.100 1583.134 .000 .982
Error 2334.900 29 80.514

Estimated Marginal Means of MEASURE_1

41

40
Estimated Marginal Means

39

38

37

36

35

1 2 3
time

A one way repeated measures ANOVA was conducted to compare scores on the Fear of
Statistics Test scores at Time 1(prior to the intervention), Time 2 (following the intervention)
and Time 3 (three month follow-up). There was a significant effect for time [Wilks’ Lambda=
.365, F(2,28 )=24.36, p<.0005, multivariate partial eta squared=.64. Inspection of the plot of
mean values indicate a steady decrease in fear scores following the intervention, and at the
three month follow-up.

8
Two-way between-groups ANOVA

5.6 For this exercise you will need to open the data file survey.sav. Follow the instructions in
Chapter 18 of the SPSS Survival Manual to conduct a two-way ANOVA to explore the impact
of sex and age group on levels of perceived stress. The three variables you will need are sex,
agegp5 and tpstress.

(a) Interpret the results. Is there a significant interaction effect? Are the two main effects
significant?

Univariate Analysis of Variance

Between-Subjects Factors

Value Label N
sex sex 1 MALES 184
2 FEMALES 249
agegp5 1 18-24 93
age 5 2 25-32 86
groups
3 33-40 82
4 41-49 95
5 50+ 77

Descriptive Statistics

Dependent Variable: tpstress total perceived stress

sex sex agegp5 age 5 groups Mean Std. Deviation N
MALES 18-24 28.18 5.619 39
25-32 25.26 4.774 38
33-40 25.50 5.177 38
41-49 25.06 4.802 35
50+ 24.71 6.157 34
Total 25.79 5.414 184
FEMALES 18-24 28.91 6.449 54
25-32 25.96 5.061 48
33-40 27.86 6.345 44
41-49 27.53 6.024 60
50+ 26.58 6.138 43
Total 27.42 6.066 249
Total 18-24 28.60 6.094 93
25-32 25.65 4.920 86
33-40 26.77 5.918 82
41-49 26.62 5.706 95
50+ 25.75 6.178 77
Total 26.73 5.848 433

Levene's Test of Equality of Error Variances a

Dependent Variable: tpstress total perceived stress

F df1 df2 Sig.
1.026 9 423 .418

Tests the null hypothesis that the error variance of the

dependent variable is equal across groups.
a. Design: Intercept+sex+agegp5+sex * agegp5

9
 Tests of Between-Subjects Effects

Dependent Variable: tpstress total perceived stress

Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
Corrected Model 839.252a 9 93.250 2.831 .003 .057
Intercept 295968.489 1 295968.489 8985.743 .000 .955
sex 277.994 1 277.994 8.440 .004 .020
agegp5 503.367 4 125.842 3.821 .005 .035
sex * agegp5 64.874 4 16.219 .492 .741 .005
Error 13932.591 423 32.938
Total 324089.000 433
Corrected Total 14771.843 432
a. R Squared = .057 (Adjusted R Squared = .037)

Post Hoc Tests

agegp5 age 5 groups
Multiple Comparisons

Dependent Variable: tpstress total perceived stress

Tukey HSD

Mean 95% Confidence Interval

(I) age 5 groups (J) age 5 groups Difference (I-J) Std. Error Sig. Lower Bound Upper Bound
18-24 18-24
25-32 2.95* .859 .006 .60 5.30
33-40 1.83 .869 .218 -.55 4.22
41-49 1.98 .837 .127 -.31 4.27
50+ 2.85* .884 .012 .43 5.27
25-32 18-24 -2.95* .859 .006 -5.30 -.60
25-32
33-40 -1.12 .886 .715 -3.54 1.31
41-49 -.97 .854 .788 -3.31 1.37
50+ -.10 .900 1.000 -2.57 2.36
33-40 18-24 -1.83 .869 .218 -4.22 .55
25-32 1.12 .886 .715 -1.31 3.54
33-40
41-49 .15 .865 1.000 -2.22 2.52
50+ 1.02 .911 .799 -1.48 3.51
41-49 18-24 -1.98 .837 .127 -4.27 .31
25-32 .97 .854 .788 -1.37 3.31
33-40 -.15 .865 1.000 -2.52 2.22
41-49
50+ .87 .880 .862 -1.54 3.28
50+ 18-24 -2.85* .884 .012 -5.27 -.43
25-32 .10 .900 1.000 -2.36 2.57
33-40 -1.02 .911 .799 -3.51 1.48
41-49 -.87 .880 .862 -3.28 1.54
50+

Based on observed means.

