Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Binary dependent variable

    Hi everyone,
    I have used binary dependent variable instead of continuous when estimating propensity score matching and I have used Psmatch2 command it will give me coefficient or odds ratio? Now how should I interpret the coefficients?
    If it gives odd ratios then why the coefficient of one of my covariates is -0.27?
    Last edited by Menal khan; 19 Jun 2023, 01:23.

  • #2
    I'm confused. Stata shows you the estimated logit model for the propensity score; it does not show the effects of covariates on the outcome variable. The coefficients reported for the logit model for the treatment variable are the usual log-odds effect. But that has nothing to do with the outcome variable.

    If Y is binary, ATE will give the effect on the probability of seeing a one, that is, P[Y(1) = 1] - P[Y(0) = 1]. Never the odds or log-odds.

    Comment


    • #3
      In my case Y is binary and i have used this command for estimating psm
      Code:
       psmatch2 treatment $xlist, out( cdev) logit
      now my question is if i apply this command how should i interpret the coefficient

      Code:
        Logistic regression                             Number of obs     =     15,704
                                                      LR chi2(9)        =     178.56
                                                      Prob > chi2       =     0.0000
      Log likelihood = -7106.8115                     Pseudo R2         =     0.0124
      
      
      
      -----------------------------------------------------------------------------------
              treatment |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
      ------------------+----------------------------------------------------------------
             l_std |   .2506631   .0463361     5.41   0.000     .1598459    .3414803
             rural |  -.3806632   .0539044    -7.06   0.000    -.4863138   -.2750126
         rec_t    |  -.2731286   .1109228    -2.46   0.014    -.4905333    -.055724
         ch_dis  |  -.0784558   .0909768    -0.86   0.388    -.2567671    .099855
        car_dis  |   .2828038   .1236093     2.29   0.022      .040534    .5250736
        m_edu  |  -.3711353   .0456674    -8.13   0.000    -.4606417   -.2816288
        ch_edu |    .029267   .0462428     0.63   0.527    -.0613673    .1199012
          ch_bo |  -.0460465   .0650594    -0.71   0.479    -.1735606    .0814676
            m_bo |   .0381401   .0549915     0.69   0.488    -.0696411    .1459214
           _cons |  -1.347321   .0673636   -20.00   0.000    -1.479351   -1.215291
      -----------------------------------------------------------------------------------
      ----------------------------------------------------------------------------------------
              Variable     Sample |    Treated     Controls   Difference         S.E.   T-stat
      ----------------------------+-----------------------------------------------------------
             cdev  Unmatched | .827330115   .805318577   .022011539   .008319025     2.65
                              ATT | .827330115    .66542889   .161901225   .096762769     1.67
      1-do these coefficient values represents the estimated association between the covariates and the log odds of the outcome, not the odds ratios??
      2-how do we interpret these coefficients?
      Last edited by Menal khan; 19 Jun 2023, 13:13.

      Comment


      • #4
        The logit results have nothing to do with your outcome, cdev. THis is the model for the treatment variable. Those coefficients are the effects on the log-odds. I've never had a good feel for what they mean. Estimate the model using the logit command and use margins, dydx(*) to get affects on the probability of treatment. This still doesn't say anything abotu the treatment effects. It only shows which variables predict treatment status.

        Comment

        Working...
        X