contrasting results

[Mark McCarthy]

Hi,

I am quite new to stata and am currently investigating the effects of regional labour demand on regional house prices. I have yearly changes of labour demand, and yearly changes of house prices.

I first performed a two-way scatter in which I observed the expected positive relationship, and then simply regressed just these labour demand changes on house prices to which I also observed a positive coefficient.

I then included both time and region fixed effects in my model to control for trends in housing that might arise over time, or within regions.

Once adding for these fixed effects, the coefficient of labour demand effect on house prices becomes negative.

(Both the positive and negative results were significant at the 1% level)

I have also then tried including controls of supply and crime rate, to which these effects are the same.

Does anyone know any reason why I am observing such results? I suggest that I continue with the fixed effects, as these trends need to be controlled for, but that gives me results that I would not expect to reserve?

Thankyou.

[Andrew Musau]

Exactly as you observe - the time effects (inflation, population growth, etc.) explain most of the variation in house prices. Once you partial out these as well as the region effects, you find a negative association between labor demand and house prices in your sample. The fixed effects results should be more reliable as they take into account the unobserved heterogeneity.

[Mark McCarthy]

Thankyou very much Andrew

[Mark McCarthy]

Can anyone advise me as to whether as well as controlling for year fixed effects I should also be controlling for regional effects.

I am currently using i.year and i.ID in my model, in which ID identifies each local authority (within the UK).

I am asking this because I am confused about the function fe, I am unsure whether I should be using this fe, or controlling for regional and time effects as shown above?

[Andrew Musau]

The region effects will be collinear with the local authority effects. So in effect, once you control for local authority effects, you have controlled for region effects.

xtset ID year
xtreg house_prices labor_demand controls i.year, fe cluster(ID)

The above includes both year effects and local authority effects.

[Mark McCarthy]

Thanks again Andrew