On Feb 14, 2004, at 2:33 AM, Olena wrote:
I have to do a general (full) White heteroscedasticity test for my
regression
with 9 variables. Number of unique variables (excluding constant) that
should
be included in the regression of squared residuals on the
cross-products of
x's is 39. When I manually run this auxiliary regression, Stata drops
2
variables. Resulting test statistic has Chi_2(37) distribtion. I get
the same
result using whitetst command which I downloaded from the web. This
command
relies on Stata's regress to keep only unique variables in the
auxiliary
regression.
But when I use imtest, white , my test statistic is reported to have
Chi_2(39)
distribution and it's value is slightly different from the one I obtain
manually.
So, my questions are: how does the imtest, white calculate the test
statistic?
And how can I make Stata not to drop variables it finds collinear?
Thank you very much!
I cannot reproduce this behavior. whitetst drops the collinear squares
and cross-products. In this example from -auto-, I have modified
whitetst to report on its _rmcoll; it drops 6 regressors, leaving 14.
ivhettest, as Mark Schaffer noted, will also produce this test. But
imtest,white generates the same test statistic with 14 d.f. I do not
see that imtest is failing to properly prune the regressor list.
Also note the 'whitetst, fitted' option; if you have a very large
number of regressors, the d.f. of the standard White test will be
prohibitively large. The alternate form, described in Wooldridge's
text, uses products of the fitted values as regressors.
. regress
Source | SS df MS Number of obs =
74
-------------+------------------------------ F( 5, 68) =
8.87
Model | 250707635 5 50141527.1 Prob > F =
0.0000
Residual | 384357761 68 5652320.01 R-squared =
0.3948
-------------+------------------------------ Adj R-squared =
0.3503
Total | 635065396 73 8699525.97 Root MSE =
2377.5
------------------------------------------------------------------------
------
price | Coef. Std. Err. t P>|t| [95% Conf.
Interval]
-------------
+----------------------------------------------------------------
mpg | -1002.389 461.152 -2.17 0.033 -1922.604
-82.1747
mpg2 | 20.10801 6.289746 3.20 0.002 7.557016
32.659
headroom | 5478.775 3711.48 1.48 0.145 -1927.368
12884.92
head2 | -712.0372 400.143 -1.78 0.080 -1510.51
86.4358
mh | -84.79946 94.36965 -0.90 0.372 -273.1112
103.5122
_cons | 13448.27 9733.056 1.38 0.172 -5973.738
32870.28
------------------------------------------------------------------------
------
. whitetst
note: __00000D dropped due to collinearity
note: __00000E dropped due to collinearity
note: __00000H dropped due to collinearity
note: __00000K dropped due to collinearity
note: __00000N dropped due to collinearity
note: __00000S dropped due to collinearity
White's general test statistic : 9.428595 Chi-sq(14) P-value = .8027
. whitetst,fitted
White's special test statistic : 3.106367 Chi-sq( 2) P-value = .2116
. ivhettest
OLS heteroskedasticity test(s) using levels and cross products of all
IVs
Ho: Disturbance is homoskedastic
White/Koenker nR2 test statistic : 9.429 Chi-sq(14) P-value =
0.8027
. imtest,white
White's test for Ho: homoskedasticity
against Ha: unrestricted heteroskedasticity
chi2(14) = 9.43
Prob > chi2 = 0.8027
Kit
*
* 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/
