Error when passing formula to mixed() with more than 60 characters

[smithdanielle]

I first noticed this problem within my own data, but have since been able to replicate it with a built-in dataset.

I can't quite figure out the pattern, but here's what I've found so far; when specifying a formula with up to four fixed terms, afex's mixed function works as expected:

library(afex)
library(languageR)

data(lexdec, package = "languageR")

mixed(RT ~ Correct*Trial + PrevType * meanWeight + (1|Subject) + (1|Word), data = lexdec)
mixed(RT ~ Correct*Trial + PrevType * meanWeight + NativeLanguage + Length + (1|Subject) + (1|Word), data = lexdec)
mixed(RT ~ Correct*Trial + PrevType * meanWeight + NativeLanguage*Length + (1|Subject) + (1|Word), data = lexdec)

However, when specifying a formula with five or more fixed terms, it seems that mixed is unable to parse any term from the fifth onward, and an error is thrown up as a result:

 mixed(RT ~ Correct + Correct:Trial + PrevType * meanWeight + NativeLanguage + Length + (1|Subject) + (1|Word), data = lexdec)
Error in parse(text = x, keep.source = FALSE) : 
  <text>:2:7: unexpected symbol
1: RT~Correct + Correct:Trial + PrevType * meanWeight + NativeLanguage + +(1 | Subject) + (1 | Word)
2: RT    Length
         ^

mixed(RT ~ Correct + Trial + PrevType * meanWeight  + NativeLanguage * Length + Frequency + (1|Subject) + (1|Word), data = lexdec)
Error in parse(text = x, keep.source = FALSE) : 
  <text>:2:7: unexpected symbol
1: RT~Correct + Trial + PrevType * meanWeight + NativeLanguage * Length + +(1 | Subject) + (1 | Word)
2: RT    Frequency
         ^

mixed(RT ~ Correct + Trial + PrevType + meanWeight  + NativeLanguage * Length + Frequency + PrevType:meanWeight + (1|Subject) + (1|Word), data = lexdec)
Error in parse(text = x, keep.source = FALSE) : 
  <text>:2:7: unexpected symbol
1: RT~Correct + Trial + PrevType + meanWeight + NativeLanguage * Length + +(1 | Subject) + (1 | Word)
2: RT    Frequency
         ^

In cases where a working formula is specified, this warning message always appears:

In addition: Warning messages:
1: In stri_c(..., sep = sep, collapse = collapse, ignore_null = TRUE) :
  longer object length is not a multiple of shorter object length

lmer is still working as expected:

lmer(RT ~ Correct + Trial + PrevType * meanWeight + 
+           Frequency + NativeLanguage * Length + (1|Subject) + (1|Word), data = lexdec)
Linear mixed model fit by REML ['lmerMod']
Formula: RT ~ Correct + Trial + PrevType * meanWeight + Frequency + NativeLanguage *      Length + (1 | Subject) + (1 | Word)
   Data: lexdec
REML criterion at convergence: -972.8595
Random effects:
 Groups   Name        Std.Dev.
 Word     (Intercept) 0.04748 
 Subject  (Intercept) 0.13547 
 Residual             0.16831 
Number of obs: 1659, groups:  Word, 79; Subject, 21
Fixed Effects:
               (Intercept)            Correctincorrect                       Trial                PrevTypeword                  meanWeight  
                  6.486675                   -0.064751                   -0.000245                   -0.009240                    0.038132  
                 Frequency         NativeLanguageOther                      Length     PrevTypeword:meanWeight  NativeLanguageOther:Length  
                 -0.048023                    0.054975                    0.003433                   -0.020189                    0.017017

All these formulas used to work in a previous version of afex (afraid I don't know exactly what I was running... whatever the most up-to-date version was in March 2015).

SessionInfo() can be seen here.

EDIT: I can confirm that mixed gives the expected output for the following formula when using afex 0.13-145 from CRAN, though I do get an error regarding object length, as above:

mixed(RT ~ Correct + Trial + PrevType * meanWeight + 
                    Frequency + NativeLanguage * Length + (1|Subject) + (1|Word), data = lexdec)
Contrasts set to contr.sum for the following variables: Correct, PrevType, NativeLanguage, Subject, Word
Numerical variables NOT centered on 0 (i.e., interpretation of all main effects might be difficult if in interactions): Trial, meanWeight, Frequency, Length
Fitting 10 (g)lmer() models:
[..........]
Obtaining 9 p-values:
[.........]
                 Effect     F ndf     ddf F.scaling p.value
1               Correct  8.15   1 1627.73      1.00    .004
2                 Trial  7.57   1 1592.43      1.00    .006
3              PrevType  0.17   1 1605.39      1.00     .68
4            meanWeight 14.85   1   75.39      1.00   .0002
5             Frequency 56.53   1   76.08      1.00  <.0001
6        NativeLanguage  0.70   1   27.11      1.00     .41
7                Length  8.70   1   75.83      1.00    .004
8   PrevType:meanWeight  6.18   1 1601.18      1.00     .01
9 NativeLanguage:Length 14.24   1 1555.49      1.00   .0002
Warning message:
In stri_c(..., sep = sep, collapse = collapse, ignore_null = TRUE) :
  longer object length is not a multiple of shorter object length

[singmann]

The reason for this bug is the (by default) 60 characters limit of deparse and not the number of factors. The following reproduces the bug:

data(obk.long)
obk2 <- obk.long
colnames(obk2) <- sapply(colnames(obk2), function(x) paste0(x, x, x, x, x, x))
mixed(valuevaluevaluevaluevaluevalue ~ treatmenttreatmenttreatmenttreatmenttreatmenttreatment * phasephasephasephasephasephase * hourhourhourhourhourhour + (1|idididididid), obk2, method = "LRT")

The problem can be pushed to only appear with a formula of length 500 characters by simply changing the width.cutoff argument to 500L. Done in 0f652d7

[singmann]

@smithdanielle Thanks a lot for reporting that. Very helpful. Please let me know if you find anything else.

[RNMahon]

Is there an easy way on the user side to adjust the width.cutoff? I have a formula with 597 characters (long variable names from another program) and was not able to open the source files to edit this line. Thank you.