Comparing wilcoxon signed rank tests and t-tests.

[Berend Wiebe]

Hi, I'm doing research on different sustainability practices between sectors. I have data at time t-1 and t+2 for different firms and want to check whether different practices are more prevalent in some sectors. I want to test if a practice is prevalent in a sector by doing either a t-test or a Wilcoxon signed rank test. If the variable changes significantly from t-1 to t+2 then it indicates that a practice is used in a sector. I was wondering if it is possible to compare different t-tests or wilcoxon signed rank tests by comparing the significance between sectors so that I can interpret that one practice is more prevalent in a sector if it is significant for one sector and not for another?
Kind regards,
Berend

[Joseph Coveney]

I was wondering if it is possible to compare different t-tests or wilcoxon signed rank tests by comparing the significance between sectors so that I can interpret that one practice is more prevalent in a sector if it is significant for one sector and not for another?

You might want to check this out before going too much farther down that path.

Why not just fit a regression model that has a sector-by-time interaction term and take a look at the marginsplot (profile plot) to see whether a consistent and cogent pattern ermerges?

[Joris Berg]

.

[Berend Wiebe]

Thank you for your reply! How should I go about that regression if the variable I want to check the chance for is ROS? I have observations of 130 firms across 5 sectors and have observations at time t-1, t+2 and t+3. I'm not sure what my dependent and independent variables should be in this situation as for the wilcoxon signed rank test and ttest my only variable would be ROS.

[Joseph Coveney]

How should I go about that regression if the variable I want to check the chance for is ROS? I have observations of 130 firms across 5 sectors and have observations at time t-1, t+2 and t+3. I'm not sure what my dependent and independent variables should be in this situation as for the wilcoxon signed rank test and ttest my only variable would be ROS.

You've upped the ante from #1 with three time points now, and so you wouldn't really be using a Wilcoxon signed-rank test or Student's t test, anyway.

I don't know what ROS is or what a suitable regression model would be for it; I assume that there are experts in economics or finance who could advise you on that.

But in principle a regression model could still have a sector-by-time interaction term just as I suggested above. With five sectors, the marginsplot might look a little crowded; nevertheless, the same approach suggested above would still be applicable, and if needed you could use multiple plots of fewer sectors each—i.e., the graphics tactic of small multiples—in order to suss out whether temporal patterns of ROS differ substantially between sectors.