Dropped the following variables: state_name, county_name, ratio_population_pc_physician_flag_low_pop
In 2019, there were some cases where we had a value for this metric, but the data quality is NA. These were almost entirely (all but 2 cases) where the ratio was 0 but population was greater than 2,000, so we were assuming these were “false zeros” and suppressing the values (the ratio values themselves should have also been NA in these cases in 2019). In the initial update of this metric we did not apply this suppression rule, but in this push I did apply it. So, now, if the population is less than 2000 and 0 PCPs are reported, we assign a ratio value of 0. If the population is greater than 2000 and 0 PCPs are reported, we assign a ratio value of NA and assume that these are "false zeros." For the later cases, the quality value is also NA. For the former cases, the sensitivity value that we calculate and use to construct quality is Infinity because of a divide by zero error. I recode these values to equal the maximum defined value of sensitivity which leads all of these observations to receive a low quality value. Note that in 2019, all of these observations (the assumed “true zeros” with 0 PCPs and pop < 2000) had high quality values. I looked at the 2019 code and could not figure out why, but I think it was an error, and may be that the way R deals with infinity values has changed since the last update. I explained this in my code as well.
There were two cases where the ratio value in 2019 was 1589:1 and the quality value is NA - these were counties (Jasper and Newton Counties, Missouri) where CHR did not have a count of primary care physicians for, but did have a ratio. Using the raw AHRF data, we have ratio values and quality values for both of these counties in all years.
Dropped the following variables:
state_name
,county_name
,ratio_population_pc_physician_flag_low_pop
In 2019, there were some cases where we had a value for this metric, but the data quality is
NA
. These were almost entirely (all but 2 cases) where theratio
was 0 but population was greater than 2,000, so we were assuming these were “false zeros” and suppressing the values (theratio
values themselves should have also beenNA
in these cases in 2019). In the initial update of this metric we did not apply this suppression rule, but in this push I did apply it. So, now, if the population is less than 2000 and 0 PCPs are reported, we assign a ratio value of 0. If the population is greater than 2000 and 0 PCPs are reported, we assign a ratio value ofNA
and assume that these are "false zeros." For the later cases, thequality
value is alsoNA
. For the former cases, thesensitivity
value that we calculate and use to constructquality
isInfinity
because of a divide by zero error. I recode these values to equal the maximum defined value ofsensitivity
which leads all of these observations to receive a lowquality
value. Note that in 2019, all of these observations (the assumed “true zeros” with 0 PCPs and pop < 2000) had highquality
values. I looked at the 2019 code and could not figure out why, but I think it was an error, and may be that the way R deals with infinity values has changed since the last update. I explained this in my code as well.There were two cases where the
ratio
value in 2019 was1589:1
and the quality value isNA
- these were counties (Jasper and Newton Counties, Missouri) where CHR did not have a count of primary care physicians for, but did have a ratio. Using the raw AHRF data, we have ratio values and quality values for both of these counties in all years.