Estimating a Two-way Random Effect Model Using lmer and plm

[ahnm2002]

I am using lmer to estimate a two-way random effect model like this y ~ x1 + x2 + (1|country) + (1|year).

When I use plm(y ~ x1 + x2, index = c("country", "year"), model = "random", effect = "twoways") I get totally different estimates. Here is a reproducible example:

if (!require("pacman")) install.packages("pacman")

pacman::p_load(stargazer, lme4,  plm)

library(gapminder)

fit.lmm <- lmer( log(lifeExp) ~ log(pop) + log(gdpPercap) + 
(1|year) + (1|country), data = gapminder)

fit.plm <- plm( log(lifeExp) ~ log(pop) + log(gdpPercap), data = 
gapminder, index = c("country","year"), model="random", effect = 
"twoways")

stargazer(fit.lmm, fit.plm, type="text")

[tappek]

No issue here. plm and lmer use different estimation techniques so one cannot expect identical results. You may read up on the methods used, e.g., in plm's Vignette as a starter https://cran.r-project.org/web/packages/plm/vignettes/A_plmPackage.html , section "plm versus nlme and lme4".

This has been asked before on SO (https://stackoverflow.com/questions/68801350/estimating-a-two-way-random-effect-model-in-r-using-lmer-and-plm), likely by the same poster; a comment there under the post already gave a similar response.