Other Revision

Question 1

•If no information is provided in the printout with respect to normality and homogeneity then…

Answer 1

It can be assumed and need not be mentioned

Question 2

If information is provided in the printout with respect to normality and homogeneity then…

Answer 2

It should be reported in the results section

Question 3

what is partial eta squared and when we would include it/ why we would, include it in a results section?

Answer 3

• Eta squared is the proportion of the total variance that is attributed to an effect.
• Partial eta-squared is the proportion of the effect + error variance that is attributable to the effect
• Measures of effect size should always be reported (see APA guidelines) if they are available

Question 4

When we go through the pre-analysis checks for the linear regression, could you indicate which normality test (Shapiro-Wilks or Kolmogorov-Smirnov) should we report and why?

Answer 4

• You can report either or the Lilliefors (but only one of them). They all compare against normal distributions but they do it in slightly different ways. The Shapiro-Wilks is more sensitive to departures from normality & all of them are ‘too’ sensitive with large sample sizes (e.g. N>100).
• In general the zskew works just fine…
• Generally F-tests are quite robust and degrade ‘gracefully’ so unless the data is really badly off then normality is a minor, but important, issue.

Question 5

•when writing the table of the means and standard deviations, will we be given a descriptive statistics table in the exam, or should we learn how to calculate them from the t-test information that is provided?

Answer 5

•The descriptive statistics will be given in the SPSS printouts, but it doesn’t do any harm to know how to calculate them from the t-test information…

Question 6

With respect to working out Cohen’s effect size for MLR using the formula f2=R2/(1-R2), assuming we use the R2 value from SPSS output [Model Summary], which model in the output do we use if it is a stepwise or hierarchical linear regression?

Answer 6

• Typically the final R2 is used.
• However, if there is a pre-analysis prediction about a specific effect then the R2 associated with that effect should be used. For example, if you expect personality to have R2=.35 then that should be entered into the f2 equation (see above)

Question 7

• Is it necessary to report in the ANOVA results section tests of the assumptions such as skew, Hartley’s F-max and Box’s M.
• Do you advise we report Levene’s Test of Homogeneity of Variances in the results section?
• Will an ANOVA be given to us that doesn’t meet the assumptions? If so, is it necessary for us to conduct data transformation and report it? How do we do this?

Answer 7

• If the normality and homogeneity information is presented in the SPSS printout then it should be reported. If it is not presented then it can be ‘assumed’.
• Levene’s test is fine and should be reported if it is available. You don’t need Hartley’s F-max if you report the Levene test.

Question 8

When reporting the main effects for ANOVA’s do you state which direction both went in and report the means?

Answer 8

When an effect is statistically significant you should always indicate the direction of the effect. The means should be reported in the text if they are not reported in the descriptive statistics table in your results section.

Question 9

•For the post hoc comparisons, should we work out the observed F for each comparison and report something like this: F(1, 33)= 2.23, p=0.306, or is the p value alone sufficient?

Answer 9

• If the values are in the SPSS printout then they should be reported (as above). If not the name of the test (e.g. Tukey, bonferroni corrected t-test, Scheffé) should be given with the associated p-value per comparison.
• Only report one kind of post hoc test. You can choose, but generally they vary in how conservative they are. (Some post hoc tests are situation specific, but they will not be in the SPSS printout).

Question 10

•How do you know what type of regression it is from the SPSS output: hierarchical, simultaneous or stepwise?

Answer 10

• If it is hierarchical then variables will be entered in blocks and all variables will be entered
• If it is stepwise then variables will be entered one at a time and not all variables may be entered.
• If it is simultaneous all the variables will be entered at the same time.

Question 11

•By the looks of the formulae the error term for two way between group ANOVA is MSS/AB for the three F-ratios). Would it be the same for a within group?

Answer 11

• In the between groups design there is only one error term
• In the within subjects (repeated measures) design there is an error term for each effect (main and interactions)
• In the mixed design there are separate error terms for within effects and a single error term for between effects (The specific number of error terms depends on the design)

Question 12

how to conduct pre-analysis checks for multiple regressions - particularly which SPSS outputs to consult for normality, homoscedasticity, and Cook’s distance for outliers and influential points?

Answer 12

• You may be given information about normality. If so it should be reported. If not normality should be mentioned as one of the pre-analysis checks
• You will not be given information about homoscedasticity or individual level Cook’s distances (too many pages…the exam would take a very long time). As above you should mention these issues when discussing pre-analysis checks.
• Further details can be found in the “C84MST Mock Exam” on moodle as well as the lecture, the example files, Field’s book and Tabachnick and Fidell’s book.