author: Koren, Miklós (https://koren.mk) date: 2024-01-23 version: 0.7.0 title: EVENTBASELINE - Correct Event Study After XTHDIDREGRESS description: | eventbaseline transforms the coefficients estimated by xthdidregress into a correct event study relative to a baseline. The reported coefficients are the average treatment effects on the treated (ATT) for each period relative to the baseline. The baseline can be either a period before the treatment or the average of the pre-treatment periods. url: https://github.com/codedthinking/eventbaseline

requires: Stata version 18

eventbaseline Correct event study after xthdidregress

Syntax

• eventbaseline, [pre(#) post(#) baseline(string) graph generate(name)]

eventbaseline transforms the coefficients estimated by xthdidregress into a correct event study relative to a baseline. The reported coefficients are the average treatment effects on the treated (ATT) for each period relative to the baseline. The baseline can be either a period before the treatment or the average of the pre-treatment periods.

The package can be installed with net install eventbaseline, from(https://raw.githubusercontent.com/codedthinking/eventbaseline/main/)

If you require a specific version, use net install eventbaseline, from(https://github.com/korenmiklos/codedthinking/eventbaseline/raw/v0.7.0/)

Options

Options

Option | Description -------|------------ pre | Number of periods before treatment to include in the estimation (default 1) post | Number of periods after treatment to include in the estimation (default 3) baseline | Either a negative number between -pre and -1 or average, or atet. If -k, the baseline is the kth period before the treatment. If average, the baseline is the average of the pre-treatment periods. If atet, the regression table reports the average of the post-treatment periods minus the average of the pre-treatment periods. Default is -1. graph (optional) | Plot the event study graph with the default settings of hetdid_coefplot. generate (optional) | Name of the frame to store the coefficients and their confidence interval.

Background

xthdidregress returns ATET between t and t-1 whenever t is before the treatment. That is, pretrends are reported as first differences, whereas actual treatment effects are reported as difference relative to the period before treatment. This can lead to misleading event study plots (Roth 2024a). The eventbaseline command transforms the coefficients into a correct event study relative to a baseline.

Remarks

The command can only be run after xthdidregress.

The command also returns, as part of e(), the coefficients and standard errors. See ereturn list after running the command. Typical post-estimation commands can be used, such as outreg2 or estout.

The reported number of observations is also corrected to exclude the treated periods outside the reported event window.

If the generate option is used, the returned frame contains the following variables: - time: the time period relative to the baseline - coef: the estimated coefficient - lower: the lower bound of the 95% confidence interval - upper: the upper bound of the 95% confidence interval

The frame is tsset by time, so tsline can be used to plot the event study.

Examples

See example.do and example.log for a full example.

``` . use "df.dta" . replace t = t + 100 . xtset i t . xthdidregress ra (y) (d), group(i) note: variable didcohort, containing cohort indicators formed by treatment variable d and group variable i, was added to the dataset.

. eventbaseline, pre(5) post(5) baseline(-1) graph

Time variable: time, -5 to 5 Delta: 1 unit

Event study relative to -1 Number of obs = 1,850

       y |       ATET   Std. err.      z    P>|z|     [95% conf. interval]

-------------+---------------------------------------------------------------- -5 | -2.31541 .2415591 -9.59 0.000 -2.788857 -1.841963 -4 | -1.310102 .2551332 -5.13 0.000 -1.810153 -.8100498 -3 | -1.256003 .284372 -4.42 0.000 -1.813362 -.6986446 -2 | -.4307123 .2619239 -1.64 0.100 -.9440736 .082649 -1 | 0 (omitted) 0 | .4105212 .2379038 1.73 0.084 -.0557617 .8768041 1 | .9888228 .2764365 3.58 0.000 .4470172 1.530629 2 | 1.859271 .2727555 6.82 0.000 1.32468 2.393862 3 | 1.865648 .2840544 6.57 0.000 1.308911 2.422384 4 | 2.591579 .2831633 9.15 0.000 2.036589 3.146569

5 | 2.923434 .2730864 10.71 0.000 2.388195 3.458674

```

``` . xthdidregress ra (y) (d), group(i) note: variable didcohort, containing cohort indicators formed by treatment variable d and group variable i, was added to the dataset.

. eventbaseline, pre(5) post(5) baseline(atet)

Event study relative to atet Number of obs = 1,850

       y |       ATET   Std. err.      z    P>|z|     [95% conf. interval]

-------------+----------------------------------------------------------------

ATET | 2.835658 .1134013 25.01 0.000 2.613396 3.05792

```

Authors

• Miklós Koren (Central European University, https://koren.mk), maintainer

License and Citation

You are free to use this package under the terms of its license. If you use it, please the software package in your work:

• Koren, Miklós. 2024. "EVENTBASELINE: Correct Event Study After XTHDIDREGRESS. [software]" Available at https://github.com/codedthinking/eventbaseline.

References

• Roth, Jonathan. 2024a. "Interpreting Event-Studies from Recent Difference-in-Differences Methods." Available at https://www.jonathandroth.com/assets/files/HetEventStudies.pdf. Last accessed January 23, 2024.
• Roth, Jonathan. 2024b. "Test Data for >Interpreting Event-Studies from Recent Difference-in-Differences Methods< [data set]." Available at https://github.com/jonathandroth/HetEventStudies/raw/master/output/df.dta Last accessed January 23, 2024.