How to plot interaction effects with confidence intervaI?

[Nuria Reguera]

Hi Experts,
I employ panel data methodology in a model type: y = b0 + b1*x1 + b2*x2 + b3*x1*x2 + control variables. I need create a graph for the interaction effect with confidence interval. Someone could help me, please?

[Carole J. Wilson]

You should begin by reading the manual entries for -margins- and -marginsplot-.

[Nuria Reguera]

Is it possible use these commands after xtreg command?

[Rich Goldstein]

yes, as a review of

help xtreg postestimation

would have shown you - please read and follow the advice in the FAQ for this forum

[Nuria Reguera]

I'm sorry but I do not understand how do it. This is the first time I do it. My model is

COMPR = b0 + b1*BUSY + b2*GENDER + b3*BUSY*GENDER + control variables

Using the coefficients from the panel data estimation and I elaborated the next graph and now I need represent the confidence interval in the graph.

Thanks in advance

[Carole J. Wilson]

You don't provide any of your Stata syntax, but here's a simple example of plotting the predicted values after a regression with an interaction. If you are not already using factor notation in your model estimation command, it is imperative that you do so when using margins. See help fvvarlist

clear
sysuse auto
reg price i.foreign##c.mpg
margins foreign, at(mpg=(10(5)50))
marginsplot

[Nuria Reguera]

The Stata syntax is:

xtreg Compromiso Gender Busy Gender_Busy Leverage ROA Size Meetings Tenure, fe

Results are attach to this post.


After this, I use coefficient for get graph using the next website: http://www.jeremydawson.co.uk/slopes.htm

Thank you very much for your help

[Carole J. Wilson]

The best way that I know how to do what you ask is described in #6. Please heed my warning about using factor variable notation in your model instead of creating the interaction variable yourself.

[Nuria Reguera]

Sorry, Carole J. Wilson but I think I can't apply factor notation because all my variables are continuos not categorical.

[Nick Cox]

#9 Not a problem. As explained e.g. at

help fvvarlist

factor variable notation applies to continuous variables too. Carole J. Wilson not only pointed this out in #6; she gave you a code example you can run and which worked.

By the way, please note our longstanding request to use full real names, e.g. given name and family name. https://www.statalist.org/forums/help#realnames and https://www.statalist.org/forums/help#adviceextras #3 explain why and what to do to fix your alias.

[Nuria Reguera]

Nick Cox I have run the code example in #6. In it the factor variable ("foreign") is a categorical variable which takes two values: domestic or foreign. However, my factor variable is continuos and when I try to run my model I obtain the error message: factor variables may not contain noninteger values.

[Bruce Weaver]

Adding to Carole's example in #6, you can use the r. contrast operator to plot the difference between foreign & domestic cars for a range of mpg values.

* Plot difference between foreign & domestic as function of mpg
quietly margins r.foreign, at(mpg=(10 50))
marginsplot, recast(line) recastci(rarea)

HTH.

[Nick Cox]

Well, I don't think that Carole J. Wilson was implying that her example mirrors your set-up exactly.

We still lack a data example and I guess the lack of such an example was behind Carole's decision to use a standard dataset in her example. Ditto for Bruce's post.

All you have to do, please, is to read the help for fvvarlist -- as suggested in #6 and again in #10 -- and just a short way down you will see examples of syntax like

     c.age             same as age

     c.age#c.age       age squared

which suggest how continuous-continuous interactions can be coded. If the interaction involves two different variables, you need two different names.

Again, you give no code in #11. The advice to show code, explicit or implicit in earlier posts, still applies.

However, I came late to this thread and haven't read all posts until just now. I am still unclear what is going on. In #5 you state that you are interested in the interaction BUSY * GENDER. You need to explain how those variables are quantified that makes you regard both of them as continuous. That is what you are saying: e.g. in #9 you declare that all of your variables are continuous. Yet the implication of #5 is that BUSY is just binary (low, high) and while there is scope for some categorization more complicated than (male. female) it is also a puzzle how data for GENDER could be continuous.