Hierarchical fractional logistic regression (random effects)

[Marco Riguzzi]

Hi Stata community

Is there a way to do cross-sectional logistic regression with a fractional outcome and random effects for clusters?

Thank you.

Marco

[Joseph Coveney]

"cross-sectional . . . random effects"

No one else has bit(ten) yet and so I'll give it a try.

It's not exactly "random effects" regression, but you could consider something like

xtgee . . ., i(<cluster ID>) family(binomial) link(logit) vce(robust)

or

glm . . ., family(binomial) link(logit) vce(cluster <cluster ID>)

for starters. Both I believe will take a fractional variable as an outcome.

Otherwise, you could try the classic variance-stabilizing arcsine transformation for proportions mentioned by B. J. Winer in his Statistical Principles in Experimental Design (1962 and 1971) and fit the transformed outcome variable with a mixed-effects regression model using, for example, mixed.

[Marco Riguzzi]

Thank you, Joseph. I will give it a try and might also try to get back to you on the interpretation of the LMM for the arcsine-transformed outcome.

I suppose, cross-sectional may not be accurate. I have survey data, one survey per individual, multiple individuals per cluster.