Unavoidable "Some initial race values are NA." error

[kaseyzapatka]

Hey and thanks so much for creating such a wonderful package!

I've done all the checks but am still getting the following the following error. I'm using model = "fBISG" and I've set impute.missing = TRUE, and skip_bad_geos = TRUE. I think the issue stems from that there are 47 more tracts in the dataset I pulled from the US Census using wru' s get_census_data function than in the dataset from which I'm trying to impute race. I would think setting the two aforementioned parameters would solve this but the error only works when I set model = "BISG". Perhaps, I could drop those 47 extra tracts? But I'm having trouble filtering the wru census data since it's a nested list. When I

Error in `predict_race()`:
! Some initial race values are NA.

         If you didn't provide initial values, check the results of calling predict_race() on the voter.file you want me to work on.

         The most likely reason for getting a missing race prediction is having a geolocation that does not match 
 
         locations on the census. If this problem persists, try impute.missing = TRUE or model = fBISG.
Hide Traceback
    ▆
 1. └─wru::predict_race(...)

Unfortunately, I wanted to use the fBISG function. do you have any thoughts on why this might be? (I can't produce a reproducible examples given the data are proprietary). Here is the code I'm running so you can see.

df_predicted_wru <-
  predict_race(
    voter.file = df_processed,
    census.geo = "tract",
    year = "2020",
    model = "fBISG",
    census.data = census_data_wru,
    age = FALSE,
    sex = FALSE,
    impute.missing = TRUE,
    skip_bad_geos = TRUE
  )

Any thoughts or help would be appreciated. Thanks!

[1beb]

Can you try re-running predict race with a race.init (a prior) that is the race distribution of your highest level of geography and let me know if that works for you?

predict_race(
  ...,
  race.init = c(...)
)

[kaseyzapatka]

Yes! But could you clarify how to specify that? I couldn't find any examples.

I've got about 4.8 million observations at the census tract level. How would I specify the initial race prior for each observation? Is it is simple 1-5 per the CRAN documentation for each observation. I assumed it might be the vector of the racial distribution for all 5 racial groups as probabilities and tried this but it didn't seem to work :

predict_race(
  ...,
race.init = c(0.2,0.2,0.2,0.2,0.2),
)