Skip to content
Prev 3817 / 5636 Next

[R-meta] possible miscalculation of Cook’s distances

Just as a brief follow-up with respect to the calculations: They are correct. Continuing with your code, Antonina, we can easily manually compute the Cook's distances here:

cooksd[1:2]

sub <- rma.mv(yi, vi,
              random = ~ 1 | study/experiment,
              data=dat, subset=-1)

(coef(res.ml) - coef(sub))^2 / vcov(res.ml)

sub <- rma.mv(yi, vi,
              random = ~ 1 | study/experiment,
              data=dat, subset=-2)

(coef(res.ml) - coef(sub))^2 / vcov(res.ml)

As you will see, these match up.

With these data, removal of a single row can have a large impact on the variance components, which can lead to considerable changes in the pooled estimate.

Best,
Wolfgang