Contrast coding (L10) Flashcards
What are the drawbacks of an F-Ratio?
Only tells us an experiment was successful, but doesn’t tell us exactly which groups differ, therefore additional analysis is needed.
What are planned contrasts/comparisons?
Hypothesis driven, planned tests to analyse where the effect between groups specifically lies.
Breaks down even further the variability explained by the model to say it was explained by specific hypothesis
‘A priori’ - before experiment
What are post-hoc tests?
Multiple tests comparing each mean against all the others.
Increases chances of a type 1 error, so uses stricter criterion for significance.
What is trend analysis?
Expecting means to increase in a specific way
What are the rules when choosing contrasts?
1) Independent contrasts (reduces type 1 error, shouldn’t reuse groups)
2) only 2 chunks (should only use 2 groups so know relationship)
3) K - 1 (one less predictor variable than you started with)
Why is the first contrast usually control vs. experimental?
Because most experiments are 1+ control group, therefore logic dictates we compare control to experimental first.
How do you code planned contrasts?
1) Groups are coded with positive or negative weights
2) Sum of weight comparison should be 0
3) Any groups not in comparison should be coded 0.
What is the contrast model?
y = b0 + b1contrast1 + b2contrast2 etc + error
Why can’t you just carry out multiple tests?
Because the comparisons aren’t independent which inflates chance of type 1 error.
How do you calculate Bonferroni corrections?
Bonferroni = a (0.05) / no. stat tests
Why do we carry out Bonferroni corrections?
Type of post hoc test;
Used to correct possible issues with p value that arise when conducting multiple tests (ie. post hocs) to make the p value more stringent.
What did Field (2009) find regarding post hoc test?
SPSS has 18 types of post hoc tests;
All assumptions met = REGWQ / Tukey
Safest option = Bonferroni
Unequal sample size = Gabriel (if similar size) and Hochbergs GT2 (not similar sizes)
Unequal variances = Games-Howell
What is a linear trend analysis?
Typical linear trend, no change in direction.
What is quadratic trend analysis?
Common linear trend, one change in direction (a curve).
What is cubic trend analysis?
Must have at least four means for trend to be seen, two changes in direction