*. The mean difference is significant at the .05 level.

10
Homogeneous Subsets
tpstress total perceived stress
a,b,c
Tukey HSD
Subset
age 5 groups N 1 2
25-32 86 25.65
50+ 77 25.75
41-49 95 26.62 26.62
33-40 82 26.77 26.77
18-24 93 28.60
Sig. .706 .159

Means for groups in homogeneous subsets are displayed.

Based on Type III Sum of Squares
The error term is Mean Square(Error) = 32.938.
a. Uses Harmonic Mean Sample Size = 86.075.

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.

Estimated Marginal Means of total perceived stress

29
sex
MALES
FEMALES

28
Estimated Marginal Means

27

26

25

24

18-24 25-32 33-40 41-49 50+

age 5 groups

The interaction effect (sex*agegp5) did not reach statistical significance[F(4, 423)=.492,
p=.741), however there was a significant main effect for sex [F(1,423)=8.44,p=.004) and age
group [F(4,423)=3.82, p=.005). Inspection of the mean scores and the plot suggest that
overall males have lower levels of perceived stress at all age levels. Overall younger people
(18 to 24 yrs) reported higher levels of stress than the other age groups. The results of this
analysis shows that although the means plot suggests the possibility of an interaction between
age and gender, it did not reach statistical significance.

11
(b) Write up this analysis and the results in a report. (Don’t forget to report the means and
standard deviations for each group.)

A two-way between groups analysis of variance was conducted to explore the impact of sex
and age on levels of perceived stress, as measured by the Perceived Stress Scale. Subjects
were divided into five groups according to their age (Group 1: 18 to 24years; Group 2: 25 to
32yrs; Group 3: 33 to 40yrs; Group 4: 41 to 49yrs; Group 5: 50yrs and above). There was no
significant interaction effect between age and sex [F(4,423)=.49, p=.74]. The main effect for
both sex [F(1,423)=8.44, p=.004, partial eta squared=.02] and age [F(4,423)=3.82, p=.005,
partial eta squared=.035] was statistically significant. Post hoc tests using Tukey’s Honestly
Significance Difference test revealed that the 18 to 24yr age group differed significantly from
the 25 to 32yr age group and the 50+ age group. All other group comparisons did not reach
statistical significance. Table XX below shows the mean scores for males and females for
each of the age groups.

Table XX
Mean and Standard Deviations for Males and Females across Age Groups

Males Females
n Mean SD n Mean SD
18-24yrs 39 28.18 5.62 54 28.91 6.45
25-32yrs 38 25.26 4.77 48 25.96 5.06
33-40yrs 38 25.50 5.18 44 27.86 6.35
41-49yrs 35 25.06 6.16 60 27.53 6.02
50+ 34 24.70 6.16 43 26.58 6.14

12
Mixed between-within subjects analysis of variance

5.7 In Chapter 19 of the SPSS Survival Manual we explored the impact of two different
intervention programs (maths skills/confidence building) on participants’ fear of statistics. We
found that both interventions were equally effective in reducing participants’ fear—that is, we
found no differences between groups—but a significant difference across the three time
periods. Repeat these analyses, but this time use confidence scores as the dependent variable.

Open the file experim.sav. You will need to use the following variables: group, conf1, conf2
and conf3.

