Dears,
I am writing to seek your advice on a methodological challenge I am facing. Social expenditure scheme shares are compositional in nature, as they are parts of a fixed whole that sum to unity. To identify the effect of social expenditure composition on income inequality, I therefore apply an additive log-ratio (ALR) transformation following Aitchison (1986).
gen alr_h_e = ln(s_h/s_e)
gen alr_sp_e = ln(s_sp/s_e)
To address potential endogeneity between social spending schemes and inequality, I use Bartik (shift-share) instruments. However, this leads to a new issue: my endogenous variables are compositional treatments expressed in log-ratio form, while the instruments are defined in levels. I am concerned about how best to address this mismatch and whether it may induce nonlinearities or heterogeneous effects in the estimation.
ivreg2 ln_gini ( alr_h_e alr_sp_e = Z_health Z_sp) L_gdppc L_trade ln_expgdp i.year , cluster(id) first savefirst
My study is panel data of SSA countries.
I am writing to seek your advice on a methodological challenge I am facing. Social expenditure scheme shares are compositional in nature, as they are parts of a fixed whole that sum to unity. To identify the effect of social expenditure composition on income inequality, I therefore apply an additive log-ratio (ALR) transformation following Aitchison (1986).
gen alr_h_e = ln(s_h/s_e)
gen alr_sp_e = ln(s_sp/s_e)
To address potential endogeneity between social spending schemes and inequality, I use Bartik (shift-share) instruments. However, this leads to a new issue: my endogenous variables are compositional treatments expressed in log-ratio form, while the instruments are defined in levels. I am concerned about how best to address this mismatch and whether it may induce nonlinearities or heterogeneous effects in the estimation.
ivreg2 ln_gini ( alr_h_e alr_sp_e = Z_health Z_sp) L_gdppc L_trade ln_expgdp i.year , cluster(id) first savefirst
My study is panel data of SSA countries.
