Why Are American Non-College Whites Killing Themselves?

From Liminal Thinking By David Gray

Summary:

• The media provided extensive coverage of a 2017 draft paper by Princeton professors Case and Deaton (“C&D”) that asserts that non-college midlife whites (“NCMW”) are experiencing increased mortality because of increasing “cumulative despair” related to their weakened status in the labor market.
• There is no doubt that mortality has increased for NCMW, or that labor conditions have weakened for non-college workers (they have for whites, blacks and Hispanics), the question is whether deteriorating labor conditions are the primary cause of increased NCMW mortality.
• The C&D paper is baffling because the authors repeat their causal story even though they seem to be well aware they offer no proof. Their paper also makes clear that they are aware of critiques made by others which include negative selection bias, the inability to explain why the same forces of “cumulative despair” have not affected non-college blacks and Hispanics, uncritical analysis of “corroborating” evidence and failure to review conflicting evidence, and failure to account for temporal and racial differences in exposure to toxicity.
• It is not difficult to come up with more compelling causal stories that don't feature unique labor market responses by whites.

The C&D Causal Story Of White Non-College Despair

The most recent C&D paper was widely (and uncritically) covered by the media including NPR, the Washington Post, and the NY Times. In general these reports sensationalized the findings and generalized them. For example, the Washington Post headline announced “a ‘sea of despair’ among white, working-class Americans” (borrowing from the author’s presentation) even though C&D only discuss those without a college degree -- not all working class Americans. Given the rise of populism in the US (and abroad), the C&D story is topical and fits a trendy narrative that populism can be solely attributed to “poor whites.”

The two graphs most often reprinted from the Case and Deaton paper are shown in Figure 1. They show that white mid-life mortality has only worsened for those without a college degree, and that non-college white mortality is now worse than mortality for all mid-life blacks.

Figure 1: Case & Deaton “Midlife” Mortality Graphs

The C&D causal story is essentially that NCMW are killing themselves via alcohol, overeating, suicide and drugs because: 1) they faced a weakened labor market for those without a college degree when they were young, 2) they experienced lower relative wages and higher probability of unemployment over their lifetime which lead to lower marriage rates and higher divorce rates, 3) they felt the pain of loss of social status and status relative to their parents’ age-equivalent absolute income, and 4) they received less social support due to declining participation in religion and other social groups.

Importantly C&D assert that this causal chain of cumulative lifetime “insult” for NCMW has progressively worsened for every succeeding birth cohort. Their causal story has to infer this in light of their findings that “despair” deaths increase more rapidly with age for younger birth cohorts (see Figure 2). Those NCMW born in 1980 were already experiencing rates of “despair” mortality in their 30’s that was not experienced by those born in 1960 until their late 40s or early 50s.

Figure 2: Case & Deaton Progressive Worsening “Despair” Mortality With Birth Cohort

In Section III of their paper they derive a “worsening birth cohort factor” that is nothing more than a numerical summary of their Figure 1.7 (Figure 2 above) that shows worsening NCMW despair mortality for successively younger birth cohorts. If you are capable of reading the chart, you would automatically know you could derive a “factor” that showed worsening experience with time. C&D explicitly say (underlining mine): “This might be the real wage at the time of entering, but it could be a range of other economic and social factors, including the general health of the birth cohort (Case et al. 2005); we deliberately treat this as a latent variable that we do not specify.”

But then C&D nevertheless conclude their paper (my underlining and parenthetical question mark): “The factor could certainly represent some force that we have not identified, or we could try to make a case that declining real wages is more fundamental than other forces. Better(?), we can see globalization and automation as the underlying deep causes. Ultimately, we see our story as about the collapse of the white, high school educated, working class after its heyday in the early 1970s, and the pathologies that accompany that decline.”

So despite admitting they have “...a latent variable that we do not specify...” that could “...certainly represent some force that we have not identified...”, they nevertheless repeat their unsupported belief that it’s probably globalization and automation killing non-college whites. That unsupported guess is what the media chose to report as a proven result.

