Predictions with GAM

Thanks for the swift reply,

I might have been a bit sloppy with describing my datasets and problem. I
showed the first model as an example of the type of GAM that I had been
able to use the predict function on. What I am looking for is how to
predict my m3:
model3<-gam(y_no~s(day,by=mapID),family=binomial, data=mergeday)

When I plot this I get 8 different graphs. Each showing me a different
habitat type with on the x-axis days and on the y-axis s(day,2,81):mapID.
With predict I was hoping to get the scale of the y-axis right for a
selection of days (for example 244,304).

I have tried to reform the script you gave me to match my dataset in m3,
but it all did not seem to work.

newd2 <- data.frame(day = rep(seq(244, 304, length = 100), 8),
                   mapID = rep(levels(mergeday$mapID), each = 100))

newd2 <- data.frame(day = rep(seq(244, 304, length = 100), 8),
                   mapID = rep(sort(unique(mergeday$mapID)),
                               each = 100))

I am guessing it must have something to do with the  by in s(day,by=mapID).
I haven't come across any examples that used a GAM with by and then used
the predict function.

(A sample of the dataset:
          mapID                  day   y_no
Urban Areas and Water    25      1
Urban Areas and Water    26      1
Early Succesional Forest  27      0
Agriculture                       28      0
Early Succesional Forest  29      0
Mature Coniferous Forest  30      0)


I am sorry that I have to bother you even more with this, and I hope that
my additional explanation about my problem might help solve it.

Sincerely yours,

Robbert Langenberg

2009/1/16 Gavin Simpson <gavin.simpson at ucl.ac.uk>

On Fri, 2009-01-16 at 12:36 +0100, Robbert Langenberg wrote:

Dear,

I am trying to get a prediction of my GAM on a response type. So that I
eventually get plots with the correct values on my ylab.
I have been able to get some of my GAM's working with the example shown
below:
*
model1<-gam(nsdall ~ s(jdaylitr2), data=datansd)
newd1 <- data.frame(jdaylitr2=(244:304))
pred1 <- predict.gam(model1,newd1,type="response")*

Hi Robert,

You want predictions for the covariate over range 244:304 for each of
your 8 mapID's, yes?

This is not tested, but why not something like:

newd2 <- data.frame(jdaylitr2 = rep(seq(244, 304, length = 100), 8),
                   mapID = rep(levels(datansd$mapID), each = 100))

Then use newd2 in your call to predict.

I am assuming that datansd$mapID is a factor in the above. If it is just
some other indicator variable, then perhaps something like:

newd2 <- data.frame(jdaylitr2 = rep(seq(244, 304, length = 100), 8),
                   mapID = rep(sort(unique(datansd$mapID)),
                               each = 100))

Does that work for you?

HTH

G

The problem I am encountering now is that I cannot seem to get it done

for

the following type of model:

*model3<-gam(y_no~s(day,by=mapID),family=binomial, data=mergeday)*

My mapID consists of 8 levels of which I get individual plots with *
plot(model3)*. When I do predict with a newdata in it just like my
first model I need all columns to have the same amount of rows or else
R will

not

except it ofcourse, the col.names need to at least include day and
mapID. This way I can not get a prediction working for this GAM, I am
confused because of this part in the model: *s(day,by=mapID).

*I have been reading through the GAM, an introduction with R book from

Wood,

S. but could not find anything about predictions with BY in the model.

I hope someone can help me out with this,

Sincerely yours,

Robbert Langenberg

      [[alternative HTML version deleted]]