Help with interpreting interaction term in logit regression output

[Pete Rhodes]

Hi, i have a dataset consisting of target prices (future expected stock prices). There are two groups which have published these target prices, professional analysts and economics students. I have run logistic regression analysis to figure out if students are less likely than analysts to meet their target prices.

The Y-variable (TPmet) is equal to one if target price is met and 0 otherwise. I have included an independent variable which is equal to 1 if student and 0 otherwise.
Furthermore i have an independent variable which expresses target price optimism (the distance between target price and todays price, TP/P), which is called TPsize.

My hypothesis is that students miss are less likely to meet their target prices than analysts, and i believe the reason for this is partly because they issue more optimistic target prices.
I therefore also included an interaction term between students and target: Stud*TPsize

I have included a picture of the output.The interaction term is significant, but I'm not quite sure how to interpret it. Can i say that students are less less likely to meet their target prices because they are more optimistic than analysts? Or does it mean that for the same level of target price optimism (same TP/P) students are less likely to meet their target prices?

[Clyde Schechter]

So, the negative coefficient for Student says that, all else equal, students are less likely than analysts to meet their target price. The negative interaction term coefficient says that when TPSize is greater (which, if I understand your post means that the target price set was even more optimistic), then the "difference" (actually an odds ratio) between the students and the analysts becomes yet a wider gulf. The word "because" simply doesn't belong anywhere in a description of these results, as this is observational data from which causal inferences cannot be made.

I think the best way to see this and express it clearly is actually with graphs. First, go back and re-run your model using factor-variable notation (-help fvvarlist-): generating product terms for interactions is by now archaic in Stata. Then you can run -margins- afterwards, and -marginsplot- after that. So something like this:

logit TPmetany i.Student##c.TPSize RM PB LogMV
margins Student, at(TPSize = (0.5 1 1.5 2))
marginsplot

The resulting graphs will show you how frequently both Students and Analysts target prices are met as functions of TPSize.

Note that the numbers 0.5, 1, 1.5, and 2 are just my estimates of interesting values of TPSize at which to examine the results. You should replace those numbers by appropriate values that span and more or less fill the range of typical or otherwise interesting values of TPSize observed in the data.

Added: In the future, please do not use screenshots to show Stata output. What you posted is just barely readable on my computer. The helpful way to show Stata output is to copy it from your log file or Stata's Results window and paste it between code delimiters here on the forum. If you do not know how to work with code delimiters, please read FAQ #12.