Like a Fish Needs a Bicycle

As the  blogger Random Critical Analysis has observed in discussions of Raj Chetty’s economic mobility studies as well as of variation in homicide rates, a surprising amount of black and white disparities can be explained (statistically if not causally) by variation in family structure. Whether or not this variation in family structure is causally responsible for these outcomes, at a geographic level, areas with higher marriage rates and lower single motherhood for black families also tend to have outcomes more similar to white national norms, while areas with lower marriage rates and higher single motherhood for white families tend to have outcomes more similar to black national norms.

This turns out to be true of women’s mortality as well- areas where a higher percentage of 45-54 year old women are married also have lower mortality for 45-54 year old women (the age group that Case & Deaton looked at in their 2015 paper), for both blacks and whites. Here are all the values for 2005 through 2015:

marriedmortality

And then the average percentage married and average mortality for larger counties (those with over 4,000 women in the racial and age group) over the ten years:

blackwhitemortality

The exceptions to the overall pattern appear to be unusually wealthy counties with a large percentage of unmarried white women (see Manhattan and San Francisco labeled above; other outliers include Santa Fe, NM and Alexandria, VA). Even so, these are highly unusual: there are 192 large counties with mortality rates lower than Santa Fe’s, and on average 72% of the 45-54 year old white women in these counties are married.  DuPage County, Illinois or Washington County, Minnesota (comfortable suburbs of Chicago and the Twin Cities respectively) or Forsyth, Georgia  are much more typical of low-mortality counties than New York or San Francisco. Counties with large black populations with lower mortality rates for black women than the national average white rate (or than the white rate in Santa Fe) include Collin County, Texas, a suburb of Dallas-Fort Worth, and Henry County, Georgia, both with over 55% of their black middle aged women married. In general, the Atlanta suburbs appear to have very low mortality for both whites and blacks, which is interesting given how badly the Atlanta suburbs did in Chetty’s  economic mobility studies. Well, money isn’t everything, but this does seem like evidence his results were in large part confounding the effects of race with those of marriage and the “geographic” effects he argued he was singling out.

This association persists even in an OLS regression at the county level (weighting by population) if you include some obvious “controls” for level of education and income:

Marriage OLS

Nonetheless, I don’t think the overall association is causal, for individuals in the short term at least (even if some studies find a modestly reduced mortality rate for married women versus single women); for one thing, places with increases in marriage rates haven’t fared much better than places with decreases, in the last ten years. Instead in most counties, mortality rates have gone up for white women and down for black women, regardless of changes in marriage rate:

a1

It seems more likely to me that marriage rates are one indicator (and perhaps driver) of the broader social and economic stability of counties, that then translates into changes in mortality over the very long term.

Update: There’s a lot of randomness in the year-to-year mortality rates, so I tried looking at the change from the average of 2005-2009 to the average of 2011-2015 for mortality and marriage rates. This should be a more reliable measure than comparing a single year measure from 2015 with 2005. Comparing these two averages, it looks like changes in marriage rates were indeed an excellent predictor of where mortality rose for white women, but not so much where they fell for black women. (I forget who called marriage rates a “one-way ratchet.”)

averagedchanges
The “Coming Apart” meets Case/Deaton Graph

This relationship between declining marriage among white women and increased mortality also appears in the regression results for average changes:

(1) (2)
VARIABLES  

Averaged Change in White Female 45-54 YO  Mortality 2005-2009 to 2011-2015
Averaged Change in BlackFemale 45-54 YO  Mortality 2005-2009 to 2011-2015
% Married in 2005 0.0401 1.235
(0.313) (0.773)
Averaged Change in Marriage Rate 2005-2010 to 2011-2015 -4.333*** 2.018
(0.941) (1.991)
Income in 2015 ($1000s) -1.315*** 1.047
(0.327) (0.911)
Income in 2005 ($1000s) 0.106 -3.742***
(0.595) (1.385)
% with Higher Education in 2005 -0.0468 0.247
(0.423) (0.818)
% with Higher Education in 2005 0.213 0.0375
(0.424) (0.758)
Mortality Rate in 2005 -0.0717** -0.0894*
(0.0320) (0.0530)
Constant 56.11* -4.551
(31.74) (57.93)
Observations 260 93
R-squared 0.333 0.159

Robust standard errors in parentheses

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

The Census CPS data is here.

The CDC Wonder data is here.  blackandwhitefemalemortality  Updated Stata code below:

clear all
set more off
use nonhispanicwomen
*this is a file with American Community Survey results from 2005-2015, for 45-54 year old non-Hispanic women, outputted by IPUMS
drop if race>2
*keeping only black and white non-hispanic women
gen racestr=”Black or African American” if race==2
replace racestr=”White” if race==1
save nonhispanicwomenformerge, replace
insheet using blackandwhitefemalemortality.csv, clear
*this is the CDC Wonder data for non-Hispanic Women
gen county5=string(countycode,”%05.0f”)
drop if race==”American Indian or Alaska Native”
drop if race==”Asian or Pacific Islander”
rename race racestr
rename county countystr
save femaleblackandwhite2003-201545-54yomortalitybycounty, replace
merge m:m county5 year racestr using nonhispanicwomenformerge
*dropping 2003-2004
drop if year<2005
save nonhispanicwomenmortality, replace
*create the death rate per 100000
gen rate=deaths/population*100000
*hieduc=education past high school
gen hieduc=(educ>7)*100
*change income into $1000s
replace incearn=incearn/1000
*Marst==1 means married and living with partner/not separated- generating for percent
gen married=(marst==1)*100
preserve

