Looping over dependent variable in a mixed model

Question

I am trying to loop over multiple variables in a mixed model (using the rptGaussian function from the rptR package) but I am unable to do it despite several efforts. I am trying the following code. I use the following code without a loop and it works fine:

(rptGaussian(Arg ~  (1|class)+(1|kit)+(1|sex),
    grname=c("class","kit","sex","Fixed"),
    data=ggm2, nboot=10, npermut=10, adjusted=FALSE)

However, when I try to loop more variables I get the error

Error in terms.default(formula) : no terms component nor attribute

I am trying the following code for the loop.

varlist<-c("var1", "var2")

blups.models <- lapply(varlist, function(x) {
  rptGaussian(substitute(i ~  (1|class)+(1|kit)+(1|sex), 
    list(i = as.name(x))), 
      grname=c("class","kit","lab","Fixed"),
      data=ggm2, nboot=10, npermut=10, adjusted=FALSE)
})

Here is a dummy data table:

sex class   kit var1    var2    var3    var4
Female  A   Cont    10.79730768 10  20  18
Female  A   Exp 11.2474347  17  1   17
Female  A   Cont    11.64820939 10  5   17
Female  A   Exp 15.62800413 20  8   4
Female  B   Cont    12.41705885 5   16  8
Female  B   Exp 12.80249244 9   10  1
Female  B   Cont    10.76949177 6   13  2
Female  B   Exp 14.71370141 7   12  11
Male    A   Cont    8.931529823 8   3   6
Male    A   Exp 10.46899683 3   12  13
Male    A   Cont    8.363257621 3   13  17
Male    A   Exp 8.753117911 10  16  10
Male    B   Cont    9.110946315 9   13  4
Male    B   Exp 9.595131886 18  10  17
Male    B   Cont    9.454670188 1   10  11
Male    B   Exp 10.59379123 11  1   3

Answer 1

In general this kind of looping is easier (IMO) with string-based solutions, especially the reformulate() wrapper function, than with substitute() .

I used read.table(header=TRUE,text="...") to read the data above and this slightly modified code for the single model:

library(rptR)
r1 <- rptGaussian(var1 ~  (1|class)+(1|kit)+(1|sex),
    grname=c("class","kit","sex","Fixed"),
    data=ggm2, nboot=10, npermut=10, adjusted=FALSE)

For multiple models:

varlist <- c("var1", "var2")

Make list of formulas:

formulas <- lapply(varlist,
          reformulate,
          termlabels="(1|class)+(1|kit)+(1|sex)")

Apply rptGaussian to formulas:

blups.models <- lapply(formulas, 
      rptGaussian,
      grname=c("class","kit","sex","Fixed"),
      data=ggm2, nboot=10, npermut=10, adjusted=FALSE)

If you want to collapse the results to a nice form, you have to figure out how to extract the results from a single fit into a data frame or similar structure. In this case the result is a rpt object and methods(class="rpt") tells you that there are only print , plot , and summary methods, but the summary() method returns an object that has lots of potentially useful bits. Here's an example:

## extract estimates and standard errors of estimates as a 1-row data frame
sumfun <- function(x) {
    ss <- summary(x)
    se.names <- paste(rownames(ss$se),"se",sep=".")
    cbind(ss$R,setNames(as.data.frame(t(ss$se)),se.names))
}

A possibly-better alternative would be to return data.frame(term=names(ss$R),rpt=unlist(ss$R),se=ss$se) (a 3-column by n-row data frame) instead.

I'm going to use dplyr::bind_rows() because it's handy, but you could use base-R tools ( do.call(rbind(...)) ) instead if you prefer.

names(blups.models) <- varlist
dplyr::bind_rows(lapply(blups.models,sumfun),
                 .id="var")

   var class       kit        sex Fixed   class.se    kit.se    sex.se Fixed.se
1 var1     0 0.1444659 0.65887365     0 0.04992624 0.2136589 0.2954982        0
2 var2     0 0.3322780 0.01734343     0 0.01981748 0.2243989 0.1158878        0

Are you sure it makes sense to do repeatability scores across sexes and other categories with small numbers of levels?