Solved
Contributor
Posts: 46

# PROC Mixed Multiple Contrast statements

I have 5 treatments (lets call A, B, C, D, E). I want to compare:

A to B

A to average of B, C, D, E

linear (not including A)

I wrote the following contrast statements, but it seems like A is being used in my linear & quadratic contrasts.

CONTRAST 'A vs B'                           trt  1  -1 0  0  0;

CONTRAST 'A vs ave of B to E'         trt  4 -1 -1 -1 -1 ;

CONTRAST 'Linear B to E'                 trt  0 -3 -1  1  3;

CONTRAST 'Quadratic B to E'            trt  0 -1  3 -3  1;

How do you make sure that in the linear & quadratic contrasts, trt A is not being used?

Any help verifying these statements would be appreciated.

Accepted Solutions
Solution
‎03-11-2013 07:50 AM
Posts: 2,655

## Re: PROC Mixed Multiple Contrast statements

These look like they should be OK.  What indication do you have that trt A is being used?  To check, add the /e option, and the L matrix coefficients will be displayed.

Steve Denham

All Replies
Solution
‎03-11-2013 07:50 AM
Posts: 2,655

## Re: PROC Mixed Multiple Contrast statements

These look like they should be OK.  What indication do you have that trt A is being used?  To check, add the /e option, and the L matrix coefficients will be displayed.

Steve Denham

Contributor
Posts: 46

## Re: PROC Mixed Multiple Contrast statements

It seems like A is being used, because when I exclude A from the data and run linear/quadratic contrasts, .

Trt diet A is much different than the other 4 trt diets. Trt A is a traditional diet and was mainly included to compare

A to B.  Thus, I think is OK to run the analysis, only using A and B.  Then re-run the analysis using only B to E?

Then, just write the paper as Trial 1 and Trial 2?

See any problems using trt B in both analysis?

Again, your time/knowledge is much appreciated.

Travis

Posts: 2,655

## Re: PROC Mixed Multiple Contrast statements

The analysis should reflect the design.  Separate analyses leads to reduced power, and the lack of indepence in the comparisons (A to B, B to the others) is lost when you do separate analyses.

When you exclude A, and run the contrasts (assuming that you rewrite them so that ordering is correct), the p value is going to change.  You change both the residual error estimate and the degrees of freedom available for the comparison.

Steve Denham

🔒 This topic is solved and locked.