For most of American history, taking care of aging parents was a hands-on job. Adult children shared a roof with mom and dad, cooked their meals, nursed them through illnesses, and lived close enough to be there in an emergency. That kind of support came with a hidden price tag for the next generation: it tied children to their parents’ hometowns, whether or not the local job market suited them.
So what happens when the government suddenly steps in and hands aging parents a monthly check? A new NBER working paper traces the arrival of Social Security in the 1930s and 1940s through the lives of recipients’ children, following them all the way to their final ZIP codes. The authors find that sons whose parents gained coverage moved farther from home, earned more over their careers, and ended their lives in wealthier neighborhoods. Daughters, whose caregiving was harder to replace with cash, saw no such gains.
A question the founding theories missed
Economists have long known that Social Security displaced private family support. Co-residence rates among the elderly plummeted as benefits rolled out. But classic models of intergenerational transfers, from Robert Barro and Gary Becker in the 1970s, largely ignored one dimension of family support: it required physical proximity.
That’s the puzzle Daniel Fetter of Dartmouth, Lee Lockwood of the University of Virginia, and Paul Mohnen of the Federal Reserve Bank of Atlanta set out to investigate. If cash benefits could substitute for shared housing and shopping trips, but not so easily for hands-on nursing care, then Social Security might have loosened a geographic leash on some adult children while leaving others still tethered. And because sons and daughters historically provided different kinds of support, the effects might split along gender lines.
Finding usable variation in a near-universal program
Studying Social Security’s effects is tricky because almost everyone eventually got it. The researchers exploited a quirk of the program’s rollout: when Social Security launched in 1935, it covered wage workers in many industries but excluded employees of nonprofits, state and local governments, and private households. Those workers weren’t brought in until 1950 or later. A janitor at a for-profit company was covered from the start; a janitor at a nonprofit hospital, church, or public school was not.
That distinction had a bigger bite for older cohorts than younger ones. A man born in 1877 who worked in a 1935-covered job could start collecting benefits at 65, in 1942. If he worked in a 1950-covered job instead, he wouldn’t accrue enough credits until around age 75. But for a man born just eleven years later, the gap between the two coverage categories nearly disappeared. This gave the authors a way to compare otherwise similar workers, same occupation, similar backgrounds, whose eligibility age differed by up to a decade because of which industry happened to employ them.
To follow the children of these men into old age, the researchers built a new dataset. They linked Social Security death records, which include a person’s last known ZIP code, to the 1920, 1930, and 1950 U.S. Censuses, along with Social Security application records that list parents’ names. Using a supervised machine-learning approach developed for the LIFE-M project, they connected about 1.6 million children from roughly 860,000 families to their fathers’ pre-1935 employment information. They also drew on the 1973 Survey of Occupational Changes in a Generation, which recorded fathers’ occupations and, for a subset of respondents, linked to actual Social Security earnings records.
What the data revealed
The first check was whether predicted eligibility actually predicted behavior. It did. Men whose 1930 jobs made them eligible earlier were more likely to file for benefits, and they filed at younger ages. And crucially, families with earlier-eligible fathers looked no different from other families on 1930 characteristics like home ownership, literacy, or urban residence. The comparison groups were genuinely comparable.
Then came the main finding. Sons whose fathers were predicted to become eligible ten years earlier, roughly the gap between a 1935-covered and 1950-covered worker born in 1877, ended their lives in ZIP codes with about 2.4 percent higher average income and 4.3 percent higher home values. Their neighborhood quality rank rose by about two percentiles. The OCG data showed these sons had higher occupational income scores in mid- and late-career, on the order of a 35 percent gain. Linked earnings records confirmed they were more likely to hit the Social Security taxable maximum during their prime working years.
For daughters, the coefficients hovered near zero.
Migration as the connecting thread
Why the gap? The authors point to geography. Sons with more Social Security-exposed fathers were about two percentage points more likely to live in a different state by 1950, and 2.5 percentage points more likely to have crossed a Census division. They were also more likely to move more than 500 miles from home. Daughters showed no such pattern.
The moves weren’t random. Sons tended to relocate to states where workers with their level of education earned more, suggesting they were finding better matches for their skills rather than just heading to bigger cities. Sons who moved across divisions were more likely to end up in the top two quintiles of the ZIP income distribution, while those who stayed in their parents’ division were more likely to end up in the bottom two. Migration and long-run outcomes moved together.
The researchers argue that this reflects the different kinds of support sons and daughters historically provided. Shared housing, which sons offered relatively more often, is fungible with cash: a parent with a guaranteed monthly check can rent an apartment and buy groceries. Hands-on caregiving, disproportionately provided by daughters, is harder to purchase on the market and harder to arrange from a distance. So Social Security freed sons to move; daughters remained anchored.
A surprising fiscal accounting
The size of the estimated gains is striking. Under the authors’ baseline assumptions, a ten-year difference in a father’s earliest eligibility age translated into about a 12 percent gain in a son’s full-career average earnings, worth roughly $130,000 in expected present value (in 2020 dollars) per son. Aggregating across the average recipient family’s children, total earnings gains were about six times the Social Security benefits the parents received.
That flips a longstanding story about early Social Security. Textbooks describe the initial generation of recipients as receiving a “windfall” at the expense of later generations. The authors argue that when you count the children’s earnings gains, the early expansions may have roughly paid for themselves. Their point estimate implies a net fiscal cost of negative 55 cents per dollar of benefits, though the confidence interval is wide enough that a modest positive cost can’t be ruled out.
The researchers also push back on the standard efficiency critique. Social Security is often faulted for reducing recipients’ labor supply by inducing early retirement. But in their accounting, the labor-supply loss from recipients was outweighed by the earnings gains of children who could now migrate to better-matched labor markets.
Why families couldn’t do this on their own
A natural question: if migration paid so well, why didn’t families arrange it themselves, with one sibling sending money home while another stayed put? The authors offer several explanations. Migration’s returns arrive gradually and unpredictably, so a son who moved couldn’t easily promise to compensate his parents or siblings in real time. Insurance markets for these risks didn’t exist. Care obligations tended to fall indivisibly on whoever lived closest. Social Security, by pooling risk across all workers and guaranteeing income until death, effectively filled in for missing markets.
Caveats worth keeping in mind
The authors are explicit that several features of their setting amplified the effects they measured. Government old-age support was nearly nonexistent before Social Security, family support was extensive, and regional wage gaps in mid-century America were unusually large, meaning the returns to leaving a poorly matched labor market were high. Those conditions don’t describe the contemporary United States, where wage differences across regions have compressed and Social Security is already near-universal. The magnitudes shouldn’t be extrapolated to marginal changes in today’s program.
The mechanism, though, may still apply where family support constrains where adult children live and work. The authors point to long-term care policy as a likely example: informal caregiving by adult children remains widespread in the U.S., and programs that substitute for it could have intergenerational effects that short-run evaluations miss.
