Statistical Concepts

Session 5: The General Linear Model

Dependent Variable - Interval
(can bend rules by using Ordinal if there are at least 5 ordered categories)

ANOVA - Independent variables are categorical (no more than 5 categories)
Regression - Independent variables are ordinal and/or interval (can use binary categorical variables)

Mixed - either use ANCOVA (covariates before factors) or Regression (create one dummy variable for each degree of freedom for each categorical factor)

N.B. By default, ANOVA also analyses interactions, but Regression does not.

ANOVA Example

288 students given essays to mark.
Mean essay marks and sample sizes by same/opposite sex and level of attractiveness.

 Same sex Opposite sex Overall n Attractiveness of author Low 22.8 23.2 22.9 96 Medium 28.6 22.9 25.8 96 High 25.1 28.0 26.6 96 Overall 25.5 24.7 25.1 288 n 144 144 288

 Source of Variation Sum of Squares Degrees of Freedom Mean Square F Ratio Same/opposite sex (A) 686.85 Attractiveness (B) 46.85 Interaction (AxB) 942.35 Residual 21053.14 Total 22729.19