General Linear Model

Within-Subjects Factors

Measure: MEASURE_1
Dependent
time Variable
1 confid1
2 confid2
3 confid3

Between-Subjects Factors

Value Label N
group type 1 maths skills 15
of class 2 confidence
15
building

Descriptive Statistics

group type of class Mean Std. Deviation N

confid1 confidence time1 maths skills 18.87 5.527 15
confidence building 19.13 5.397 15
Total 19.00 5.369 30
confid2 confidence time2 maths skills 20.00 4.660 15
confidence building 23.73 5.970 15
Total 21.87 5.594 30
confid3 confidence time3 maths skills 24.07 4.543 15
confidence building 26.00 5.782 15
Total 25.03 5.203 30

Box's Test of Equality of Covariance Matrices a

Box's M 8.522
F 1.254
df1 6
df2 5680.302
Sig. .275

Tests the null hypothesis that the observed covariance

matrices of the dependent variables are equal across groups.
a.
Design: Intercept+group
Within Subjects Design: time

13
 Multivariate Tests b

Partial Eta
Effect Value F Hypothesis df Error df Sig. Squared
time Pillai's Trace .752 40.897a 2.000 27.000 .000 .752
Wilks' Lambda .248 40.897a 2.000 27.000 .000 .752
Hotelling's Trace 3.029 40.897a 2.000 27.000 .000 .752
Roy's Largest Root 3.029 40.897a 2.000 27.000 .000 .752
time * group Pillai's Trace .207 3.534a 2.000 27.000 .043 .207
Wilks' Lambda .793 3.534a 2.000 27.000 .043 .207
Hotelling's Trace .262 3.534a 2.000 27.000 .043 .207
Roy's Largest Root .262 3.534a 2.000 27.000 .043 .207
a. Exact statistic
b.
Design: Intercept+group
Within Subjects Design: time

Mauchly's Test of Sphericity b

Measure: MEASURE_1

a
Epsilon
Within Approx. Greenhouse- Huynh-
Subjects Effect Mauchly's W Chi-Square df Sig. Geisser Feldt Lower-bound
time .573 15.059 2 .001 .701 .753 .500

Tests the null hypothesis that the error covariance matrix of the orthonormalized transformed dependent
variables is proportional to an identity matrix.
a. May be used to adjust the degrees of freedom for the averaged tests of significance. Corrected
tests are displayed in the Tests of Within-Subjects Effects table.
b.
Design: Intercept+group
Within Subjects Design: time

Tests of Within-Subjects Effects

Measure: MEASURE_1
Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
time Sphericity Assumed 546.467 2 273.233 35.383 .000 .558
Greenhouse-Geisser 546.467 1.401 390.038 35.383 .000 .558
Huynh-Feldt 546.467 1.505 363.097 35.383 .000 .558
Lower-bound 546.467 1.000 546.467 35.383 .000 .558
time * group Sphericity Assumed 45.089 2 22.544 2.919 .062 .094
Greenhouse-Geisser 45.089 1.401 32.182 2.919 .082 .094
Huynh-Feldt 45.089 1.505 29.959 2.919 .079 .094
Lower-bound 45.089 1.000 45.089 2.919 .099 .094
Error(time) Sphericity Assumed 432.444 56 7.722
Greenhouse-Geisser 432.444 39.230 11.023
Huynh-Feldt 432.444 42.140 10.262
Lower-bound 432.444 28.000 15.444

14
 Tests of Within-Subjects Contrasts

Measure: MEASURE_1
Type III Sum Partial Eta
Source time of Squares df Mean Square F Sig. Squared
time Linear 546.017 1 546.017 52.526 .000 .652
Quadratic .450 1 .450 .089 .767 .003
time * group Linear 10.417 1 10.417 1.002 .325 .035
Quadratic 34.672 1 34.672 6.867 .014 .197
Error(time) Linear 291.067 28 10.395
Quadratic 141.378 28 5.049

Levene's Test of Equality of Error Variances a

F df1 df2 Sig.

confid1 confidence time1 .000 1 28 .986
confid2 confidence time2 1.718 1 28 .201
confid3 confidence time3 .873 1 28 .358

