Adding region level and time dummies to fixed effects models

[Kyle Patel]

Hello,
I am new to this forum. I hope you are all well.
I am using a fixed effects model to find out the effects of transitioning in and out of the labour market on mental health scores. My supervisor has advised me to include region level (urban or rural) dummies as well as yearly level dummies (I have 7 waves of data). My thought process so far would be to create a dummy for urban/rural - taking 1 if ==urban, 0 otherwise. And for the yearly dummies, to create 7 dummies, for example w1 if ==year 1, 0 otherwise. and then adding all these variables to my regression. Is this a logical way of going about this? Thank you.

[Clyde Schechter]

The logic is fine. But there is no need to explicitly create indicator ("dummy") variables in your data set. If you already have a year variable that simply contains the calendar year, adding i.year to the list of regressors in your estimation command will cause Stata to create virtual indicators "on the fly" for your regression. Similarly if you already have a variable that takes on two values, one for urban and the other for rural, you can just prefix that variable with i. in your regression. This is known as factor variable notation. Read more about it by running -help fvvarlist-.

Added: you might want to ask your advisor what the rationale for including year indicators is. Mental health is not, to my knowledge, subject to calendar-year shocks of any importance. There may be particular years that affect mental health, such as years of economic recession, or perhaps the pandemic years, but I am not aware of any regular year-to-year general effect on mental health. Perhaps I'm simply unaware of it? Anyway, it's worth asking.

[Carlo Lazzaro]

Kyle:
welcome to this forum.
As an aside to Clyde's helpful explanation, are you sure that mental health scores cannot cause per se transitions in/out the labour market (i.e., reverse causality aka nasty cause of endogeneity)?

[Clyde Schechter]

Yes, Carlo raises an excellent point!

[Kyle Patel]

Thank you both for your comments. Clyde, that information will help me thank you!
Carlo: the concern for reverse causality is definitely there, in most academic literature they tend to use terms such as 'unemployment is associated with a decrease of mental health scores by ____ amount' suggesting that the reverse causality is hard to overcome. I will have to mention this in my limitations of my work!

[Carlo Lazzaro]

Kyle:
a word of advice: be 100% sure that your mentor/supervisor/professor shares this approach (i.e., that the risk of reverse causality can simply be mentioned as a limitation of your study).
Nasty reviewers can state that, due to the risk of endogeneity, your coefficients are unreliable.

[Kyle Patel]

Thank you for the words of advice Carlo, I will double check with my supervisor.