When I fit a mediation model using mediation::mediate, like this:
library(mediation)
set.seed(30)
df <- data.frame(x = runif(100),
x2 = runif(100),
y = runif(100), m = runif(100),
p_id = sample(1:5, 100, replace = TRUE),
item = sample(LETTERS, 100, replace = TRUE))
fit.totaleffect <- lmer(y ~ (1|item) + (1|p_id) + x + x2, data = df)
fit.mediator <- lmer(m ~ (1|item) + (1|p_id) + x + x2, data = df)
fit.dv <- lmer(y ~ (1|item) + (1|p_id) + x + x2 + m, data = df)
results <- mediation::mediate(fit.mediator, fit.dv, treat=c('x1', 'x2'), mediator='m')
I get the error, "mediate does not support more than two levels per model".
In another answer, someone says:
"The mediate function in the mediation package takes only a binary
mediator or a numeric mediator. In your case, it seems that your
mediator is categorical but contains more than 2 levels. You can
either convert it to numeric or dummy code it."
However, this does not apply to my data. My data seems to be suitable (the mediator is numeric), based on this.
So what is wrong?
(NB. my actual data doesn’t raise the boundary (singular) warning, but otherwise has the same qualities as the dummy data above).
>Solution :
The code that generates the error is:
out.group <- names(model.y@flist)
n.out <- length(out.group)
if (n.out > 1) {
stop("mediate does not support more than two levels per model")
}
If we look at model.y@flist you’ll see that it has the following values:
> fit.dv@flist
$item
[1] U A A F A Z L B V W C C X P J W X Z A Z A B O N J T K M L A S I V O B Q B T F M S L A N B
[46] O P A Y K U T T Z R O E Y M A B Z S Y U A V Z F J H K X D K V H G S S W E T S M I O Z W R
[91] H N Q M C U T O Y R
Levels: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
$p_id
[1] 3 3 1 3 3 3 1 1 2 4 1 5 4 1 3 3 4 2 1 4 2 4 1 2 4 4 1 5 5 2 2 1 4 1 1 5 3 3 3 2 5 5 1 5 3
[46] 1 2 1 3 5 2 5 4 3 4 3 5 3 2 5 1 1 2 2 5 2 2 1 2 2 1 3 2 3 3 2 4 5 3 5 4 1 3 4 4 1 1 5 5 3
[91] 5 1 4 3 3 4 4 3 2 5
Levels: 1 2 3 4 5
attr(,"assign")
[1] 1 2
The "levels" in the error refer not to the levels of a factor, but to the random effects in the model. The error is getting tripped because there is more than one random effect (indicating more than two hierarchical levels, because the first level is accounted for by the idiosyncratic observation-level error).