0

I am interested in analyzing the difference in outcome(total bank deposits) among banks that were in states that implemented bail out policy and banks that were in state that did not implement bail out policy. So the intervention is, bail out policy.

  • total bank deposits (banks in state without bailout policy)
  • total bank deposits (banks in state with bailout policy)

The states that did implement bailout policy , implemented these policies starting from 1992. There were banks in these states since 1980s, prior to implementing the policy.

  • total bank deposits (banks in state prior to bailout policy)

Since my goal is to compare the effect of policy implementation on banks, which group does the bank that resided in state prior to policy implementation belong to ? Can I move these observations (total bank deposits) from banks that were in state prior to bail out policy into the group of banks that never resided in state that implemented the bail out policy ?

  • total bank deposits (banks in state without bailout policy) + total bank deposits (banks in state prior to bailout policy) ??

Curious to know if there there any pitfalls or drawbacks due to moving , total bank deposits (banks in state prior to bailout policy) to the total bank deposits (banks in state without bailout policy) category ?

Ahir Bhairav Orai
  • 37
  • 5
  • If some of the same banks existed before and after the bailout policy, then it seems more efficient to do a paired test Before vs After for those banks, understanding the part of any difference might be due to different economies at different times and part due to change in bailout policy. // Separately, perhaps do a two-sample test on contemporaneous banks in states with and without bailout---now understanding that some of the difference may be due to bailout policies and some due to other differences among states. – BruceET Nov 22 '21 at 03:52

1 Answers1

0

Banks in treated states are always considered treated in DID. This is the case with repeated cross-sections as well as the panel version.

dimitriy
  • 31,081
  • 5
  • 63
  • 138
  • your statement `Banks in treated states are always considered treated in DID` , does this imply that bank in treated states but prior to receiving treatment can be moved to control category ? – Ahir Bhairav Orai Nov 22 '21 at 06:09
  • No, they are always labeled as treated, so cannot be moved to control. – dimitriy Nov 22 '21 at 06:28
  • but some banks in the treatment category did not received treatment until 1992. – Ahir Bhairav Orai Nov 22 '21 at 06:53
  • DID calculates the change for ever-treated banks over time minus the change for never-treated banks over time. If you move the treated state banks to the untreated group in the pre-treatment period, you will have no way to calculate that first difference. – dimitriy Nov 22 '21 at 07:29
  • This removes any preexisting differences between the two types of banks that do not change over time, isolating the bailout effect. – dimitriy Nov 22 '21 at 07:34
  • If you click on the DID tag under your question, you can confirm what I wrote. – dimitriy Nov 22 '21 at 07:38
  • so is my understanding correct that, we need to compare the slopes of in `pre treatment group` with control group and then `post treatment group` with control group ? – Ahir Bhairav Orai Nov 22 '21 at 08:46
  • See DD here (ignore DDD) for [intuition](https://stats.stackexchange.com/a/341653/7071) and [here for the regression implementation](https://stats.stackexchange.com/a/549496/7071). – dimitriy Nov 22 '21 at 15:03
  • @dimitry, after going through that example, I like to point out few things 1) If I restrict the intuition exercise to comparison between elders (65 and older) in Florida vs elder (65 and older) in state other than Florida the DD process makes sense. 2) The second part considering younger than 65year old creates confusion. It is fundamentally wrong to consider younger generation as controls when the test subjects are clearly 65 and older. 3) How to deal with controls some of who may go on to become cases is not clear from that intuition example. – Ahir Bhairav Orai Nov 22 '21 at 22:51
  • I am not sure if I can say it any other way: the variable is ever_treated, not treated in this period. It is constant across time within a bank. So banks are either treated or not, but not both. There is no transition. – dimitriy Nov 23 '21 at 18:53