Hi All,
I am estimating the relationship between stunting and childhood chronic illness aimed at assessing whether stunting predisposes children to developing chronic illness. So, I regressed the child's binary chronic illness status in 2017 on the child's height-for-age z-scores (measuring stunting) from 2012. One reviewer suggested that there is a circular relationship between stunting and chronic illness- that unobserved factors affecting the development of chronic illness also affect stunting, so I decided to use ivprobit and employed mother's height and chronic illness status as instruments for stunting. I had argued initially that the circular relationship was valid only when the relationship between stunting and chronic illness is analyzed contemporaneously, and that in this case where I am 'predicting' the probability of the child developing chronic illness based on current stunting status that wouldn't hold, but the reviewer thought otherwise. The estimated model and corresponding output are included below. I have 3 key concerns that I need help with:
1. The Wald test of exogeneity suggests that stunting is exogenous (at 5% significance level). Is this sufficient to justify a return to estimate a standard probit model?
2. The reduced form equation (first stage regression) has covariates from 2017 on the right-hand side. I know that it is not a causal model, but my intuition tells me it is wrong to regress height-for-age z-scores from 2012 on values of covariates in 2017.
3. How do I account for the panel structure in my model? Since I am using information from 2 periods with same children at different times. Does it help to include the means of the time-varying covariates?
Thanks
I am estimating the relationship between stunting and childhood chronic illness aimed at assessing whether stunting predisposes children to developing chronic illness. So, I regressed the child's binary chronic illness status in 2017 on the child's height-for-age z-scores (measuring stunting) from 2012. One reviewer suggested that there is a circular relationship between stunting and chronic illness- that unobserved factors affecting the development of chronic illness also affect stunting, so I decided to use ivprobit and employed mother's height and chronic illness status as instruments for stunting. I had argued initially that the circular relationship was valid only when the relationship between stunting and chronic illness is analyzed contemporaneously, and that in this case where I am 'predicting' the probability of the child developing chronic illness based on current stunting status that wouldn't hold, but the reviewer thought otherwise. The estimated model and corresponding output are included below. I have 3 key concerns that I need help with:
1. The Wald test of exogeneity suggests that stunting is exogenous (at 5% significance level). Is this sufficient to justify a return to estimate a standard probit model?
2. The reduced form equation (first stage regression) has covariates from 2017 on the right-hand side. I know that it is not a causal model, but my intuition tells me it is wrong to regress height-for-age z-scores from 2012 on values of covariates in 2017.
3. How do I account for the panel structure in my model? Since I am using information from 2 periods with same children at different times. Does it help to include the means of the time-varying covariates?
Code:
ivprobit morbid_2017 age i.female i.race i.location pcy i.water i.toilet (haz_2012 = mother_height mother_chronic)
Code:
Fitting exogenous probit model
Iteration 0: log likelihood = -683.53033
Iteration 1: log likelihood = -669.18999
Iteration 2: log likelihood = -668.82397
Iteration 3: log likelihood = -668.82352
Iteration 4: log likelihood = -668.82352
Fitting full model
Iteration 0: log likelihood = -7800.4883
Iteration 1: log likelihood = -7800.3061
Iteration 2: log likelihood = -7800.3051
Iteration 3: log likelihood = -7800.305
Probit model with endogenous regressors Number of obs = 3,765
Wald chi2(10) = 43.36
Log likelihood = -7800.305 Prob > chi2 = 0.0000
------------------------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------------------------+----------------------------------------------------------------
haz_2012 | -.28906 .1213478 -2.38 0.017 -.5268973 -.0512227
age | -.005219 .0149409 -0.35 0.727 -.0345026 .0240646
1.female | -.0921963 .0735743 -1.25 0.210 -.2363994 .0520068
|
race |
Coloured | -.2093019 .1205054 -1.74 0.082 -.4454881 .0268844
Asian/indian | .6856803 .3292159 2.08 0.037 .0404291 1.330932
White | .7565883 .2985863 2.53 0.011 .1713699 1.341807
|
location |
Urban | .202714 .1001315 2.02 0.043 .0064599 .398968
pcy | .0289666 .0477688 0.61 0.544 -.0646586 .1225917
|
water |
Tap water | -.1640133 .1063639 -1.54 0.123 -.3724827 .0444561
|
toilet |
Flush toilet | -.0338975 .1048557 -0.32 0.746 -.2394109 .1716159
_cons | -1.876283 .3446661 -5.44 0.000 -2.551816 -1.20075
-------------------------------+----------------------------------------------------------------
corr(e.haz_2012,e.morbid_2017)| .4434656 .1992688 -.009634 .7454822
sd(e.haz_2012)| 1.608509 .0185383 1.572582 1.645257
------------------------------------------------------------------------------------------------
Instrumented: haz_2012
Instruments: age 1.female 2.race 3.race 4.race 1.location pcy 1.water 1.toilet mother_height
mother_chronic
------------------------------------------------------------------------------------------------
Wald test of exogeneity (corr = 0): chi2(1) = 3.69 Prob > chi2 = 0.0547
