Chapter 5 Placebo Tests: False Timing

Our second major type of placebo test for difference-in-differences estimator is false timing. We can accomplish this by changing our \(t\) to \(t-n\).

5.1 False Timing

For our trade example, our treatment date is the year 2000, but let’s set the date to 1995.

gen post2 = .
replace post2 = 0 if year < 1995
replace post2 = 1 if year >= 1995

gen treat_post2=treatment*post2

We should expect no statistically significant difference between treatment and comparison for our placebo test.

xtdidregress (trade) (treat_post2), group(countrynum) time(year)

Treatment and time information

Time variable: year
Control:       treat_post2 = 0
Treatment:     treat_post2 = 1
-----------------------------------
             |   Control  Treatment
-------------+---------------------
Group        |
  countrynum |       132         21
-------------+---------------------
Time         |
     Minimum |      1990       1995
     Maximum |      2006       2000
-----------------------------------

Difference-in-differences regression                     Number of obs = 2,814
Data type: Longitudinal

                           (Std. err. adjusted for 153 clusters in countrynum)
------------------------------------------------------------------------------
             |               Robust
       trade | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
ATET         |
 treat_post2 |
   (1 vs 0)  |     253834   83426.64     3.04   0.003     89008.47    418659.5
------------------------------------------------------------------------------
Note: ATET estimate adjusted for panel effects and time effects.
Note: Treatment occurs at different times.

We reject the null hypothesis that the parameter is equal to zero, so our placebo test fails. There is something going on with these countries that may increase trade relative to other countries besides the trade policy change.