coefplot and predicted values

[Alexander Wuttke]

Dear all,

I want to plot the predicted outcomes of several experimental groups from a linear regression analysis. #

I use -margins- with -at()- to specify the values of interest.
However -coefplot- (SJ, Ben Jann) plots the coefficients of the regression preceding margins and does not plot the value predicted by margins.
-margins- followed by -marginsplots- does plot the desired coefficients but obviously only for one -margins- result at a time.

I have found it that -matrix list r(table)- contains the information I want to plot. But I did not figure out how to tell -coefplot- to use this matrix (and the matrices of the other preceding -margin- commands)

To clarify what I hope for, here is a reproducible example. In my case the explanatory variable are three dichotomous variables for the experimental conditions.

sysuse auto, clear
ssc install coefplot

eststo clear
regress mpg foreign turn
eststo: margins, at(foreign=0 turn=40)
eststo: margins, at(foreign=1 turn=40)
eststo: margins, at(foreign=0 turn=41)
eststo: margins, at(foreign=1 turn=41)
marginsplot //Plots the desired estimated (of the last calculation)
coefplot est1 est2 est3 est4
matrix list r(table)

I am grateful for any advice.

[Clyde Schechter]

I am not a user of coefplot, so I'm not entire surely what your desired result looks like. But I think I get what you want. Have you tried:

sysuse auto, clear

regress mpg foreign turn

margins, at(foreign = (0 1) turn = (40 41))
marginsplot

[Alexander Wuttke]

Yes, thank you Clyde, this would be the more elegant coding. And it allows to use marginsplot because -margins- is only executed once.
However, in my actual case this leads to undesired combinations. This is why I want to run margins several times.

I am puzzled. What I am trying to do seems to be a very easy application, I thought.

[Clyde Schechter]

Well, the other thing is that -margins- (and -marginsplot-, too) have a saving() option so that you can create data sets containing the points. Then you can put all those data sets together and use -graph twoway- to construct your plot.

Also, if the problem is that you are over-generating combinations of variable levels that are not of interest, you can just use multiple -at()- options in the same -margins- command to get precisely what you want. For example:

sysuse auto, clear

regress mpg rep78 turn

margins, at(rep78 = (1 2) turn = (40 41)) at(rep78 = (3 4) turn = (50 51))
marginsplot

This involves four values each of rep78 and turn, but produces only the selected 8 pairings of interest. You can get any set of combinations you want this way.

[Alexander Wuttke]

I was unaware of margins, saving().
Thank you!
Since I had to calculate predicted values of different regression models, I saved the various results of margins, appended the datasets and plotted them with -eclplot- (ssc, Newton)

Thanks!