Dear colleagues,
I research the drivers of regional eco-innovation. I have panel data for 2010-2015 for over 70 regions. Since the dependent variable (eco-innovation) ranges from 0 to 1, I am trying to apply generalized linear models, fractional regression models. Specifically, I built regressions according to Papke & Wooldridge (2008) "Panel data methods for fractional response variables with an application to test pass rates" - i.e. in addition to the explanatory variables themselves, I added time averages for these variables (average over the last 5 years):
where shareorg_ecoinn6_1 is eco-innovation;
ind_envir_index soc_envir_index envir_index illness_infect - independent variables (current values);
ind_envir_index_maver5 soc_envir_index_maver5 envir_index_maver illness_infect_maver5 - 5-year time averages of explanatory variables;
i.year - dummy years
Have I correctly reproduced the model presented in the article? Do I understand correctly that this is a kind of analogue of the fixed effects model?
glm math4 lavgrexp alavgrexp lunch alunch lenroll alenroll y96-y01 if year>1994, fa(bin) link(probit) cluster(distid)
For some variables, the question of endogeneity associated with a reverse causality arises, i.e. illness_infect can affect shareorg_ecoinn6_1 and vice versa. Is it then possible to add lags of the explanatory variables together with their time averages to mitigate endogeneity? Doesn't this break the logic of the model?
For example:
I research the drivers of regional eco-innovation. I have panel data for 2010-2015 for over 70 regions. Since the dependent variable (eco-innovation) ranges from 0 to 1, I am trying to apply generalized linear models, fractional regression models. Specifically, I built regressions according to Papke & Wooldridge (2008) "Panel data methods for fractional response variables with an application to test pass rates" - i.e. in addition to the explanatory variables themselves, I added time averages for these variables (average over the last 5 years):
Code:
glm shareorg_ecoinn6_1 ind_envir_index soc_envir_index envir_index illness_infect ind_envir_index_maver5 soc_envir_index_maver5 envir_index_maver illness_infect_maver5 i.year, fam(bin) link(probit) vce(robust)
Code:
xtgee shareorg_ecoinn6_1 ind_envir_index soc_envir_index envir_index illness_infect ind_envir_index_maver5 soc_envir_index_maver5 envir_index_maver illness_infect_maver5 i.year, fam(bin) link(probit) vce(robust)
ind_envir_index soc_envir_index envir_index illness_infect - independent variables (current values);
ind_envir_index_maver5 soc_envir_index_maver5 envir_index_maver illness_infect_maver5 - 5-year time averages of explanatory variables;
i.year - dummy years
Have I correctly reproduced the model presented in the article? Do I understand correctly that this is a kind of analogue of the fixed effects model?
glm math4 lavgrexp alavgrexp lunch alunch lenroll alenroll y96-y01 if year>1994, fa(bin) link(probit) cluster(distid)
For some variables, the question of endogeneity associated with a reverse causality arises, i.e. illness_infect can affect shareorg_ecoinn6_1 and vice versa. Is it then possible to add lags of the explanatory variables together with their time averages to mitigate endogeneity? Doesn't this break the logic of the model?
For example:
Code:
glm shareorg_ecoinn6_1 ind_envir_index_LAG1 soc_envir_index_LAG1 envir_index_LAG1 illness_infect_LAG1 ind_envir_index_maver5 soc_envir_index_maver5 envir_index_maver illness_infect_maver5 i.year, fam(bin) link(probit) vce(robust)
