Factorial Anova Flashcards
Background
More than one iv in the same analysis, if 2 way, means 2 ivs ect. E.g. having a clinical vs non sample and then taking a drug vs not. Looks at how variables interact e.g. effect of one iv may depend on the level of another like drug only effect clinical sample
Assumptions
Data independent, normality, homo of variance, one dv, diff ps in all conditions - if just violated, careful at claiming big effects
Results
For for tests of between subject effects, for for ivs, each has f ratio, df and p value. Write descriptions for graph, results of assumptions tests then results: F(model df, error df)= f value p=0.05. Write the effect size, looks similar to p called partial eta squares, looks like weird n squared
How interactions of the ivs might look on a line graph
2 lines going in diff directions or crossing over after intervention e.g gradients difff. If no interaction, lines in parallel, even if start in diff places
Follow up tests/ planned contrasts
If anova sig, do more to explore effects. More systematic, side for testing spec hyp. Lesss likely to make type 1 as corrected and not type 2 as not over corrected. Further dividing variance explained by model and looks at chunks ex0ained by conditions and comparing them
Types of contrasts
Deviation: compares mean of each condition to overall mean. Simple: compares mean of each condition to either fires tor last condition. Helmert: compares mean of one condition to average of all other conditions
Difference: compares mean of condition to average of all previous conditions. Repeated: compares sequential pairs of conditions
Polynomial: looks for trends in the data
Most likely to use top 4, choice of contrast depends on hyp or previous data
Table to look at is contrast results (k matrix) and find sig row
Custom contrasts
Rules: must be independent/ test unique hyp so can’t run comparisons twice e.g. 1 and 2 but then not 1, 2 and 3, only 2nchunks compared at once. Always 1 fewer contrasts than number of groups. Every contrast needs a weight and + - on each side so add up to 0. To remove a group from the contrast, give it a 0
Custom contrast example
Contrast one is Democracy or none and conditions are full, flawed, hybrid or authoritarian so full is 1, flawed is 1, hybrid -1, authoritarian -2. If only looking at 2 so 1, -1, 0, 0. Contrast test table has t value, df and p,
Effect sizes
Now standard practice to report
Two main types: effect size based on diffs in means, scaled by variance or effect sizes that tell you what proportion of variance has been explained by the test
Called eta squared (one way anova) or partial eta squared (factorial anovas), scaled between 0 and 1 . have to tick estimates of effect sizes