collapse (mean) rate incearn married hieduc population (firstnm) countystr year, by(county5 racestr)
gen blackrate=rate if racestr==”Black or African American”
gen whiterate=rate if racestr==”White”
drop if married==0

twoway (scatter blackrate married if population>4000 ,msymbol(Oh)) (scatter whiterate married if population>4000,msymbol(Oh) ) (lowess blackrate married if population>4000) (lowess whiterate married if population>4000) , title(Black and White 45-54 YO Women’s Mortality) subtitle(vs Percent of Female Respondents Married in County) caption(Average 45-54 YO female Mortality and Marriage Rate by County for 2005-2015) note(Mortality rates from CDC Wonder & Marriage rates from Census CPS) ytitle(Mortality per 100000) xtitle(Percent of 45-54 YO Female Respondents Married)

graph export marriedmortality.png, replace

restore
collapse (mean) rate incearn married hieduc population (count)numobs=pernum, by(county5 racestr year)
*drop if 5 or fewer respondents in a county/race/year cell
drop if numobs<6

gen blackrate=rate if racestr==”Black or African American”
gen whiterate=rate if racestr==”White”
reg blackrate incearn married hieduc [aw=population], vce(cluster county5)
outreg2 using blackandwhitefemalemortality.doc, replace
reg whiterate incearn married hieduc [aw=population], vce(cluster county5)
outreg2 using blackandwhitefemalemortality.doc, append

twoway (scatter blackrate married, msymbol(Oh)) (scatter whiterate married, msymbol(Oh)) (lowess rate married), title(Black and White 45-54 YO Women’s Mortality) subtitle(vs Percent of Female Respondents Married in County) caption(Average 45-54 YO female Mortality and Marriage Rate by County for 2005-2015) note(Mortality rates from CDC Wonder & Marriage rates from Census CPS) xtitle(Percent of 45-54 YO Female Respondents Married) ytitle(Mortality per 100000)
graph export marriedallyearscounties.png, width(1000) replace
drop if year==.
reshape wide rate numobs incearn married hieduc population whiterate blackrate, i(county racestr) j(year)
gen whitechange= whiterate2015-whiterate2005
gen avgwhchg=(whiterate2011+whiterate2012+whiterate2013+whiterate2014+whiterate2015)/5-(whiterate2005+whiterate2006+whiterate2007+whiterate2008+whiterate2009)/5
gen avgblchg=(blackrate2011+blackrate2012+blackrate2013+blackrate2014+blackrate2015)/5-(blackrate2005+blackrate2006+blackrate2007+blackrate2008+blackrate2009)/5

gen avgmarchg=(married2011+married2012+married2013+married2014+married2015)/5-(married2005+married2006+married2007+married2008+married2009)/5
*drop small cells
drop if (numobs2005+numobs2006+numobs2007+numobs2008+numobs2009+numobs2010+numobs2011+numobs2012+numobs2013+numobs2014+numobs2015)<20
drop if population2015<4000&racestr==”Black or African American”
drop if population2015<8000&racestr==”White”

twoway (scatter avgblchg avgmarchg if population2015>4000 ,msymbol(Oh)) (scatter avgwhchg avgmarchg if population2015>8000,msymbol(Oh) ) (lowess avgblchg avgmarchg if population2015>4000) (lowess avgwhchg avgmarchg if population2015>8000) , title(Change in Black and White 45-54 YO Women’s Mortality 2005-2009 to 2011-2015, size(small)) subtitle(vs Chg % of Female Respondents Married in County 2005-2009 to 2011-2015, size(small)) caption(Average 45-54 YO female Mortality and Marriage Rate 2005-2015 for larger counties) note(Mortality rates from CDC Wonder & Marriage rates from Census CPS) ytitle(Chg in mortality per 100000) xtitle(Change in % of 45-54 YO Female Respondents Married, size(small))
graph export averagedchanges.png, replace
gen incchange=incearn2015-incearn2005
gen educch=hieduc2015-hieduc2005
reg avgwhchg married2005 avgmarchg incearn2015 incearn2005 hieduc2005 hieduc2015 whiterate2005 [aw=population2015], vce(cluster county5)
outreg2 using changemort.doc, replace
reg avgblchg married2005 avgmarchg incearn2015 incearn2005 hieduc2005 hieduc2015 blackrate2005 [aw=population2015], vce(cluster county5)
outreg2 using changemort.doc, append
twoway (scatter avgblchg married2015 if population2015>4000 ,msymbol(Oh)) (scatter avgwhchg married2015 if population2015>8000,msymbol(Oh) ) (lowess avgblchg married2015 if population2015>4000) (lowess avgwhchg married2015 if population2015>8000) , title(Change in Black and White 45-54 YO Women’s Mortality 2005-2010 to 2010-2015, size(small)) subtitle(vs Chg % of Female Respondents Married in County 2005-2010 to 2010-2015, size(small)) caption(Average 45-54 YO female Mortality and Marriage Rate 2005-2015 for larger counties) note(Mortality rates from CDC Wonder & Marriage rates from Census CPS) ytitle(Chg in mortality per 100000) xtitle(Change in % of 45-54 YO Female Respondents Married, size(small))

2 thoughts on “Like a Fish Needs a Bicycle

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s