I have identified 17 studies for use in a meta-analysis. In order to make sure that I analyze the data correctly, I had a couple of questions regarding how to approach two of them.

First, one study uses two different Cognitive-Behavioural techniques (one more cognitive and one more behavioural) and uses the same outcome measure to test the difference between groups. No statistical difference was demonstrated between the two and pre-post scores were used to measure effects without the use of control groups. I am interested in CBT techniques overall, so would it be best to pool the scores together?

Also, two studies were conducted whereby the first one used a delayed-treatment group as a control for an RCT, comparing an outcome measure after the first group received treatment. Four years later, the data from those two groups were combined and then divided between 'depressed' and 'non depressed' and included follow-up data for up to two years after treatment. This means that there is one study that uses an RCT model with fewer participants in the treatment group or a pre-post design with more participants. I am particularly interested in the impact of self-esteem interventions on depression, so my gut instinct is to use the 'depressed' group from the latter study. Would that be reasonable or would it be better to use an RCT instead? For that matter, would it be best to perform a t-test on the effect sizes between the two?

Thank you for any suggestions.

  • $\begingroup$ Can you expand on your second question as on the face of it they would not be addressing the same scientific/clinical question? $\endgroup$ – mdewey Jun 11 '16 at 14:00
  • $\begingroup$ What I would like to know how to best calculate the relevant effect size to use in my meta-analysis, aimed at the effect of CBT on self-esteem and depression. I see three possibilities, but I do not know which would be most valid. I could either use the data from the first study, which is an RCT comparing two independent groups with a lower number of participants or I could use a single-group design with a larger number of participants. For the single-group, I could then either use all of the participants together or just the participants in the 'depressed' group. $\endgroup$ – Dan K Jun 13 '16 at 10:32
  • $\begingroup$ Most of the other studies have participants who are clinically depressed, differentiating between the 'depressed' and 'not depressed' groups could provide an effect size more in line with the rest of the studies involved. I'm aware, though, that there are two issues. First, I don't have the inter-item correlations, which means that I would have to estimate it. Second, being a single-group design will probably lead to a higher effect size than would be expected from an RCT, thus possibly over-inflating the summary effect. $\endgroup$ – Dan K Jun 13 '16 at 10:35

Your Answer

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

Browse other questions tagged or ask your own question.