psacalc with support for reghdfe

A minor revision to psacalc to add support for reghdfe.

psacalc by Emily Oster is a Stata command (available from SSC: ssc install psacalc which implements the method outlined in Oster (2017) to calculate treatment effects and relative degree of selection under proportional selection of observables and unobservables (see help psacalc after installation for details).

The current version of psacalc (22 Oct 2020) supports linear models estimated using regress, areg, and xtreg; however, the popular reghdfe is not supported. See this Statalist discussion.

In this repo I propose a simple edit to the psacalc.ado file available from SSC which adds support for reghdfe. You can easily see what I have changed from the version on SSC by inspecting the commit history of this repo.

When using areg or reghdfe, my edits also allow the user to treat all covariates as nuisance parameters thus calculating treatment effects and relative degree of selection based only on the fixed effects. For example:

clear all

// make some data
set obs 2000
set seed 101
generate f = rnormal() // the individual fixed effect
generate id = _n
expand 10 // create 10 years of data
bysort id: generate year=_n
generate x1 = rnormal()+2*f
generate x2 = 2*rnormal()+f
generate u = 3*rnormal()+5*f
generate y = x1 + x2 + u

// psacalc2 is allowed after reghdfe
reghdfe y x1 x2, absorb(id)
psacalc2 delta x1

// psacalc2 is also allowed if the only controls are those in `absorb()`
reghdfe y x1, absorb(id)
psacalc2 delta x1

// psacalc2 treating all parameters as nuisance but those in `absorb()`
reghdfe y x1 x2, absorb(id)
psacalc2 delta x1, mcontrol(x2)

All my edits are based on the assumption that the s calculated by areg and reghdfe are comparable researchers users should always verify that any estimation software is doing what they think it is before proceeding. Which is to say, any errors in my edits are my own, but you own any errors of your own.

I welcome any comments or improvements.

Future work:

• Allow psacalc to treat the fixed-effects absorbed by reghdfe as nuisance parameters. psacalc currently allows this for any control, but only for fixed effects estimated with xtreg.

References:

Oster, Emily (2017). "Unobservable Selection and Coefficient Stability: Theory and Validation", Journal of Business Economics and Statistics, DOI: 10.1080/07350015.2016.1227711