[R-meta] Testing interaction in rma.mv()

[Viechtbauer, Wolfgang (SP)]

Dear Stefanou,

Just include training_hr and time as 'main effects', no interaction. The intercept then referes to pre-testing occasion (with 0 training hours), the coefficient for time to the average difference between the post-testing occasion versus pre-testing occasion (when 0 training hours are provided at post-testing; you could rescale the variable to make the intercept correspond to a different number of training hours), and the coefficient for training_hr indicates how this average difference changes for a one-unit increase in training_hr.

Best,
Wolfgang

>-----Original Message-----
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On
>Behalf Of Stefanou Revesz
>Sent: Thursday, 17 March, 2022 5:39
>To: R meta
>Subject: [R-meta] Testing interaction in rma.mv()
>
>Dear Meta Community,
>
>My studies measure the effect of a training program to improve
>teachers' self-efficacy.
>
>We have coded for the 'length of the training sessions' (training_hr)
>up until each testing occasion (time).
>
>At the pre-testing occasion, since no training is provided, we coded 0
>hrs, and for subsequent testing occasions, we coded whatever hours
>reported in the studies.
>
>The problem is that now we CAN'T use 'training_hr * time' in our
>`model`. Because 'training_hr' at time0 is just '0' for all studies
>(see below).
>
>Q: Is our current coding of 'training_hr' wrong? Sh/Could we code
>'training_hr' differently for such an interactive model?
>
>Any help would be highly appreciated,
>Stefanou
>
># Data and code
>d <- read.csv("https://raw.githubusercontent.com/fpqq/w/main/f.csv")
>
>subset(d, time == "Baseline")$training_hr
>#[1] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
>
>model <- rma.mv(gi~training_hr*time, v_gi, random = ~1|study/obs, data = d)