4

I am currently trying to plot a the outcome of an lme4::lmer function, following this tutorial. I've tried the code in the tutorial, and it works as intended. In the tutorial a column pred_dist is added to the dataset for the fitted function before plotting. When I try this with my own data, I get the following error:

Error: Problem with `mutate()` column `pred_dist`.
ℹ `pred_dist = fitted(model0)`.
ℹ `pred_dist` must be size 15 or 1, not 17192.
ℹ The error occurred in group 1: sgroup = 578, group = 1.

Sample data

My code:

# Loaded libraries
library(dplyr)
library(ggplot2)
library(lme4)
library(lmerTest)
library(lattice)

# My lmer model. myData is fairly 
model0 <- lmer(outcome ~ (1|group), data=myData, REML = FALSE)
summary(model0)

myData %>% 
    # save predicted values
    mutate(pred_dist = fitted(model0))

What I am trying to understand is: What is causing this error message, and how do I resolve it?

Update:

As per bouncyball's comment I added ungroup() before mutate(). This worked for my initial model, but not the subsequent. I tried the following:

> model1 <- lmer(outcome ~ predictor + (1|group), myData, REML=FALSE) 

> plotVar$pred_dist = fitted(model1). 

Error: Assigned data 'value' must be compatible with existing data. 
✖ Existing data has 17192 rows. 
✖ Assigned data has 16794 rows. 
ℹ Only vectors of size 1 are recycled.

> plotVar %>% ungroup(.) 
          %>% mutate(pred_dist = fitted(model1)) 

Error: Problem with 'mutate()' column 'pred_dist'. 
ℹ 'pred_dist = fitted(model1)'. 
ℹ 'pred_dist' must be size 17192 or 1, not 16794.
Pål Bjartan
  • 793
  • 1
  • 6
  • 18
  • 1
    better off doing it this way: `myData$pred_dist = fitted(model0)` – bouncyball Dec 01 '21 at 13:49
  • Thanks, that worked! Do you have an idea why using `mutate()` generated the error? – Pål Bjartan Dec 01 '21 at 13:53
  • 3
    I think `myData` is a grouped data frame. if you put `ungroup()` before the `mutate` call what happens? – bouncyball Dec 01 '21 at 14:10
  • 1
    Is there any chance we could have a [mcve]? – Ben Bolker Dec 01 '21 at 14:50
  • @bouncyball `ungroup()` worked on my first example, but neither solutions worked on subsequent ones (see "Update"). @BenBolker I've tried reducing it as much as possible. I am unable to reproduce the error with smaller datasets, as I am not entirely sure what's causing the issue. That's why I included link to sample dataset that should be able to reduce the errors. – Pål Bjartan Dec 01 '21 at 15:22

1 Answers1

6

My very strong guess is that there are two different things going on here: (1) grouping (2) NA values.

Let's make up an example that is both grouped and contains NA values:

library(dplyr)
library(lme4)
ss <- sleepstudy |> group_by(Subject)
ss$Days[1:5] <- NA

m0 <- lmer(Reaction ~ Days + (Days | Subject), ss)

Try the original code:

ss |> mutate(pred = fitted(m0))

ℹ pred must be size 10 or 1, not 175.
ℹ The error occurred in group 1: Subject = 308.

Try with ungroup():

ss |> ungroup() |> mutate(pred = fitted(m0))

ℹ pred = fitted(m0).
ℹ pred must be size 180 or 1, not 175.

We still get an error, but note that the sizes are different.

Now update the model using na.action = na.exclude (this could have been done in the first model, or can be done by setting options(na.action = "na.exclude"):

m1 <- update(m0, na.action = na.exclude)
ss |> ungroup() |> mutate(pred = fitted(m1))
Ben Bolker
  • 211,554
  • 25
  • 370
  • 453
  • 1
    I've been looking at your code example, and can't help but notice you use `|>` as a pipe operator instead of `%>%`. I've never seen this syntax before, but the documentation says it's part of `base`. It's certainly a lot quicker to write. Can `|>` and `%>%` be used fully interchangeably? – Pål Bjartan Dec 09 '21 at 13:48
  • 3
    `|>` is new as of R version 4.0. It is *not* interchangeable: https://stackoverflow.com/questions/67633022/what-are-the-differences-between-rs-new-native-pipe-and-the-magrittr-pipe . I like it because it comes without dependencies, although in this case that doesn't matter since we're using `dplyr` anyway. – Ben Bolker Dec 09 '21 at 15:41