I am trying to run a function which calculates the marginal effects for different mixed effects models, based on two different main predictors (var1 vs. var2). The original code can be found here: https://stats.idre.ucla.edu/r/dae/mixed-effects-logistic-regression/. Below is a reproducible example:
I create a dataframe (ex):
time <- seq(from = 1, to = 500, by =1)
var1 <- factor(sample(0:1, 500, replace = TRUE))
var2 <- factor(sample(0:1, 500, replace = TRUE))
var3 <- sample(1:500, 500, replace = TRUE)
group <- rep(1001:1005, 500)
out <- sample(0:1, 500, replace = TRUE)
group <- as.factor(group)
ex <- data.frame(time,var1,var2,var3,group,out)
Run the models:
m1a <- glmer(out ~ time + var1 + (1|group), data=ex, family = binomial(link = "logit"), nAGQ = 1,
control = glmerControl(calc.derivs = FALSE))
m1b <- glmer(out ~ time + var2 + (1|group), data=ex, family = binomial(link = "logit"), nAGQ = 1,
control = glmerControl(calc.derivs = FALSE))
Create subsets of the data with only the predictors for complete cases:
sub1a <- na.omit(ex[, c("time", "var1", "group")])
sub1b <- na.omit(ex[, c("time", "var2", "group")])
I cannot attach my data frame, ex, because R says var1 and var2 are masked. Therefore, the only way I know to refer to the variables is using $. However, every function I create produces a wrong or null result. I first tried:
marg <- function(v1, v2, d, m) {
biprobs <- lapply(levels(v1), function(var) {
v2[ ] <- var
lapply(time, function(ti) {
d$time <- ti
predict(m, newdata = d, type = "response")
})
})
plotdat <- lapply(biprobs, function(X) {
temp <- t(sapply(X, function(x) {
c(M=mean(x), quantile(x, c(.25, .75)))
}))
temp <- as.data.frame(cbind(temp,time))
colnames(temp) <- c("PP", "Lower", "Upper", "Dayssince")
return(temp)
})
plotdat <- do.call(rbind, plotdat)
}
result1 <- marg(ex$var1, sub1a$var1, sub1a, m1a)
Although this creates a data frame, it produces the same predicted probabilities for each level of var1 (0 vs. 1) at a given time (1-500), which is not what I want. So then I tried:
marg <- function(v, d, m) {
biprobs <- lapply(levels(ex$v), function(var) {
d$v[ ] <- var
lapply(time, function(ti) {
d$time <- ti
predict(m, newdata = d, type = "response")
})
})
.....
}
result2 <- marg(var1,sub1a, m1a)
This produces a null result. I also tried, which produces a null result:
marg <- function(d1,v,d2,m) {
biprobs <- lapply(levels(d1$v), function(var) {
d2$v[ ] <- var
lapply(time, function(ti) {
d2$time <- ti
predict(m, newdata = d2, type = "response")
})
})
......
}
result3 <- marg(ex,var1,sub1a,m1a)
I also tried creating a new object to input directly into the function:
v1 <- ex$var1
marg <- function(d, m) {
biprobs <- lapply(levels(v1), function(var) {
.....
})
})
That also produces a null result. How do I refer to different variables in an unattached data frame?? The code works with direct inputs, so it's a matter of correctly defining the function arguments. I appreciate any help!
