Event Study advice

[Kate Ritchie]

I am using a difference-in-difference method and have been advised to use an event study and coefficient plot to visually analyse the parallel trends assumption. My panel spans from 2000-2023 and the treatment year is 2014. I have used 2013 as the base year that is 'dropped' when estimating coefficients for each year as this is the final pre-treatment year, therefore when creating the timetoevent variable 2013 = 0 (i.e., 2014 = 1 and 2000 = -13). When writing my code and specifying the number of lags and leads I used lag(10) and lead(13) and STATA produced output for all of these. I am of the understanding that lead1 is the dropped year (when time to event = 0) and lag0 is therefore my treatment year 2014. However, when assigning years to their corresponding lag or lead there are too many post-treatment lags and too few pre-treatment leads for the panel - am I misinterpreting which lead/lag is which (if i assigned from the starting point lag0 = 2014, lag10 = 2024 which is outside my dataset but STATA still produces the output)? For the pre-treatment number of leads to make sense it would mean lead13 is 2000 and lead2 is 2011, does this mean there is not an coefficient estimate for 2012 or 2013, or have I misaligned these years as there are too many post-treatment lags? I have been gone over this so many times and cannot make sense of it, any advice as to where I have gone wrong here would be greatly appreciated.

[Andrew Musau]

With your setup, you have:

clear
set obs `=2023-2000+1'
gen year= 1999+_n
gen eventyear= 2014
gen timetoevent= year-eventyear
gen description= cond(timetoevent<0, "lead" + string(abs(timetoevent)), "lag"+string(timetoevent))

Res.:

. l, sep(0)

     +---------------------------------------+
     | year   eventy~r   timeto~t   descri~n |
     |---------------------------------------|
  1. | 2000       2014        -14     lead14 |
  2. | 2001       2014        -13     lead13 |
  3. | 2002       2014        -12     lead12 |
  4. | 2003       2014        -11     lead11 |
  5. | 2004       2014        -10     lead10 |
  6. | 2005       2014         -9      lead9 |
  7. | 2006       2014         -8      lead8 |
  8. | 2007       2014         -7      lead7 |
  9. | 2008       2014         -6      lead6 |
 10. | 2009       2014         -5      lead5 |
 11. | 2010       2014         -4      lead4 |
 12. | 2011       2014         -3      lead3 |
 13. | 2012       2014         -2      lead2 |
 14. | 2013       2014         -1      lead1 |
 15. | 2014       2014          0       lag0 |
 16. | 2015       2014          1       lag1 |
 17. | 2016       2014          2       lag2 |
 18. | 2017       2014          3       lag3 |
 19. | 2018       2014          4       lag4 |
 20. | 2019       2014          5       lag5 |
 21. | 2020       2014          6       lag6 |
 22. | 2021       2014          7       lag7 |
 23. | 2022       2014          8       lag8 |
 24. | 2023       2014          9       lag9 |
     +---------------------------------------+

So you would omit 'lead1', or time to event= -1, which is the year immediately preceding the event year when doing the estimation.

[Kate Ritchie]

Thank you!