Moderator and double moderator graph and coefficient

[Rico Nederveen]

Dear Statalists,

I am struggling to understand what the coefficient of my moderator and double moderator means in my regression so I want to create a graph of it. I found some previous threats on this, but do not get it to work for me.

My variables are:
DependentV: bonus (in dollars)
Independent: Return on Equity (ROE)
moderator 1: ownership (in percentage) (OWN)
moderator 2: favorable shareholder vote (in percentage) (VOTE)

I created new variables for the moderators: gen ROE_OWN = ROE * OWN
and gen ROE_OWN_VOTE = ROE_OWN * VOTE

wat I want is a graph that shows the first interaction with bonus in dollars on the Y-axis, ROE on the x-axis. Then in it 3 lines, one for high (e.g. -1* standard dev), medium (mean) and low (-1 * SD).

I am not sure how to get this graph and how to handle this with the double moderator.

I hope you can help.

Best regards,


Rico

[Clyde Schechter]

Your question is very unclear. So I'm going to largely disregard what you've asked and just show you how to create the kind of graphs that most people frequently want to create when they have interaction models. Perhaps it will turn out to be what you wanted. If not, perhaps seeing what it produces will help you clarify your question, if only by pointing out in what way it differs from this "response."

What people often want in these models is to show a series of graphs of bonus vs ROE at several different levels of OWN and VOTE.

So, first you need to identify what are the interesting values of ROE, OWN and VOTE. Knowing nothing about the data, I'm going to assume that the interesting values of ROE are something like 0.10 0.20 and 0.30, that those of OWN are 20 40 and 60, and that those of VOTE are 10 20 30 40 and 50.

Then you would do this:

regress bonus c.ROE##c.(OWN VOTE)
margins, at(ROE = (0.10 0.20 0.30) OWN = (20 40 60) VOTE = (10(10)50))
marginsplot, xdimenison(ROE)

Then in it 3 lines, one for high (e.g. -1* standard dev), medium (mean) and low (-1 * SD).

I have no idea what this refers to. For which variable's mean and standard deviation are these lines wanted? Or did you mean that the interesting values of all of these variables are at their means and +/- 1 standard deviation? Or something else?

[Rico Nederveen]

Thank you for taking the time to answer despite my poor explanation.
I think it will be more clear if I split the question in two.

Lets first focus on the effect of ROE on Bonus is moderated by Ownership. So on the Y-axis I want Bonus and on the x-axis ROE.
What I meant by the text you quoted was, within the graph I would like to see 3 vectors (i suppose is what they are called). One vector that shows the development if ownership is high, one vector that shows effect of ROE on Bonus if ownership level is medium and one if it is high.

I got an r(198) option xdimenison() not allowed

[Clyde Schechter]

OK. Then something like this.

regress bonus c.ROE##c.(OWN VOTE)
summ OWN
local low = r(mean) - r(sd)
local medium = r(mean)
local high = r(mean) + r(sd)
margins, at(ROE = (0.10 0.20 0.30) OWN = (`low' `medium' `high'))
marginsplot

I don't know why you got that error message about xdimension() not being allowed. What version of Stata are you using? Maybe in version 13 or earlier they didn't have this option--I can't remember. I know it's part of -marginsplot-'s array of options in versions 14 and 15. Or perhaps you forgot to put the comma between marginsplot and xdimension()? That would be a syntax error.

Anyway, with the -margins- command written as above, with ROE first inside the -at()- option, -marginsplot- will put ROE on the x-axis by default, so we don't need the -xdimension()- specification.

Note, by the way, that -marginsplot- accepts nearly all -graph twoway- options, so you can do pretty much whatever you like to customize the appearance of the graph.

[Rico Nederveen]

Thank you Clyde, this worked! I just have to tweak it a little bit. I think I can manage the second part on my own.