Comparing regression coefficients - Different samples, different DVs, same predictor...

[Bob Hernandez]

I am regressing work performance on a focal IV, X, in each of three samples of individuals. Each sample was taken from a different occupation, so the performance outcome is different across the three groups. But the predictor X is the same. Two of the occupations are characterized by low levels of third variable Y, while the other occupation is high in Y.

I'd like to compare the magnitude of the X effect on performance across occupations low in Y and high in Y. Setting aside the fact that any conclusions drawn about the influence of Y on the X --> Perf relationship would be extremely tentative, what approach would be most appropriate to compare the different regression coefficients? Simple Wald tests? SUR? I'm a bit lost here.

Edit: I neglected to mention a further complication--across the three samples, the regressions include different covariates, in addition to the common predictor, X.

[Carlo Lazzaro]

Bob:
have you taken a look at Example #2 under -sureg- entry in Stata 13.1 .pdf manual?

[Bob Hernandez]

Thank you Carlo. But in my data, some of the covariates used in one sample are undefined for all individuals in the other samples). Given this, I wonder if SUR is inappropriate? I get an insufficient observations error when I attempt to run -sureg-.

[Carlo Lazzaro]

Bob:
as an aside, can't you simply reduce the comparison to the set of predictors that are common to all the samples (using SUR as in Example # 1 under -sureg- entry in Stata 13.1 .pdf manual)?

[Bob Hernandez]

Carlo: Yes, I could, although I would lose several variables and would have to standardize the different performance outcomes and treat them as a single variable. But perhaps this is the closest I can get to a valid contrast.

[Carlo Lazzaro]

Bob:
I would second your last thoughts on this issue.