[R] Returning NA from lm

[Prof Brian Ripley]

See ?na.exclude

On Fri, 23 Jan 2009, Neil Beddoe wrote:

> Hi.
>
> I need to apply run a regression analysis for groups of data of fixed length:100 As, 100 Bs, 100 Cs etc.
>
> eg
>
> x
> Key     Value
> A       1
> A       21.2
> A       4
> A       6.5
> ...repeat 96 times with differing values of A
> B       1
> B       2.3
> B       NA
> B       6.5
> ...repeat 96 times with differing values of B
> etc
>
> I run these against a linear model using tapply(data$Value, data$Key,FUN=regr,100) where
> regr<-function(x,w)
> {
>        #run the model against the last w values of x
>        lm((x[length(x)-w):length(x)]~myModel(w))
> }
> In the results, I want to return NA for any Key group where one or more of the values is NA.  If I run the above I get a regression structure ignoring the missing values and returning values for data that contains NA.  Using na.action=na.fail or na.action=NULL causes the whole tapply function to fail and I get nothing.  Is there a way I can get lm to return NA if any of the values in the data are NA but valid numbers for complete data?
>
> I realise that I could remove the groups with NAs but I'm running the regressions over multiple time periods and most of the data groups will have a full complement of data for at least some of these periods. It becomes a pain to manage NAs if I do that.
>
> Sorry if the above is a little unclear.
>
> Thanks
>
> Neil

-- 
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