[Limdep Nlogit List] Using Probit to Approximate Panel Fractional Response Model
Petrolia, Daniel
d.petrolia at msstate.edu
Wed May 18 02:55:45 AEST 2016
Hi: I would like to add year dummies to the panel fractional response model (NLOGIT command "fractional"), but this cannot work because the year dummies are time-varying and the model wants to add the population averages of these to the model, which results in a singular X matrix. As an alternative, I manually created the group average variables for my time-varying variables, and estimated a probit model, including a constant and the group average variables, and specifying "; cluster = N" where N is the number of observations in each group. The results are very similar to that given by the fractional command.
My question is if I am interpreting my findings correctly, i.e., that this is a legitimate alternative to fractional (i.e., is a legitimate way to estimate Papke & Wooldridge's (2008) model).
Below are the code and results I obtained.
Thanks,
Dan
|-> sample; all $
|-> CREATE ; crsmean = Group Mean (crs, Pds = 23) $
|-> CREATE ; edumean = Group Mean (edu, Pds = 23) $
|-> CREATE ; hhmean = Group Mean (hh1000, Pds = 23) $
|-> CREATE ; incmean = Group Mean (inc1000, Pds = 23) $
|-> CREATE ; pptmean = Group Mean (ppt, Pds = 23) $
|-> fractional
; Lhs=PIFRATE
;Rhs =
MISS,COAST,CRS,EDU,HH1000,INC1000,
SLP,PPT,ELEV,STRDN,
AVZONES,BCZONES,
;Pds=23
$
(starting values omitted)
Normal exit: 21 iterations. Status=0, F= 297.6328
-----------------------------------------------------------------------------
Fractional Response Model - Panel Data
Dependent variable PIFRATE
Log likelihood function 297.63281
Estimation based on N = 6739, K = 18
Inf.Cr.AIC = -559.3 AIC/N = -.083
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
PIFRATE| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Time Invariant Variables in Conditional Mean........................
Constant| -2.18313* 1.19857 -1.82 .0685 -4.53229 .16602
MISS| .03114 .11942 .26 .7943 -.20293 .26520
COAST| .32190 .23260 1.38 .1664 -.13398 .77778
SLP| -.07457** .03515 -2.12 .0339 -.14345 -.00569
ELEV| -.00068 .00094 -.72 .4734 -.00253 .00117
STRDN| -.16654 .14083 -1.18 .2370 -.44256 .10949
AVZONES| 1.31919*** .22331 5.91 .0000 .88151 1.75686
BCZONES| -.49804*** .17758 -2.80 .0050 -.84609 -.14998
|Time Varying Variables in Conditional Mean..........................
CRS| .31734*** .10665 2.98 .0029 .10830 .52638
EDU| -.00573 .00451 -1.27 .2042 -.01456 .00311
HH1000| -.01686*** .00511 -3.30 .0010 -.02688 -.00685
INC1000| .00488** .00221 2.21 .0274 .00054 .00921
PPT|-.62915D-04*** .1480D-04 -4.25 .0000 -.91926D-04 -.33904D-04
|Group Means of Time Varying Variables...............................
CRS| .07390 .18779 .39 .6939 -.29415 .44196
EDU| .00906 .00653 1.39 .1651 -.00373 .02186
HH1000| .01368** .00635 2.15 .0312 .00123 .02613
INC1000| .00542 .00514 1.06 .2909 -.00464 .01549
PPT| .00032 .00088 .36 .7156 -.00140 .00204
--------+--------------------------------------------------------------------
nnnnn.D-xx or D+xx => multiply by 10 to -xx or +xx.
***, **, * ==> Significance at 1%, 5%, 10% level.
Model was estimated on May 17, 2016 at 11:50:45 AM
-----------------------------------------------------------------------------
|-> probit
; Lhs=PIFRATE
;Rhs =
one,MISS,COAST,CRS,EDU,HH1000,INC1000,
SLP,PPT,ELeV,STRDN,
AVZONES,BCZONES,
crsmean,edumean,hhmean,incmean,pptmean
; cluster = 23
$
Normal exit: 7 iterations. Status=0, F= 1092.854
+---------------------------------------------------------------------+
| Covariance matrix for the model is adjusted for data clustering. |
| Sample of 6739 observations contained 293 clusters defined by |
| 23 observations (fixed number) in each cluster. |
+---------------------------------------------------------------------+
-----------------------------------------------------------------------------
Probit Model for Fractional Data
Dependent variable PIFRATE
Log likelihood function -1092.85352
Restricted log likelihood -1415.83949
Chi squared [ 17](P= .000) 645.97194
Significance level .00000
McFadden Pseudo R-squared .2281233
Estimation based on N = 6739, K = 18
Inf.Cr.AIC = 2221.7 AIC/N = .330
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
PIFRATE| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Index function for probability......................................
Constant| -2.17156* 1.18004 -1.84 .0657 -4.48438 .14127
MISS| .02969 .11862 .25 .8024 -.20280 .26217
COAST| .32345 .23429 1.38 .1674 -.13574 .78265
CRS| .31306*** .10597 2.95 .0031 .10537 .52074
EDU| -.00572 .00434 -1.32 .1875 -.01422 .00279
HH1000| -.01645*** .00496 -3.32 .0009 -.02617 -.00674
INC1000| .00479** .00220 2.18 .0296 .00047 .00910
SLP| -.07395** .03473 -2.13 .0332 -.14202 -.00588
PPT|-.64055D-04*** .1463D-04 -4.38 .0000 -.92720D-04 -.35390D-04
ELEV| -.00066 .00094 -.70 .4818 -.00251 .00119
STRDN| -.16587 .13906 -1.19 .2330 -.43843 .10668
AVZONES| 1.32265*** .22223 5.95 .0000 .88709 1.75820
BCZONES| -.50058*** .17533 -2.86 .0043 -.84421 -.15695
CRSMEAN| .08047 .18524 .43 .6640 -.28260 .44354
EDUMEAN| .00910 .00637 1.43 .1526 -.00337 .02158
HHMEAN| .01359** .00628 2.16 .0305 .00128 .02591
INCMEAN| .00550 .00512 1.07 .2831 -.00454 .01553
PPTMEAN| .00031 .00086 .36 .7181 -.00138 .00201
--------+--------------------------------------------------------------------
nnnnn.D-xx or D+xx => multiply by 10 to -xx or +xx.
***, **, * ==> Significance at 1%, 5%, 10% level.
Model was estimated on May 17, 2016 at 11:50:45 AM
-----------------------------------------------------------------------------
More information about the Limdep
mailing list