CH 14. Factorial ANOVA

Factorial ANOVA

2+ IV. Categorical

Tests for Interaction
-Main Effects; the level of one IV regardless of level of other IV.
- Interaction; talking about the unique combination of both IVs (4 hours and time out of class).

Looking for a unique condition.

Main effects

A main effect is an outcome that is a consistent difference between levels of a factor.

Interaction effects

An interaction effect exists when differences on one factor depend on the level you are on another factor.

Running a Factorial ANOVA

Analyze>GLM>Univariate>both IV's go in fixed factors.

SPSS automatically assumes you want a factorial design. It automatically creates interaction term.

Contrasts>no need to run contrasts if IV only has 2 levels.

Plots> main effect:
IVs(separately)>Horizontal>add>error bars

Plots>interaction:
-Multiple lines in the graph. In general, if a variable has a natural order to it, put on horizontal axis.
-Variables with fewer categories goes in 'separate lines'.

Post hoc: no post hoc for variable with 2 levels

EM Means: not going to be any different that the descriptive means because there is no covariate. Essentially, we are just asking for another set of means, but in a different format. This format helps in interpreting the graphs a bit more. And we can get more options for post hoc. Select all>compare means

Options>descriptives>homogeneity.

Helmert contrast

compares one group to combined other groups.

Levene's Test in Factorial ANOVA

Levene is comparing all groups.

Testing for homogeneity of variance across all groups.

If you violate levee's, you have to look at the 6 SD in the descriptives, find high/low to get Hartley's Fmax

Pairwise comparison

in Goggles.sav::

no need because this study is a 2x3 factorial design.

Simple effects analysis

following an ANOVA, a follow-up test to a significant interaction, comparing individual cells

At placebo is there a significant difference between the two

At low is there " " "

At high is there " " "

Can only be done through syntax, and is used to control for type 1 error.

controlling for Type 1 error

alpha .05

100 correlations only 5 will be significant by chance.

Reducing 5/100 = .20 --- much more sobering number, 1/20 will be significant.