Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: RE: Using ivhettest to test for heterogeneity


From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   st: RE: RE: Using ivhettest to test for heterogeneity
Date   Tue, 28 Feb 2012 16:33:07 -0000

Andreas, 

> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of 
> [email protected]
> Sent: 27 February 2012 08:38
> To: [email protected]
> Subject: st: RE: Using ivhettest to test for heterogeneity
> 
> Dear Mark,
> 
> Thank you for explaining me the usage of the two different commands.
> Unfortunately, I cannot infer whether the two commands give the same 
> results from your toy auto example because I miss the -ivhettest- 
> command below.

Sorry, my fault - I should have added the following for the example from
the toy auto dataset:

. ivhettest, ivcp nr2

OLS heteroskedasticity test(s) using levels and cross products of all
IVs
Ho: Disturbance is homoskedastic
    White/Koenker nR2 test statistic    :   5.890  Chi-sq(2) P-value =
0.0526

> But kindly have a look at the following example using
> my own dataset. There is a marked difference between the results after
> calling -imtest, white- and -ivhettest, ivsq nr2-. I would be happy to
> know where this difference must likely come from.

And the rest of the toy auto example above answers your question: you
need to use the -ivcp- rather than the -ivsq- option.  The White general
test uses the full set of instruments in levels, their squares, and
their cross-products.

I note, by the way, that in your regression you have 98 observations.
The general White test can rapidly use up degrees of freedom, because
the degrees of freedom correspond to the unique elements of the VCV.  In
your case, the general White test has 89 degrees of freedom, and with 98
observations you're not going to get much power.  This is probably why
the general White test doesn't reject homoskedasticity in your example,
whereas the way you called -ivhettest- (using the -ivsq- option and
hence ignoring the cross-products) gave you a test stat with fewer
degrees of freedom (24 instead of 89) and more power.

