For my project, which is examining the impact of the Affordable Care
Act and Medicaid expansion on socioeconomic outcomes in the US, I
am relying on data from the US Census Bureau’s American Community
Survey (ACS) series of data, as well as on CDC data. The ACS data
include public health insurance coverage, which includes Medicaid
coverage, and is split out by gender and age group; it also contains data
in separate tables on socioeconomic indicators, such as educational
attainment, median income, percentage of the population below the
poverty line, and for health outcomes data. The Census Bureau makes
it possible to look at data by county, and from 2009-2019, to capture
one year prior to the ACA’s passage, and then to look for statistically
significant changes. An important indicator to append to my dataset
will be whether the state had signed on to Medicaid expansion in a
given year. I am using a binary variable for this because each county
will have 11 records, one for each year under review. I will also be
using data from the CDC on health outcomes, such as death rates, and
rates of conditions such as heart disease and diabetes. For both
Census and CDC data, I need to download the data one year at a time,
and I plan to merge these large datasets using SAS Enterprise Miner or
other tools with more capacity than Excel. Merging the data should be
made easier by the fact that both the CDC and the Census record
county names the same way, and the ID that the CDC uses seems to be
derived from the Census geographic codes.
There are about 3200 counties or county-level jurisdictions (e.g.
parishes in Louisiana) in the US, so each of the datasets described
above will have about 3200 records per year; covering 11 years means
that the final, merged dataset should have 35,000+ records. The public
health insurance dataset contains 12 variables, some of which are
aggregations of the original Census data. Much of the data is also more
narrowly divided that I need it to be—for example, insurance is divided
into 5-6 age groups, but I really only need two: for people 25 and
under (and able to use their parents’ insurance) and over 25. I’ve used
Excel for interim data preparation steps, such as aggregating age group
data. A preliminary data analysis of 2020 health insurance coverage
data reveals that population density varies greatly at the county level.
For example, Los Angeles County, CA had a population of nearly 10
million in 2020, while Loving County, TX, had a population of 117 (see
the bar chart, which shows the counties’ populations in descending
order). Accordingly, the data will need to be normalized, and I plan to
use percentages of the population where ever possible to avoid the
skewing that would come from population variation. Something
interesting in the preliminary data analysis I did using R is that the
mean value for Medicaid expansion status in 2020 is 0.6448. This is
interesting because that variable is binary; this means that about 65
percent of counties are in states with expansion, and 35 percent are
not. However, only 12 states (24 percent of states) have not signed on
to the expansion; this means there are relatively more counties in
those states, than in states with expansion. This is one of the reasons
to examine this dynamic at the county level, rather than at the state
level—a state-level analysis could somewhat overestimate the impact
of expansion.
