R help - Jun 2007 - Linear model predictions, differences in class

Hi,

I am using R to fit statistical models to data were the observations are
means of the original data.  R is used to calculate the mean before fitting
the model.  My problem is: When R calculates the means using tapply, the
class of the means differs from the class of the original data, which gives
me trouble when I want to use the original data to calculate model
predictions.  Here is a simple example that demonstrates the problem:
> data.in<-read.table('example.dat',header=TRUE)
>
> #Here are the data:
> data.in
  location    x      y
1        A  17.2  28.46
2        A 91.7 143.33
3        A 93.6 148.05
4        B 95.8 150.28
5        B 54.9  89.49
6        B 51.1  82.51
7        C 53.9  88.46
8        C 40.3  63.62
9        C 38.5  64.46
 >
> attach(data.in)
>
> #Calculate means by variable "location":
> data.mn<-data.frame(xm = tapply(x,location,mean), ym
tapply(y,location,mean))
> detach(data.in)
>
> #Here are the means:
> data.mn
        xm       ym
A 67.50000 106.6133
B 67.26667 107.4267
C 44.23333   72.1800
>
> #Fit the model:
> mod1<-lm(ym ~ xm, data.mn)
>
> mod1
Call:
lm(formula = ym ~ xm, data = data.mn)

Coefficients:
(Intercept)           xm
      5.633        1.505
> #R will make "predictions" using the data.mn data frame:
> predict(mod1,newdata =  data.mn)
        A         B         C
107.19260 106.84153  72.18587
>
> #But, even if new variables are created in the original data
> #with names that match those names used in the regression:
 > data.in$xm<-data.in$x
> data.in$ym<-data.in$y
> data.in
  location    x      y   xm     ym
1        A 17.2  28.46 17.2  28.46
2        A 91.7 143.33 91.7 143.33
3        A 93.6 148.05 93.6 148.05
4        B 95.8 150.28 95.8 150.28
5        B 54.9  89.49 54.9  89.49
6        B 51.1  82.51 51.1  82.51
7        C 53.9  88.46 53.9  88.46
8        C 40.3  63.62 40.3   63.62
9        C 38.5  64.46 38.5  64.46
>
> #R will not use data.in to make predictions:
> predict(mod1,newdata = data.in)
Error: variable 'xm' was fitted with class "other" but class
"numeric" was
supplied
>
> data.in$xm
[1] 17.2 91.7 93.6 95.8 54.9 51.1 53.9 40.3 38.5
> data.mn$xm
       A        B        C
67.50000 67.26667 44.23333
>
Is there a way to make these variables have the same class?  Or, is there
something other than "tapply" that will work better for this?

Thanks!

	[[alternative HTML version deleted]]

tapply gives an array: you want to use as.vector() on its result.

On Tue, 19 Jun 2007, John Phillips wrote:
> Hi,
>
> I am using R to fit statistical models to data were the observations are
> means of the original data.  R is used to calculate the mean before fitting
> the model.  My problem is: When R calculates the means using tapply, the
> class of the means differs from the class of the original data, which gives
> me trouble when I want to use the original data to calculate model
> predictions.  Here is a simple example that demonstrates the problem:
>
>> data.in<-read.table('example.dat',header=TRUE)
>>
>> #Here are the data:
>> data.in
>  location    x      y
> 1        A  17.2  28.46
> 2        A 91.7 143.33
> 3        A 93.6 148.05
> 4        B 95.8 150.28
> 5        B 54.9  89.49
> 6        B 51.1  82.51
> 7        C 53.9  88.46
> 8        C 40.3  63.62
> 9        C 38.5  64.46
> >
>> attach(data.in)
>>
>> #Calculate means by variable "location":
>> data.mn<-data.frame(xm = tapply(x,location,mean), ym >
tapply(y,location,mean))
>> detach(data.in)
>>
>> #Here are the means:
>> data.mn
>        xm       ym
> A 67.50000 106.6133
> B 67.26667 107.4267
> C 44.23333   72.1800
>>
>> #Fit the model:
>> mod1<-lm(ym ~ xm, data.mn)
>>
>> mod1
>
> Call:
> lm(formula = ym ~ xm, data = data.mn)
>
> Coefficients:
> (Intercept)           xm
>      5.633        1.505
>
>> #R will make "predictions" using the data.mn data frame:
>> predict(mod1,newdata =  data.mn)
>        A         B         C
> 107.19260 106.84153  72.18587
>>
>> #But, even if new variables are created in the original data
>> #with names that match those names used in the regression:
> > data.in$xm<-data.in$x
>> data.in$ym<-data.in$y
>> data.in
>  location    x      y   xm     ym
> 1        A 17.2  28.46 17.2  28.46
> 2        A 91.7 143.33 91.7 143.33
> 3        A 93.6 148.05 93.6 148.05
> 4        B 95.8 150.28 95.8 150.28
> 5        B 54.9  89.49 54.9  89.49
> 6        B 51.1  82.51 51.1  82.51
> 7        C 53.9  88.46 53.9  88.46
> 8        C 40.3  63.62 40.3   63.62
> 9        C 38.5  64.46 38.5  64.46
>>
>> #R will not use data.in to make predictions:
>> predict(mod1,newdata = data.in)
> Error: variable 'xm' was fitted with class "other" but
class "numeric" was
> supplied
>>
>> data.in$xm
> [1] 17.2 91.7 93.6 95.8 54.9 51.1 53.9 40.3 38.5
>> data.mn$xm
>       A        B        C
> 67.50000 67.26667 44.23333
>>
>
> Is there a way to make these variables have the same class?  Or, is there
> something other than "tapply" that will work better for this?
>
> Thanks!
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch 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.
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595