2
$\begingroup$

To conduct a meta-analysis, I want to calculate the effect sizes (standardized mean differences) of several studies. In some of my studies, not all the information needed is given:

  1. In two studies, the only information I can extract are the pre-post ANOVAs (within-group-contrasts) for the treatment group as well as for the control group. No means, standard deviations or between-group-contrasts are reported.

Is it possible to calculate effect sizes based on this information?

  1. In one study, the standard deviations are missing. I found this post related to my question: Can I calculate effect size using F, p and N (sample size)?. One per "[...] I think several metrics, for example log response ratio of means (ROM in metafor) does not require SD to calculate ES. The studies can be later on weighted using sample sizes instead of the variance (e.g. de Graaff et al. 2006; van Groenigen et al. 2006)"(Can I calculate effect size using F, p and N (sample size)?) Is this an adequate way of dealing with the problem?
$\endgroup$
  • $\begingroup$ Note that the response ratio of means requires that the dependent variable is a true ratio scale, otherwise a ratio of the means is meaningless. The literature on ROM has not always made that sufficiently clear. $\endgroup$ – dbwilson Sep 18 '17 at 17:33
  • $\begingroup$ 1 in body of question is not clear to me. What is the information (actual) available to you? $\endgroup$ – Subhash C. Davar Sep 19 '17 at 12:03
1
$\begingroup$

For your case 1 it looks like you may have to resort to asking the original authors if they can let you have the information you need. If they are fairly recent studies you have a non-zero chance of success here as people are often helpful.

For your case 2 you also have the option of imputing the standard deviations. If you have several studies using the same measure then use a typical value, for some meaning of typical. If you have no comparable studies but the measure has bee widely used then again you may be able to convince your audience of the typical value. If all else fails use a range of values and see if it alters your conclusions.

$\endgroup$

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.