Problem #1: Lagged Selection Bias

C&D explicitly warn “ It is important not to focus on those with less than a high school degree, a group that has grown markedly smaller over time, and is likely to be increasingly negatively selected on health.” Curiously, given critiques C&D acknowledge from Gelman, Bound and others (see Footnote 1) regarding their work, they don’t explicitly answer the question of why the same danger doesn’t arise from their focus on those without any college experience.[1]

In C&D’s defense, while NCMW grew smaller as a percent of all whites from 55% in 1990 to 37% in 2000, their percentage held steady through 2015. Thus rising NCMW despair mortality would not seem to be directly attributable to negative selection related to group size. However group size may not be the only indicator of negative selection.

Consider that the NCMW despair mortality increase started in 2000 for a cohort of 50 to 54 year olds that did not go to college from 1967 to 1971. Yes, that would have been during the Vietnam war when 10% of that generation -- primarily men -- served in Vietnam. Of those that served, 88% were Caucasian and 76% were from the lower middle/working class. Notably as to future mortality, 58,000 of those that served never came home.

Compare that cohort to the NCMW cohort that was 50 to 54 years old in 2015 when mortality had worsened. This cohort did not go to college from 1982 to 1986. They had no wars to contend with and by that time, two additional student loan programs had been initiated, arguably making it easier to attend college. Female college enrollment steadily grew from the late 60’s with the passing of Title IX in 1972. Women’s enrollment equalled that of men for the first time in 1980. Yet despite increased access, this later group of men and women still didn’t attend college. One should at least be suspicious of increasing negative selection at work. At a minimum, it is difficult to confidently draw conclusions from comparisons between two cohorts linked only by non-college education when the availability of a college education was increasing.

Problem #2: Why Weren't Blacks and Hispanics Similarly Affected?

C&D make a (very) small effort to explain why their narrative story doesn’t apply to non-Hispanic blacks (and no effort to explain why it doesn’t apply to Hispanics) :

“Social upheaval may have taken different forms, on average, for African Americans. Black kin networks, though often looser, may be more extensive and more protective, as when grandmothers care for children. Black churches provide a traditional and continuing source of support. As has often been noted, blacks are no strangers to labor market deprivations, and may be more inured to the insults of the market.”

Their authors’ presentation also puts forth the suggestion (in the form of a question) that perhaps Blacks had greater access to Federal anti-poverty programs.

Remember that 3 key components of the C&D causal story relate to the successively worsening cumulative impact of lower wages, higher probability of unemployment and lower likelihood of spousal support for non-college individuals. Figures 3, 4 and 5 present US Census data indicating that “midlife” non-college blacks and Hispanics ended up in weaker positions and suffered similar -- if not worse -- deterioration than whites did in these 3 components over the period NCWM despair mortality increased.[2]

Figure 3: 50-54 Year Old Median Household Income/Person In 1999 Dollars

Figure 4: 50-54 Year Old Percent Employed

Figure 5: 50-54 Year Old Percent Living In Household With Spouse Present

Looking at these figures, it is very hard to imagine why -- if socioeconomic pressure is the primary cause of increased alcohol, drug and suicide mortality -- African Americans and Hispanics who suffered at least as much whites would not have experienced a similar (or any -- in the case of blacks) increase in “despair” mortality. To argue that blacks and Hispanics are “used to it” and whites are not seems, in fact, to be an argument that race is more determinative of mortality than socioeconomic pressure.

Problem #3: Uncritical Analysis of “Corroborating” Data & Failure To Review Conflicting Data

Corroborating:

Reports Of Unspecified Pain And Distress

C&D cite “corroborating” evidence to support their causal story of ever worsening cumulative pain felt by NCMW by noting increased reports of pain and use of pain medication, especially by men who are out of the labor force. C&D state: “Since the mid-1990s (when questions on pain and mental health began to be asked annually in the National Health Interview Survey), middle-aged whites’ reports of chronic pain and mental distress have increased, as have their reports of difficulties with activities of daily living (Case and Deaton 2015a).”

