Announcement

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

  • First-Stage F-Statistics for IV with Control Function Approach

    Hello everyone,

    I am estimating a model using the Poisson Pseudo Maximum Likelihood (PPML) estimator with a control function approach (Testing and Correcting for Endogeneity in Nonlinear Unobserved Effects Models) to address endogeneity. However, due to the specific nature of my model, I cannot directly use the standard IV commands (e.g., ivregress) or user-written IV programs in Stata.

    I would like to check for weak instruments using the first-stage F-statistics offered by weakivtest (from SSC). I can assume a hypothetical linear model for the second stage with a first-stage reduced form that is identical to what I have in my control function approach.

    My question is: Can I use the first-stage F-statistic from a traditional linear IV model (e.g., 2SLS) to assess the strength of my instruments, even though my final estimation is with PPML using the control function approach?

    This is how I implement my IV with control function (I skipped the part when I adjust the standard errors)

    HTML Code:
    xtreg y2 i.city##c.log_price i.city##c.log_price x1 x2 , fe vce(cluster id)
    predict double u2h_fe, e 
    ppmlhdfe y y2 u2h_fe x1 x2, absorb(time id) cluster(id)
    What I have in mind is to implement this IV model to only get F-stats:

    HTML Code:
    ivreghdfe y (y2=i.city##c.log_price i.city##c.log_price) x1 x2 , absorb(months home_id) cluster(home_id)
    weakivtest
    P.S. In the new update, ivreghdfe (from SSC) is compatible with weakivtest, although it takes very long time to run.

  • #2
    The F test is only on the first stage, so that will work.

    Comment

    Working...
    X