I have several issues and questions concerning the use of the Heckman correction.
First things first, I am trying to estimate a labor supply curve using SOEP Data from the DIW Berlin.
My first step was to estimate the wage using variables for experience and all the others that I will later use in my Heckman command.
Code:
reg NET_INCOME logjobbtenure EXPERIENCE_FULLTIME EXPERIENCE_PARTTIME YEARS_EDUCATION MATERNITY_LEAVE male age agesq west whitecollardummy intdivmonth NETINCOTHERS rentleasmonth unemploymentbenefitmonth DEGREE_HANDICAP predict wageestimate
Is this an issue for the results that come out of this? Typically observability would depend on being employed which it does not strictly in this case.
For now I moved on just to see whether the command will work with the following code:
Code:
. heckman laborsupply lohndach intdivmonth NETINCOTHERS rentleasmonth age agesq youngchildren, > select(D_EMPLOYMENTSTAT= lohndach intdivmonth NETINCOTHERS rentleasmonth unemploymentbenefit > month age agesq DEGREE_HANDICAP youngchildren) twostep note: two-step estimate of rho = 1.8198941 is being truncated to 1 Heckman selection model -- two-step estimates Number of obs = 48,293 (regression model with sample selection) Selected = 47,449 Nonselected = 844 Wald chi2(7) = 3366.08 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------------------ | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------------------+---------------------------------------------------------------- laborsupply | wageestimate | .313529 .006188 50.67 0.000 .3014007 .3256574 intdivmonth | -.0011759 .0037377 -0.31 0.753 -.0085016 .0061498 NETINCOTHERS | -.0371831 .0023763 -15.65 0.000 -.0418406 -.0325256 rentleasmonth | -.0167795 .0060393 -2.78 0.005 -.0286163 -.0049428 age | 9.35933 3.492967 2.68 0.007 2.513241 16.20542 agesq | -.1707992 .0394581 -4.33 0.000 -.2481357 -.0934626 youngchildren | -169.0883 14.80611 -11.42 0.000 -198.1078 -140.0689 _cons | 943.5435 78.96384 11.95 0.000 788.7772 1098.31 -------------------------+---------------------------------------------------------------- D_EMPLOYMENTSTAT | wageestimate | .0004075 .0000221 18.46 0.000 .0003642 .0004507 intdivmonth | -2.58e-06 .0000116 -0.22 0.823 -.0000253 .0000201 NETINCOTHERS | -1.66e-06 4.72e-06 -0.35 0.725 -.0000109 7.59e-06 rentleasmonth | -.0000624 .0000157 -3.97 0.000 -.0000933 -.0000316 unemploymentbenefitmonth | -.0015518 .000115 -13.50 0.000 -.0017771 -.0013265 age | .143686 .0074721 19.23 0.000 .129041 .158331 agesq | -.0015521 .0000881 -17.61 0.000 -.0017248 -.0013793 DEGREE_HANDICAP | -.0014723 .0011021 -1.34 0.182 -.0036323 .0006877 youngchildren | -.4121195 .0430961 -9.56 0.000 -.4965863 -.3276526 _cons | -1.367453 .1416507 -9.65 0.000 -1.645083 -1.089823 -------------------------+---------------------------------------------------------------- /mills | lambda | 898.6195 115.8124 7.76 0.000 671.6313 1125.608 -------------------------+---------------------------------------------------------------- rho | 1.00000 sigma | 898.61947 ------------------------------------------------------------------------------------------
My Rho, which as I understand is the correlation between the two error terms is 1 which just does not seem right, but I could not find any clues pertaining to why this might be the case.
I have seen that often researchers estimate these things for subgroups for example women, but I couldn't figure out why.
Hope you can help me and my description of the data is sufficient.
Kindest regards,
R. Gerlitzky
0 Response to Heckman Correction Rho=1 and some other issues
Post a Comment