how do you get the upper and lower confidence intervals for generalized linear mixed models?

Question

I am trying to put confidence intervals into a tibble so I can graph them; however, I keep getting an error message saying an incorrect number of dimensions. I have pasted my model, along with the code I am using to attempt to extract the upper and lower confidence limits. Please let me know to get around this.

mod6 <- glmer(count.inv ~ mon.sst.levels + (1 | Year), family = poisson(link = "log"), data = model_set)

tib <-  tibble(
  levels = names(exp(fixef(mod6b))),
  coef = exp((fixef(mod6b))), 
  lower = exp((fixef(mod6b)))[,1], 
  upper = exp((fixef(mod6b)))[,2])

Error in exp((fixef(mod6b)))[, 1] : incorrect number of dimensions

If you could provide a reproducible example, that would help with determining why a model with your data resulted in an "incorrect number of dimensions" error — MBorg, May 11 '21 at 13:49
@MBorg in general it's best practice to provide a [mcve], but in this case the problem is pretty clear by inspection if you know what you're looking at. `fixef(mod6b)` is a vector, so `exp(fixef(mod6b))` is also a vector, so trying to refer to its first column (via `[,1]`) throws an error ... — Ben Bolker, May 11 '21 at 14:13

Ben Bolker · Answer 1 · 2021-05-11T00:39:22.693

How about using broom.mixed::tidy()? (hint, fixef() returns a vector, not an object with multiple columns: how did you get the idea for the code shown in your example?)

library(lme4)
library(broom.mixed)
library(dplyr)
## built-in example from `?glmer`
m1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
                  family = binomial, data = cbpp)
tidy(m1, effects="fixed", conf.int=TRUE, conf.method="profile",
     exponentiate=TRUE) %>% select(term, estimate, conf.low, conf.high)

? Results:

  term        estimate conf.low conf.high
  <chr>          <dbl>    <dbl>     <dbl>
1 (Intercept)    0.247   0.149      0.388
2 period2        0.371   0.199      0.665
3 period3        0.324   0.165      0.600
4 period4        0.206   0.0820     0.449

If you leave out conf.method="profile" you'll get faster, but less accurate, Wald confidence intervals.

You may also be interested in the dotwhisker package.

how do you get the upper and lower confidence intervals for generalized linear mixed models?

1 Answers1