Specifying nested random effects in "brms"

Question

I am new to using "brms" and am encountering an issue when specifying the model formula. The error occurs in the specification of the random intercept (three variables/terms denoting the nested, hierarchical group structure). The error message I encounter is shown below:

Setting 'rescor' to FALSE by default for this model
Error in str2lang(x) : <text>:1:17: unexpected ')'
1: ~ var4/var3/var1)
                    ^

Could someone please lend a hand in resolving this issue? Any help would be greatly appreciated. My code is shown below:

result <- brm(
  bf(
    mvbind(var9, var10, var15, var16) ~ var2 + var13 + var14 + var19 + var20 + var34 + var46 + var47 + var50 + var51 + var52 + var56 + var57 + var59 + var60 + var62 + var65 + var66 + var89 + var90 + var91 + var92 + var93 + var94 + cor_ar(var2 + 1 | var4/var3/var1),
           zi ~ var2 + var5 + var114 + cor_ar(var2 + 1 | var4/var3/var1)
    ),
  data = y,
  family = zero_inflated_negbinomial(),
  iter = 1000,
  chains = 4,
  future = TRUE)

Other aspects of the model formula that I am unsure of are:

specification of autoregressive covariance structure using cor_ar()
setting future = TRUE for parallel computation of chains.

Please let me know if these are specified correctly. Thank you!

(1) does it work if you leave out cor_ar()? (2) What is str(y$var2) ? (3) cor_ar() is deprecated, you should use ar() instead, see here (4) you may have to expand your nested structure explicitly, i.e. into three separate components (var4, var4:var3, var4:var3:var1) — Ben Bolker
– Ben Bolker, Commented Apr 16, 2024 at 20:29
PS I would suggest trying to debug his by working up from simpler examples ... — Ben Bolker
– Ben Bolker, Commented Apr 16, 2024 at 20:54
Thank you for your help, @BenBolker (1) it works without cor_ar() but the execution halts due to other issues, possibly syntax-related. (2) str(y$var2) is a factor with 165 levels. (3) Thank you for the info. I will look into how to implement it in the model formula. (4) If I expand my model formula to make the nesting explicit, would the following be correct: (var2 | var4) + (var2 | var4:var3) + (var2 | var4:var3:var1)? — StatisticsFanBoy
– StatisticsFanBoy, Commented Apr 16, 2024 at 21:50

Ben Bolker · Accepted Answer · 2024-04-16 22:43:49Z

The beginning of a simulation setup for testing brm syntax. This is not working right now because, for ar(), values of time within groups must be unique. It's hard to construct the nested factor structure and make sure that time values are unique within groups ... or at least not obvious ...

Nevertheless I thought I'd leave this here temporarily in case it's useful.

library(brms)
library(glmmTMB)
library(dplyr)
nn <- 1000
ng <- 50
set.seed(101)
s <- function() factor(sample(1:ng, size = nn, replace = TRUE))
dd <- data.frame(x1 = rnorm(nn), x2 = rnorm(nn),
     x3 = rnorm(nn), f1 = s(), f2 = s(), f3 = s())
dd <- (dd
   |> group_by(f1, f2, f3)
   |> mutate(time = factor(seq(n())))
)

form <- y ~ x1 + x2 + x3 + 
    ar1(0 + time|f1) + ar1(0 + time|f1:f2) + ar1(0 + time|f1:f2:f3)
dd$y <- simulate_new(
    form[-2], ## formula without LHS
    ziformula = ~1,
    newdata = dd,
    family = gaussian,
    newparams = list(beta = 1:4,
            theta = rep(1, 6),
            betad = 1,
            betazi = 1))[[1]]

brm( y ~ x1 + x2 + x3 + 
    ar(time = time, gr = f1),
    family = gaussian,
    data = dd)

Collectives™ on Stack Overflow

Specifying nested random effects in "brms"

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related