pre

SAS Output: PROC SURVEYREG Results, All Covariates

                            The SURVEYREG Procedure

                Regression Analysis for Dependent Variable re78

                                  Data Summary

                      Number of Observations           614
                      Sum of Weights             329.58393
                      Weighted Mean of re78         6027.8
                      Weighted Sum of re78       1986665.6


                                 Fit Statistics

                           R-square           0.05578
                           Root MSE           6917.62
                           Denominator DF         613


                             Tests of Model Effects

                    Effect       Num DF    F Value    Pr > F

                    Model             9       2.49    0.0086
                    Intercept         1       0.32    0.5696
                    treat             1       0.55    0.4601
                    age               1       0.00    0.9572
                    educ              1       8.20    0.0043
                    black             1       0.56    0.4549
                    hispan            1       0.13    0.7232
                    nodegree          1       0.11    0.7441
                    married           1       0.21    0.6490
                    re74              1       0.10    0.7535
                    re75              1       0.64    0.4242

        NOTE: The denominator degrees of freedom for the F tests is 613.

                      Estimated Regression Coefficients

                                      Standard
         Parameter      Estimate         Error    t Value    Pr > |t|

         Intercept    -2458.7982    4321.26255      -0.57      0.5696
         treat          758.4891    1026.06245       0.74      0.4601
         age              3.0047      55.99413       0.05      0.9572
         educ           748.8193     261.55961       2.86      0.0043
         black         -762.6813    1019.88558      -0.75      0.4549
         hispan         610.6271    1723.24917       0.35      0.7232
         nodegree       535.0055    1638.20712       0.33      0.7441
         married        491.7587    1079.99447       0.46      0.6490
         re74             0.0570       0.18142       0.31      0.7535
         

Estimating the program effect using linear regression

Users may be wondering whether using twang and weighting to adjust for differences between groups yields different results than the more familiar regression approaches to adjusting for group differences on observed covariates. We now compare our weighted estimates of the program effect to results from a more traditional analysis in which the program effect is estimated by a linear model with a treatment indicator and linear terms for each of the covariates. PROC REG is the standard procedure for fitting such models in SAS. Recall that "sasin" is the for the folder where the Lalonde dataset set is stored.