Calcite | Level 5

## Contrast Statement for repeated measures of ANOVA

What could be an example where a planned contrast might be preferable to a standard post hoc test like Tukey?

1 ACCEPTED SOLUTION

Accepted Solutions

## Re: Contrast Statement for repeated measures of ANOVA

I know this is going to sound kind of rude, but a preplanned contrast is always preferable to a post hoc test, no matter which post hoc test is chosen.  Preplanned contrasts use your prior knowledge of comparisons of interest and so are more powerful than post hoc tests, which are essentially throwing all the comparisons you could possibly make into a pot, and seeing which rise to the top.  Something I learned in grad school: Tukey was a guru of exploratory analysis, thus consider any post hoc test (especially one name for Tukey) as exploratory.  That leaves preplanned contrasts as confirmatory.  Does that make sense at all?

SteveDenham

3 REPLIES 3

## Re: Contrast Statement for repeated measures of ANOVA

I know this is going to sound kind of rude, but a preplanned contrast is always preferable to a post hoc test, no matter which post hoc test is chosen.  Preplanned contrasts use your prior knowledge of comparisons of interest and so are more powerful than post hoc tests, which are essentially throwing all the comparisons you could possibly make into a pot, and seeing which rise to the top.  Something I learned in grad school: Tukey was a guru of exploratory analysis, thus consider any post hoc test (especially one name for Tukey) as exploratory.  That leaves preplanned contrasts as confirmatory.  Does that make sense at all?

SteveDenham

Diamond | Level 26

## Re: Contrast Statement for repeated measures of ANOVA

Typically, one uses a CONTRAST statement to test the hypothesis that

b1*level1 + b2*level2 + b3*level3 ... is equal to zero

Note that if all of the b1 b2 b3 ... are 0 or +1 or –1, you would not use a CONTRAST statement at all, as these can usually be handled via LSMEANS or SLICE. The only real use (in my mind) of a CONTRAST statement is for the case where you don't have all b1 b2 b3 equal to –1, 0, +1, such as

0.5*level1 – 1.0*level2 + 0.5*level3 ...

which tests to see if (the average of the means of level1 and level3) is equal to the mean of level2 (if that is a meaningful thing to you)

By the way, @zahidhasandipu, your title and the text of your message seem to be completely unrelated. I have responded to your title while @SteveDenham has responded to the text of your message.

--
Paige Miller
Calcite | Level 5

## Re: Contrast Statement for repeated measures of ANOVA

I was looking for an example where a planned contrast might be preferable to a standard post hoc test, apologies for the wrong title!

Discussion stats
• 3 replies
• 654 views
• 3 likes
• 3 in conversation