Notably, C&D did not report on increased reports of pain by race. That would seem to be critical in examining Problem #2 as to why MCNW despair mortality did not increase for blacks and hispanics.

It is unlikely to be completely coincidental that there was an increased reporting of pain and the fact that those in economic distress have increasingly looked to Federal Disability benefits as seen in Figure 6. To qualify for disability, many applicants referenced unspecified pain.

“Another driver of DI growth is a lower threshold for what constitutes a “disability” that prevents someone from working. In 1961, the largest share of new DI recipients (25.7%) suffered from life-threatening illnesses like heart disease. By 2011, however, subjective ailments like musculoskeletal problems (33.8%) and mental illness (19.2%) had risen to the top. At the same time, high-tech tools that determine disease have made it easy to find a second (or third) opinion that supports the patient’s case.”

Note that “disability” has mysteriously declined as the economy recovered from the 2008 Great Recession. C&D would have been well served to at least question the value of their corroborating evidence in the context of the economic incentive to self-report mental distress, chronic pain and difficulty with activities.

Figure 6: Disability Applications And Award

Mortality And Chinese Imports

The best confirming evidence supporting the C&D thesis that economic pressure causes increased mortality is found in Autor, Dorn and Hansen's 2017 study (referenced by C&D) of declining marriage in areas that lost manufacturing jobs (which are disproportionately non-college and white) due to increased Chinese imports:

“...we show that trade shocks lead to a differential rise in mortality from drug and alcohol poisoning, liver disease, diabetes, and lung cancer among young men relative to young women. The proportional rise in mortality from these causes is substantial: a one-unit shock more than doubles the relative male death rate from drug and alcohol poisoning.”

However, even here, note that this only refers to an increase in male mortality relative to female mortality and C&D’s data indicates mortality increasing equally for both male and female NCMW. Also the timing is a bit off as:

“For the period of 1990 - 2000, our data indicate a mean rise of Chinese import penetration of 0.94 percentage points, 60 percent of which accrued to male employment and 40 percent to female employment. In the subsequent 2000 - 2007 period, when Chinese import penetration accelerated sharply, import penetration rose by an additional 1.33 percent, with 65 percent of this rise accruing to male employment.”

Based on this economic pressure, NCMW despair mortality should have: 1) impacted men more than women, 2) started increasing before 2000, and 3) accelerated (rather than steadily increased as shown in the left panel of Figure 1). In general, Chinese import penetration can at best explain about a third of the decline in the percentage of young women married, the percent of children in two parent households, and the increase in out of wedlock births (see Table 9 of Autor, Dorn and Hansen). The authors conclude “The inference that we draw, consistent with our priors, is that rising China trade is a contributor to, but not the primary driver of the broader demographic shifts on which we focus.” As Chinese imports are a leading cause of declining employment for NCMW, that conclusion calls into question C&D’s assertion that economic pressures are the dominant cause of NCMW mortality increases.

Marriage and Non-College Alcohol, Suicide & Drug Mortality

Finally, C&D suggest the "despair" sequence for NCMW is poor labor prospects lead to increased likelihood of never marrying or being divorced. That social isolation in turn leads to alcohol, suicide or drug mortality.  Table 1 shows the percentage of all deaths due to alcohol, suicide and drugs for non-college non-Hispanic whites by marital status in 2015. 

Table 1: 2015 Non-College, Non-Hispanic Whites: Percent of Deaths Due to Alcohol, Suicide and Drugs By Age Group

In C&D's context, there appear to be nearly as many "despairing" married NCMW as those that are single and divorced. Marriage is not as strongly protective as C&D imply.

