Testing for Anticipation Effects in Difference-in-Differences

Summary

The document asks how to interpret a test for anticipation effects in a Difference-in-Differences study. It distinguishes this concern from the parallel-trends assumption, which the author describes as commonly assessed by jointly testing whether lead coefficients are zero. The question refers to an assumption in a paper by Borusyak and to an event-time specification in which the event year is shifted earlier.

The author labels estimates for successive relative years and asks whether a joint null test of the pre-event coefficients would show that an anticipation effect exists. No coefficient values, test statistic, p-value, or answer are included, so the document does not establish the result. It is a methodological question about interpreting event-time estimates and selecting the coefficients for a test. The exact test depends on the timing convention and study design, details that the text does not fully describe.

Key ideas

• The document separates the no-anticipation assumption from the parallel-trends assumption in Difference-in-Differences.
• It describes testing parallel trends through a joint test of lead coefficients.
• It asks whether several pre-event coefficients should be tested jointly for anticipation effects.
• No estimates or test results are supplied, so the presence of anticipation cannot be determined from the document.

Tags

# How to explain the " no anticipation effect" testing result in Diff-in-Diff?


# How to explain the " no anticipation effect" testing result in Diff-in-Diff?












Regarding Difference-in-Difference, the main assumption is the parallel trend satisfication. Regarding the parallel trend test, just simply prove the joint null test of leads coefficients equalling to zero.However, there is another assumption of concern is "no anticipation effect". I read the paper of Borusyak, 2021 and I saw that he identified it in the Assumption 2 (page 8).

After running the test, I have the result is as below ( in this case I shift the actual event year to 3 years before.

tau0 represents the average estimate for relative year = -3, tau1 relative toyear = -2, ..., tau3 for relative year = 0

I do not know how to explain this result properly about whether the anticipation effect exists? Should we use the joint null test for tau0, tau1, and tau2?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.