I have a five condition dataset where all conditions share variablespostscore,prescore, and reading.time.1.

However condition 2 additionally has reading.time.2 and conditions 3-5 additionally have trials and score (but not reading.time.2).

These differences in variables derive from the purpose of the conditions

  • Condition 1: Control
  • Condition 2: Stronger Control (extra reading time equivalent to treatment)
  • Condition 3: Treatment 1
  • Condition 4: Treatment 2
  • Condition 5: Treatment 3

All conditions have approximately 60 participants each (N=300).

So far I have analyzed the data 2 ways

  1. Model A:postscore ~ condition * prescore * reading.time.1 (using data for all five conditions). I found significant differences here b/w two conditions.
  2. Model B:postscore ~ condition * prescore * reading.time.1 * trials * score (using data from just conditions 3-5). I found significant differences here b/w all three conditions.

My question: I'm wondering if there's a way to combine all predictors into one model across all five conditions using some kind of regression. If not, should I be looking at multiple sample/group SEM

  • $\begingroup$ Could you clarify some of the elements of your design/measurement? It's difficult to tell what some variables refer to (e.g., "trials", which sounds like an index for repeated measurement but might not be) and how seemingly related variables are distinct from one another (e.g., "score" vs. "prescore" vs. "postscore")? $\endgroup$
    – jsakaluk
    Commented Jan 24, 2017 at 1:02
  • $\begingroup$ prescore and postscore are proportion correct on pre- and post-tests of knowledge. trial and score are treatment variables that refer to the number of trials needed to meet a learning criterion and the score obtained across those trials (proportion correct). $\endgroup$ Commented Jan 25, 2017 at 4:51


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.