HTH,
Mark

> 
> . reg COC DRANK LBAGE D_LBAGE LNMV LNBM LNLEV CapInt ROA 
> AssTr LTG LVOL MAFE if FIN==0 
> 
>       Source |       SS       df       MS              Number 
> of obs =      98
> -------------+------------------------------           F( 12, 
>    85) =   24.23
>        Model |  .156474018    12  .013039502           Prob > 
> F      =  0.0000
>     Residual |   .04574489    85  .000538175           
> R-squared     =  0.7738
> -------------+------------------------------           Adj 
> R-squared =  0.7418
>        Total |  .202218908    97  .002084731           Root 
> MSE      =   .0232
> 
> --------------------------------------------------------------
> ----------------
>          COC |      Coef.   Std. Err.      t    P>|t|     
> [95% Conf. Interval]
> -------------+------------------------------------------------
> ----------------
>        DRANK |  -.2832065   .1170839    -2.42   0.018    
> -.5160006   -.0504123
>        LBAGE |  -.0152769   .0087299    -1.75   0.084    
> -.0326344    .0020806
>      D_LBAGE |    .032221   .0134307     2.40   0.019     
> .0055172    .0589248
>         LNMV |  -.0046663   .0022202    -2.10   0.039    
> -.0090806    -.000252
>         LNBM |   .0030857   .0060793     0.51   0.613    
> -.0090016     .015173
>        LNLEV |   .0093984   .0140586     0.67   0.506    
> -.0185538    .0373506
>       CapInt |  -.0209075   .0201706    -1.04   0.303    
> -.0610121    .0191972
>          ROA |  -.0582245   .0346929    -1.68   0.097    
> -.1272033    .0107543
>        AssTr |  -.0037709   .0039292    -0.96   0.340    
> -.0115832    .0040414
>          LTG |   .0009717   .0000776    12.53   0.000     
> .0008175     .001126
>         LVOL |  -.0011873   .0096773    -0.12   0.903    
> -.0204284    .0180538
>         MAFE |   .0030178   .0005758     5.24   0.000     
> .0018729    .0041626
>        _cons |   .2751544   .0723573     3.80   0.000     
> .1312886    .4190201
> --------------------------------------------------------------
> ----------------
> 
> . 
> end of do-file
> . 
> . imtest, white
> 
> White's test for Ho: homoskedasticity
>          against Ha: unrestricted heteroskedasticity
> 
>          chi2(89)     =     97.10
>          Prob > chi2  =    0.2613
> 
> Cameron & Trivedi's decomposition of IM-test
> 
> ---------------------------------------------------
>               Source |       chi2     df      p
> ---------------------+-----------------------------
>   Heteroskedasticity |      97.10     89    0.2613
>             Skewness |      17.64     12    0.1272
>             Kurtosis |       1.82      1    0.1769
> ---------------------+-----------------------------
>                Total |     116.56    102    0.1537
> ---------------------------------------------------
> 
> . ivhettest, ivsq nr2
> OLS heteroskedasticity test(s) using levels and squares of IVs
> Ho: Disturbance is homoskedastic
>     White/Koenker nR2 test statistic    :  58.500  Chi-sq(24) 
> P-value = 0.0001
> 
> 
> 
> 
> 
> [email protected] wrote: ----- 
> To: <[email protected]>
> From: "Schaffer, Mark E" 
> Sent by: [email protected]
> Date: 02/24/2012 11:51PM
> Subject: st: RE: Using ivhettest to test for heterogeneity
> 
> Andreas,
> 
> > -----Original Message-----
> > From: [email protected] 
> > [mailto:[email protected]] On Behalf Of 
> > [email protected]
> > Sent: 24 February 2012 20:14
> > To: [email protected]
> > Subject: st: Using ivhettest to test for heterogeneity 
> > 
> > Dear Statalist users, 
> >  
> > I am concerned with using White's general heteroskedasticity 
> > test when running an instrumental variables regression. There 
> > is an option "nr2" which can be used with -ivhettest- after 
> > the -ivreg2- command that displays the White-Koenker test 
> > statistic.
> 
> This is a test of heteroskedasticity in the IV equation (or 
> "2nd stage"
> if you prefer).
> 
> > However, when I run the first-stage regression 
> > manually and use the command "imtest, white", I get a very 
> > different value for the test statistic.
> 
> Because it's a test of heteroskedasticity in the first-stage 
> regression.
> 
> > The help file for -ivhettest- claims that this command can 
> > also be used for OLS regressions.
> 
> It can; see the example in the help file.  You estimate using OLS
> (-regress- or -ivreg2-), then call -ivhettest-.
> 
> > But why do the -ivhettest- 
> > and -imtest- commands yield different test statistics?
> 
> They should be the same.  Estimate using OLS and -regress-, and then
> compare -imtest, white- with -ivhettest, ivsq nr2-.  You 
> should get the
> same result.  Below is an example using the toy auto dataset.
> 
> Cheers,
> Mark
> 
> 
> . reg mpg weight
> 
>       Source |       SS       df       MS              Number of obs =
> 74
> -------------+------------------------------           F(  1,    72) =
> 134.62
>        Model |   1591.9902     1   1591.9902           Prob > F      =
> 0.0000
>     Residual |  851.469256    72  11.8259619           R-squared     =
> 0.6515
> -------------+------------------------------           Adj R-squared =
> 0.6467
>        Total |  2443.45946    73  33.4720474           Root MSE      =
> 3.4389
> 
> --------------------------------------------------------------
> ----------
> ------
>          mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf.
> Interval]
> -------------+------------------------------------------------
> ----------
> ------
>       weight |  -.0060087   .0005179   -11.60   0.000    -.0070411
> -.0049763
>        _cons |   39.44028   1.614003    24.44   0.000     36.22283
> 42.65774
> --------------------------------------------------------------
> ----------
> ------
> 
> . imtest, white
> 
> White's test for Ho: homoskedasticity
>          against Ha: unrestricted heteroskedasticity
> 
>          chi2(2)      =      5.89
>          Prob > chi2  =    0.0526
> 
> Cameron & Trivedi's decomposition of IM-test
> 
> ---------------------------------------------------
>               Source |       chi2     df      p
> ---------------------+-----------------------------
>   Heteroskedasticity |       5.89      2    0.0526
>             Skewness |       4.92      1    0.0266
>             Kurtosis |       1.70      1    0.1919
> ---------------------+-----------------------------
>                Total |      12.51      4    0.0139
> ---------------------------------------------------
> 
> > Is the 
> > -ivhettest- a combined heteroskedasticity test that is 
> > computed from individual test statistics of the first-stage 
> > and reduced form regression in the case of 2SLS?
> >  
> > Any help would be greatly appreciated!
> >  
> > Andreas Zweifel
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
> > 
> 
> 
> -- 
> Heriot-Watt University is a Scottish charity
> registered under charity number SC000278.
> 
> Heriot-Watt University is the Sunday Times
> Scottish University of the Year 2011-2012
> 
> 
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 


-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index