/xtevent

Stata package -xtevent-

Primary LanguageStataMIT LicenseMIT

xtevent

GitHub release (latest by date)

Description

xtevent is a Stata package to estimate linear panel event-study models. It includes three commands: xtevent for estimation; xteventplot to create event-study plots and; xteventtest for post-estimation hypotheses testing.

  • Last version: 3.0.0 (23feb2024)
  • Current SSC version: 2.2.0 (15mar2023)

Updates

  • Version 3.0.0 (23feb2024):

    • Increase default replications for sup-t confidence intervals: #153
    • Change the name of the option to omit the label for the value of the dependent variable from nominus1label to nonormlabel: #152
    • Fixed bugs present in version 2.2.0: #155, #159, #166
    • Clarifications in the documentation: #152, #161, #163
    • See here for the complete update list.

  • Version 2.2.0 (15mar2023):

    • Add cohort and control_cohort to obtain estimates using Sun and Abraham's (2021) method.
    • Add repeatedcsoption and get_unit_time_effectscommand to estimate event-studies in repeated cross-section settings.
    • Add noestimate option to savek() to generate event-time dummies without estimating the regression model.
    • Enable clustered and robust standard errors for IV estimation without reghdfe.
    • Fixed bugs present in version 2.1.1.
    • See here for the complete update list.

  • Version 2.1.1 (12aug2022):

    • Fixed bugs present in version 2.1.0.
    • Updates in the help files and other documentation.
    • See here for the complete update list.

  • Version 2.1.0 (1aug2022):

    • Adds diffavg option to xtevent to obtain the difference between the average post-event and pre-event coefficient estimates.
    • Adds textboxoption option to xteventplot to specify characteristics for displaying the p-values of the pre-trend and leveling-off tests.
    • Fixed bugs present in version 2.0.0
    • See here for the complete update list.

  • Version 2.0.0 (24jun2022):

    • To produce equivalent results as with xtevent 1.0.0, where the default was to impute the endpoints, the user should use impute(stag). The impute option imputes missing values in the policyvar following different rules. For instance, specifying impute(stag) indicates the program to check before imputing if the policyvar follows staggered adoption. For a detailed explanation of the impute option, see this detailed example.
    • The option nonstaggered has been depreciated. The default option is now not to impute missing values or endpoints. You should now choose any of the imputation rules in the impute option. To get results using imputation consistent with staggered adoption, as in version 1.0.0 you should use impute(stag).
    • Now the option trend allows for trend adjustment by either OLS or GMM.
    • Fixed several bugs present in version 1.0.0.
    • See here for the complete update list.

Installation

To install version 2.2.0 from SSC:

ssc install xtevent

To update from an older version:

adoupdate xtevent, update

To install the last version in this repository, use the github command:

First, install the github command:

net install github, from("https://haghish.github.io/github/")

Then execute:

cap github uninstall xtevent
github install JMSLab/xtevent

The github command will also install all the necessary dependencies.

If you have an older version and want to update:

github update xtevent

To install using net:

cap ado uninstall xtevent
net install xtevent, from("https://raw.githubusercontent.com/JMSLab/xtevent/master")

To get started

help xtevent

Examples

Using xtevent 2.2.0

xtevent

*** setup
webuse nlswork, clear
* year variable has many missing observations
* Create a time variable that ignores the gaps
by idcode (year): gen time=_n
xtset idcode time


*Generate a policy variable that follows staggered adoption
by idcode (time): gen union2=sum(union)
replace union2=1 if union2>1 
order time union union2, after(year)

*** Examples
*Estimate a basic event study with clustered standard errors 
*Impute the policy variable assuming no unobserved changes
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union) w(3) cluster(idcode) impute(nuchange)
	   
*Omit unit and time fixed effects
*Impute the policy variable verifying staggered adoption
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union2) w(3) cluster(idcode) nofe note impute(stag)

* Bring back unit and time fixed effects
*Adjust for a pre-trend by estimating a linear trend by GMM
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
      pol(union2) w(3) cluster(idcode) trend(-2, method(gmm)) ///
	    impute(stag)
			
*Freyaldenhoven, Hansen and Shapiro (2019) estimator with proxy variables
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
      pol(union) w(3) vce(cluster idcode) proxy(wks_work) ///
	    impute(stag)
			          
*reghdfe and two-way clustering
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union2) w(3) impute(stag) cluster(idcode year) reghdfe ///
            proxy(wks_work) 
            
*Sun and Abraham (2021) Estimator
*Generate the variable that indicates cohort
gen timet=year if union2==1
by idcode: egen time_of_treat=min(timet)
*Generate the variable that indicates the control cohort 
*We use the never-treated units as the control cohort 
gen never_treat=time_of_treat==.
*estimate the event-time coefficients with the Sun-and-Abraham (2021) Estimator
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure, ///
            policyvar(union2) window(3) impute(stag) vce(cluster idcode) ///			
            reghdfe cohort(time_of_treat) control_cohort(never_treat)

xteventplot

*** Setup
webuse nlswork, clear
* year variable has many missing observations
* Create a time variable that ignores the gaps
by idcode (year): gen time=_n
xtset idcode time

*** Examples 
*Basic event study with clustered standard errors 
*Impute the policy variable assuming no unobserved changes
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union) w(3) cluster(idcode) impute(nuchange) 

* Simple plot
xteventplot

*Plot smoothest path in confidence region
xteventplot, smpath(line)

*Freyaldenhoven, Hansen and Shapiro (2019) estimator with proxy variables
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union) w(3) vce(cluster idcode) impute(nuchange) ///
            proxy(wks_work)

*Dependent variable, proxy variable, and overlay plots
xteventplot, y
xteventplot, proxy
xteventplot, overlay(iv)
xteventplot

xteventtest

*** setup
webuse nlswork, clear
xtset idcode year

*** examples 
*Basic event study with clustered standard errors. 
*Impute the policy variable assuming no unobserved changes
xtevent ln_w age c.age#c.age ttl_exp c.ttl_exp#c.ttl_exp tenure , ///
            pol(union) w(3) cluster(idcode) impute(nuchange) 

*Test some coefficients to be equal to 0 jointly
xteventtest, coefs(1 2)

*Test that the sum of all pre-event coefficients is equal to 0
xteventtest, allpre cumul

*Test whether the coefficients before the event follow a linear trend
xteventtest, linpretrend

*Tests that the coefficients for the earliest 2 periods before the event are equal to 0
xteventtest, overidpre(2)

Video tutorial:

Our YouTube channel, Linear Panel Event-Study Design, contains a video series discussing xtevent and the accompanying paper, Visualization, Identification, and Estimation in the Panel Event-Study Design.

Citation

Simon Freyaldenhoven, Christian Hansen, Jorge Pérez Pérez, and Jesse M. Shapiro. "Visualization, Identification, and Estimation in the Linear Panel Event-Study Design." NBER Working Paper No. 29170, August 2021; forthcoming in Advances in Economics and Econometrics: Twelfth World Congress.

Simon Freyaldenhoven, Christian Hansen, Jorge Pérez Pérez, Jesse M. Shapiro, and Constantino Carreto. "Estimation and Visualization in the Linear Panel Event-Study Design." Article to accompany Stata package, October 2023.