Hello everyone,
I am currently working on a project focused on the impacts of subsidies (ln_cum_bdg) on innovation in a specific technology and I am looking at the moderation effect that industrial structure (ind_int_4scale measures integration and conc_1scale measures concentration) at regional level has on this impact. My dependent variable is patents (ccs_pat), so I started by employing poisson model (with the command ppmlhdfe). Following you find the estimated model:
For clarity, remaining variables are controls such as a regional innovation score, a dummy for a specific CCS policy introduced amidst the panel period in EU, per capita real GDP, wages in the industrial sector and a pre sample mean of the dependent variable to act as a fixed effect (as per Blundell et al. 1995). Another important point is that while the dependent variable is specific innovation at regional level, the subsidies are not to innovation, rather to commercial projects and they are at national level.
This modelling and the data suffer from a series of shortcomings related to endogeneity. In particular, I am worrying about potential simultaneous shocks that affected both patenting and the application of subsidies at regional or national level. I tried to instrument the potentially endogenous variable of subsidies with data on regional elections: election dummy (election_dummy) and left-to-right distance measure (left_right_new_3). I also thought about using lagged country-level means of dependent variables as instruments (country_y_ccs, country_y_conc and country_y_ind_int). This can be an unwise choice even if the relation is strong and significant, please give me a feedback on this if you can as well.
My issue is that ivpoisson does not provide for interaction commands, and while I would follow the standard two stages in a 2sls framework, I know that ivpoisson works with a GMM estimator and does not really have a "standard" two stage framework. My question to you is: how can I recreate the model you see above using ivpoisson and instrumenting both the endogenous variable and its interaction in the correct way? Below I show you what I have done so far (but results are not good: the model does not converge).
Thank you for the help in advance.
I am currently working on a project focused on the impacts of subsidies (ln_cum_bdg) on innovation in a specific technology and I am looking at the moderation effect that industrial structure (ind_int_4scale measures integration and conc_1scale measures concentration) at regional level has on this impact. My dependent variable is patents (ccs_pat), so I started by employing poisson model (with the command ppmlhdfe). Following you find the estimated model:
Code:
ppmlhdfe ccs_pat L.ln_cum_bdg_dep c.L.ln_cum_bdg_dep#c.L.ind_int_4scale c.L.ln_cum_bdg_dep#c.L.conc_1scale L.conc_1scale L.ind_int_4scale i.perf_score eu_policy gdp_pc_new ind_wage_ pre_sample_ccs, abs(i.year i.country) cluster(country)
This modelling and the data suffer from a series of shortcomings related to endogeneity. In particular, I am worrying about potential simultaneous shocks that affected both patenting and the application of subsidies at regional or national level. I tried to instrument the potentially endogenous variable of subsidies with data on regional elections: election dummy (election_dummy) and left-to-right distance measure (left_right_new_3). I also thought about using lagged country-level means of dependent variables as instruments (country_y_ccs, country_y_conc and country_y_ind_int). This can be an unwise choice even if the relation is strong and significant, please give me a feedback on this if you can as well.
My issue is that ivpoisson does not provide for interaction commands, and while I would follow the standard two stages in a 2sls framework, I know that ivpoisson works with a GMM estimator and does not really have a "standard" two stage framework. My question to you is: how can I recreate the model you see above using ivpoisson and instrumenting both the endogenous variable and its interaction in the correct way? Below I show you what I have done so far (but results are not good: the model does not converge).
Code:
reg ln_cum_bdg_dep L.left_right_new_3 L.election_dummy L.eu_policy L.gdp_pc_new L.ind_wage_ L.perf_score L.country_y_ccs L.country_y_conc L.country_y_ind_int i.country i.year, robust predict ln_cum_bdg_dep_hat
Code:
gen int_ind_int = ln_cum_bdg_dep*ind_int_4scale gen int_conc= ln_cum_bdg_dep*conc_1scale
Code:
gen int_ind_int_hat = ln_cum_bdg_dep_hat*ind_int_4scale gen int_conc_hat= ln_cum_bdg_dep_hat*conc_1scale
Code:
ivpoisson gmm ccs_pat (L.ln_cum_bdg_dep L.int_ind_int L.int_conc = L2.left_right_new_3 L2.election_dummy L2.int_ind_int_hat L2.int_conc_hat L2.country_y_ccs L2.country_y_conc L2.country_y_ind_int i.year i.country) L.ind_int_4scale L.conc_1scale L.eu_policy L.gdp_pc_new L.ind_wage_ L.perf_score, vce(cluster country)
Comment