Announcement

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

  • Choosing Instrument Variables in ivregress model

    Hi. I have a model that is theoretically and proven by prior researches has a potential endogeneity issue (corporate governance). Therefore I need to use instrument variables (IV) in running my regression.

    I understand that stata is able in suggesting whether the variables/instruments are:
    - endogenous (estat endog)
    - strong/good (estat firststage)
    - valid (estat overid) where this is for more than one IV available for one endogenous variable

    After testing few variables to be the IV for my regression (there are 5 potential IV that I choose based on theories and my sampling condition), I noticed that there are few combinations of these IV that could give me a good result which fulfil the above 3 criteria.

    For example, my chosen IV are x1, x2, x3, x4 and x5:

    The 1st combination of x1, x3, x4 and x5 = meet all criteria.
    The 2nd combination of x1, x3 and x5 also = meet all criteria
    The 3rd combination of x2 and x5 = meet all criteria
    The 4th combination of x1 and x2 = meet all criteria
    I have also tested each of the IV separately (without any combination) and the variable is exogenous. However, I could not rely on this because the variables in my models are very much prone to endegeneity issue (accordingly to the theory)

    My question are:
    (1) How can I choose the best combination to use in my regression?
    (2) If I can find one IV that could meet all criteria, can I just stop combining and just use that particular IV since the result is stable (stable means any combination of IV will also give the same result, hence I decided to take only one IV instead of a combination of few IV)?
    (3) Is there any 'understanding' that using more IV will give a better model? (I don't think so because our aim is to get a stable model with a stable result - but I just want to clarify on this point)

    Many thanks.

  • #2
    You'll get a more helpful answer if you follow the FAQ on asking questions - provide Stata code in code delimiters, Stata output, and sample data.

    If you have multiple sets of IVs that appear to work, I'd be inclined to check that the results are similar across the four combinations - which is what I would mean by saying the results are stable.

    Comment

    Working...
    X