Page 1 of 1

Calculating Standard Errors for Random Effects

Posted: Tue Apr 27, 2010 2:27 pm
by NicolaBretscher
Dear forum,

I would like to know how the standard errors for random effects are calculated by MLwiN. Can anyone help?

I am running a multi-level model using R: the function lmer (from the lme4 package) produces standard errors for fixed effects but does not provide standard errors for random effects. This is an intentional omission on the part of the developer, Douglas Bates. However, I was hoping that if I knew how these standard errors were calculated, in MLwiN for example, then I could implement this myself.

My data set has upwards of 250,000 pupils from roughly 9000 schools in 41 countries. My model is as follows:
lmer.out <- lmer( formula = ZSCMAT ~ ZPV1MATH + Quad_ZPV1 + SchoolAverageAbility + (1 + ZPV1MATH + Quad_ZPV1 + SchoolAverageAbility| COUNTRY) + (1 + ZPV1MATH + Quad_ZPV1 | School), data = mydata)

ie. fixed effects: ZPV1MATH (maths ability measure); quadratic of ZPV1MATH; SchoolAverageAbility
with random effects at country level for each independent variable and random effects at school level for ability and quadratic ability only.

Thanks in advance for any help, Nicola Bretscher