Statistical test for repeated measures ANOVA with covariates We have 10 participants who have undergone a cold pressor test (for pain) in 3 conditions (no drug, drug A, drug B). It is a repeated measure design. The stimulus is continued for 1 minute after they start reporting pain. The length of time before they report pain is their 'tolerance / threshold'. Each participant, in each session, gets the stimulus (i.e., 1 minute of cold after they start reporting pain, i.e., after the pain threshold has been reached). The pain score reported on VAS (a continuous scale) is the main outcome. However their tolerance / threshold is a covariate (their pain scores may be related to their tolerance times which would otherwise not be controlled for). 
How should this data be dealt with? Should I analyse the threshold and pain report (VAS) as 2 separate 'independent' outcomes? Or should I stick with pain report as the main outcome (and use Pain threshold as a covariate? If the latter, how do I do it? Using RMANOVA in SPSS? 
 A: No advice on doing this in SPSS, but it appears to me that the pain threshold is in the causal path, I.e., it is affected by the drug. 
If so, analysis of covariance techniques whereby you compare adjusted means of VAS -- predictions of VAS at the same pain threshold -- could be very misleading because you wouldn't expect the same mean pain threshold with each drug. 
Continuing with the idea that pain threshold is in the causal path... One option is to simply ignore the pain threshold data. But the downside is that it could account for a lot of variation in the response variable, so the power of your tests are reduced. 
Another option is to adjust the covariate for treatment effects before fitting the model, then use the adjusted covariate instead of the original one. The ANCOVA textbook by Rutherford (publ by Wiley) describes this technique. Then you get the reduction of error variance without obtaining misleading predictions. The basic technique is to fit the treatment model with pain threshold as the response, then the adjusted pain threshold values are the residuals from this model.
I don't use SPSS a so I can't tell you how to fit a covariance model with repeated measures, but it ought to be possible.
