Dear Statalists,
I am analyzing the impact of local labor market variables (differentials in wage and in unemployment, measured as the ratio between the wage in region i over the national average wage) on interregional migration using a panel dataset composed of 128 europeans regions (in 13 countries) between 1998-2013. I used a fixed effect model with region and time fixed effects. My unit of observation is the region.
To account for heteroskedaticity and potential autocorrelation, I used the option fe, vce (robust). But due to the fact that my data are nested by country, leading to potential cross-correlation between regions of the same country, I though of trying the option fe, vce(cluster country) as well. I am however struggling to define if clustering at the country level makes sense, because one can also argue that regions from different countries may have more in common in terms of unemployment experience and regional income than regions from the same country.
#1. On which criterion should I compare my two sets of estimates? How can I pick the most appropriate model?
#2. Is clustering with a small number of cluster (13) justifiable methodologically? (Kézdi, Gabor. 2004. “Robust Standard Error Estimation in Fixed-E↵ects Panel Models.”, suggested that clustering with less than 50 clusters may be even worse than not clustering at all.)
# 3. In my case, should I also consider a hierarchical model which will allow me to include a country specific fixed effect to take into account the macroeconomic performance of the country (which may influence the migration choice).
Thank you a lot in advance,
Best,
Randa
I am analyzing the impact of local labor market variables (differentials in wage and in unemployment, measured as the ratio between the wage in region i over the national average wage) on interregional migration using a panel dataset composed of 128 europeans regions (in 13 countries) between 1998-2013. I used a fixed effect model with region and time fixed effects. My unit of observation is the region.
To account for heteroskedaticity and potential autocorrelation, I used the option fe, vce (robust). But due to the fact that my data are nested by country, leading to potential cross-correlation between regions of the same country, I though of trying the option fe, vce(cluster country) as well. I am however struggling to define if clustering at the country level makes sense, because one can also argue that regions from different countries may have more in common in terms of unemployment experience and regional income than regions from the same country.
#1. On which criterion should I compare my two sets of estimates? How can I pick the most appropriate model?
#2. Is clustering with a small number of cluster (13) justifiable methodologically? (Kézdi, Gabor. 2004. “Robust Standard Error Estimation in Fixed-E↵ects Panel Models.”, suggested that clustering with less than 50 clusters may be even worse than not clustering at all.)
# 3. In my case, should I also consider a hierarchical model which will allow me to include a country specific fixed effect to take into account the macroeconomic performance of the country (which may influence the migration choice).
Thank you a lot in advance,
Best,
Randa
Code:
xtset reg_id year xtreg netmig_rate_dom1 lnunempratio lngdpratio lnpop_dens i.year, fe vce(robust)
Code:
xtreg netmig_rate_dom1 lnunempratio lngdpratio lnpop_dens i.year, fe vce(cluster country)
