Hi all. I have a methodological question concerning the use of instrumental variables. I'm working on a paper that uses individual-level survey data. To that data, I added variables from the municipalities where individuals in the survey data live. They include geographical variables, which are: altitude, longitude, distance to the state’s capital, and anthropic use of soil.
The instrumented variables are: number of media outlets in a municipality per 100k people (ln_veiculos_por_100k_hab), whether a municipality has a tv station or radio station (tvcom_radio_am_fm), and the interaction between the two variables (veiculos_100k_x_tvcom_radio).
The dependent variable is whether the survey respondent turned out to vote in the last general election (turnout_first_round).
I want to use these geographical variables as instruments in my regression.
I ran a IV-probit model using the aforementioned geographical variables as instruments (see model’s output below).
The question is: Is there any objection to the use of "higher-level" variables (i.e., variables not measured at the individual level) as instrumental variables in models with individual-level data?
Thank you
The instrumented variables are: number of media outlets in a municipality per 100k people (ln_veiculos_por_100k_hab), whether a municipality has a tv station or radio station (tvcom_radio_am_fm), and the interaction between the two variables (veiculos_100k_x_tvcom_radio).
The dependent variable is whether the survey respondent turned out to vote in the last general election (turnout_first_round).
I want to use these geographical variables as instruments in my regression.
I ran a IV-probit model using the aforementioned geographical variables as instruments (see model’s output below).
The question is: Is there any objection to the use of "higher-level" variables (i.e., variables not measured at the individual level) as instrumental variables in models with individual-level data?
Thank you
Code:
. ivprobit turnout_first_round pol_interest intensity_follow_media retrosp_eval_socio party_id unemployed sexo educ ec_casado beneficiary_bf rel_evang rel_cat color_nowhite (ln_veiculos_por_100k_hab tvcom_radio_am_fm veiculos_100k_x_tvcom_radio = altitude dist_cap_est longitude area_hec_anthropic_perc), twostep Checking reduced-form model... Two-step probit with endogenous regressors Number of obs = 2,243 Wald chi2(15) = 40.84 Prob > chi2 = 0.0003 --------------------------------------------------------------------------------------------- | Coef. Std. Err. z P>|z| [95% Conf. Interval] ----------------------------+---------------------------------------------------------------- ln_veiculos_por_100k_hab | 4.757881 1.610617 2.95 0.003 1.601129 7.914633 tvcom_radio_am_fm | 5.142635 1.553713 3.31 0.001 2.097415 8.187856 veiculos_100k_x_tvcom_radio | -6.554101 2.044757 -3.21 0.001 -10.56175 -2.546451 pol_interest | .3006396 .0847307 3.55 0.000 .1345704 .4667087 intensity_follow_media | -.1738803 .0868592 -2.00 0.045 -.3441211 -.0036394 retrosp_eval_socio | -.0281594 .0512776 -0.55 0.583 -.1286617 .072343 party_id | .3859058 .1455576 2.65 0.008 .100618 .6711935 unemployed | -.3206427 .1794491 -1.79 0.074 -.6723564 .0310711 sexo | .1277034 .1189566 1.07 0.283 -.1054472 .360854 educ | .0572311 .0306147 1.87 0.062 -.0027725 .1172348 ec_casado | .1395974 .1284862 1.09 0.277 -.112231 .3914258 beneficiary_bf | .0892701 .1381738 0.65 0.518 -.1815455 .3600858 rel_evang | -.1692903 .1862543 -0.91 0.363 -.5343419 .1957614 rel_cat | .1692262 .1701128 0.99 0.320 -.1641889 .5026412 color_nowhite | -.0886158 .13524 -0.66 0.512 -.3536814 .1764498 _cons | -1.968669 .8197188 -2.40 0.016 -3.575288 -.3620493 --------------------------------------------------------------------------------------------- Instrumented: ln_veiculos_por_100k_hab tvcom_radio_am_fm veiculos_100k_x_tvcom_radio Instruments: pol_interest intensity_follow_media retrosp_eval_socio party_id unemployed sexo educ ec_casado beneficiary_bf rel_evang rel_cat color_nowhite altitude dist_cap_est longitude area_hec_anthropic_perc --------------------------------------------------------------------------------------------- Wald test of exogeneity: chi2(3) = 41.46 Prob > chi2 = 0.0000 . overid Test of overidentifying restrictions: Amemiya-Lee-Newey minimum chi-sq statistic 1.071 Chi-sq(1) P-value = 0.3008