Lecture 7- Contrast Coding Flashcards
The f-statistic tests
The overall fit of the model
Model parameters tell us about
Specific differences between means
Dummy coding compares
Each category to the baseline
What are the options for breaking down categorical predictors
- Orthogonal contrasts
- Post hoc tests
- Trend analysis
Features of orthogonal contrasts
- Hypothesis driven
- Planned a priori
- Control type 1 error rate
Features of post hoc tests
- Not planned (not hypothesis driven)
- Compare all pairs of means
- Multiple t-tests adjusted for the number of tests
When is a Trend analysis useful
For ordered means
With planned contrasts, the variability explained by the model (SSM) is due to
Participants being assigned to different groups
With planned contrasts, the variability of the model (SSM) sometimes represents
An experimental manipulation
With planned contrasts, the variability of the model (SSM) can be broken down further to test
Specific hypotheses about which groups might differ
To control type 1 error rates contrasts must be
Independent
With independent contrasts, if a group is singled out in a contrast then that group
Should not be used in any subsequent contrasts
With k-1 contrasts you should always end up with
One less contrast than the number of groups
With only 2 chunks each contrast should compare
Only 2 chunks of variation
The first contrast will usually compare
Any control conditions (chunk 1) with any experimental ones (chunk 2)
Groups coded with positive weights compared to groups coded with
Negative weights
The sum of the weights for a comparison should be
Zero
If a group is not involved in a comparison, assign it a weight of
Zero
For a given contrast, the initial weight assigned to the groups in one chunk of variation should be
Equal to the number of groups in the opposite chunk of variation
How to get the final weight
Divide the initial weights by the number of groups with non-zero weights
In the absence of a specific hypotheses compare
All pairs of means to see where the specific differences lie
How to avoid inflation in the type 1 error rate in post hoc tests
Adjust the alpha (or test statistic) to be more conservative
What does trend analysis (polynomial contrasts) do
- Test for trends in the means
- Only makes sense for ordered groups