Tests the null hypothesis that the error variance of the dependent variable is equal
across groups.
a.
Design: Intercept+group
Within Subjects Design: time

Tests of Between-Subjects Effects

Measure: MEASURE_1
Transformed Variable: Average
Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
Intercept 43428.100 1 43428.100 619.488 .000 .957
group 88.011 1 88.011 1.255 .272 .043
Error 1962.889 28 70.103

15
 Estimated Marginal Means of MEASURE_1

26
type of class
maths skills
confidence
building
Estimated Marginal Means

24

22

20

18

1 2 3
time

(a) Is there a significant interaction effect between type of intervention (group) and time?

The interaction between type of intervention and time is significant (p=.043). An inspection of
the plot suggests that the confidence building group showed greater improvement in
confidence levels following the intervention than the maths skills group.

(b) Is there a significant main effect for the within-subjects independent variable, time?

The interaction effect for group by time is significant, therefore it is not really appropriate to
interpret the main effect. The impact of one variable (eg. Time) is dependent on the level of
the other variable (group).

(c) Is there a significant main effect for the between-subjects independent variable, group
(maths skills/confidence building)?

The interaction effect for group by time is significant, therefore it is not appropriate to
interpret the main effect. The impact of one variable (eg. Time) is dependent on the level of
the other variable (group).

16
Multivariate analysis of variance

5.8 How does MANOVA differ from ANOVA?

Multivariate analysis of variance is an extension of analysis of variance for use when there is
more than one dependent variable.

5.9 In Chapter 20 of the SPSS Survival Manual it is recommended that you check the
Mahalonobis distances before proceeding with MANOVA. What does this allow you to check
for?

Mahalonobis distances is a test of multivariate normality.

5.10 Which assumption is Box’s M Test used to assess?

Box’s M Test is used to assess the homogeneity of variance-covariance matrices.

5.11 Follow the procedure detailed in Chapter 20 of the SPSS Survival Manual to perform a
MANOVA to explore positive and negative affect scores for the three age groups (18-29yrs,
30-44yrs, 45+yrs). The three variables you will need are tposaff, tnegaff, agegp3. Remember
to check your assumptions.

General Linear Model

Between-Subjects Factors

Value Label N
agegp3 age 1 18-29 147
3 groups 2 30-44 153
3 45+ 135

Descriptive Statistics

agegp3 age 3 groups Mean Std. Deviation N

tposaff total positive affect 18-29 33.33 7.409 147
30-44 33.59 7.316 153
45+ 34.13 7.017 135
Total 33.67 7.247 435
tnegaff total negative affect 18-29 20.65 7.346 147
30-44 19.37 6.616 153
45+ 18.09 7.076 135
Total 19.40 7.072 435

17
 Box's Test of Equality of Covariance Matrices a

Box's M 2.703
F .448
df1 6
df2 4335850.466
Sig. .847

Tests the null hypothesis that the observed covariance

matrices of the dependent variables are equal across groups.
a. Design: Intercept+agegp3

Multivariate Tests c

Partial Eta
Effect Value F Hypothesis df Error df Sig. Squared
Intercept Pillai's Trace .976 8661.453a 2.000 431.000 .000 .976
Wilks' Lambda .024 8661.453a 2.000 431.000 .000 .976
Hotelling's Trace 40.192 8661.453a 2.000 431.000 .000 .976
Roy's Largest Root 40.192 8661.453a 2.000 431.000 .000 .976
agegp3 Pillai's Trace .021 2.340 4.000 864.000 .054 .011
Wilks' Lambda .979 2.347a 4.000 862.000 .053 .011
Hotelling's Trace .022 2.354 4.000 860.000 .052 .011
Roy's Largest Root .022 4.709b 2.000 432.000 .009 .021
a. Exact statistic

b. The statistic is an upper bound on F that yields a lower bound on the significance level.
c. Design: Intercept+agegp3

Levene's Test of Equality of Error Variances a

F df1 df2 Sig.

tposaff total positive affect .350 2 432 .705
tnegaff total negative affect .970 2 432 .380

