(Bloomberg View) -- Stress, I've long suspected, may explain whylifespans have been lengthening for high-incomeAmericans but have remained the same or even shortened forlow-income and middle-income people.

A new analysis from the Hamilton Project released today addsimportant evidence: Biomarkers of stress have risen much morerapidly for low-income people than for high-income ones.

Related: Americans suffering from financialPTSD

The most recent national data show that life expectancy declinedin 2015.

Although these statistics are not broken down by income, it’s agood bet that the greatest declines were concentrated among lower-and middle-income people, because for decades gaps in lifeexpectancy by income have been widening -- as a National Academy ofSciences panel I co-chaired documented.

Our panel was unable to explain why this has been happening,however.

A sharper decline in smoking among high-income people is onepartial explanation, but this and other identifiable factors can'taccount for most of the growing longevity gap.

And so I've come to suspect that the hidden factor is stress: Asincome inequality has increased, life has become substantially morestressful for low- and moderate-income families than it is forhigh-income families, and prolonged exposure to such stress isliterally killing people.

Related: Employee well-being predictions for2017

This has always been just a theory. While there has been plentyof indirect evidence, including the opioid epidemic, there has beenlittle direct empirical support.

One recent study did find an association between workplacestress and a decline in life expectancy, and the link wasespecially strong for low-income workers. But that study generatedits results from several sets of assumptions, and so was onlysuggestive.

The new Hamilton Project analysis, however, noticeablystrengthens the evidence. (I am on Hamilton's advisorycouncil.)

Diane Schanzenbach, Megan Mumford, Ryan Nunn and Lauren Bauerexamined changes over time in the relationship between health andincome. And, using laboratory measurements of health biomarkersfrom the National Health and Nutrition Examination Survey, theyconstructed a new measure of stress load.

That survey is particularly insightful for the purpose at handbecause it collects health measurements along with more traditionalsocioeconomic and related data, and because its data go backseveral decades.

The Hamilton team looked at the measurements from 1976-80 and2009-14 to see how things had changed.

Beginning with self-reported health, the researchers found adecline for all income groups, but for high-income people, thechange was small and not statistically significant.

For low-income people, the decline was much larger andstatistically significant.

The researchers next examined the data on biomarkers that areassociated with long-term stress. Note that this approach does notmeasure the external environment, but rather how one internalizesit.

The measurements involved have all been identified in themedical literature as predictors of illness and mortality: bloodpressure, triglycerides and cholesterol (markers of cardiovascularrisk); creatinine (kidney function); and albumin (liverfunction).

The researchers then combined these into a single stress index,based on how much each variable affected self-reported health.

Two features of the results stand out. First, across the board,stress loads have grown heavier since the late 1970s. Second, theincrease has been substantially larger for low-income people thanfor high-income people.

We may never have definitive proof of the stress theory toexplain the widening life expectancy gap.

But these new results -- which will come as no surprise toanyone who has read "Hillbilly Elegy," by J.D. Vance -- are strongevidence that stress load plays a crucial role in stagnant ordeclining life expectancy among America's low- and middle-incomefamilies.

This column does not necessarily reflect the opinion of theeditorial board or Bloomberg LP and its owners.

Copyright 2018 Bloomberg. All rightsreserved. This material may not be published, broadcast, rewritten,or redistributed.