Search code examples
rvarsmoothinggam

R Extraxt time varying smoothing parameters from gam

I am trying to extract the time varying smoothing parameter of my gam.

This is the model:

gam1 <- gam(as.numeric(A5_1)~s(tt,k=k)+s(tt,by=A5_1L,k=k)+s(tt,by=A5_2L,k=k)+
            s(tt,by=A5_5L,k=k)+s(tt,by=A5_9L,k=k),data=data_subset)
gam2 <- gam(as.numeric(A5_2)~s(tt,k=k)+s(tt,by=A5_1L,k=k)+s(tt,by=A5_2L,k=k)+
            s(tt,by=A5_5L,k=k)+s(tt,by=A5_9L,k=k),data=data_subset)
gam5 <- gam(as.numeric(A5_5)~s(tt,k=k)+s(tt,by=A5_1L,k=k)+s(tt,by=A5_2L,k=k)+
            s(tt,by=A5_5L,k=k)+s(tt,by=A5_9L,k=k),data=data_subset)
gam9 <- gam(as.numeric(A5_9)~s(tt,k=k)+s(tt,by=A5_1L,k=k)+s(tt,by=A5_2L,k=k)+
            s(tt,by=A5_5L,k=k)+s(tt,by=A5_9L,k=k),data=data_subset)

summary(gam1) look like this:

enter image description here

I can create plots showing smoothing parameters over time like this:

plot(gam1, select=2,ylim=c(-3,1),rug=F,xlab="time points",
     ylab=substitute(paste("Joy",italic("(t-1)"), "on Joy",italic("(t)"))))

Plot is looking like this:

plot_gam

So far I found the predict.gam() function:

predict.gam(gam1, type = "terms")

The predict function gives me:

predict_gam

(nrows=103)

However the output doesn't match the plot. The line in the plot start around 0.2, whereas the smoothing Parameter in from predict.gam() is around 1.1. How can I extract the correct smoothing parameters over time?

In genereal I want these parameters, so that i can create different qgraphs() over time. One qgraph at the beginning, middle and end. Like network analysis plots. If there is a direct way of creating multiple qgraph() plots from time varying gam, I will gladly take this as well.

Maybe I will even use qgraph.animate() once I created a matrix of those Parameters.

Thanks a lot!

Solution

  • It won't look like the plot because the plot is showing the estimated effect for a 100 ordered values over the observed range of tt. The predict() call is giving you back the fitted values for your data because you didn't supply anything to the newdata argument.

    Two options are to:

    1. Save the output from plot.gam(): plt <- plot.gam(....)

      Now plt will contain an object with all the data used to create the plot. If you plot all smooths rather than just one, you'll get a list back with a data object per smooth.

    2. Prepare some new data to predict at and pass that to predict() as newdata. You'll need to provide data for tt plus all the by variables used in the model. expand.grid() is useful for this.