lmer() with 'na.action=na.exclude'; error with summary()

John Maindonald <john.maindonald at ...> writes:

The following demonstrates the issue:

library(DAAG)
science.lmer <- lmer(like ~ sex + PrivPub + (1 | school) + 
+                      (1 | school:class), data = science,  
+                      na.action=na.exclude) 
summary(science.lmer)
Linear mixed model fit by REML ['lmerMod']
[snip]

Scaled residuals: 
Error in quantile.default(resids) : 
 missing values and NaN's not allowed if 'na.rm' is FALSE
## Suppress details of residuals

summary(science.lmer, show.resids=FALSE)
 This is a confusion between the arguments of the summary method
(summary.merMod) and the *print method* (print.summary.merMod).
We should add a warning to summary that says it is discarding
unused arguments (the ... in the model definition is only there
for compatibility with the summary() generic method).

 print(summary(science.lmer), show.resids=FALSE)

works fine.
That is a subtlety that I had not contemplated.  It is not in forming
the summary object, but in printing it that the problem is evident.
I doubt that there is an intention for summary.merMod() to throw an error
lmer() has been called with 'na.action=na.exclude?.  
It should certainly not 
throw an error with the argument 'show.resids=FALSE?.
  Now I'm wondering whether the correct behaviour when there are
NAs in the (extended) residuals is 

 1 omit NAs from the residual quantile calculation (easiest)
I see no reason not to choose (1),  At the time we prepared the 3rd edition of 
'Data Analysis and Graphics using R?, this was the behaviour. My understanding 
of ?na.action=na.exclude? has been that the result is as for ?na.action=na.omit?, 
except that residuals match up with the original observations, with NAs inserted
as necessary.  In other words, the change from  ?na.action=na.omit? to
?na.action=na.exclude? is all about wharf appears when one explicitly
calculates and returns residuals, not about what happens when summary 
information (quantiles, not the residuals themselves!) is printed.

This strategy also simplifies documenting and describing what is done.
 2 return NA values for all quantiles of the residual
 3 return the quantiles plus a statement of the number of NAs.

Any good reason not to just do #1?
John Maindonald             email: john.maindonald at anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.
_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

lmer() with 'na.action=na.exclude'; error with summary()

Thread (5 messages)