I am running a mixed-effects model in R and I want to have random intercepts for two sampling levels, country and site. The variable site is perfectly nested within the level country. So, is it necessary to model a separate term for the random intercept of country, or will simply modeling the error for site be suitable?

I.e., would the model

lmer(outcome ~ predictor + (1|country) + (1|site))

have any benefit over the model

lmer(outcome ~ predictor + (1|site))
  • $\begingroup$ I think you meant to use random intercepts in your title? Do you have multiple sites for each country? $\endgroup$ May 15, 2018 at 21:57
  • $\begingroup$ Thanks @IsabellaGhement. I did mean random intercepts. And I do have random sites for each country. I should add that I am mostly interested in getting good estimates for my (fixed) predictor variable. $\endgroup$ May 16, 2018 at 19:09
  • 1
    $\begingroup$ There may be a country characteristic that effects your predictor. I would add the country level, though I'm relatively new to MLM. $\endgroup$
    – dankernler
    May 16, 2018 at 20:07
  • 2
    $\begingroup$ If site is nested within country, then the correct specification of the random effects would be: (1|country/site). For more information, see here. Your current specification is for crossed random effects. $\endgroup$ May 17, 2018 at 7:14


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.