Variable lengths differ in R (linear modelling with lme4)

Question

My input file:

Treat1  Treat2  Batch   gene1    gene2
High    Low     1       92.73    4.00
Low     Low     1       101.85   6.00
High    High    1       136.00   4.00
Low     High    1       104.00   3.00
High    Low     2       308.32   10.00
Low     Low     2       118.93   3.00
High    High    2       144.47   3.00
Low     High    2       189.66   4.00
High    Low     3       95.12    2.00
Low     Low     3       72.08    6.00
High    High    3       108.65   2.00
Low     High    3       75.00    3.00
High    Low     4       111.39   5.00
Low     Low     4       119.80   4.00
High    High    4       466.55   11.00
Low     High    4       125.00   3.00

There are tens of thousands of additional columns, each with a header and a list of numbers, same length as "gene1" column.

My code:

library(lme4)
library(lmerTest)

# Import the data.
mydata <- read.table("input_file", header=TRUE, sep="\t")

# Make batch into a factor
mydata$Batch <- as.factor(mydata$Batch)

# Check structure
str(mydata)

# Get file without the factors, so that names(df) gives gene names.
genefile <- mydata[c(4:2524)]

# Loop through all gene names and run the model once per gene and print to file.
for (i in names(genefile)){
    lmer_results <- lmer(i ~ Treat1*Treat2 + (1|Batch), data=mydata)
    lmer_summary <- summary(lmer_results)
    write(lmer_summary,file="results_file",append=TRUE, sep="\t", quote=FALSE)
}

Structure:

'data.frame':     16 obs. of  2524 variables:
$ Treat1          : Factor w/ 2 levels "High","Low": 1 2 1 2 1 2 1 2 1 2 ...
$ Treat2          : Factor w/ 2 levels "High","Low": 2 2 1 1 2 2 1 1 2 2 ...
$ Batch           : Factor w/ 4 levels "1","2","3","4": 1 1 1 1 2 2 2 2 3 3 ...
$ gene1           : num  92.7 101.8 136 104 308.3 ...
$ gene2           : num  4 6 4 3 10 3 3 4 2 6 ...

My error message:

Error in model.frame.default(data = mydata, drop.unused.levels = TRUE, formula = i ~ : variable lengths differ (found for 'Treat1') Calls: lmer ... -> eval -> eval -> -> model.frame.default Execution halted

I have tried to examine all objects involved and cannot see any differences in variable lengths and I have also made sure there are no missing data. Running it with na.exclude doesn't change anything.

Any idea of what is going on?

A starting place for you is to add a print statement to figure out what column is causing the problem. Since we don't have your data, it may be quite difficult to diagnose the problem. — lmo
You don't have a column i in your data. Thus, lmer looks outside of the data and finds the character vector i. Try lmer_results <- lmer(as.formula(paste0(i, " ~ Treat1*Treat2 + (1|Batch)")), data=mydata) — Roland

Ben Bolker Ben Bolker · Accepted Answer · 2016-05-26T15:11:04

@Roland's diagnosis (lmer is looking for a variable called i, not a variable whose name is i: obligatory Lewis Carroll reference) is correct, I think. The most immediate way to handle this would be with reformulate(), something like:

for (i in names(genefile)){
    form <- reformulate(c("Treat1*Treat2","(1|Batch)"),response=i)
    lmer_results <- lmer(form, data=mydata)
    lmer_summary <- summary(lmer_results)
    write(lmer_summary,file="results_file",
           append=TRUE, sep="\t", quote=FALSE)
}

On second thought, you should be able to speed up your computations significantly using the built-in refit() method, which refits a model for a new response variable: suppose for simplicity that the first gene is called geneAAA:

wfun <- function(x)  write(summary(x), 
       file="results_file", append=TRUE, sep="\t",quote=FALSE)
mod0 <- lmer(geneAAA ~ Treat1*Treat2 + (1|Batch), data=mydata)
wfun(mod0)
for (i in names(genefile)[-1]) {
    mod1 <- refit(mod0,mydata[[i]])
    wfun(mod1)
}

(By the way, I'm not sure your write() command does anything sensible ...)

Variable lengths differ in R (linear modelling with lme4)

1 Answers