Skip to content

[R-meta] Meta-regression question

3 messages · Hellen Mirr, Wolfgang Viechtbauer

Original
Hellen Mirr
# 
Hello

Apologies if this has already been answered and thank you in advance for
your help!

We are doing a multilevel meta-analysis and would like to run a
meta-regression to see differences across performance as a function of a
grouping variable we have.

Our dataset structure is as follows.
We have two patient groups (A and B) and their performance on a series of
tasks across several studies. All these tasks are of specific categories,
e.g. Computing, Imagery, Reading - etc. in the variable "task_category"

We have fit a multi-level meta as follows:
model1 <-rma.mv(yi, vi, random = ~ 1 | Study/Task, tdist=TRUE, data=df)
and then applied RVE as some studies reported multiple effect sizes for the
same patients
robust_model<- robust(model1, cluster = df$Study, adjust = TRUE) and also
with clubsandwich
coef_test () function for better small-sample adjustments.

We would now like to run a meta-regression to see if there are differences
in performance based on the category of tasks (which is coded in the
variable "task_category")

Would this be appropriate? As I am getting a bit confused and not sure this
is the right approach
metaregression_model <-rma.mv(yi, vi, random = ~ 1 | Study/Task, tdist =
TRUE, data = df,
                                 method = "REML",
                                 mods = ~ (task_category))


Thank you!

Best,
Hellen
Wolfgang Viechtbauer
# 
Dear Hellen,

Just to clarify: Is 'Task' the same as 'task_category'? Or is task_category 'broader', such that there may be multiple tasks within the same study that fall under the same category?

In the latter case, you could even consider the model:

random = ~ 1 | Study/task_category/Task

But whether this is sensible also depends on how often you actually have multiple tasks for the same category within studies (if this occurs only rarely, then distinguishing these two levels is difficult).

Regardless, yes, your last model does make sense, but you don't need the parentheses, so 'mods = ~ task_category' is sufficient (assuming that 'task_category' is either a factor or a character variable; if it is numeric, then you need 'mods = ~ factor(task_category)').

Best,
Wolfgang
Hellen Mirr
# 
Dear Wolfgang,

Thank you very much for your reply, your explanations are always super
useful!

To answer your question, the "task_category" variable is indeed  'broader'
than that of "task", because there are multiple tasks within the same study
that do indeed fall under the same category

Thanks again for the help!


On Mon, Nov 9, 2020 at 12:58 PM Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote: