Announcement

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

  • twowayfeweights: first difference vs fixed effect

    I'm attempting to apply Chaisemartin and D’Haultfoeuille's twowayfeweights command to check the weights on a DiD regression with two-way fixed effects, and wanted to check my intuition.

    When I run:
    Code:
    twowayfeweights Y G T D, type(fds)
    which is for a first difference model, I get the output I'm expecting,


    Under the common trends, treatment monotonicity, and if groups' treatment effect does not change over time,
    beta estimates a weighted sum of 8708 LATEs.
    2912 LATEs receive a positive weight, and 5796 receive a negative weight.
    The sum of the positive weights is equal to 1.234136.
    The sum of the negative weights is equal to -.23413596.
    beta is compatible with a DGP where the average of those LATEs is equal to 0,
    while their standard deviation is equal to .4681197.
    beta is compatible with a DGP where those LATEs all are of a different sign than beta,
    while their standard deviation is equal to 2.4338346.
    But as I have more than 2 time periods, and a large number of groups (about 2200) relative to time periods (4) I'm using a fixed effects regression instead of first differences. But when I run
    Code:
    twowayfeweights Y G T D, type(fes)
    which is the fixed effects model, I get


    Under the common trends, treatment monotonicity, and if groups' treatment effect does not change over time,
    beta estimates a weighted sum of 0 LATEs.
    0 LATEs receive a positive weight, and 0 receive a negative weight.
    The sum of the positive weights is equal to ..
    The sum of the negative weights is equal to ..
    beta is compatible with a DGP where the average of those LATEs is equal to 0,
    while their standard deviation is equal to ..
    as my output. Since the coefficients would be the same under FE and FD, am I correct in thinking that I would get the same results for the LATE weights, and thus don't need to worry that I'm only getting weights for one of them? If not, does anyone know why I'm getting this weird output for the FE model? Thanks!

  • #2
    I haven’t tried this command. But I do know with T > 2 FE and FD are not the same. That doesn’t explain the odd output for FE, though.

    Comment


    • #3
      Originally posted by Jeff Wooldridge View Post
      I haven’t tried this command. But I do know with T > 2 FE and FD are not the same. That doesn’t explain the odd output for FE, though.
      thanks Jeff!

      Comment


      • #4
        Noah: Is your intervention staggered or does it occur at the same time for all units?

        Comment


        • #5
          Originally posted by Jeff Wooldridge View Post
          Noah: Is your intervention staggered or does it occur at the same time for all units?
          it occurs at the same time for all units

          Comment


          • #6
            Then I’d just estimate separate effects in different time periods. How big is T?

            Comment


            • #7
              Originally posted by Jeff Wooldridge View Post
              Then I’d just estimate separate effects in different time periods. How big is T?
              T= 4, with 2 periods before treatment and 2 periods after treatment

              Comment


              • #8
                I believe I figured out what the problem was, and can confirm that the weights are different for FD and FE. Thanks again Jeff!

                Comment

                Working...
                X