When I first read Anne Case and Angus Deaton’s 2015 article about rising morbidity and mortality among middle-aged non-Hispanic white Americans, I had a fairly simple mental model that I thought explained a good deal of what they were observing:
a) Americans had experienced an unprecedented rise in obesity and physical inactivity.
b) White Americans might be particularly susceptible to chronic pain and depression as a result of that obesity and inactivity, perhaps with current or former smoking exaggerating these outcomes for lower income whites.
c) Middle-aged whites responded to the chronic pain and depression by self-medicating with alcohol and by obtaining prescriptions for opiate painkillers.
d) More isolated and likely to live alone and in less contact with their neighbors than previous generations, addiction and alcoholism could more easily spin out of control, with rising suicide rates concomitant:
Even though reading Sam Quinones’s Dreamland has changed my thinking about overdoses, I still think that obesity and inactivity are important to the overall mortality trend, and I’ll stand by the little doodle I made last Fall to explain my thinking:
In fact, the 2012 rate of obesity is one of the very best predictors of the 2013 death rates among the 45-54 year old non-Hispanic whites that Case & Deaton observed: here’s the county-level local regression:
Even focusing in on overdoses among all age cohorts rather than all-cause mortality, as we have been in the recent posts, obesity still appears to have a fair amount of predictive power, though it looks to be less predictive in 2015 than it was in 2010:
But if we control for 2010 overdose rates, obesity does not appear materially to change our estimates of the effects of Medicaid expansion on 2015 overdose rates:
|VARIABLES||Overdose rate-2015||Overdose rate-2015||Overdose rate-2015|
|Overdose Rate 2010||0.834***||0.780***|
|Population in 2010 (in 100,000s)||-1.40e-07|
|White income in $1000s||-0.0275|
|Median income in $1000s||0.00207|
|Percent White in County||0.0533|
|Percent 16 and older Employed||0.106|
|Percent adult white obesity||0.192*||0.167|
Robust standard errors in parentheses. Standard Errors account for clustering at state level.
*** p<0.01, ** p<0.05, * p<0.1
Nor has there been any significant divergence between Medicaid expansion states and Holdout states in obesity: the Holdout states have somewhat higher obesity, and have had such at least since 2000:
Obesity may be important in broader mortality trends (and, as commenters at Marginal Revolution recently suggested, perhaps in declining employment rates in some of the same communities), but doesn’t explain the divergence between Medicaid expansion and Holdout states in overdose rates.
County-level 2012 obesity and state obesity rates 1999-2015 here obesity.
*Graph unweighted state mean obesity for each group of states:
*Get the list of Medicaid holdouts
insheet using medicaidholdouts2.txt, clear
ren v1 state
save medicaidholdouts2.dta, replace
use medicaidholdouts2, clear
merge 1:m state using odyearstate
recode _merge 2=1 3=0
replace medstatus=”Medicaid Expansion” if medicaidexpansion==1
save medicaidholdouts2.dta, replace
use obesitystate, clear
drop ageadjustedpercent ageadjustedlowerconfidencelimit ageadjustedupperconfidencelimit
destring ageadjustedobesitypercent*, force replace
reshape long ageadjustedobesitypercent, i(state) j (year)
drop if state==””
drop if year==.
drop if year<1999
save obesitybyyear, replace
rename ageadjustedobesitypercent obesitypercent
destring obesitypercent, force replace
merge m:m state using medicaidholdouts2, gen(m5)
collapse (mean)rate obesity, by(year medicaidex)
gen obesexp=obesity if medicaid
gen obesnoexp=obesity if medicaid==0
drop if medicaid==.
tab year obesexp
twoway (scatter obesexp year) (scatter obesnoexp year), title(Obesity Rate over time) subtitle (for Medicaid expansion and Holdout states) legend (label(2 “Holdout states”) label(1″Medicaid expansion”)) ytitle(Age adjusted percent Obese)
graph export obesity.png, replace