Translating proc mixed to lmer - SAS to R

Question

I have the following SAS code that I would like to write in R. I know the class statement is redundant in R (not necessary).

proc mixed data=in_data;
class G F K kal;
model conc=;
random G F K(F) kal(G*F*K);
ods output covparms=out.cov_out;
run;

I tried the below code, with no luck.

fit <- lmer( conc ~ (1 | G) + (1 | F) + (1 | K/F) + (1 | kal/G:F:K) , sample_1)

with the following output. I was hoping not to get a value for kal or K.

summary(fit)

Random effects:
 Groups    Name        Variance  Std.Dev.
 G:F:K:kal (Intercept) 1.421e-04 0.011921
 F:K       (Intercept) 1.326e-05 0.003641
 F         (Intercept) 6.548e-05 0.008092
 kal       (Intercept) 9.852e-06 0.003139
 K         (Intercept) 1.272e-05 0.003567
 G         (Intercept) 2.165e-03 0.046527
 Residual              4.647e-04 0.021557

Answer 1

Your formula translates to:

conc ~ value is modelled using G + F fixed effects (K|F) random slope of K varying on G and (Z|G/F/K) random slope of Z varying on K nested in F nested in G. Also, while you use \ rather then :, this translates to: (Z|G) + (Z|G:F) + (Z|G:F:K). You do not use 0 + or - 1 in your definition, so intercept is included.

So your model translates to: conc ~ 1 + G + F + (1 + K|F) + (1 + Z|G) + (1 + Z|G:F) + (1 + Z|G:F:K). Is this what you wanted?

What may be problematic is that in your definition K is both the random slope and grouping variable for random effects - is it for purpose?

Check an article by Bates et al. (in press) on lme4 and formulas in this package.