Skip to content

[R-meta] RVE: degrees of freedom < 4

2 messages · Cátia Ferreira De Oliveira, James Pustejovsky

Cátia Ferreira De Oliveira
# 
Hello,

I hope you are all doing well and that you will get a nice break during the
festive season.
Following from Tanner-Smith and colleagues' article entitled "Handling
Complex Meta-analytic Data Structures Using Robust Variance Estimates: a
Tutorial in R" where they focus on the issue of having a very number of
degrees of freedom (<4), they make the case that we should not trust the
p-values as they are not reliable. I assume the same can be said for the
remaining information as this small number often reflects a lack of power
or influential points, i.e. we wouldn't trust the estimates either, correct?

Thank you!

Best wishes,

Catia
1 day later
James Pustejovsky
# 
Hi Catia,

The issue is with drawing inferences. When the degrees of freedom are
low (less than 4, say), the point estimates and the sampling variance
estimates should still be unbiased (or close to unbiased, for the
sampling variances). However, with low degrees of freedom, the
sampling variance estimate will be very noisy. Therefore, we don't
have a good way to judge the *uncertainty* in the point estimate.

James

On Sat, Dec 11, 2021 at 1:59 PM C?tia Ferreira De Oliveira
<cmfo500 at york.ac.uk> wrote: