This is my first post, so im testing here before posting in General forum!!
In STATA 14 we estimate the following double hurdle model
Which gives nice and interpretable results.
However we would like to allow the variance to have heteroscedasticity, and specify the model to allow for that as:
The model converges without problems, but the second hurdle parameter coefficients changes tremendously, especially when we calculate the marginal effects. Looking at the parameter for self the parameter value increases from 0.23 to -5.81 and the marginal effects (ystar) change from -3.01 to 18.03 and unconditional effects from 0.129 (strongly insignificant) to 57.8 and strongly significant. What is going on??
Our output without het():
And with the marginal effects using margins for ystar and e (we do not show the partial effects for the probability as that is unchanged between the two estimations.
And with heteroscedasticity specified
In STATA 14 we estimate the following double hurdle model
Code:
churdle lin y self turen age,select(self turen female) ll(0) margins, dydx(self turen age female) predict(ystar(0,.)) margins, dydx(self turen age) predict(e(0,.))
However we would like to allow the variance to have heteroscedasticity, and specify the model to allow for that as:
Code:
churdle lin y self turen age,select(self turen female) het(self turen age) ll(0) margins, dydx(self turen age female) predict(ystar(0,.)) margins, dydx(self turen age) predict(e(0,.))
The model converges without problems, but the second hurdle parameter coefficients changes tremendously, especially when we calculate the marginal effects. Looking at the parameter for self the parameter value increases from 0.23 to -5.81 and the marginal effects (ystar) change from -3.01 to 18.03 and unconditional effects from 0.129 (strongly insignificant) to 57.8 and strongly significant. What is going on??
Our output without het():
Code:
------------------------------------------------------------------------------
y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
y |
self | .2306694 2.871907 0.08 0.936 -5.398164 5.859503
turen_d | 106.8683 139.0496 0.77 0.442 -165.6638 379.4005
age | -4.750032 3.039761 -1.56 0.118 -10.70785 1.20779
_cons | 884.4703 391.2746 2.26 0.024 117.5863 1651.354
-------------+----------------------------------------------------------------
selection_ll |
self | -.0079584 .0021007 -3.79 0.000 -.0120756 -.0038411
turen_d | -.0192124 .1045648 -0.18 0.854 -.2241556 .1857307
female | .0836791 .0652191 1.28 0.199 -.044148 .2115062
_cons | .8016719 .2816518 2.85 0.004 .2496446 1.353699
-------------+----------------------------------------------------------------
lnsigma |
_cons | 6.757771 .0462606 146.08 0.000 6.667102 6.84844
-------------+----------------------------------------------------------------
/sigma | 860.7213 39.81747 786.1139 942.4094
------------------------------------------------------------------------------
And with the marginal effects using margins for ystar and e (we do not show the partial effects for the probability as that is unchanged between the two estimations.
Code:
Expression : Mean estimates of y*= max{a, min(y,b)}, predict(ystar(0,.))
dy/dx w.r.t. : self turen_d age female
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
self | -3.011951 1.06357 -2.83 0.005 -5.096509 -.9273921
turen_d | 18.22074 52.31436 0.35 0.728 -84.31352 120.755
age | -1.138989 .7240076 -1.57 0.116 -2.558018 .2800397
female | 32.25111 25.15053 1.28 0.200 -17.04303 81.54524
------------------------------------------------------------------------------
Code:
Expression : Conditional mean estimate of dependent variable in (a,b), predict(e(0,.))
dy/dx w.r.t. : self turen_d age
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
self | .1293251 1.61034 0.08 0.936 -3.026883 3.285533
turen_d | 59.91585 77.94611 0.77 0.442 -92.85571 212.6874
age | -2.66311 1.691031 -1.57 0.115 -5.977471 .6512505
------------------------------------------------------------------------------
And with heteroscedasticity specified
Code:
------------------------------------------------------------------------------
y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
y |
self | -5.816994 1.497549 -3.88 0.000 -8.752137 -2.881851
turen_d | 679.1555 105.7595 6.42 0.000 471.8708 886.4402
age | -.3096104 2.079937 -0.15 0.882 -4.386213 3.766992
_cons | 1370.061 120.9306 11.33 0.000 1133.042 1607.081
-------------+----------------------------------------------------------------
selection_ll |
self | -.0079584 .0021007 -3.79 0.000 -.0120756 -.0038411
turen_d | -.0192124 .1045648 -0.18 0.854 -.2241556 .1857307
female | .0836791 .0652191 1.28 0.199 -.044148 .2115062
_cons | .8016719 .2816518 2.85 0.004 .2496446 1.353699
-------------+----------------------------------------------------------------
lnsigma |
self | .0494853 .0009241 53.55 0.000 .047674 .0512965
turen_d | .4566014 .1064034 4.29 0.000 .2480547 .6651482
age | .0108982 .0021244 5.13 0.000 .0067345 .0150619
------------------------------------------------------------------------------
Code:
Expression : Mean estimates of y*= max{a, min(y,b)}, predict(ystar(0,.))
dy/dx w.r.t. : self turen_d age female
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
self | 18.03894 2.089843 8.63 0.000 13.94292 22.13496
turen_d | 369.8748 82.58536 4.48 0.000 208.0105 531.7391
age | 5.273477 1.044157 5.05 0.000 3.226965 7.319988
female | 51.33115 39.9712 1.28 0.199 -27.01096 129.6733
------------------------------------------------------------------------------
Code:
Expression : Conditional mean estimate of dependent variable in (a,b), predict(e(0,.))
dy/dx w.r.t. : self turen_d age
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
self | 57.82585 2.816197 20.53 0.000 52.3062 63.34549
turen_d | 922.2891 125.2199 7.37 0.000 676.8626 1167.716
age | 13.2504 2.562389 5.17 0.000 8.228212 18.27259
------------------------------------------------------------------------------
