Calculating cohen's d from regression coefficient?

[Patrick]

Hello everybody,

I would like to perform a meta-analysis using the Hedges and Olkin
approach for which I need Cohen's d or Hedge's g as the effect size
values.

One of my selected studies reports regression coefficients only (beta
and standard error).
Is there any way to transform these into Cohen d or Hedges g
(irrespective of the fact that betas are adjusted values)?

Could not find the appropriate formula. I hope that somebody of you
can help.

Greetings
Patrick

[Ray Koopman]

Post by Patrick
Hello everybody,
I would like to perform a meta-analysis using the Hedges and Olkin
approach for which I need Cohen's d or Hedge's g as the effect size
values.
One of my selected studies reports regression coefficients only (beta
and standard error).
Is there any way to transform these into Cohen d or Hedges g
(irrespective of the fact that betas are adjusted values)?
Could not find the appropriate formula. I hope that somebody of you
can help.
Greetings
Patrick

If one of the variables is dichotomous then t = b/SE[b] is the
same as the pooled-variance t for testing the difference between
the two group means. Then look up how to convert t to d or g.

[Patrick]

Great, thank you!

[Patrick]

Btw, it is also possible to convert Odds Ratios into Cohen d's?

[Patrick]

Btw, is it also possible to convert Odds Ratios into Cohen's d's?

[Ray Koopman]

Post by Patrick
Btw, is it also possible to convert Odds Ratios into Cohen's d's?

In general, no. But for two homoscedastic logistic distributions,
dichotomized at the same threshhold value, d = ln[OR]*sqrt[3]/pi.

[t***@gmail.com]

Post by Ray Koopman

Post by Patrick
Btw, is it also possible to convert Odds Ratios into Cohen's d's?

In general, no. But for two homoscedastic logistic distributions,
dichotomized at the same threshhold value, d = ln[OR]*sqrt[3]/pi.

Hey there Ray - i know this is an old thread but I have a new question: is it possible to convert the regression coefficient into a correlation coefficient? I am also carrying out a meta-analyis and facing this problem.

Best,

Tor

[Bruce Bradbury]

For a bivariate regression it is straightforward: If you have an estimate of b in
y = a + bX + e
then the correlation coefficient is
r = b sX / sY
where sX is the standard deviation of X and sY the standard deviation of Y.

[Rich Ulrich]

Post by t***@gmail.com

Post by Ray Koopman

Post by Patrick
Btw, is it also possible to convert Odds Ratios into Cohen's d's?

In general, no. But for two homoscedastic logistic distributions,
dichotomized at the same threshhold value, d = ln[OR]*sqrt[3]/pi.

Hey there Ray - i know this is an old thread but I have a new question: is it possible to convert the regression coefficient into a correlation coefficient? I am also carrying out a meta-analyis and facing this problem.

I will comment. After a question. Are you doing epidemiology,
or something else with extreme Ns? - or just something where
a study happens to use logistic regression?

The OR is a good measure of effect size when the sample Ns are
widely different, whereas the correlation r is not.

Aren't you using d, anyway? (If not, why not?)