[R-meta] SMD formula for between-groups pre/post data
m@th|@@@b||mont @end|ng |rom gm@||@com
Tue May 9 22:27:43 CEST 2023
I am conducting a meta analysis of RCTs that explores the efficacy of
mindfulness on several physical health outcomes.
All my main outcomes are measured on a continuous scale. For most of
the studies, I have been able to gather means and SD of the two groups
at pre- and post-tests (or, when this information was not available,
means/SD change). However, in about 20% of the trials, no information
on pre-test or mean change was reported.
I would like to combine the studies with pre/post data with those that
have reported only post data. I plan to make a meta regression to
assess whether these two types of information lead to differences in
effect size. However, I am not sure which is the best SMD formula to
use for studies with pre/post data in this particular situation.
I understand that if I wanted to pool the ‘between-groups post-test
SMD’ with the ´within-group pre/post-tests SMD’, I would need to
compute Cohen d-av for the pre/post group (ie, SMCR in escalc).
However, because my meta-analysis is restricted to between group
comparisons (either group differences on post data or group
differences in score change from pre to post test), I wondered whether
computing d-av for studies with pre/post data was still the best
choice (or whether dz should be preferred).
# code for the studies with pre/post or mean change data
dav_grp1=metafor::escalc(m1i=grp1_pre, m2i=grp1_post, sd1i=grp1sd_pre,
dav_grp2=metafor::escalc(m1i=grp2_pre, m2i=grp2_post, sd1i=grp2sd_pre,
d = dav_grp1$yi - dav_grp2$yi
var_d = dav_grp1$vi + dav_grp2$vi
I initially planned to use the formula reported in Borenstein’s book
to estimate the between-groups pre-post SMD but I saw in escalc code
that this formula was not accurate. I thus prefer asking for advice.
Many thanks for the help!
University of Lille, France
PS: I have dug through the list to explore whether this question has
already been raised but I was not able to find a similar one.
More information about the R-sig-meta-analysis