I have several data related questions I am hoping to get answers to. I have a 4-week intervention (no control group) and all participants are given and pre and post test assessment consisting of two highly reliable, valid, and correlated measures (BAI and BDI). Here are the questions I am trying to answer:

  1. Is the observed difference in BAI / BDI scores between pre and post tests significant?
  2. Is the observed difference in BAI / BDI scores between pre and post test due to chance or my 4-week intervention?
  3. Does the number of sessions attended have an impact on the change between pre and post test scores (e.g. is there a magic # of weeks to attend, do only those who complete all 4 get significant change, etc.)

    (Note: I'm not sure this one is answerable from data but thought it wouldn't hurt to ask on here)

  4. Is there a way to tell if EXTENDING the intervention from 4 to 6 weeks would have an impact of improving post test scores?
  • $\begingroup$ You have no placebo, no control group, the number of sessions attended may be correlated with significant individual traits (dropout NOT at random)... You are going to have some hard time persuading folks that your anxiety/depression treatment works, and if it does this is not just because of Hawthorne effect... $\endgroup$ Commented Mar 29, 2013 at 19:28

1 Answer 1


1: This looks like a task for a paired t-test (assuming either normality or a big enough sample size). Note: this will compare the means. For investigating the probability that post is bigger than pre (depending on your needs, this may be a more relevant measure), you could try a probabilistic index model.

2: Since you don't have a control group, there's no telling whether your intervention had anything to do with the differences. At most, through Q3: if "more treatment" means better results, this may be indicative.

3: Create dummy variables representing number of sessions >0, >1, >2 and >3 (I'm assuming 1 session per week here). Now build a regression model having the differences as outcome and the dummies as covariates.

4: If you've found anything significant in Q3, then assuming your model is correct and will hold outside of the region investigated (this is a strong assumption), you can try and make statements on that. But look into the rest first. I'm going to be bold and say that you are likely to have not enough data to get solid results...

A final word of warning: you should really apply multiple testing correction (so don't just use the (arbitrary) 5% error levels).


Your Answer

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

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