------------------------------------------------------------------------------- log: C:\Temp\web\class\stata\truncreg.smcl log type: smcl opened on: 11 Apr 2008, 09:23:04 This example uses the Mroz data on labor force participation. First we estimate a linear regression using the observations for which women are actually working for pay. The second regression treats whrs as coming from a truncated normal distribution. The lower truncation point is set to 0 using the ll(0) option. This will cause Stata to estimate the truncated regression using only those observations that have positive whrs. You can see that the coefficients on kids less than 6 and kids between 6 and 18 have changed dramatically in the two regressions. Which should you believe? Well, it depends on your purpose. As the Stata manual states: If you are interested in the mean of wife's working hours conditional on the subsample of market laborers, least squares is appropriate. If you are interested in the mean of wife's working hours regardless of market or nonmarket labor status the usual least squares estimates are quite misleading. In other words, the least squares results shouldn't be applied to women who are not working in the labor force: the truncreg results are probably better. . webuse laborsub, clear . regress whrs kl6 k618 wa we if whrs > 0 Source | SS df MS Number of obs = 150 -------------+------------------------------ F( 4, 145) = 2.80 Model | 7326995.15 4 1831748.79 Prob > F = 0.0281 Residual | 94793104.2 145 653745.546 R-squared = 0.0717 -------------+------------------------------ Adj R-squared = 0.0461 Total | 102120099 149 685369.794 Root MSE = 808.55 ------------------------------------------------------------------------------ whrs | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- kl6 | -421.4822 167.9734 -2.51 0.013 -753.4748 -89.48953 k618 | -104.4571 54.18616 -1.93 0.056 -211.5538 2.639668 wa | -4.784917 9.690502 -0.49 0.622 -23.9378 14.36797 we | 9.353195 31.23793 0.30 0.765 -52.38731 71.0937 _cons | 1629.817 615.1301 2.65 0.009 414.0371 2845.597 ------------------------------------------------------------------------------ . truncreg whrs kl6 k618 wa we, ll(0) (note: 100 obs. truncated) Fitting full model: Iteration 0: log likelihood = -1205.6992 Iteration 1: log likelihood = -1200.9873 Iteration 2: log likelihood = -1200.9159 Iteration 3: log likelihood = -1200.9157 Iteration 4: log likelihood = -1200.9157 Truncated regression Limit: lower = 0 Number of obs = 150 upper = +inf Wald chi2(4) = 10.05 Log likelihood = -1200.9157 Prob > chi2 = 0.0395 ------------------------------------------------------------------------------ whrs | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- kl6 | -803.0042 321.3614 -2.50 0.012 -1432.861 -173.1474 k618 | -172.875 88.72898 -1.95 0.051 -346.7806 1.030578 wa | -8.821123 14.36848 -0.61 0.539 -36.98283 19.34059 we | 16.52873 46.50375 0.36 0.722 -74.61695 107.6744 _cons | 1586.26 912.355 1.74 0.082 -201.9233 3374.442 -------------+---------------------------------------------------------------- /sigma | 983.7262 94.44303 10.42 0.000 798.6213 1168.831 ------------------------------------------------------------------------------ . mfx Marginal effects after truncreg y = Fitted values (predict) = 1051.5263 ------------------------------------------------------------------------------ variable | dy/dx Std. Err. z P>|z| [ 95% C.I. ] X ---------+-------------------------------------------------------------------- kl6 | -803.0042 321.36 -2.50 0.012 -1432.86 -173.147 .173333 k618 | -172.875 88.729 -1.95 0.051 -346.781 1.03058 1.31333 wa | -8.821123 14.368 -0.61 0.539 -36.9828 19.3406 42.7867 we | 16.52873 46.504 0.36 0.722 -74.617 107.674 12.64 ------------------------------------------------------------------------------ . log close log: C:\Temp\web\class\stata\truncreg.smcl log type: smcl closed on: 11 Apr 2008, 09:31:37 -------------------------------------------------------------------------------