Linear regression with two non-nested random effects (mixed)

[Edlira Shehu]

Dear Statalists,

I am estimating a linear model using the -mixed command in Stata 14.I want to specify to non-nested random intercepts (I have multiple observations for two different variables, advertisers and websites, which are non-nested).

I have specified the model as: mixed DV IV || _all: R.id_ads||id_websites:

My question is: what does the R. after the _all: exactly do? If I estimate the same model without the R. (i.e., mixed DV IV || _all: id_ads||id_websites: ) the estimates vary substantially. Any hint on what causes these differences would be highly appreciated and very helpful.

Thank you very much in advance.

Regards,

Eda

[Clyde Schechter]

The R. tells Stata to create a random effect at each level of id_ads. Without the R, your specification treats id_ads as if it were a continuous variable and Stata tries to fit a model with a random slope on id_ads. The version with R is what you want, given your description of the problem.