repeated measures anova assumptions

[LucyB]

I am conducting a repeated measures anova with the follwing code:

proc glm data=data;

  class exposure;

  model pre post = exposure;

  repeated time 2;

run;

I am wondering if the normality assumptions on the outcome need to be for the change in the post - pre scores here? Do I output residuals like a normal anova to assess this? if there is a violation, do I transform both the pre and the post scores?