Eiji, you have two distinct questions here.
The standard errors are different because xtivreg2 uses a small sample
adjustment which accounts for the degrees of freedoms lost by estimating
your control variables and all the fixed effects. ivreg2 by default does
not use a small sample adjustment (i.e. standard errors are just divided by
N) although you can get the small sample adjusted standard errors using the
'small' option. The choice here is really a matter of taste. IV is an
asymptotic estimator and thus a small sample adjustment is unnecessary. If
you do want a small sample adjustment using ivreg2 you will have to manual
adjust the standard errors because of the additional degrees of freedom lost
by mean-differencing your data (i.e. the number of fixed effects).
I am surprised that the Sargan statistics are not identical since this does
not depend on any small sample adjustment. I am a co-author of both ivreg2
and overidxt. overidxt is a simple implementation of the sargan statistic
and I have a high degree of confidence that the code is correct. I and my
co-authors will have to take a closer look at the ivreg2 code because it is
much more complicated. We will report back to the list once we have more
information on this.
Steve
> -----Original Message-----
> From: Eiji Mangyo [SMTP:[email protected]]
> Sent: Wednesday, August 13, 2003 4:09 PM
> To: [email protected]
> Subject: st: overid test in ivreg2 vs overidxt
>
> Dear Statalisters,
>
> I have a 2-year panel household data set. I estimated the theoretically
> same
> model using two different Stata commands. First, using xtivreg, fe and
> second, manually differencing within household and using ivreg2. I got the
> same coefficient estimates but different standard errors and different
> overid test statistics. (Result tables copied below; 1st stages omitted)
>
> 1) Are the standard errors of the coefficient estimates different because
> xtivreg, fe used households with both single observation and two repeated
> observations while differencing and ivreg2 used only household with two
> repeated observations?
>
> 2) Is the Sargan statistic given by overidxt different from the Sargan
> statistic given in ivreg2 for the same reason as above?
>
> Thank you very much for your help.
> Eiji
>
> xtivreg dprome2 mefor afor mesrh asrh maxedu /*
> > */ y93 lny93 heny93 sdy93 hby93 hny93 gxy93 gzy93 city93 suby93
> towy93 /*
> > */ (lnpcdef=drainq1-drainq4 drainqland1-drainqland4 /*
> > */ dsunq1-dsunq4 dsunqland1-dsunqland4 /*
> > */ drainqvill1-drainqvill4 dsunqvill1-dsunqvill4),fe
>
> Fixed-effects (within) IV regression Number of obs =
> 850
> Group variable: ohhid Number of groups =
> 581
>
> R-sq: within = 0.0438 Obs per group: min =
> 1
> between = 0.0006 avg =
> 1.5
> overall = 0.0024 max =
> 2
>
> Wald chi2(17) =
> 943.39
> corr(u_i, Xb) = -0.3811 Prob > chi2 =
> 0.0000
>
>
> ----------------------------------------------------------------------
> --------
> dprome2 | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
>
> -------------+--------------------------------------------------------
> --------
> lnpcdef | -.0829731 .0578167 -1.44 0.151 -.1962917
> .0303454
> mefor | .014233 .022224 0.64 0.522 -.0293253
> .0577912
> afor | -.0363992 .0359746 -1.01 0.312 -.1069081
> .0341096
> mesrh | -.0880036 .03345 -2.63
> 0.009 -.1535644 -.0224427
> asrh | .1225672 .047207 2.60 0.009 .0300431
> .2150912
> maxedu | .0141301 .0174315 0.81 0.418 -.020035
> .0482953
> y93 | -.0033069 .062804 -0.05 0.958 -.1264005
> .1197867
> lny93 | .137499 .092804 1.48 0.138 -.0443935
> .3193915
> heny93 | -.0247196 .10157 -0.24 0.808 -.2237932
> .174354
> sdy93 | -.1033524 .093967 -1.10 0.271 -.2875244
> .0808195
> hby93 | -.1834472 .092521 -1.98
> 0.047 -.3647851 -.0021094
> hny93 | -.0731059 .1022382 -0.72 0.475 -.2734891
> .1272773
> gxy93 | -.0030924 .0728227 -0.04 0.966 -.1458223
> .1396375
> gzy93 | -.0967301 .0762305 -1.27 0.204 -.2461392
> .052679
> city93 | .026772 .0498719 0.54 0.591 -.0709751
> .1245191
> suby93 | -.0511917 .0635875 -0.81 0.421 -.175821
> .0734375
> towy93 | .0035864 .0529958 0.07 0.946 -.1002835
> .1074562
> _cons | .3330647 .4183751 0.80 0.426 -.4869353
> 1.153065
>
