Plotting results from nonlinear effects

[Anna Manzoni]

I am looking at how my outcome y evolves over time assuming such evolution may differ by x (binary).
I assume a nonlinear effect of time, which I operationalize as a quadratic effect.
Therefore I have time and time squared in my model, and I interact x with time, so my model looks like
xtreg y time##x timesquare##x, i(pid)

How can I plot y trajectories over time by x?
In a model with a linear effect of time or where I treat time with dummies for each time point I would use “marginsplot” as below…but I cannot figure out how to correctly apply that in the case of a quadratic effect.
xtreg y x##c.time, i(pid) /*or i.time if I want dummies for each time*/
margins x, at(time=(0(5)50))
marginsplot

thanks. Any suggestions is welcome.
Anna

[Joseph Coveney]

How can I plot y trajectories over time by x?
In a model with a linear effect of time . . . I would use “marginsplot” as below…but I cannot figure out how to correctly apply that in the case of a quadratic effect.

Use the factor variable notation for the quadratic term, as well.

xtreg y i.x##c.time##c.time, i(pid) re
margins x, at(time = (0(5)50)
marginsplot

[Anna Manzoni]

thanks! this is very helpful and works great.

I am now stuck with trying to plot effects if I use splines for time
So I create 4 splines for time as:

mkspline time1 2 time2 10 time3 20 time4=time

and then add time in my regression as
xtreg y x##time1 x##time2 x##time3 x##time4

however, I have yet to succeed at plotting trajectories over time by x when I use those time splines..
If anyone has any suggestions on how to use margins or anything else to plot that, thanks!

[Joseph Coveney]

There is some advice here and here.

[Maarten Buis]

Also [here](http://www.maartenbuis.nl/wp/inter_q...quadr_new.html)