Oster 2019

[Maximilian Ehinger]

Dear All,
I am trying to estimate Oster 2019 selection of observables.
I am using


reg ln_tot_conf arm_share Votes_Pc $geo subregion_fe_* if insamp ==1, cluster(adm1_en)
local rmax = 1.3*`e(r2)'
di `rmax'
psacalc delta arm_share, mcontrol(subregion_fe_1- subregion_fe_20) rmax(`rmax') beta(0)


The command works, I am just wondering if I a doing everything right as it should be

My goal is to: For the restricted model we use a specification without any controls except subregion fixed effects. We first consider our baseline specification as the fully controlled model .
=> As in my code I am wondering if I am doing correctly what I described above as I am not sure with the mcontrol.
I am unsure about: I’m unsure about mcontrol(). Since psacalc uses the last regression as Controlled, and Uncontrolled =Is the uncontrolled the Controlled minus mcontrol(), Should then mcontrol() list the variables that are in Full but not in Restricted — i.e., $geo Votes_Pc — not the FE?

[Andrew Musau]

psacalc is from SSC, as you are asked to explain (FAQ Advice #12). Since the command provides the estimated coefficient and the R² statistic for both the controlled and uncontrolled models, you can replicate the regressions and answer your own question. Without any arguments specified within mcontrol(), the uncontrolled model is simply the model with an intercept only. If you want this to be the model with fixed effects only, you need to include the fixed effects within mcontrol().