下载diff命令
https://mpra.ub.uni-muenchen.de/43943/1/MPRA_paper_43943.pdf 例子
https://mpra.ub.uni-muenchen.de/43943/1/MPRA_paper_43943.pdf
DIFF: Stata module to perform Differences in Differences estimation

help diff
-----------------------------------------------------------------------------------------------------------

Title

    diff --       Difference in differences estimation


Syntax

        diff outcome_var [if] [in] [weight] ,[ options]


Description

    diff performs several difference in differences (diff-in-diff) treatment effect estimations of a
    given outcome variable from a pooled baseline and follow up dataset: Single Diff-in-Diff,
    Diff-in-Diff accounting for covariates, Kernel Propensity Score Matching diff-in-diff, and the
    Quantile Diff-in-Diff.  diff is also suitable for estimating repeated cross sections diff-in-diff
    (including the kernel option).


Options

    options               Description
    -----------------------------------------------------------------------------------------------------
    Model - Required
      period(varname)     Indicates the binary period variable (0: base line; 1: follow up).
      treated(varname)    Indicates the binary treatment variable (0: controls; 1:treated).

    Optional
      cov(varlist)        Specifies the pre-treatment covariates of the model. When option kernel is
                            selected these variables are used to estimate the propensity score.
      kernel              Performs the Kernel-based Propensity Score Matching diff-in-diff. This option
                            generates the variable _weights containing the weights derived from the
                            Kernel Propensity Score Matching, and _ps when the Propensity Score is not
                            supplied in pscore(varname), following Leuven and Sianesi (2014).  This
                            option requires the id(varname) of each unit or individual except under the
                            repeated cross section rcs) setting.
      id(varname)         Option kernel requires the supply of the identification variable.
      bw(#)               Supplied bandwidth of the Kernel function. The default bandwidth is 0.06.
      ktype(kernel)       Specifies the type of the Kernel function. The types are epanechnikov (the
                            default), gaussian, biweight, uniform and tricube.
      rcs                 Indicates that the kernel is set for repeated cross section. This option does
                            not require option id(varname). Option rcs strongly assumes that covariates
                            in cov(varlist) do not vary over time.
      qdid(quantile)      Performs the Quantile Difference in Differences estimation at the specified
                            quantile from 0.1 to 0.9 (quantile 0.5 performs the QDID at the medeian). You
                            may combine this option with kernel and cov. qdid does not support weights
                            nor robust standard errors.  This option uses [R] qreg and [R] bsqreg for
                            bootstrapped standard errors
      pscore(varname)     Supplied Propensity Score.
      logit               Specifies logit estimation of the Propensity Score. The default is Probit.
      support             Performs diff on the common support of the propensity score given the option
                            kernel.
      addcov(varlist)     Indicates additional covariates in addition to those specified in the
                            estimation of the propensity score.

    SE/Robust
      cluster(varname)    Calculates clustered Std. Errors by varname.
      robust              Calculates robust Std. Errors.
      bs                  performs a Bootstrap estimation of coefficients and standard errors.
      reps(int)           Specifies the number of repetitions when the bs is selected. The default are 50
                            repetitions.

    Balancing test
      test                Performs a balancing t-test of the difference in the means of the covariates
                            between the control and treated groups in period == 0. The option test
                            combined with kernel performs the balancing t-test with the weighted
                            covariates.  See [R] ttest

    Reporting
      report              Displays the inference of the included covariates or the estimation of the
                            Propensity Score when option kernel is specified.
      nostar              Removes the inference stars from the p-values.

    Exporting results
      export(filename)    Exports the output table into the working directory in csv format.  See [R] cd
                            for details.

    -----------------------------------------------------------------------------------------------------


Example

    1. Diff-in-Diff with no covariates.
    We use the dataset form Card & Krueger (1994)*.

         use "http://fmwww.bc.edu/repec/bocode/c/CardKrueger1994.dta"

         diff fte, t(treated) p(t)

    For bootstrapped std. err.:

         diff fte, t(treated) p(t) bs rep(50)

    2. Diff-in-Diff with covariates.

         diff fte, t(treated) p(t) cov(bk kfc roys)
         diff fte, t(treated) p(t) cov(bk kfc roys) report
         diff fte, t(treated) p(t) cov(bk kfc roys) report bs

