Dear all,
I have encountered a possible glitch with -xi- for which I want to check I am not going
mad.
The specific instance is using -xi- in models with two equations, such as -heckman- or
in my case, a double hurdle. I have found, repeatedly, that if the same variable appears
in both the structural and the selection equation then -xi- reports interaction in the
second equation, but seems to omit the base coefficient of the interacted variable for
the second equation, seemingly constraining it to be the same as the structural
equation.
I have included a report below as an indication of what I mean. Apologies for layout,
but the issue can be seen - note the absence of bls in the selection equation. Can
others replicate this, or am I doing something wrong? A clear work-around is to
manually interact, but I firstly want to check that I am correct.
. xi: heckman bgi i.country*bls , sel (d=i.country*bls)
i.country _Icountry_1-3 (naturally coded; _Icountry_1 omitted)
i.country*bls _IcouXbls_# (coded as above)
i.country*bls _IcouXblsa# (coded as above)
Iteration 0: log likelihood = -836.29474 (not concave)
...
Iteration 48: log likelihood = 11.05681
Iteration 49: log likelihood = 11.056816
Iteration 50: log likelihood = 11.056818
Iteration 51: log likelihood = 11.056818 (backed up)
Heckman selection model Number of obs = 299
(regression model with sample selection) Censored obs = 71
Uncensored obs = 228
Wald chi2(5) = 1.66e+07
Log likelihood = 11.05682 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
bgi |
_Icountry_2 | .0618531 .0358143 1.73 0.084 -.0083417 .1320479
_Icountry_3 | -.0597964 .0274365 -2.18 0.029 -.113571 -.0060219
bls | .0015758 3.87e-07 4075.57 0.000 .001575 .0015766
_IcouXbls_2 | .0125649 .007275 1.73 0.084 -.0016938 .0268236
_IcouXbls_3 | .026978 .0043983 6.13 0.000 .0183575 .0355984
_cons | .0603059 .017187 3.51 0.000 .0266199 .0939918
-------------+----------------------------------------------------------------
d |
_Icountry_2 | .396696 .2036728 1.95 0.051 -.0024952 .7958873
_Icountry_3 | .0895229 .1558116 0.57 0.566 -.2158622 .3949079
_IcouXblsa2 | .0554417 .0413973 1.34 0.180 -.0256955 .1365789
_IcouXblsa3 | .0131956 .0249932 0.53 0.598 -.0357901 .0621813
_cons | .3238787 .1000618 3.24 0.001 .1277612 .5199963
-------------+----------------------------------------------------------------
/athrho | 18.29885 310.525 0.06 0.953 -590.3189 626.9166
/lnsigma | -1.736847 .0477675 -36.36 0.000 -1.830469 -1.643224
-------------+----------------------------------------------------------------
rho | 1 2.07e-13 -1 1
sigma | .1760748 .0084106 .1603383 .1933557
lambda | .1760748 .0084106 .1595902 .1925593
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0): chi2(1) = 100.81 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
Many thanks for your thoughts,
Julian
---------------------------------------------------------
Julian A. Fennema
CERT (Centre for Economic Reform and Transformation)
School of Management and Languages
Heriot-Watt University
Riccarton
Edinburgh EH14 4AS
Tel: +44 (0)131 451 3492
Fax: +44 (0)131 451 3296
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
