# st: testnl p values are same as those of a coefficient

 From Kristine Watson To statalist@hsphsun2.harvard.edu Subject st: testnl p values are same as those of a coefficient Date Mon, 1 Aug 2005 20:40:40 -0700 (PDT)

```Hello.  I have a quick question. I'm running a fixed
effect regression with an AR(1) process.  I was
surprised by some of my test results.  Testnl gave the
exact same p value for my test as for one of the
coefficients included in the test.  Using xtregar in
Stata8SE, I find the t stat and p value for variable
A. I then run testnl A/(1-B-C).  The p values for A
and the test were the same (to the 3rd decimal point).
Using nlcom, I found they have different coefficient
estimates - so I believe the test is specified
correctly.  This issue isn't pervasive in my testing,
but did arise more than once.  Has anyone ever seen
anything like this?  Is it anything to worry about?
If it matters, two lags of the dependent variable are
in the model.

My code and output is below if that helps.

Many thanks for your time - Kristine

.xtregar percent logsize entire lagperc lag2 maxcg
hq_pi inc_pi, fe

FE (within) regression with AR(1) disturbances  Number
of obs      =     18018
Group variable (i): mgrno                       Number
of groups   =      1199

R-sq:  within  = 0.8117
Obs per group: min = 1
between = 0.9288
avg = 15.0          overall = 0.8367
max =  21
F(8,16811)         =
9061.29
corr(u_i, Xb)  = 0.8229
Prob > F           =    0.0000

------------------------------------------------------------------------------
percent |      Coef.   Std. Err.      t    P>|t|

-------------+----------------------------------------------------------------
sz|   .0139895   .0012666    11.04   0.000
ent|   1.046695   .0636388    16.45   0.000
lag1|     .07843   .0068122    11.51   0.000
lag2|    .361661   .0067388    53.67   0.000

max|   -2.79592   .2133297   -13.11   0.000
A|  -.0030597   .0070038    -0.44   0.662
B|  -.0195817   .0049677    -3.94   0.000
two|  -.0037839   .0042401    -0.89   0.372
_cons |   .0130677   .0013891     9.41   0.000

.testnl ( _b[A]/[1-_b[lag1]-_b[lag2]] = 0)
(1)  _b[A]/[1-_b[lag1]-_b[lag2]] = 0

F(1, 16811) =        0.19
Prob > F =        0.6622

____________________________________________________