Announcement

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

  • Questions regarding etregress command

    Hi all,

    I have a few questions regarding etregress command, since I intend to incorporate that into a potential project in the future.

    To better illustrate, consider this admittedly ridiculous model.

    Suppose I wish to regress student test scores on a binary variable that corresponds to whether or not a student smokes.
    Etreg allows me to specify another relationship for the binary smoking variable (since the probability of a student can be dependent on his/her environment, income level, etc)

    Am I right to specify the equation as follows:
    global X " a whole slew of variables that correspond to Individual characteristics"
    global Z "characteristics that might be related to someone smoking, but not necessarily related to test scores, which is why these variables only enter in the treatment equation"
    global Environment "Environment factors like quality and income level of neighbourhood, etc"

    etreg math_score $X, tr(binary_smoking = $X $Environment $Z)

    If this is how I ought to specify my model, I then have several follow-up questions.

    1) Can I include average test scores as an independent variable in the treatment regression (since I believe probability of smoking is related to how smart you are, potentially) ?

    2) Is this test equivalent to a two-stage least squares or simultaneous equation model (reg3 command) ? I ask this because my coefficients for the binary_smoking is well-behaved under etreg (i.e. plausible coefficient values). I get very weird values under "ivregress 2sls" and "reg3" commands, even though I have a semi-decent instrument. Under the etreg command, I get coefficients of a magnitude of 1-2, which is believable since the dependent variable is standardised test scores, but under ivregress and reg3, I get coefficients in the order of -10 and -20, which isn't plausible.

    3) Lastly, are there any specific econometric/statistics paper that I can be directed to, to further read up on the theoretical underpinnings of this etreg command? The closest thing I've come up with is the Control function approach, popularized by Heckman; but I'm not using the cfunc option (mostly because I can't seem to get it to converge).
    Last edited by Hisham Yacob; 28 Aug 2019, 18:23.

  • #2
    Anyone?

    Comment


    • #3
      Hi Hisham
      So some thoughts on your problem:

      1) Can I include average test scores as an independent variable in the treatment regression (since I believe probability of smoking is related to how smart you are, potentially) ?

      What has the literature done in the past? It seems to me that adding average scores will just generate an endogeneity problem

      2) Is this test equivalent to a two-stage least squares or simultaneous equation model (reg3 command) ? I ask this because my coefficients for the binary_smoking is well-behaved under etreg (i.e. plausible coefficient values). I get very weird values under "ivregress 2sls" and "reg3" commands, even though I have a semi-decent instrument. Under the etreg command, I get coefficients of a magnitude of 1-2, which is believable since the dependent variable is standardised test scores, but under ivregress and reg3, I get coefficients in the order of -10 and -20, which isn't plausible.

      For why you didnt get similar results may well depend on how you are using the commands and the data. It would be more useful if you provide your outputs to see if anything wrong has been done

      3) Lastly, are there any specific econometric/statistics paper that I can be directed to, to further read up on the theoretical underpinnings of this etreg command? The closest thing I've come up with is the Control function approach, popularized by Heckman; but I'm not using the cfunc option (mostly because I can't seem to get it to converge).

      Look into the reference section of the manual of etregress. They have a very comprehensive suggested readings.

      Comment


      • #4
        Originally posted by FernandoRios View Post
        Hi Hisham
        So some thoughts on your problem:

        1) Can I include average test scores as an independent variable in the treatment regression (since I believe probability of smoking is related to how smart you are, potentially) ?

        What has the literature done in the past? It seems to me that adding average scores will just generate an endogeneity problem

        2) Is this test equivalent to a two-stage least squares or simultaneous equation model (reg3 command) ? I ask this because my coefficients for the binary_smoking is well-behaved under etreg (i.e. plausible coefficient values). I get very weird values under "ivregress 2sls" and "reg3" commands, even though I have a semi-decent instrument. Under the etreg command, I get coefficients of a magnitude of 1-2, which is believable since the dependent variable is standardised test scores, but under ivregress and reg3, I get coefficients in the order of -10 and -20, which isn't plausible.

        For why you didnt get similar results may well depend on how you are using the commands and the data. It would be more useful if you provide your outputs to see if anything wrong has been done

        3) Lastly, are there any specific econometric/statistics paper that I can be directed to, to further read up on the theoretical underpinnings of this etreg command? The closest thing I've come up with is the Control function approach, popularized by Heckman; but I'm not using the cfunc option (mostly because I can't seem to get it to converge).

        Look into the reference section of the manual of etregress. They have a very comprehensive suggested readings.


        Thank you so much. I've been Google-ing for the past couple days. Didn't think to scroll all the way down to the end of the help file for etregress (since I usually skip the examples portion).

        Comment

        Working...
        X