Tests the null hypothesis that the error variance of the dependent variable is equal across
groups.
a. Design: Intercept+agegp3

Tests of Between-Subjects Effects

Type III Sum Partial Eta

Source Dependent Variable of Squares df Mean Square F Sig. Squared
Corrected Model tposaff total positive affect 47.346a 2 23.673 .450 .638 .002
tnegaff total negative affect 463.048b 2 231.524 4.709 .009 .021
Intercept tposaff total positive affect 492175.882 1 492175.882 9347.172 .000 .956
tnegaff total negative affect 162755.374 1 162755.374 3310.007 .000 .885
agegp3 tposaff total positive affect 47.346 2 23.673 .450 .638 .002
tnegaff total negative affect 463.048 2 231.524 4.709 .009 .021
Error tposaff total positive affect 22746.985 432 52.655
tnegaff total negative affect 21241.743 432 49.171
Total tposaff total positive affect 515910.000 435
tnegaff total negative affect 185499.000 435
Corrected Total tposaff total positive affect 22794.331 434
tnegaff total negative affect 21704.791 434
a. R Squared = .002 (Adjusted R Squared = -.003)

b. R Squared = .021 (Adjusted R Squared = .017)

18
 Estimated Marginal Means of total positive affect

34.2

34
Estimated Marginal Means

33.8

33.6

33.4

33.2

18-29 30-44 45+

age 3 groups

Estimated Marginal Means of total negative affect

21

20.5
Estimated Marginal Means

20

19.5

19

18.5

18

18-29 30-44 45+

age 3 groups

19
The results of Box’s test of equality of covariance matrices indicate no violation of the
assumption (p=.85)

The results of Levene’s test of equality of error variances indicate that we have not violated
the assumption for either of our dependent variables (p=.71, p=.38).

Inspection of the results shown in Multivariate tests indicate a significant result overall
[Wilks’ Lambda=.98, F(4, 862)=2.35, p=.05].

The Tests of Between Subjects Effects table indicates a significant result for Total Negative
Affect [F(2,432)=4.71, p=.009, partial eta squared=.02], but not for Total Positive Affect
[F(2,432)=.45, p=.64, partial eta squared=.002]. Inspection of the mean scores for each age
group indicates a steady decrease in levels of negative affect across the three age groups (18-
29yrs mean=20.65, SD=7.35; 30-44yrs mean=19.37, SD=6.62; 45+yrs mean=18.09,
SD=7.07).

20
Analysis of covariance

5.12 Under what circumstances would you want to consider using analysis of covariance?

Analysis of covariance is used when you wish to compare groups, while controlling for
additional variables that you suspect might be influencing scores on the dependent variable.

5.13 What issues do you need to consider when you are selecting possible covariates?

Covariates need to be chosen with a good understanding of background theory and previous
research in your research area. The covariates need to be continuous variables, measured
reliably and correlate significantly with the dependent variable. The covariate must be
measured before the treatment or experimental manipulation is conducted.

5.14 Using the experim.sav data file, perform the appropriate analyses (including assumption
testing) to compare the confidence scores for the two groups (maths skills, confidence
building) at time 2, while controlling for confidence scores at time 1. The variables you will
need are group, conf1, conf2.

Univariate Analysis of Variance

Between-Subjects Factors

Value Label N
group type 1 maths skills 15
of class 2 confidence
15
building

Tests of Between-Subjects Effects

Dependent Variable: confid2 confidence time2

Type III Sum
Source of Squares df Mean Square F Sig.
Corrected Model 466.737a 3 155.579 9.178 .000
Intercept 196.989 1 196.989 11.621 .002
group 2.067 1 2.067 .122 .730
confid1 348.104 1 348.104 20.536 .000
group * confid1 17.644 1 17.644 1.041 .317
Error 440.730 26 16.951
Total 15252.000 30
Corrected Total 907.467 29
a. R Squared = .514 (Adjusted R Squared = .458)

The above output is used to assess the assumption of homogeneity of regression slopes. The
interaction term (group*confid1) is not significant (p=.317), therefore we have not violated
the assumption and can then proceed with the ANCOVA analysis.

