I am running a mixed effects model on my dataset ,
library(lme4)
data(cake)
each dataset is a subset of a larger dataset
subset(cake, recipe=="A")
subset(cake, recipe=="B")
subset(cake, recipe=="C")
I am using dlply to run my mixed effects model on each subset
MxM1 <- plyr::dlply(cake,
"recipe",
function(x)
lmer(angle ~ 1 + (1|replicate) + temperature,
data=x))
How do I estimate the predicted value ?
I tried the ldply function,
cake$pred <- ldply(MxM1, as.data.frame(predict))$value
but I am not sure if the estimated predicted values are correctly merged with the observation in the dataset. Because if this my dataset
head(cake)
replicate recipe temperature angle
1 1 A 175 42
2 1 A 185 46
3 1 A 195 47
4 1 A 205 39
5 1 A 215 53
6 1 A 225 42
I like the predicted value to match the observation for which it is predicting the value, and I am not confident if the output from ldply(MxM1, as.data.frame(predict))$value aligns accurately with the observations in the dataset. In other words my worry is the output from ldply(MxM1, as.data.rame(predict))$value might be out of sequence with the observations in the dataset. Any suggestions how to resolve this issue and add the predicted value accurately back into the dataset is much appreciated.
I'm not sure if it's correct to calculate separate models for the recipes, but you may have your reason.
You can fit and predict by recipe in one step. If you use na.action=na.exclude you can directly cbind the predictions.
res <- by(cake, cake$recipe, \(x) {
fit <- lmer(angle ~ (1 | replicate) + temperature, data=x, na.action=na.exclude)
cbind(x, yhat=predict(fit, na.action=na.exclude))
}) |> do.call(what=rbind)
head(res)
# replicate recipe temperature angle temp yhat
# A.1 1 A 175 42 175 39.40195
# A.2 1 A 185 46 185 41.80195
# A.3 1 A 195 47 195 41.06861
# A.4 1 A 205 39 205 43.80195
# A.5 1 A 215 53 215 48.93528
# A.6 1 A 225 42 225 45.33528