corinne-riddell
commented
7 years ago States where the bayes smoothing model tracks the impute 5 model (and also sits very close to impute 1 and impute 9, implying that it doesnt really matter how you model, at least in terms of LE and LE gap: Alabama, California, Florida, Georgia, Illinois, Louisiana, Michigan, Mississippi, New Jersey, New York, North Carolina, Ohio, South Carolina, Texas --> 71% of the US black population total.
Note: The Bayes model didn't initially run for some states with big black populations, including: Maryland, Massachusetts, Pennsylvania, and Virginia. We think it had something to do with setting up the computing cluster -- KTM will re-run these soon, but I suspect the Bayes smoothing model to match all the imputation models (as above), bringing the total to 85% of the US black population.
List of US States by size of black population.
Things to note:
When does the trends start to differ? -trend is kept but size of gap is smaller in the bayes smoothing. this might be because the bayes smoothing imputes numbers < 5 more often when there is missing which might make sense, especially if the population size to begin with is small in the strata/
-States that underwent a lot of population growth in the black population (Nevada) will still have high missingness for the oldest ages because of the population distribution (less old people).
-In terms of imputation, incorrectly imputating the missing values in the youngest age groups can have the largest affects -- in the Toson report, replacing all 0s with 0.63 of a death can reduced the estimated LE by 1 year in a population of size 20,000, by 2.4 years in size 10,000.