Hello everyone
I am a sort of beginner applying causal inference and the potential outcome framework. I have been working in a paper about the effect of irrigation technologies on a series of outcomes, namely net income and input usage, among others. My dataset is a cross-section with 335 observations. So far, I have employed the doubly robust estimators (AIPW and IPWRA) for controlling for selection bias and confounding.
The technology adoption literature states that adoption might be potentially endogenous (unobservables are affecting the decision to apply a technology). Thus, I used the endogenous treatment effects command (eteffects) and tested for endogeneity afterward. The results for both tests were no rejection of the null hypothesis, so there is no correlation between treatment unobservables and outcome, so I continued using AIPW and IPWRA. (PART 1 below).
However, I came across extended regression models (ERM), which also allows estimating the ATE and ATET. So I made the same as in eteffects, setting my treatment as endogenous. The results, in this case, show that the treatment is endogenous. (PART 2 below).
The names of the variables are a kind of Spanglish (sorry).. some of them are in Spanish and others in English. The treatment is riego_efic.
Could anyone explain to me the main differences between both approaches, or at least how the endogeneity test works in eteffects and erm???
Thanks in advance
Cristian
1. Using effects (omitted output)
Here the results (last 2 lines of the output)
Endogeneity test
2. Using ERM (omitted output)
I am a sort of beginner applying causal inference and the potential outcome framework. I have been working in a paper about the effect of irrigation technologies on a series of outcomes, namely net income and input usage, among others. My dataset is a cross-section with 335 observations. So far, I have employed the doubly robust estimators (AIPW and IPWRA) for controlling for selection bias and confounding.
The technology adoption literature states that adoption might be potentially endogenous (unobservables are affecting the decision to apply a technology). Thus, I used the endogenous treatment effects command (eteffects) and tested for endogeneity afterward. The results for both tests were no rejection of the null hypothesis, so there is no correlation between treatment unobservables and outcome, so I continued using AIPW and IPWRA. (PART 1 below).
However, I came across extended regression models (ERM), which also allows estimating the ATE and ATET. So I made the same as in eteffects, setting my treatment as endogenous. The results, in this case, show that the treatment is endogenous. (PART 2 below).
The names of the variables are a kind of Spanglish (sorry).. some of them are in Spanish and others in English. The treatment is riego_efic.
Could anyone explain to me the main differences between both approaches, or at least how the endogeneity test works in eteffects and erm???
Thanks in advance
Cristian
1. Using effects (omitted output)
Code:
eteffects (ln_income_ha_pond Edad Educacion Exp_agr credito caudal_total Sup_prod_2018 owner tipoturno pago2 difacceso2 wua grupo1 grupo2) (riego_efic Edad Educacion Exp_agr credito caudal_total Sup_prod_2018 owner tipoturn > o pago2 difacceso2 wua grupo1 grupo2 Riego_asesor), vce(robust) aequations
Code:
TEOM0 |
_cons | -.4536561 .3415186 -1.33 0.184 -1.12302 .2157081
--------------+----------------------------------------------------------------
TEOM1 |
_cons | -.9158318 .5301846 -1.73 0.084 -1.954975 .1233109
-------------------------------------------------------------------------------
Code:
estat endogenous
Test of endogeneity
Ho: treatment and outcome unobservables are uncorrelated
chi2( 2) = 4.37
Prob > chi2 = 0.1125
2. Using ERM (omitted output)
Code:
eregress ln_income_ha_pond Edad Educacion Exp_agr credito caudal_total Sup_prod_2018 owner tipoturno pago2 difacceso2 wua grupo1 grupo2, entreat(riego_efic = Riego_asesor Edad Educacion Exp_agr credito caudal_total Sup_prod_2018 owner tipoturno pago2 difacceso2 wua grupo1 grupo2,nointeract )
Code:
---------------------------------------+---------------------------------------------------------------- corr(e.riego_efic,e.ln_income_ha_pond)| -.7689006 .1174924 -6.54 0.000 -.9187507 -.4254372