21
Univariate Analysis of Variance

Between-Subjects Factors

Value Label N
group type 1 maths skills 15
of class 2 confidence
15
building

Descriptive Statistics

Dependent Variable: confid2 confidence time2

group type of class Mean Std. Deviation N
maths skills 20.00 4.660 15
confidence building 23.73 5.970 15
Total 21.87 5.594 30

Levene's Test of Equality of Error Variances a

Dependent Variable: confid2 confidence time2

F df1 df2 Sig.
.136 1 28 .715

Tests the null hypothesis that the error variance of the

dependent variable is equal across groups.
a. Design: Intercept+confid1+group

Tests of Between-Subjects Effects

Dependent Variable: confid2 confidence time2

Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
Corrected Model 449.093a 2 224.546 13.227 .000 .495
Intercept 200.700 1 200.700 11.822 .002 .305
confid1 344.560 1 344.560 20.296 .000 .429
group 95.102 1 95.102 5.602 .025 .172
Error 458.374 27 16.977
Total 15252.000 30
Corrected Total 907.467 29
a. R Squared = .495 (Adjusted R Squared = .457)

Estimated Marginal Means

type of class

Dependent Variable: confid2 confidence time2

95% Confidence Interval
type of class Mean Std. Error Lower Bound Upper Bound
maths skills 20.086a 1.064 17.902 22.269
confidence building 23.648a 1.064 21.465 25.831
a. Covariates appearing in the model are evaluated at the following
values: confid1 confidence time1 = 19.00.

22
Inspection of the table ‘Levene’s Test of Equality of Error Variances’ indicate we have not
violated the assumption concerning the equality of variances (p=.715).

The Tests of Between-Subjects Effects table results indicate a significant effect for group
(p=.025). There is a significant difference in confidence scores for the confidence building
and maths skills groups, after controlling for confidence scores administered prior to the
treatment program.

5.15 Perform a two-way analysis of covariance to explore the question: Does gender influence
the effectiveness of the two intervention programs designed to increase participants’
confidence in being able to cope with statistics training? You will need to assess the impact of
sex and type of intervention (group) on confidence at time 2, controlling for confidence scores
at time 1.

Univariate Analysis of Variance

Between-Subjects Factors

Value Label N
group type 1 maths skills 15
of class 2 confidence
15
building
sex 1 male 15
2 female 15

Descriptive Statistics

Dependent Variable: confid2 confidence time2

group type of class sex Mean Std. Deviation N
maths skills male 22.25 4.301 8
female 17.43 3.823 7
Total 20.00 4.660 15
confidence building male 21.29 5.880 7
female 25.88 5.515 8
Total 23.73 5.970 15
Total male 21.80 4.931 15
female 21.93 6.364 15
Total 21.87 5.594 30

Levene's Test of Equality of Error Variances a

Dependent Variable: confid2 confidence time2

F df1 df2 Sig.
2.277 3 26 .103

Tests the null hypothesis that the error variance of the

dependent variable is equal across groups.
a. Design: Intercept+confid1+group+sex+group * sex

23
 Tests of Between-Subjects Effects

Dependent Variable: confid2 confidence time2

Type III Sum Partial Eta
Source of Squares df Mean Square F Sig. Squared
Corrected Model 826.891a 4 206.723 64.139 .000 .911
Intercept 55.142 1 55.142 17.109 .000 .406
confid1 556.942 1 556.942 172.800 .000 .874
group 92.813 1 92.813 28.797 .000 .535
sex .815 1 .815 .253 .620 .010
group * sex 377.226 1 377.226 117.040 .000 .824
Error 80.576 25 3.223
Total 15252.000 30
Corrected Total 907.467 29
a. R Squared = .911 (Adjusted R Squared = .897)

Estimated Marginal Means

1. type of class

Dependent Variable: confid2 confidence time2

95% Confidence Interval
type of class Mean Std. Error Lower Bound Upper Bound
maths skills 19.855a .465 18.898 20.811
confidence building 23.381a .465 22.424 24.339
a. Covariates appearing in the model are evaluated at the following
values: confid1 confidence time1 = 19.00.

