0
$\begingroup$

The following model was estimated:

Linear regression                             Number of obs =     10,288
                                            F(45, 10242)      =     118.71
                                            Prob > F          =     0.0000
                                            R-squared         =     0.3259
                                            Root MSE          =     .58052

                    |               Robust
              ljbot |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
 ------------------------+----------------------------------------------------------------
             greens |  -.1290424   .0360989    -3.57   0.000    -.1998033   -.0582815
             libdem |  -.0211495   .0314561    -0.67   0.501    -.0828096    .0405105
           leftwing |  -.0418696   .0184481    -2.27   0.023    -.0780314   -.0057078
         apolitical |   .0541413   .0161846     3.35   0.001     .0224163    .0858663
               ukip |   .0143972    .037146     0.39   0.698    -.0584162    .0872106
                bnp |   .3696505    .069015     5.36   0.000     .2343675    .5049335

Where ljbot indicates the log of overtime hours worked. The independent variables of interest, greens, libdem, leftwing, apolitical, ukip and bnp, are all dummy variables derived from a categorical variable indicating a respondent's political party support. Support for the conservative party acts as the reference group. (A number of controls were also included in the model).

I have found a suitable instrumental variable 'mfgirl' (a dummy variable where 1 = has female children, and 0 = has male children only).

I estimated

ivregress 2sls ljbot greens libdem (leftwing = mfgirl) apolitical ukip bnp, vce(robust)

(and included all controls)

Instrumental variables (2SLS) regression     Number of obs   =      8,712
                                              Wald chi2(45)   =    4198.35
                                              Prob > chi2     =     0.0000
                                              R-squared       =     0.2495
                                              Root MSE        =     .61392

                    |               Robust
              ljbot |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]  

           leftwing |  -.7397816   .2952977    -2.51   0.012    -1.318554   -.1610086
             greens |  -.5214898   .1758845    -2.96   0.003    -.8662171   -.1767625
             libdem |  -.3840828   .1594964    -2.41   0.016    -.6966899   -.0714757
         apolitical |  -.3148761    .160717    -1.96   0.050    -.6298757    .0001235
               ukip |  -.3700357   .1704412    -2.17   0.030    -.7040943   -.0359772
                bnp |  -.0658923    .186908    -0.35   0.724    -.4322253    .3004406

Is it sufficient to place the instrument on leftwing only?

When I run

ivregress 2sls ljbot (greens libdem leftwing apolitical ukip bnp = mfgirl), vce(robust)

(Including all controls)

I receive the message:

equation not identified; must have at least as many instruments not in the regression as there are instrumented variables r(481);

I have run the 2sls regression multiple times with the instrument placed on a different dummy each time. The Durbin-Wu-Hausman (DWH) test (estat endog) reveals the same exact answer no matter which one dummy I place the instrument on (variables are endogenous).

Please let me know if it is sufficient to place the instrument on one of the dummies only.

Thank you.

$\endgroup$
  • $\begingroup$ not clear to me why you want to use your instrument on only one dummy ? are you interested only in leftwing causing ljbot ? $\endgroup$ – oDDsKooL Jul 21 '17 at 6:44
  • $\begingroup$ You need at least as many instruments as there are endogenous variables, hence the error message. In your case, you would need 6 instruments for the six party votes. $\endgroup$ – Andy Jul 21 '17 at 7:59
  • $\begingroup$ @Andy Thank you for the replies. I thought only one instrument is needed for one variable because the six dummies all stem from the same categorical variable. I'm guessing this isn't the case and I need to control for endogeneity for each dummy. When I try 'ivregress 2sls ljbot (greens libdem leftwing apolitical ukip bnp = mfgirl mfgirl mfgirl mfgirl mfgirl mfgirl), vce(robust)' I still get the message 'equation not identified; must have at least as many instruments not in the regression as there are instrumented variables'. Any advice for moving forward? $\endgroup$ – L.Emma Jul 22 '17 at 8:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.