Is there a way to get standard errors for the coef.merMod method of the lme4 package, or is it nonsense to compute SE for the coefficients?
Example:
library(lme4)
fit <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
coef(fit)
$Subject
(Intercept) Days
308 253.6637 19.6662579
309 211.0065 1.8475828
310 212.4449 5.0184061
330 275.0956 5.6529547
331 273.6653 7.3973914
332 260.4446 10.1951153
333 268.2455 10.2436615
334 244.1725 11.5418620
335 251.0714 -0.2848731
337 286.2955 19.0955699
349 226.1950 11.6407002
350 238.3351 17.0814910
351 255.9829 7.4520288
352 272.2687 14.0032993
369 254.6806 11.3395026
370 225.7922 15.2897506
371 252.2121 9.4791309
372 263.7196 11.7513157
The coef.merMod is simply the sum of ranef + fixef for each explanatory variable for each level of each grouping factor.
With the arm package, you can get standard errors for fixed and random effects with se.fixef and se.ranef (and se.coef, which is, however, only a concatanation of the two first outputs).
My question is: is the SE of coefficients simply coef(ranef) + coef(fixef)? Or do the sums of random and fixed coefficients have no real standard error?
