In my analysis I compare 3 treatment groups with a pre- and posttest (emotionality scores 0 - 100), so I want to use the repeated measures. Unfortunately, the pretest scores differ significantly amoungst groups which makes me think that simply running the test would not be valid. When I transfer the scores to relative ones, I get uninterpretable data due to the method of measurement (0 = not emotional, 100 = very emotional): a baseline score of 40 and postscore of 80 should not be interpreted as equal to a baseline score of 2 and a postscore of 4.

Any ideas on how to tackle this problem?

  • $\begingroup$ Was this a randomized experiment? If no, e.g. stratifying the analysis by propensity score may be an option. $\endgroup$ – Björn Jan 13 '16 at 12:27
  • $\begingroup$ It was a randomized experiment indeed. $\endgroup$ – Guest with a question Jan 13 '16 at 12:45

Probably the best method is a multilevel model, with person as a random effect and a random intercept. While ordinary regression has this model:

$Y = X \beta + \epsilon $

a multilevel model is:

$Y = X \beta + Z \gamma + \epsilon$

These models have been extensively discussed here and in many other places.

  • $\begingroup$ Thank you for your input. With multilevel model I assume you mean that I should use a mixed factor analysis? Because that was what I meant by repeated measures, I use SPSS and add Condition as Between-Subjects Factor, my bad. Could you perhaps elaborate a bit more as in how to run the analysis? I'm sorry to ask but am quite new to statistics. $\endgroup$ – Guest with a question Jan 13 '16 at 11:26
  • $\begingroup$ I have no idea what a mixed factor analysis is and I don't use SPSS. But I will elaborate in my answer. $\endgroup$ – Peter Flom Jan 13 '16 at 12:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.