Solved – Plotting predicted values with lmer

Question

I would like to plot a prediction graph in R using this model :

mod7<- lmer(log(BAI) ~ LogSt(Hegyi, calib = Hegyi) + log(BA)+ Number_graft + (1|Tree_label)  + (1|Year)+(1|Site), mydata, REML = FALSE)

all variables are continuous except number graft is categorial.

I used this code :

mydata$predlmer = predict(mod7)
ggplot(mydata,aes(x= LogSt(Hegyi), y = log(BAI), color = as.factor(Number_graft )))+
  geom_point() +
  geom_line(aes(y = predlmer), size = 1)

And I obtained this plot :

I don't understand how to smooth and overlay the lines. Is there anyone who can help me ?

Thank you !

Answer 1

Here is a minimal example using a dataset from lme4. The modelr library has some handy functions for doing this. The strategy is to create a different dataset which has all the combinations of predictors you want to predict and plot for. data_grid from modelr does this by taking the Cartesian product of a grid of the variables in your dataset and then converts that to a tibblle.

library(tidyverse)
library(lme4)


sleepstudy

model<-lmer(Reaction ~ Days + (Days + 1|Subject), data = sleepstudy)

predicted_values<- modelr::data_grid(sleepstudy, Days, Subject) %>% 
                   modelr::add_predictions(model)


predicted_values %>% 
  ggplot(aes(Days, pred, color = Subject))+
  geom_line()+
  geom_point(data = sleepstudy, aes(Days, Reaction, color = Subject))

The result is