[R-meta] Reasonable rma.mv() syntax for a longitudinal meta-analysis

Simon Harmel @|m@h@rme| @end|ng |rom gm@||@com
Wed May 5 02:17:21 CEST 2021

Hello List Members,

I'm doing a meta-analysis where the effect sizes (yi) are computed as the
difference in pre-post changes of a treatment group and that in a control
group. That is:

yi = (standardized mean change of a treatment) - (standardized mean change
of a control)

As a result, I have a categorical variable (post_id) that indicates the
pre-post interval associated with each "yi" (`1` = pre to post 1;  `2` =
pre to post 2, . . .).

Almost all studies have more than 2 treatment groups, a couple have more
than one control group, and about 15 studies have used more than 1 outcome
but these related outcomes within studies don't represent the same thing
across these 15 studies. In each study, a "yi" for each unique situation
(e.g., having 2 control groups, 2 or more outcomes etc.) has been computed.

I was wondering given my data structure, is my syntax reasonable?

Many thanks, Simon


dat <- read.csv("https://raw.githubusercontent.com/ilzl/i/master/pr.csv")

dat <- transform(dat, post_id = factor(post_id))

V <- impute_covariance_matrix(vi = dat$vi, cluster = dat$studyID, r = .7)

mv1 <- rma.mv(yi ~ 0 + post_id + x1 + x2, V = V, struct= "AR", random =
list(~post_id|studyID, ~1|esID), data = dat)

	[[alternative HTML version deleted]]

More information about the R-sig-meta-analysis mailing list