2. sex

Dependent Variable: confid2 confidence time2

95% Confidence Interval
sex Mean Std. Error Lower Bound Upper Bound
male 21.783a .465 20.826 22.740
female 21.453a .465 20.495 22.410
a. Covariates appearing in the model are evaluated at the
following values: confid1 confidence time1 = 19.00.

3. type of class * sex

Dependent Variable: confid2 confidence time2

95% Confidence Interval
type of class sex Mean Std. Error Lower Bound Upper Bound
maths skills male 23.750a .645 22.422 25.078
female 15.959a .688 14.543 17.375
confidence building male 19.816a .688 18.400 21.232
female 26.947a .640 25.629 28.265
a. Covariates appearing in the model are evaluated at the following values: confid1
confidence time1 = 19.00.

24
 Estimated Marginal Means of confidence time2

27.5
sex
male
female

25
Estimated Marginal Means

22.5

20

17.5

15

maths skills confidence building

type of class

An inspection of the plot of mean scores suggests the possibility of an interaction between
gender and type of intervention in terms of confidence scores. Females in the Confidence
building group showed higher confidence scores at Time 2, than those who received the
Maths skills intervention. Males however who participated in the Maths skills intervention
showed higher mean scores than those who were in the Confidence Building group. This is
supported by the results in the Tests of Between Subjects Effects table. The group*sex
interaction term is statistically significant [F(1,25)=117.04, p<.0005].

25
Non-parametric statistics

5.16 What is the difference between parametric techniques and non-parametric techniques?

The parametric tests (eg. T-tests, ANOVA) make assumptions about the population the sample
has been drawn from. Non-parametric techniques do not have such stringent requirements
and do not make assumptions about the underlying population distribution.

5.17 What factors would you consider when choosing whether to use a parametric or a non-
parametric technique?

You need to consider the levels of measurement of your data. If you have nominal or ordinal
scaled data you should use a suitable non-parametric, rather than parametric technique.

5.18 For each of the following parametric techniques indicate the non-parametric alternative
(if one exists).

(a) one-way between-groups ANOVA Kruskal-Wallis Test

(b) Pearson’s product-moment correlation Spearman Rank Order Correlation
(c) independent samples t-test Mann-Whitney Test
(d) multivariate analysis of variance No equivalent
(e) one-way repeated measures ANOVA Friedman Test
(f) paired samples t-test Wilcoxon Signed Rank Test
(g) partial correlation No equivalent

5.19 Choose and perform the appropriate non-parametric test to address each of the following
research questions.

(a) Using the survey.sav data file find out whether smokers are significantly more stressed
than non-smokers. The variables you will need are smoke and total perceived stress (tpstress).

Mann-Whitney Test

(b) Using the survey.sav data file compare the self-esteem scores across the three different age
groups (18-29yrs, 30-44yrs, 45+yrs). The variables you will need are tslfest and agegp3.

Kruskal-Wallis Test.

(c) Using the survey.sav data file explore the relationship between optimism and negative
affect. The variables you will need are toptim and tnegaff.

Spearman Rank Order Correlation

(d) Using the survey.sav data file explore the association between education level and
smoking. The variables you will need are educ2 and smoke. Check the codebook and the
questionnaire in the appendix of the SPSS Survival Manual for details on these two variables.

Chi square test for independence

26
(e) Using the experim.sav data file compare the depression scores at time 1 and the depression
scores at time 2. Did the intervention result in a significant change in depression scores? The
variables you will need are depress1 and depress2.

Wilcoxon Signed Rank Test

(f) Using the experim.sav data file compare the depression scores for the three time periods
involved in the study (before the intervention, after the intervention and at the three-month
follow up). The variables you will need are depress1, depress2 and depress3.

Friedman Test

27