Using "gamplot" for generalized additive models (GAM)

[Itai Magodoro]

Dear Forum,

I will much appreciate guidance about how to use the "gamplot" function for generalized additive models. I have created 2 generalized additive model:

1. gam yvar1 xvar, f(gaussian) link(identity) df(n)

2. gam yvar2 xvar, f(binomial) link(logit) df(n)

Where y-variable yvar1 has normal distribution, and yvar2 has a binomial distribution. The x-variable is a continuous variable.

I would like to use the GAM plotter, "gamplot xvar, nopres". What additional code should I use, or how should I change my yvar such that the y-axis plots the precited mean (for yvar1) and, separately, the predicted probability (for yvar2)?

I am aware that gam stores "GAM_mu", the predicted yvar.

Thanks in advance.

Itai

[Joseph Coveney]

I would like to use the GAM plotter . . . or how should I change my yvar such that the y-axis plots the precited mean (for yvar1) and, separately, the predicted probability (for yvar2)?
I am aware that gam stores "GAM_mu", the predicted yvar.

I believe that you cannot use -gamplot- to do what you're looking to do. Instead, just use Stata's regular graphing commands to plot the fitted versus the function of predictor, which is also left behind in the dataset by -gam-. Try something like that below. The first is for the Gaussian case and the second is for the binomial.

.ÿ
.ÿversionÿ17.0

.ÿ
.ÿclearÿ*

.ÿ
.ÿquietlyÿsysuseÿauto

.ÿ
.ÿglobalÿGAMDIRÿ`c(pwd)'\

.ÿ
.ÿgamÿweightÿdisplacement,ÿfamily(gaussian)ÿlink(identity)ÿdf(3)

74ÿrecordsÿmerged.

GeneralizedÿAdditiveÿModelÿwithÿfamilyÿgauss,ÿlinkÿident.

Modelÿdfÿÿÿÿÿ=ÿÿÿÿÿ4.000ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNo.ÿofÿobsÿ=ÿÿÿÿÿÿÿÿ74
Devianceÿÿÿÿÿ=ÿÿÿ6877280ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿDispersionÿ=ÿÿÿ98246.8
-------------------------------------------------------------------------
ÿÿÿÿÿÿweightÿ|ÿÿÿdfÿÿÿÿLin.ÿCoef.ÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿÿÿÿÿGainÿÿÿÿP>Gain
-------------+-----------------------------------------------------------
displacementÿ|ÿÿ3.000ÿÿÿ7.573259ÿÿÿ.3994655ÿÿÿÿ18.958ÿÿÿÿ19.386ÿÿÿÿ0.0001
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿÿÿÿ1ÿÿÿÿ3019.46ÿÿÿÿ36.4371ÿÿÿÿ82.868ÿÿÿÿÿÿÿÿÿ.ÿÿÿÿÿÿÿÿÿ.
-------------------------------------------------------------------------
Totalÿgainÿ(nonlinearityÿchisquare)ÿ=ÿÿÿÿ19.386ÿ(2.000ÿdf),ÿPÿ=ÿ0.0001

.ÿ
.ÿ//ÿresponseÿversusÿfunctionÿofÿpredictor
.ÿgraphÿtwowayÿ///
>ÿÿÿÿÿÿÿÿÿlineÿGAM_muÿs_displacement,ÿsortÿlcolor(black)ÿ||ÿ///
>ÿÿÿÿÿÿÿÿÿscatterÿweightÿs_displacement,ÿmcolor(black)ÿmfcolor(white)ÿmsize(small)ÿ///
>ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿytitle(Weightÿ(lbs.))ÿylabel(ÿ,ÿangle(horizontal)ÿnogrid)ÿlegend(off)

.ÿquietlyÿgraphÿexportÿGausGAM.png

.ÿdropÿ*_displacementÿ*_mu

.ÿ
.ÿgamÿforeignÿmpg,ÿfamily(binomial)ÿlink(logit)ÿdf(3)

74ÿrecordsÿmerged.

GeneralizedÿAdditiveÿModelÿwithÿfamilyÿbinom,ÿlinkÿlogit.

Modelÿdfÿÿÿÿÿ=ÿÿÿÿÿ3.793ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNo.ÿofÿobsÿ=ÿÿÿÿÿÿÿÿ74
Devianceÿÿÿÿÿ=ÿÿÿ77.7278ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿDispersionÿ=ÿÿÿÿÿÿÿÿÿ1
-------------------------------------------------------------------------
ÿÿÿÿÿforeignÿ|ÿÿÿdfÿÿÿÿLin.ÿCoef.ÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿÿÿÿÿGainÿÿÿÿP>Gain
-------------+-----------------------------------------------------------
ÿÿÿÿÿÿÿÿÿmpgÿ|ÿÿ2.793ÿÿÿ.1595858ÿÿÿ.0521688ÿÿÿÿÿ3.059ÿÿÿÿÿ0.849ÿÿÿÿ0.6016
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿÿÿÿ1ÿÿÿ-.976156ÿÿÿÿ.282028ÿÿÿÿ-3.461ÿÿÿÿÿÿÿÿÿ.ÿÿÿÿÿÿÿÿÿ.
-------------------------------------------------------------------------
Totalÿgainÿ(nonlinearityÿchisquare)ÿ=ÿÿÿÿÿ0.849ÿ(1.793ÿdf),ÿPÿ=ÿ0.6017

.ÿ
.ÿlowessÿforeignÿs_mpg,ÿlineopts(lcolor(black)ÿlpattern(dash))ÿ///
>ÿÿÿÿÿÿÿÿÿmcolor(black)ÿmsize(vsmall)ÿjitter(3)ÿjitterseed(787217005)ÿ///
>ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿaddplot(lineÿGAM_muÿs_mpg,ÿsortÿlcolor(black)ÿlpattern(solid)ÿ///
>ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿtitle("")ÿytitle(ProbabilityÿofÿForeignÿOrigin)ÿ///
>ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿylabel(ÿ,ÿangle(horizontal)ÿnogrid)ÿlegend(off))

.ÿquietlyÿgraphÿexportÿLogitGAM.png

.ÿ
.ÿexit

endÿofÿdo-file


.

For the binomial GLM case, you might also want to consider plotting the linear prediction (in the logit transform as estimated).

[Itai Magodoro]

Thank you very much Joseph. Much appreciated.