st: 2SLS with probit in the first stage

[Renuka Metcalfe <rm18203@yahoo.co.uk>]

Dear Kit

Thanks to Kit and Nick for your help. I have used
-xtivreg2- as Kit kindly suggested. I would be
grateful, if you would let me know if the training
variable is endogenous. I had also tried -treatreg-,
the partial results are below. I also tried -xtivreg2-
with 
-xtivreg2 pay x1 x2..., fe small 

but the results were mostly the same.

.xtivreg2 pay  ysed  (b4a2 = m2hat) j1a1 k1a2-k1a5
k2a1-k2a5 k4a1 dis2 s2a2 s2a3 s4a2 k3a1 k3a3 k3a4 d1a1
z2-z6  rk2 ns1-ns4 ns6-ns12 fn1-fn6  fna1-fna6 rk2
aha1  ka2-ka4 in1 ca1, fe
Warning - singleton groups detected.  45
observation(s) not used.
Warning - collinearities detected
Vars dropped:  various explanatory variables

FIXED EFFECTS ESTIMATION
------------------------
Number of groups =      1303                    Obs
per group: min =         2
                                                      
        avg =      11.1
                                                      
        max =        25

IV (2SLS) estimation
--------------------

Estimates efficient for homoskedasticity only
Statistics consistent for homoskedasticity only

                                                     
Number of obs =    14434
                                                     
F( 21, 13110) =    85.16
                                                     
Prob > F      =   0.0000
Total (centered) SS     =  3095.893194               
Centered R2   =   0.0979
Total (uncentered) SS   =  3095.893194               
Uncentered R2 =   0.0979
Residual SS             =  2792.790724               
Root MSE      =    .4612

The coeff. etc. of training is:
Training -.114 std.error .38 t-ratio -0.34
                
Underidentification test (Anderson canon. corr. LM
statistic):          10.456
                                                  
Chi-sq(1) P-val =    0.0012
------------------------------------------------------
Weak identification test (Cragg-Donald Wald F
statistic):               10.447
Stock-Yogo weak ID test critical values: 10% maximal
IV size             16.38
                                         15% maximal
IV size              8.96
                                         20% maximal
IV size              6.66
                                         25% maximal
IV size              5.53
Source: Stock-Yogo (2005).  Reproduced by permission.
------------------------------------------------------------------------------
Sargan statistic (overidentification test of all
instruments):           0.000
                                                
(equation exactly identified)
------------------------------------------------------------------------------
Instrumented:         training Included instruments:
x3 x4 etc.
Excluded instruments: m2hat
Dropped collinear:    x1 x2 etc.

 treatreg pay  x1 x2 x3 x4, treat(b4a2 = job)

Iteration 0:   log likelihood = -19767.494  
Iteration 1:   log likelihood =  -19766.92  
Iteration 2:   log likelihood = -19766.868  
Iteration 3:   log likelihood = -19766.868  

Treatment-effects model -- MLE                   
Number of obs   =      14479
                                                  Wald
chi2(57)   =    5742.51
Log likelihood = -19766.868                       Prob
> chi2     =     0.0000

b4a2         |coeff.               t-ratio
     job|   .2000315   .0281674     7.10   
 _cons |   .3459879   .0117911    29.34   ------------
/athrho |  -.0313172   .0878785    -0.36 
/lnsigma |  -.6999465   .0061073  -114.61           
rho |   -.031307   .0877924      -.2007904    
sigma|   .4966119    .003033                         
lambda |  -.0155474   .0436247                     
LR test of indep. eqns. (rho = 0):   chi2(1) =    0.11
  Prob > chi2 = 0.7383
------------------------------------------------------

I would be grateful if anyone would let me know if
-treatreg- is appropriate in my case. Where I have a
cross-section data and the data are grouped across
workplaces so I used a random effects GLS in my
original regression. One of my independent variable
training is potentially endogenous. It is a binary
variable. dummy = 1 if they have had training and 0
otherwise. 

Thank you




      __________________________________________________________
Sent from Yahoo! Mail.
A Smarter Inbox. http://uk.docs.yahoo.com/nowyoucan.html
*
*   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/