[R] Using Predict and GLM

[Prof Brian Ripley]

On Mon, 28 Jan 2008, Sherri Rose wrote:

> Dear R Help,
>
> I read through the archives pretty extensively before sending this
> email, as it seemed there were several threads on using predict with
> GLM.  However, while my issue is similar to previous posts (cannot
> get it to predict using new data), none of the suggested fixes are
> working.
>
> The important bits of my code:
>
> set.seed(644)
> n0=200 #number of observations
> W1=rnorm(n0,mean=2,sd=2) #Use rnorm to generate W1
> W2=rnorm(n0,mean=3,sd=8) #Use rnorm to generate W1
> Aprob=matrix(.2, nrow=n0, ncol=1) #generating the probability of A
> #generating probability of A dependant on W1
> for(i in 1:n0){
> if (W1[i]>1.5) {Aprob[i]=0.4}
> }
> A=matrix(rbinom(n0, 1, Aprob), nrow=n0, ncol=1) #generating the 0/1
> exposure
> Yprob=1/(1+exp(-(10*A-5*(W1)^2+2*W2)))
> Y=matrix(rbinom(n0, 1, Yprob), nrow=n0, ncol=1) #generating the 0/1
> exposure
> zero=data.frame(rep(0, n0))
>
>
> Q=glm(cbind(Y, 1-Y) ~ A + W1 + W2, family='binomial')
> QA=predict(Q, newdata=as.data.frame(A))
> Q0=predict(Q,newdata=(A=zero))
>
> I've tried many variations of the last line (Q0) to get the predicted
> values when A=0 with no luck.  With this code,  I get errors that my
> A=zero is a list even though I made it into a data frame.  This is
> the version of the code (after my reading) that *should* work for
> predict once I can get it to accept that it is not a list.

What do you think (A=zero) is?  It is the data frame zero, which is a list 
but does not have variable names 'A', 'W1' and 'W2', and as a side effect 
assigned 0 to A.  I suspect you meant == but then you would have had a 
logical matrix.  (This is one of the consequences of allowing '=' as an 
assignment operator.)

> With other variants of the line that will run but are not
> syntactically correct, my QA and Q0 are the same.

Code that is not syntactically correct cannot be parsed and so cannot be 
run.

Your intentions are not clear to me.  Why are you using one-column 
matrices?  Is

A <- rbinom(n0, 1, Aprob)
mydat <- data.frame(A, W1, W2)
Q <- glm(cbind(Y, 1-Y) ~ A + W1 + W2, family='binomial', data=mydat)
predict(Q, subset(mydat, A==0))

what you intended?

>
> Any guidance would be appreciated!
>
> Sherri
>
>
>
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> 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.
>

-- 
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