    3. Kernel Propensity Score Diff-in-Diff.

         diff fte, t(treated) p(t) cov(bk kfc roys) kernel id(id)
         diff fte, t(treated) p(t) cov(bk kfc roys) kernel id(id) support
         diff fte, t(treated) p(t) cov(bk kfc roys) kernel id(id) support addcov(wendys)
         diff fte, t(treated) p(t) kernel id(id) ktype(gaussian) pscore(_ps)
         diff fte, t(treated) p(t) kernel id(id) ktype(gaussian) pscore(_ps) bs reps(50)

    3. Kernel Propensity Score Diff-in-Diff (Repeated Cross Section - rcs).

         diff fte, t(treated) p(t) cov(bk kfc roys) kernel rcs
         diff fte, t(treated) p(t) cov(bk kfc roys) kernel rcs support
         diff fte, t(treated) p(t) cov(bk kfc roys) kernel rcs support addcov(wendys)
         diff fte, t(treated) p(t) kernel rcs ktype(gaussian) pscore(_ps)
         diff fte, t(treated) p(t) cov(bk kfc roys) kernel rcs support addcov(wendys) bs reps(50)

    4. Quantile Diff-in-Diff.

         diff fte, t(treated) p(t) qdid(0.25)
         diff fte, t(treated) p(t) qdid(0.50)
         diff fte, t(treated) p(t) qdid(0.75)
         diff fte, t(treated) p(t) qdid(0.50) cov(bk kfc roys)
         diff fte, t(treated) p(t) qdid(0.50) cov(bk kfc roys) kernel id(id)
         diff fte, t(treated) p(t) qdid(0.50) cov(bk kfc roys) kernel rcs

    5. Balancing test of covariates.

         diff fte, t(treated) p(t) cov(bk kfc roys wendys) test
         diff fte, t(treated) p(t) cov(bk kfc roys wendys) test id(id) kernel
         diff fte, t(treated) p(t) cov(bk kfc roys wendys) test kernel rcs

Saved results

    diff saves the following list of scalars in r():

      r(mean_c0)      mean of output_var of the control group in period == 0
      r(mean_t0)      mean of output_var of the treated group in period == 0
      r(diff0)        difference of the mean of output_var between treated and control groups in period
                       == 0
      r(mean_c1)      mean of output_var of the control group in period == 1
      r(mean_t1)      mean of output_var of the treated group in period == 1
      r(diff1)        difference of the mean of output_var between treated and control groups in period
                       == 1
      r(did)          differences in differences - Treatment Effect
      r(se_c0)        Standard error of the mean of output_var of the control group in period == 0
      r(se_t0)        standard errors of the mean of output_var of the treated group in period == 0
      r(se_d0)        standard Errors of the difference of output_var between the treated and control
                       groups in period == 0
      r(se_c1)        standard errors of the mean of output_var of the control group in period == 1
      r(se_t1)        standard errors of the mean of output_var of the treated group in period == 1
      r(se_d1)        standard errors of the difference of output_var between the treated and control
                       groups in == 0
      r(se_dd)        standard errors of the difference in difference

References

Single diff-in-diff:

    Card, D., Krueger, A. "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New
        Jersey and Pennsylvania".  The American Economic Review, Vol. 84, No. 4 (Sep., 1994), pp.
        772-793.

Kernel diff-in-diff:

    Heckman, J., Ichimura, H., Todd, P. "Matching As an Econometric Evaluation Estimator". The Review of
        Economic Studies, Vol. 65, No. 2 (Apr., 1998), pp. 261-294.

    Leuven, E., Sianesi, B., 2014. "PSMATCH2: Stata module to perform full Mahalanobis and propensity
        score matching, common support graphing, and covariate imbalance testing, Statistical Software
        Components". Boston College Department of Economics.

Kernel diff-in-diff (repeated cross section):

    Blundell, R., Dias, M. "Alternative Approaches to Evaluation in Empirical Microeconomics". Journal of
        Human Resources, Vol. 44, No. 3 (Jun., 2009), pp. 565-640.

Quantile diff-in-diff:

    Meyer, B., Viscusi, W., . "Workers' Compensation and Injury Duration: Evidence from a Natural
        Experiment". The American Economic Review, Vol. 85, No.3 (Jun., 1995), pp. 322-340.

Author

    Juan M. Villa
    Brooks World Poverty Institute
    The University of Manchester
    juan.villalora@manchester.ac.uk

    Please cite as: Villa, Juan M. 2011. DIFF: Stata Module to Perform Differences in Differences
        Estimation. Statistical Software Components. Boston College Department of Economics.

    Update: November - 2014. Click here periodically to get the lastest version.
    *Acknowledgements to Kit Baum for valuable comments. The Kernel matching is based on the command
        psmatch2 developved by Edwin Leuven and Barbara Sianesi.