> -------------+--------------------------------------------------------
> --------
> sigma_u | .25929132
> sigma_e | .22144028
> rho | .57825123 (fraction of variance due to u_i)
>
> ----------------------------------------------------------------------
> --------
> F test that all u_i=0: F(580,252) = 1.22 Prob > F
> = 0.0325
>
> ----------------------------------------------------------------------
> --------
> Instrumented: lnpcdef
> Instruments: mefor afor mesrh asrh maxedu y93 lny93 heny93 sdy93
> hby93 hny93
> gxy93 gzy93 city93 suby93 towy93 drainq1 drainq2
> drainq3
> drainq4 drainqland1 drainqland2 drainqland3
> drainqland4 dsunq1
> dsunq2 dsunq3 dsunq4 dsunqland1 dsunqland2
> dsunqland3
> dsunqland4 drainqvill1 drainqvill2 drainqvill3
> drainqvill4
> dsunqvill1 dsunqvill2 dsunqvill3 dsunqvill4
>
> overidxt
>
> Test of overidentifying restrictions: 17.241 Chi-sq(14) P-value =
> .2436
>
>
> ivreg2 ddprome2 dmefor dafor dmesrh dasrh dmaxedu /*
> > */ dy93 dlny93 dheny93 dsdy93 dhby93 dhny93 dgxy93 dgzy93 dcity93
> dsuby93 dto wy93 /*
> > */ (dlnpcdef= ddrainq1-ddrainq4 ddrainqland1-ddrainqland4 /*
> > */ ddsunq1-ddsunq4 ddsunqland1-ddsunqland4 /*
> > */ ddrainqvill1-ddrainqvill4 ddsunqvill1-ddsunqvill4)
>
>
> Instrumental variables (2SLS) regression
> ----------------------------------------
>
> Number of obs
> =
> 269
> F( 16, 238)
> =
> 1.84
> Prob > F
> =
> 0.0268
> Total (centered) SS = 25.45874667 Centered R2
> =
> 0.0293
> Total (uncentered) SS = 25.84544851 Uncentered R2
> =
> 0.0438
> Residual SS = 24.71404235 Root MSE
> =
> .303107
>
>
> ----------------------------------------------------------------------
> --------
> ddprome2 | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
>
> -------------+--------------------------------------------------------
> --------
> dlnpcdef | -.0829732 .0559599 -1.48 0.138 -.1926526
> .0267063
> dmefor | .014233 .0215103 0.66 0.508 -.0279264
> .0563924
> dafor | -.0363992 .0348193 -1.05 0.296 -.1046438
> .0318453
> dmesrh | -.0880035 .0323758 -2.72
> 0.007 -.151459 -.0245481
> dasrh | .1225671 .045691 2.68 0.007 .0330144
> .2121199
> dmaxedu | .0141301 .0168717 0.84 0.402 -.0189378
> .0471981
> dy93 | (dropped)
> dlny93 | .137499 .0898237 1.53 0.126 -.0385522
> .3135502
> dheny93 | -.0247196 .0983082 -0.25 0.801 -.2174001
> .1679609
> dsdy93 | -.1033525 .0909493 -1.14 0.256 -.2816099
> .0749049
> dhby93 | -.1834473 .0895498 -2.05
> 0.041 -.3589616 -.0079329
> dhny93 | -.0731059 .0989549 -0.74 0.460 -.267054
> .1208421
> dgxy93 | -.0030924 .0704841 -0.04 0.965 -.1412387
> .1350539
> dgzy93 | -.0967301 .0737825 -1.31 0.190 -.2413411
> .0478809
> dcity93 | .026772 .0482703 0.55 0.579 -.067836
> .12138
> dsuby93 | -.0511917 .0615455 -0.83 0.406 -.1718186
> .0694352
> dtowy93 | .0035864 .0512939 0.07 0.944 -.0969478
> .1041205
> _cons | -.0033069 .0607871 -0.05 0.957 -.1224474
> .1158336
>
> ----------------------------------------------------------------------
> --------
> Sargan statistic (overidentification test of all instruments):
> 5.456
> Chi-sq(14) P-val
> =
> 0.97839
>
> ----------------------------------------------------------------------
> --------
> Collinearities detected among instruments: 10 instrument(s) dropped
> Instrumented: dlnpcdef
> Instruments: dmefor dafor dmesrh dasrh dmaxedu dy93 dlny93 dheny93
> dsdy93
> dhby93 dhny93 dgxy93 dgzy93 dcity93 dsuby93 dtowy93
> ddrainq1
> ddrainq2 ddrainq3 ddrainq4 ddrainqland1 ddrainqland2
> ddrainqland3 ddrainqland4 ddsunq1 ddsunq2 ddsunq3
> ddsunq4
> ddsunqland1 ddsunqland2 ddsunqland3 ddsunqland4
> ddrainqvill1
> ddrainqvill2 ddrainqvill3 ddrainqvill4 ddsunqvill1
> ddsunqvill2
> ddsunqvill3 ddsunqvill4
>
> ----------------------------------------------------------------------
> --------
>
> *
> * 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/
The information contained in this document is intended only for the
addressee and is not necessarily the views nor the official
communication of the Department of Labour. All final/official papers
which are sent from the Department will be sent by non-electronic
means, on appropriate letterhead, signed by authorised personnel.
*
* 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/
