Wk 7 - ANOVA 2

Question 1

What do follow-up tests tell us? (x1)

Answer 1

Where a difference found in the omnibus ANOVA is among the means of diff groups

Question 2

What is the diff between a priori and post-hoc tests?

Answer 2

A priori are in advance, based on specific predictions

Post-hoc are when you find unplanned/predicted significance

Question 3

What is the primary advantage of linear contrasts over pairwise tests?

Answer 3

They correct for the high family-wise error rate that would result from running multiple t-tests

Question 4

What are the rules for assigning weights to means in linear contrast?

Answer 4

1. Can only test two chunks at a time, eg mean of 2 v 3 groups
2. Groups coded with positive values compared against negative
3. Sum of the weights for comparison must equal zero
4. Group not involved in comparison gets zero (removes from comparison)
5. Weights assigned to groups in one ‘chunk’ of variation need to equal number of groups in other chunk – eg placebo, weighted +2, because there’s 2 x -1 in other chunk (2 treatment groups)

Question 5

What are orthogonal contrasts?

Answer 5

Linear contrasts that have more power due to using the smallest number of tests
Gives non-redundant info from independent questions on data set

Question 6

What rules do we follow to check orthogonality? (x4)

Answer 6

Same as for linear, but also
Products of coefficients down the column (for pairs of comparisons) also = 0
No more than k - 1 comparisons
SS for all contrasts = SStreat

Question 7

How does the Tukey test differ from the Scheffe?

Answer 7

Tukey adjusts alpha as though you’re doing all possible pair-wise comparisons (each group to each other)
Scheffe as if you’re also going to do all non-pairwise (grouped conditions against each other)

Question 8

What is the Bonferroni adjustment? (x2)

Should be used for… (x1)

Answer 8

Corrects alpha to reduce familywise error in multiple comparisons, by giving adjusted t’ for comparison in
Square it to give F’
Linear contrasts

Question 9

What are linear contrasts? (x1)

And benefit?

Answer 9

Comparing of groups of means against each other

More power to detect diffs - doing less comparisons, so less adjustment of alpha

Question 10

How do you conduct linear contrasts using the F ratio?

Answer 10

You can convert data from linear contrast, L, into SScontrast
Then F = MScontrast/MSerror (from ANOVA table)
Find the root of F to give t’

Question 11

How do you do a Scheffe test?

Answer 11

F calculation stays the same, but for comparison:
Multiply original F-critical by k - 1
Gives more conservative Fcrit, which is necessary in post-hoc to avoid capitalising on chance