We're calculating the overall effect size of 3 types of judgment on 3 types of violent outcomes. Each study in this multilevel meta-analysis has effect sizes for multiple types of judgment and/or multiple types of violence. Understandably no independence of effect sizes. This is why we're using the rma.mv() function in the R meta package.
Now, the question arises to calculate the average effect size per judgment for each violence type. I haven't been able to find guidance thus far as to what the best way to do this is. The most logical way to assess a single effect size per judgment for each violent outcome would be to subset it, pairing judgment with outcome. This would be unencumbered by the other outcomes or judgments. An example is embedded below.
Viol.A<- rma.mv(y, v, random = list(~ 1 | Study, ~ + 1 | Effect), tdist=TRUE, data=df, subset = Outcome==2 & Judgment ==3)
Translation is the overall effect size of Judgment A Score (Judgment ==3) on Violent Offending (Outcome ==2)
Colleagues are wondering about subsetting only the outcome variable, and then putting all judgment types in the same rma.mv() to arrive at the effect size for each judgment for that outcome. Intuitively, it doesn't sit completely right. I have the sense that this output is somehow dependent on the other judgments? An example of doing it this way is embedded below.
Viol <- rma.mv(y, v, random = list(~ 1 | Study, ~ + 1 | Effect), tdist=TRUE, data=df, subset = Outcome==2, mods = ~ A+B+C)
mods = ~ the three Judgments (Instrument A, Instrument B, Instrument C).
Both operations give different effect sizes! Happy to be wrong with my intuition, as long as I can get feedback on which option is correct, and why? Much appreciated, Stack community!
