Population growth in panel data

[Loris Di Stefano]

Dear Statalist Community,

I have a simple and short question to you.
I am studying the effects of a certain treatment on the population density growth of municipalities.
If the population of the inquired country grew by 20 M during the period investigated, how do I account for that?
Do I add a variable measuring the population growth as a control or is this rising trend already caught by the time or municipality fixed-effects I add to the model?

As always thank you in advance

Loris

[Clyde Schechter]

There are several ways of handling this, and the choice among them is determined by the organization of your data and your specific research question. Adding a variable measuring population growth as a control would be among the possibilities, although I think that would only occasionally be the preferred approach. If you say more about the nature and organization of your data, more specific advice might be forthcoming.

[Loris Di Stefano]

My data is on the municipal-level and comprises the years 1936-2011.
Municipal-year-level observations;
Municipality:= 1,..,1420;
year:= 1936,..,2011.
the treatment is given by being victim of a seismic event
The treatment variable EQ is a binary variable identifying the treatment in municipality m in year t

I have data on the population (population density, population growth etc...) for each year
my main estimation

xtreg popdensitygrowth treatvariable $control i.year, fe  vce(cluster)

The main issue is that as soon as I add the populaion growth as a control variable in the equation my coefficient turns from negative to positive, which does not make much sense ( as if my coefficients are negative at the beginning why should they turn positive once I do account for a population growth)

[Clyde Schechter]

Isn't population density = population / area? And area does not change over time. So change in population density is proportional to change in population within each municipality, the proportionality constant being the reciprocal of the municipality's area. Indeed, growth in population is precisely the mechanism by which growth in population density occurrs. So I would certainly expect that adding population density to the model would interfere with estimating effects on population density and effectively obliterate the ability to detect other effects: you are controlling for the causal pathway!. This seems like a badly mis-specified model to me. Am I missing something?

[Loris Di Stefano]

ahahah you're not, just noticed my mistake and miscalculation. Thak you for your patience and sorry