Conflicting:

Chetty, Stepner and Abraham’s 2016 paper The Association Between Income and Life Expectancy in the United States, 2001-2014 is a careful study of US geographical variations in the significant gap in life expectancy between the poor and the wealthy. They looked at where life expectancy for the bottom income quintile is the longest and where it is the shortest. They found (my underlining):

“Correlational analysis of the differences in life expectancy across geographic areas did not provide strong support for 4 leading explanations for socioeconomic differences in longevity: differences in access to medical care (as measured by health insurance coverage and proxies for the quality and quantity of primary care), environmental differences (as measured by residential segregation), adverse effects of inequality (as measured by Gini indices), and labor market conditions (as measured by unemployment rates). Rather, most of the variation in life expectancy across areas was related to differences in health behaviors, including smoking, obesity, and exercise.”

They also found that neither religion or the percentage of blacks correlated with regional variations in life expectancy for the poor. While they are studying a different issue, their results would not seem to be a good fit with C&D’s causal story and its heavy reliance on labor markets, social isolation, and race.

Also, presumably if NCMW are killing themselves out of despair, their unhappiness should show up in other ways. Pew Researchers looking at data from 2005 through 2008 found happiness in this time period had no correlation to education or race. Similar to the Chetty, Stepner and Abraham findings, the largest correlation was with health.

Figure 7: Pew Research On US Happiness 2005 - 2008

Problem #4: What About Temporal And Racial Differences In Exposure To Toxins?

Figure 8 shows the 280% increase in total deaths due to opioids (including illegal variants and heroin) since 2002 (In 2015 the CDC reported that 91 Americans died every day from opioids). According to the CDC (underlining is mine), “Since 1999, the amount of prescription opioids sold in the U.S. nearly quadrupled,2 yet there has not been an overall change in the amount of pain that Americans report.3,4” Notably for the international mortality differences that C&D refer to, “...Americans take 195 milligrams of hydrocodone, the active ingredient in Vicodin and OxyContin, per capita annually, whereas the figure in Germany is 28 milligrams per capita.”

Figure 8: Overdose Deaths From Opioids

So temporally we have a fit to the 2000 - 2014 time period, we see the smooth mortality increase for both men and women (again see Figure 1 above), and a reason why young cohorts born in 1980 are seeing similar mortality to those born in 1960 as both cohorts were exposed to this new killer at the same time (see Figure 2 above).

Racially, we also can tie this to increased white mortality. Not surprisingly, whites have greater access to opioids given their greater access to medical care, the fact that doctors are also predominantly white, and because their incomes are higher than African Americans' and Hispanics':

“A brand-new study led by Astha Singhal of Boston University’s dental school found that non-Hispanic blacks and non-Hispanic whites who came to emergency rooms complaining of fractures, toothaches or kidney stones — “definitive” conditions a doctor could see — were equally likely to be sent home with a prescription for opioids. Racial disparities emerged, however, with respect to “non-definitive” conditions. In those cases, African Americans were 33 percent less likely than whites to get an opioid, the study found.”

Differences in racial exposure to other drugs like crystal meth (higher for whites) and cocaine (higher for blacks) have also been cited. Notably deaths due to cocaine did not increase in the 2000-2015 period. Deaths from Benzodiazepines (the key ingredient in Xanax and Valium) have increased by 430% since 2002 and overdose admissions are higher for whites. Benzodiazepines are the cause of most drug-induced suicides.

Beyond drug use, whites are more likely to have smoked in their lifetime than blacks or Hispanics (and all college graduates smoke substantially less than non-college individuals).[4] Lower rates of smoking among Hispanics is one of the principal reasons cited to explain the “hispanic mortality paradox” (the fact that Hispanic life expectancy exceeds that of whites and African Americans despite low levels of income).[5] Whites use more smokeless tobacco than blacks and Hispanics.[4] The percentage of deaths due to alcohol are higher for whites than blacks, but not Hispanics. Suicide is lowest among Hispanics (for those that are Roman Catholic, suicide is a grave sin).

