Error with staggered using Castle dataset

Question

Error with staggered using Castle dataset

Closed this issue 3 years ago · 4 comments

I'm not sure what this error means, but could use the help.

# Castle Data
castle <- haven::read_dta("https://github.com/scunning1975/mixtape/raw/master/castle.dta")

# replace NA treatment_date with Inf
castle$treatment_date = ifelse(is.na(castle$treatment_date), Inf, castle$treatment_date)

staggered::staggered(
    df = castle,
    y = "l_homicide",
    i = "sid",
    t = "year",
    g = "treatment_date",
    estimand = "simple"
)
#> Error in if (varhat_conservative < 0) {: missing value where TRUE/FALSE needed

^{Created on 2021-07-15 by the reprex package (v2.0.0)}

Answer 1 · 2021-07-16T04:20:55.000Z

Should be fixed now. Issue was that Cohort treated in 2009 had a single unit, and that covariance was "breaking" and returning NaN. Now, I set it to 0 as covariance of a "constant" should be zero.

Answer 2 · 2021-07-16T06:05:33.000Z

I think we should drop observations with N_g = 1, because variance is not estimable. Pretending that var(Y(g)) = 0 doesn't seem like a good solution to me

…

On Thu, Jul 15, 2021, 10:21 PM pedrohcgs ***@***.***> wrote: Should be fixed now. Issue was that Cohort treated in 2009 had a single unit, and that covariance was "breaking" and returning NaN. Now, I set it to 0 as covariance of a "constant" should be zero. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE6EXFFS6Z3SR4EHAIZ4NL3TX6XTFANCNFSM5ANU2HFQ> .

Answer 3 · 2021-07-17T00:12:55.000Z

After talking with @jonathandroth, we converged that we should drop cohorts with a singleton observation. Just pushed this to master.

@kylebutts let us know if this works!

Answer 4 · 2021-07-17T16:46:50.000Z

Works for me!

# Castle Data
castle <- haven::read_dta("https://github.com/scunning1975/mixtape/raw/master/castle.dta")

# replace NA treatment_date with Inf
castle$treatment_date = ifelse(is.na(castle$treatment_date), Inf, castle$treatment_date)

staggered::staggered(
    df = castle,
    y = "l_homicide",
    i = "sid",
    t = "year",
    g = "treatment_date",
    estimand = "simple"
)
#> Warning in staggered::staggered(df = castle, y = "l_homicide", i = "sid", :
#> The treatment cohort g = 2009 has a single cross-sectional unit. We drop this
#> cohort.
#>    estimate         se  se_neyman
#> 1 0.1461919 0.03035112 0.03958199

^{Created on 2021-07-17 by the reprex package (v2.0.0)}