effectsize

effectsize copied to clipboard

Hmm this paper seem to suggest that such negative values should be replaced with 0?

The results presented in this article assure researchers that the common practice of replacing a negative estimate with zero for some prominent estimators is better than the MLE.

(p. 406)

Should we do this?

Originally posted by @mattansb in https://github.com/easystats/effectsize/issues/519#issuecomment-1279285963

That's the Shieh article? Yes we should do that (considering also that users do it anyway)

Yeah, I agree with Brenton. We should replace them with 0.

The negative values can especially be confusing because the CI values are, IINM, never below 0 in {effectsize} output.