Finally there are significant geographic variations in mortality that can be related to factors like diet (e.g., fried foods) and exercise. If you want to see the amazing geographic variations in mortality by cause, check out http://www.worldlifeexpectancy.com/usa-health-maps. Mortality is complex and not subsumable under a simple story of labor disadvantage.

So Let’s Tell a Different Story (That Actually Matches The Data)...

As an exercise, let's emulate C&D and make up a story to explain the increase in deaths due to alcohol, suicide and drugs (“ASD”) over time and by race depicted in Figure 8 from Shannon M. Monnat (it is an apples to apples comparison, not one conditioned on education for whites only).

Figure 8: Alcohol, Suicide & Drug Mortality By Race and Period

Figure 9 is also from Monnat and breaks down ASD by age. As a thought experiment, imagine Figure 9 is a graph of the probability of experiencing auto accidents by age instead of a graph of ASD by age. It would not be surprising that younger drivers have more accidents (and women have fewer accidents than men), and that racial groups with more access to cars (due to income) would also be more likely to have auto accidents.

Figure 9: ASD Death By Age: 2010-2014

N

So here's a story that "explains" the charts:

1. New effective killers are created that are increasingly available from 2000 - 2015 that have a high correlation with alcohol and suicide (opioids and Benzodiazepines).
2. As documented by the Astha Singhal study referenced above and other sources: medical access to the new killers is greatest for whites, followed by blacks and finally Hispanics.
3. Access to the new killers due to higher income is greatest for Whites, Hispanics and then blacks (see Figure 3). But real incomes declined over this time for all 3 groups (thus declining drug mortality for non-Hispanic blacks may reflect their lowest and declining income).
4. Propensity for reckless behavior is highest among the young,
5. Propensity for reckless behavior is also highest among the most negatively selected group (non-college whites most negatively selected, with both African Americans and Hispanics showing less negative selection).[5]
6. Propensity for reckless behavior is higher for men than women.

I guarantee that one can fit “factors” (as C&D did) to the data that would be consistent with the story outlined above as deriving factors is just numerical re-telling of the relationships shown in the charts. This story also explains the C&D findings of increased mortality with successive birth cohorts (older cohorts were not exposed to the new killers until recently and younger cohorts are more disposed to use them). It also provides a credible reason that use among whites is higher (medical access, income, negative selection) without making any reference to white despair or worsening labor markets.

Does that mean worsening labor markets play no role? No -- the Autor, Dorn and Hansen's 2017 study suggests labor markets might play a modest role (as they do in explaining declining marriage rates etc.). It just means labor markets and “white despair” are quite likely not the primary drivers.

How differently the C&D mortality data would have been covered if NPR, NYT and WaPO had chosen the headline: “Whites’ Privileged Access To Drugs Is Killing Them.”

[1] See Gelman & Auerbach here; also see Bound here and Harris here.

[2] Data available from IPUMS-USA, University of Minnesota, www.ipums.org. It is freely available subject to their licensing agreement.

[3] See here.

[4] See Tables 2.25B, 2.30B here.

[5] Mortality is lowest for the 36% of Hispanics not born in the US -- some hypothesize the physical nature of their jobs may play a role (along with some positive selection bias due to immigration).

[6] A simple proxy for negative selection for non-college individuals is the percentage they comprise of their race. The smaller percentage, the more negative selection. For whites the percentage of non-college individuals aged 50 - 54 remained stable at approximately 37% from 2000 - 2015; for blacks it declined from 54% to 48% over the same period; for Hispanics it declined from 56% to 49%. Thus by this logic whites would show the most negative selection. Blacks and Hispanics would show less (but deteriorating) negative selection.

Transparent and reproducible: I generated Figures 3, 4 and 5 from US Census data from IPUMS-USA, and Table 1 from CDC data. They can be generated by using the free, publicly-available R program and the R code available on github in “CDCdeathRate.r" and "CDCdata.r" (to download the CDC mortality data) to analyze the publicly available data I used.