[R] Differing Variable Length Inconsistencies in Random Effects/Regression Models

[Mark Difford]

Hi Aditi,

Parts of _your_ code for the solution offered by Jerome Goudet are wrong;
see my comments.

> famfit<-lmer(peg.no~1 + (1|family), na.action=na.omit, vcdf)   ## use:
> na.action=na.exclude
> resfam<-residuals(famfit) 
> for( i in 1:length(colms)) 
+ { 
+ print ("Marker", i) 
+ regfam<-abline(lm(resfam~i))    ## you need to use:
abline(lm(resfam~colms[,i]))
+ print(regfam) 
+ }

Corrected code:
famfit<-lmer(peg.no~1 + (1|family), na.action=na.exclude, vcdf)
resfam<-residuals(famfit) 
for( i in 1:length(colms)) 
{ 
print ("Marker", i) 
regfam<-abline(lm(resfam~colms[,i]))
}

This should work.

Regards, Mark.


A Singh wrote:
> 
> 
> Dear All,
> 
> I am quite new to R and am having a problem trying to run a linear model 
> with random effects/ a regression- with particular regard to my variable 
> lengths being different and the models refusing to compute any further.
> 
> The codes I have been using are as follows:
> 
> vc<-read.table("P:\\R\\Testvcomp10.txt",header=T)
>>> attach(vc)
>>
>> family<-factor(family)
>> colms<-(vc)[,4:13] ## this to assign the 10 columns containing marker
>> data    to a new variable, as column names are themselves not in any
>> recognisable sequence
>>
>> vcdf<-data.frame(family,peg.no,ec.length,syll.length,colms)
>> library(lme4)
> 
>>> for (c in levels(family))
>> + {    for (i in 1:length(colms))
>> +        { fit<-lmer(peg.no~1 + (1|c/i), vcdf)
>> +        }
>> +    summ<-summary(fit)
>> +    av<-anova(fit)
>> +    print(summ)
>> +    print(av)
>> + }
>>
>> This gives me:
>>
>> Error in model.frame.default(data = vcdf, formula = peg.no ~ 1 + (1 +  :
>>  variable lengths differ (found for 'c')
> 
> 
> I had posted a similar message on the R mixed model list a few days ago, 
> with respect to my fundamental methods, and Jerome Goudet had kindly 
> referred me to an alternative approach using residuals obtained from a 
> random effects model in lmer(), and then doing regressions using those 
> [residuals being the dependent variable and my marker data columns the 
> independent variable].
> 
> The code for that is as follows:
> 
>  vc<-read.table("P:\\R\\Text 
> Files\\Testvcomp10.txt",header=T,sep="",dec=".",na.strings=NA,strip.white=T)
>> attach(vc)
>>
>> family<-factor(family)
>> colms<-(vc)[,4:13]
>>
>> names(vc)
>  [1] "male.parent"  "family"       "offspring.id" "P1L55"        "P1L73" 
> 
>  [6] "P1L74"        "P1L77"        "P1L91"        "P1L96"        "P1L98" 
> 
> [11] "P1L100"       "P1L114"       "P1L118"       "peg.no" 
> "ec.length"
> [16] "syll.length"
>>
>> vcdf<-data.frame(family, colms, peg.no, ec.length, syll.length)
>>
>> library(lme4)
> 
>> famfit<-lmer(peg.no~1 + (1|family), na.action=na.omit, vcdf)
>> resfam<-residuals(famfit)
>> for( i in 1:length(colms))
> + {
> + print ("Marker", i)
> + regfam<-abline(lm(resfam~i))
> + print(regfam)
> + }
> 
> This again gives me the error:
> 
> 
> [1] "Marker"
> Error in model.frame.default(formula = resfam ~ i, drop.unused.levels = 
> TRUE) :
>   variable lengths differ (found for 'i')
> 
> 
> My variables all have missing values somewhere or the other. The missing 
> values are not consistent for all individuals, i.e some individuals have 
> some values missing, others have others,
>  and as much as I have tried to use na.action=na.omit and na.rm=T, the 
> differing variable length problem is dogging me persistently..
> 
> I also tried to isolate the residuals, save them in a new variable (called 
> 'resfam' here), and tried to save that in the data.frame()->vcdf, that I 
> had created earlier.
> 
> The problem with that was that when the residuals were computed, lmer() 
> ignored missing data in 'peg.no' with respect to 'family', which is 
> obviously not the same data missing for say another variable E.g. 
> 'ec.length'- leading again to an inconsistency in variable lengths. 
> Data.frame would then not accept that addition at all to the previous set.
> This was fairly obvious right from the start, but I decided to try it 
> anyway. Didn't work.
> 
> I apologise if the solution to working with these different variable 
> lengths is obvious and I don't know it- but I don't know R that well at
> all.
> 
> My data files can be downloaded at the following location:
> 
> <http://www.filesanywhere.com/fs/v.aspx?v=896d6b88616173be71ab> (excel-
> .xlsx)
> 
> <http://www.filesanywhere.com/fs/v.aspx?v=896d6b88616174a76e9e>
> (.txt file)
> 
> 
> Any pointers would be greatly appreciated, as this is holding me up loads.
> 
> Thanks a ton for your help,
> 
> Aditi
> 
> 
> 
> ----------------------
> A Singh
> Aditi.Singh at bristol.ac.uk
> School of Biological Sciences
> University of Bristol
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> 

-- 
View this message in context: http://www.nabble.com/Differing-Variable-Length-Inconsistencies-in-Random-Effects-Regression-Models-tp24502146p24505495.html
Sent from the R help mailing list archive at Nabble.com.