Wald test rreg

[Samantha St. Amand]

Hello,

I want to test if coefficients on a given variable estimated in two different sample periods are statistically different. I know I can test this using - suest - if I use OLS, however I am using robust least squares - rreg - (error message: "suest requires that predict allow the score option r(322);").


Does anyone know if there is a way to perform this test from the estimates of rreg, or if there is any other potential work-around?

Thank you,
Sam

[Rich Goldstein]

your question is not completely clear to me; however, at a guess, you want an interaction between your variable of interest and an indicator (dummy) variable for sample period

[Carlo Lazzaro]

Samantha:
elaborating a bit on Rich's helpful advice, you may want to consider something along the following lines:

. sysuse auto.dta
(1978 Automobile Data)


. rreg price c.length##i.foreign

   Huber iteration 1:  maximum difference in weights = .77390259
   Huber iteration 2:  maximum difference in weights = .33008227
   Huber iteration 3:  maximum difference in weights = .1899654
   Huber iteration 4:  maximum difference in weights = .04505219
Biweight iteration 5:  maximum difference in weights = .2645494
Biweight iteration 6:  maximum difference in weights = .25876406
Biweight iteration 7:  maximum difference in weights = .0628047
Biweight iteration 8:  maximum difference in weights = .00809507

Robust regression                               Number of obs     =         74
                                                F(  3,        70) =      29.73
                                                Prob > F          =     0.0000

----------------------------------------------------------------------------------
           price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
          length |   30.88081   8.185924     3.77   0.000      14.5545    47.20712
                 |
         foreign |
        Foreign  |  -16244.41   3548.186    -4.58   0.000    -23321.05   -9167.777
                 |
foreign#c.length |
        Foreign  |   109.1124   20.40389     5.35   0.000     68.41817    149.8067
                 |
           _cons |  -1137.252   1613.747    -0.70   0.483    -4355.768    2081.264
----------------------------------------------------------------------------------

As an aside, please note that -rreg- has seemingly become a bit side-tracked during the last years.

[Nick Cox]

Carlo's gentle and diplomatic last sentence is an important signal. See e.g. http://www.statalist.org/forums/foru...reg-vce-robust and its references for more.

rreg remains part of official Stata but I've not seen any defence of it as a method of choice since 1990 or so.

Positively, there are better methods of robust regression, although I have not checked whether any allows a Wald test. What is often key is that robust regression usually moves so far from mainstream methods that such tests no longer make sense.

search Verardi, author

would give you a way in to more recent commands.