How to attain predictive response values used for vis.gam plots?

user_19899231 0
Viewed 23

Is it possible to get the predicted response values that are used to plot 3D contour plots in vis.gam. I want to know what the predicted response for each grid cell is in the plot. I am using the MGCV package and GAM function. My model is:

pose_mod <- gam(POSEtrans ~ s(Elevation) + s(Heatload) + 
                  s(Precip_Spring_Fall_Total) + s(PSSP6) + s(BRTE) +
                  TSF + s(PointID, bs = 're'), 
                family=betar(link='logit'), method = "REML",
                data = iData) 

vis.gam(pose_mod, view=c("BRTE", "Elevation"),
        type = 'response',theta=45,ticktype="detailed", n.grid = 20)
gam
1 Answers

The easiest way that I know is to use my {gratia} package (you'll need the development version on GitHub, which you can install using the intructions on the GitHub site: https://gavinsimpson.github.io/gratia/ or if you can't build from sources, use the binary builds from R-universe: https://gavinsimpson.r-universe.dev/ui#package:gratia).

library("gratia")

ds <- data_slice(pose_mod, BRTE = evenly(BRTE, n = 20),
                 Elevation = evenly(Elevation, n = 20))

This follows the conventions in {mgcv} and vis.gam(), such that you get the same grid of points. The you could predict() (from {mgcv} or use fitted_values() from {gratia} with ds to get the predicted expectation of the response for each grid cell:

fitted_values(pose_mod, data = ds, scale = "response")
user_4298460
Related