lecture 18 - dealing with multiple factors - 2-way ANOVA Flashcards
1
Q
Beyond one factor
A
- In many situations it is important to consider more than one variable. Especially where the effect of one “thing” might differ as a function of another. For example:
- Does a drug work the same for women and men?
- Do different forms of persuasion work the same in “East” and “West”?
- ANOVA can “extend” to this sort of situation – to simultaneously ask:
- Is there an effect of factor 1 (while holding factor 2 constant)?
- Is there an effect of factor 2 (while holding factor 1 constant)?
- Does the effect of factor 1 differ depending on factor 2?
Does the effect of factor 2 differ depending on factor 1? (Note –these last two things are mathematically identical).
2
Q
Main effects and interactions
A
- Imagine an experiment comparing the effects of (A) audio vs visual distractors, and (B) relevant vs irrelevant distractors, on a comprehension task.
- So, 4 groups: Audio & Relevant; Audio & Irrelevant; Visual & Relevant; Visual & Irrelevant.
- Asking about audio vs visual (holding relevance constant) is examining the “main effect” of modality.
- Asking about relevant vs irrelevant (holding modality constant) is examining the “main effect” of relevance.
- Asking whether modality and relevance are independent of each other is examining the interaction.
Remember the chi-square contingency analysis? That was also about interactions.
3
Q
look at graphs in notes
A
4
Q
2-way between subject ANOVA
A
- Return to mood and season example – but now have data from UK and UAE.
- Appears effect of season v different across countries – but needs formal analysis
Do different forms of persuasion work the same in “East” and “West”?
- Appears effect of season v different across countries – but needs formal analysis
graphs in notes
5
Q
2-way between subject ANOVA - maths
A
- As with 1-way, MS = SS/df – just doing this for both main effects and the interaction.
- As before F = MSeffect / MSresidual
- Again, now doing separately for each main effect and interaction.
Same MSresidual for each factor/interaction.
- Again, now doing separately for each main effect and interaction.
table in notes
6
Q
2-way between subject ANOVA
reporting
A
- Report as: Between subject ANOVA revealed no significant effect of Season, F(3,152) = 0.60, p = 0.613, a significant effect of Country, F(1,152) = 4.88, p = 0.029, and a significant interaction between them, F(3,152) = 6.06, p < 0.001
Data is made up, but “makes sense” – winter is icky in the UK (but summer is nice), while summer is stupidly hot in the UAE: so different effect of season on mood countries is expected.
table in notes
7
Q
Within-subject 2-way ANOVA
A
- As with 1-way ANOVA, can do on within-subject data if have data from every subject in every condition.
Imagine 40 people tested in the four different conditions described before - Audio & Relevant; Audio & Irrelevant; Visual & Relevant; Visual & Irrelevant.
examples, tables and graphs in notes
- Note – different “residuals” for each factor. Due to way variance due to subject is “taken out” as a function of what factor/interaction is being examined.
Still compute F-ratio the same way MSeffect / MSresidual (and report the same way)
8
Q
Going beyond today’s material
A
- Presented between-subject and within-subject 2-way ANOVA today. Can also combine between and within-subject factors (Called mixed ANOVA).
- For example, within-subject factor of Season, between-subject factor of Country.
- ANOVA (between, within, or mixed) can actually go beyond 2 factors (again, not covered here).
- But ANOVA gets increasingly hard to interpret as the number of factors increases.
But in all types of ANOVA, logic remains the same, examine F = MSeffect / MSresidual (where “effect” includes main effects and interactions, and “residual” can depend on what effect is examined).
- But ANOVA gets increasingly hard to interpret as the number of factors increases.