Psychology 202a Advanced Psychological Statistics December 8, 2015 The plan for today • • • • The nested F test Another two-way ANOVA example Unbalanced designs Random effects The nested F test • Definition of nested models – One model is nested within another if it is possible to change the more complex model into the simpler one by constraining parameters. • Testing for change when a model is nested within another The nested F test (cont.) • • • • SSchange = SScomplex model - SSsimple model dfchange = dfcomplex model - dfsimple model MSchange = SSchange / dfchange Fchange = MSchange / MSerror, complex model Using the nested F test • Illustration using SAS • We identify the parameters representing the interaction and do a nested F test. • Then we identify the parameters representing the main effect of practice and do a nested F test. 85 Another Example of Two-way ANOVA F 80 Kelly Nils F 75 Midterm Mean M M Male Students Female Students ANOVA with unbalanced data • In general, if the lack of balance is not telling us something profound about the world… • …then the Type III sum of squares provides a solution to the problem of unbalanced designs. • The TA sex example. 85 Kelly & Nils, Revisited M 80 F F 75 Midterm Mean Nils M Male Students Kelly Female Students ANOVA with random effects • Distinguish between fixed effects and random effects. • If both are present, then we have a mixedeffects ANOVA. Fixed-effects Factors • It may be easiest to understand random-effects factors by contrasting them with the more familiar fixed-effects factors. • Fixed-effects factors are those in which the populations to which we wish to generalize are precisely the levels represented in our analysis. • Examples: – – – – Eysenck’s levels of processing Eysenck’s age groups Our three practice conditions Our two reward conditions Random-effects Factors • Sometimes the levels represented in our analysis are an arbitrary sample of a universe of possible levels. • When that is true, we refer to the factor as a random-effects factor. Examples of Random-effects Factors • Patients in hospitals – If the interest is in how means vary across hospitals in general, hospital is a random effect. • Pupils taught by different teachers – If the interest is in how student outcome means vary across teachers in general, teacher is a random effect. It’s the question that matters. • Note that both of those examples could be fixed effects under the right circumstances: – An administrator of a group of hospitals is interested in comparing patient outcomes in her particular hospitals. – A high school principal is interested in comparing student achievement for his particular teachers. The null hypothesis changes • If Factor A is a random effect, the null hypothesis changes. • Fixed effects: H 0 : 1 2 k • Random effects: H0 : A 0 2 Two-way ANOVA • In two-way ANOVA, the presence of a randomeffects factor changes more than the null hypothesis. – If factor A is a random effect, test factor B using the interaction mean square in the denominator of the F statistic. – If factor B is a random effect, test factor A using the interaction mean square in the denominator of the F statistic. Next time • Review for the final exam. • A study guide has been posted; come to class prepared to ask questions.