POLS the right analysis with a moderating variable?

[Gloria Nuesse]

Hello!

For my bachelor thesis I am researching whether women are more satisfied with their job when they have a female or male superior. For this I am including a moderator variable "quality of supervisor/subordinate relationship" which includes 6 questions on for example to what extend the subordinate perceives that the supervisor respects them as a person.
My dependent variable is job satisfaction (measured on 4 point likert scale) and the main independent variable is the gender of the subordinate.

I want to apply the POLS because job satisfaction is an ordinal variable and due to its simplicity its computation (compared to ordered probit). I am using survey data from the EWCS 2015 (containing over 40.000 observations).

Could anyone please assess whether my methodological decision is the right one?

[Daniel Schaefer]

POLS as in Pooled OLS? Personally, I rarely consider the computational complexity of a regression when selecting a model. Even with 40,000 observations, you shouldn't expect most models to take more than a few seconds to estimate, so the time complexity shouldn't be a central issue. Ordered logit/probit would be my first thought given the outcome variable.

That said, there is quite a bit that goes into model selection, and while some models might be better than others, there might not always be one right modeling choice. You might want to follow up with your instructor on this choice since they are the best to advise.

Sounds like a great undergraduate paper. Good luck!

[Clyde Schechter]

FWIW:

For my bachelor thesis I am researching whether women are more satisfied with their job when they have a female or male superior.

sounds like your study should include women subjects only and the key explanatory/predictive variable should be the sex of their supervisor.

But
[quote]...and the main independent variable is the gender of the subordinate.[/code]
sounds to me like the participants in the study are the supervisors, and you are explaining somebody's job satisfaction (supervisor's? subordinate's?) based on the subordinate's sex.

The latter would be an interesting question in its own right, but it is not the question you say you are trying to answer.

And I couldn't agree more with Daniel Schaefer that this is a question most appropriately addressed to your thesis advisor.

[Carlo Lazzaro]

Gloria:
welcome to this forum.
As an aside to Daniel and Clyde's wise advice, please note that if you're really dealing with survey data, you should take a look at the -svy:- prefix.

[Gloria Nuesse]

Thank you Daniel Schaefer for your quick answer! In my research proposal, I proposed using ordered probit model as well as POLS to ensure robustness. My supervisor is unfortunately not an expert herself in econometric analysis (ordered probit) and would not be able to guide me. However, she expressed doubts if ordered probit model would be suitable with a moderating variable. Do you think that the ordered probit model would also work specifically with including a moderating variable?


Oh pardon me. I have made some mistakes in the description. My key independent variable is indeed "gender of superior". Thank you for noticing Clyde Schechter

[Gloria Nuesse]

Thank you Carlo Lazzaro! Could I ask what you mean with -svy:- prefix?

[Carlo Lazzaro]

Gloria:
see -help svy- and related entry in Stata .pdf manual.

[Daniel Schaefer]

I don't know of a reason why you can't estimate moderation in an ordered probit, though I suppose it might be technically difficult to interpret the second order effect correctly. Of course, your supervisor might have good reasons for giving you that advice that I'm not aware of.

Most of our statistical methods assume simple random sampling, however, surveys often aren't simple random samples. -svy:- will let you account for the complex survey design. Be sure to check the docs and look closely at the subpop() option if you have a women only sub-population.