On Saturday, June 28, 2003, at 02:33 AM, Mark wrote:
Does anybody have any ideas on how to interpret this? Are the robust
SEs usable anyway? Is the robust var-cov matrix still usable?
FWIW, I enclose the RATS equivalent, which does not result in anything
strange in the VCV with the dummy. (Notice the drastic shrinkage of the
standard error of the dummy with the robust option--also note that the
reported VCE is a combination covariance and correlation matrix, which
explains its asymmetry). Strangely, the robust SEs do not agree even in
the absence of the dummy (i.e. if you just regress weight on length in
Stata, you do not get these SEs). Of course, we cannot take for granted
that either package is right or wrong, but I could swear that in the
past they have agreed on the robust errors to a simple regression. I'm
not sure why they differ now (I have recently updated RATS, so it is
possible that they have made some changes as well).
UNIX RATS 5.03. Run on Jun 28 2003
(c) 1992-2003 Thomas A. Doan. All rights reserved
* RATS PROGRAM BEGINS
alloc 74
open data testr.raw
data(unit=data,org=obs) / $
weight length
set dum = (t.eq.1)
tab
Series Obs Mean Std Error Minimum
Maximum
WEIGHT 74 3019.45945946 777.19356714 1760.00000000
4840.00000000
LENGTH 74 187.93243243 22.26633990 142.00000000
233.00000000
DUM 74 0.01351351 0.11624764 0.00000000
1.00000000
linreg(vcv) weight
# length dum constant
Linear Regression - Estimation by Least Squares
Dependent Variable WEIGHT
Usable Observations 74 Degrees of Freedom 71
Centered R**2 0.894947 R Bar **2 0.891988
Uncentered R**2 0.993555 T x R**2 73.523
Mean of Dependent Variable 3019.4594595
Std Error of Dependent Variable 777.1935671
Standard Error of Estimate 255.4257563
Sum of Squared Residuals 4632204.5048
Regression F(2,71) 302.4262
Significance Level of F 0.00000000
Durbin-Watson Statistic 1.460092
Variable Coeff Std Error T-Stat
Signif
************************************************************************
*******
1. LENGTH 33.018490 1.342694 24.59122
0.00000000
2. DUM -26.004876 257.182747 -0.10111
0.91974466
3. Constant -3185.434217 254.135841 -12.53438
0.00000000
Covariance\Correlation Matrix of Coefficients
LENGTH DUM Constant
LENGTH 1.8028279 0.0102270053 -0.9930568750
DUM 3.5315670 66142.9653394 -0.0238300984
Constant -338.8575583 -1557.5198329 64585.0256699
linreg(robusterrors) weight
# length constant
Linear Regression - Estimation by Least Squares
Dependent Variable WEIGHT
Usable Observations 74 Degrees of Freedom 72
Centered R**2 0.894932 R Bar **2 0.893473
Uncentered R**2 0.993554 T x R**2 73.523
Mean of Dependent Variable 3019.4594595
Std Error of Dependent Variable 777.1935671
Standard Error of Estimate 253.6640262
Sum of Squared Residuals 4632871.5500
Durbin-Watson Statistic 1.469279
Variable Coeff Std Error T-Stat
Signif
************************************************************************
*******
1. LENGTH 33.019878 1.253349 26.34531
0.00000000
2. Constant -3186.046574 237.059123 -13.43988
0.00000000
linreg(vcv,robusterrors) weight
# length dum constant
Linear Regression - Estimation by Least Squares
Dependent Variable WEIGHT
Usable Observations 74 Degrees of Freedom 71
Centered R**2 0.894947 R Bar **2 0.891988
Uncentered R**2 0.993555 T x R**2 73.523
Mean of Dependent Variable 3019.4594595
Std Error of Dependent Variable 777.1935671
Standard Error of Estimate 255.4257563
Sum of Squared Residuals 4632204.5048
Durbin-Watson Statistic 1.460092
Variable Coeff Std Error T-Stat
Signif
************************************************************************
*******
1. LENGTH 33.018490 1.253151 26.34836
0.00000000
2. DUM -26.004876 29.564608 -0.87959
0.37907885
3. Constant -3185.434217 237.135428 -13.43297
0.00000000
Covariance\Correlation Matrix of Coefficients
LENGTH DUM Constant
LENGTH 1.5703885 0.0747329227 -0.9922415126
DUM 2.7687749 874.0660749 -0.1981307449
Constant -294.8610367 -1389.0582112 56233.2110397
Thanks
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/
