1
$\begingroup$

I am performing a meta analysis on longitudinal cohort studies reporting outcomes in cognitive performance as a function of adherence to a nutritional variable. Most of them provide "standardized means differences" or "standardized beta regression coefficients" which I can easily convert to a correlation coefficient.

However, there are some studies that only report the results of their regression analysis as raw means differences (95% CI, p value) derived by using the highest tertile (vs. lowest and middle) as reference for comparison. These means differences are expressed in transformed scores, either T or Z scores.

How can I compute a Cohen's d coefficient with this information?

My first thought was to calculate SE/SD from the CI but given the fact that they are asymmetrical I don't think that makes any sense. Should I just divide the value by the SD of T-Scores (SD = 10) or Z-Scores (SD = 1)?

$\endgroup$
1
$\begingroup$

Divide the mean difference by 10 for the T-scores and 1 for the Z-scores. The only issue here is that you are effectively dividing by the overall standard deviation rather than the within groups pooled standard deviation. For small values of $d$ this won't make much difference. For larger values it will. You can compute the pooled standard deviation for the T-Score values as $$ s_{pooled} = \sqrt{ \frac{10^2(N-1) - \frac{\left(\overline{X}_1^2 + \overline{X}_2^2 - 2 \overline{X}_1 \overline{X}_2 \right)} {N}}{N}} . $$ For the z-score, drop the $10^2$ from the equation.

$\endgroup$

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.