It’s True for Counties, Too

The strongest argument against my contention in my last post that deeper reductions in uninsured rates, particularly among young white people, in ACA states fueled faster increases in opioid overdose rates is that this is a  regional phenomenon. Northeastern states are experiencing the most quickly worsening opioid crisis, you could argue, and they happen also to be more liberal, but the correlation with state policy is just a coincidence.


To see if this is just an artifact of regional differences, I took county-level mortality data from CDC Wonder and county-level data on change in uninsured rates from The CDC censors a lot of the data (presumably because the cells are too small and it represents some kind of disclosure risk), but still has drug-related mortality rates for over 500 of the 3000 counties for 2015, about 300 of which are in ACA/Medicaid expansion states and 200 in the holdouts. If the association I’m positing is real, there should be an association within states as well as between them. There is, at least in the expansion states.


Here’s a within-state (ie, state-level fixed effects) regression of the relationship between the decrease in uninsured rate in ACA Expansion states and the 2015 overdose rate. Within states, counties with larger decreases in the uninsured rate also had larger increases in the overdose rate, even when controlling for the 2013 overdose rate. A county with a one percentage point higher decrease in the uninsured rate had a crude death rate from drug-related causes in 2015 of about 2 per 100,000 higher, relative to other counties in the same state. When controlling for 2013 overdose rates, a percentage point higher decrease in the uninsured rate had a crude drug-related death rate in 2015 of 0.4 per 100,000 higher. My guess is these relationships would be stronger if we were looking at white uninsured rates and white death rates only, and if we could filter by age.

(1) (2) (3)
Within-State Regression in Expansion States [includes state-level fixed effects] 2015 Overdose Rate 2015 Overdose Rate 2015 Overdose Rate
Change in uninsured Rate from 2013 to 2016 -1.949*** -0.392**
(0.286) (0.171)
2013 Overdose Rate 1.004*** 0.974***
(0.0394) (0.0412)
Constant 7.784*** 4.633*** 1.703
(2.637) (0.841) (1.527)
Observations 343 290 290
R-squared 0.569 0.875 0.877

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1


It’s possible there’s an economic confound here, for example if poorer counties had bigger jumps in insurance rates, that is, and that poverty led to bigger jumps in overdose rates, even when controlling for 2013. Including detailed demographic data by county would help.


On the other hand, here is the unweighted county-level change in drug-related death rates from 2010 to 2015 in states with full ACA implementation and Medicaid holdout states. It seems to me there is something going on.


3 thoughts on “It’s True for Counties, Too

  1. Just a couple of questions:

    Would it be better to use change in overdose rates rather than rates in one year as you did with the states data?
    It looks like 5-6 of these counties out the 500 (?) are not like the others. What does the curve look like if these are removed? How large are these counties? That is, rates of 75-150 per 100,000 are high rate, but how many deaths do these account for?
    What proportion of total deaths are these? What is the trend in total mortality in the expansion and non-expansion states? That is, did Medicaid save more lives than it cost?

    Liked by 1 person

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s