Money Given to Poor Mothers Slowed One Measure of Aging in Their Children, and Only That One

In maternity wards in New Orleans, Omaha, the Twin Cities and New York City, women who had just given birth were handed a debit card and told it would be topped up on the same day every month until their child was ready for school. There were no conditions and no paperwork. A random draw decided how much the card would carry: $333 a month for some of the mothers, $20 for the rest.

Every mother enrolled was living below the federal poverty line when her child was born. The draw is the whole point. Children who grow up poor become adults who fall ill earlier and die younger, but income travels with so much else that the effect of money by itself has been close to impossible to isolate. Baby's First Years, the trial behind those cards, was built to pull the two apart, and the larger gift raised a family's income by roughly 18%.

On September 8, in Nature Human Behaviour, Laurel Raffington of the Max Planck Institute for Human Development in Berlin and her colleagues reported what the money did to the children's epigenome, the chemical tags that sit on DNA and turn genes up or down. No early-childhood trial aiming at later health had used a measure of this kind as an outcome, the paper notes.

At the age-four visit, mothers and children gave saliva samples. Usable DNA came back from 735 of the children. The team scored it with DunedinPACE, a formula built from adults tracked over many years in a New Zealand birth cohort, which reads the tags as a rate: how many years of physical decline a body is accumulating per calendar year. Children in the higher-cash group scored lower on it by 0.17 of a standard deviation, a small step against the ordinary spread among these children. The range of values the data are consistent with runs from −0.32 to −0.01, and P = 0.037, which clears the customary line for calling a result statistically significant.

Because the amounts were assigned at random, that gap can be laid at the money's door rather than at whatever else separates a household with more income from one with less. "Correlations alone do not tell you whether this process can be slowed by targeted measures," Raffington said in the Max Planck Society's German-language announcement of the study, translated here. "The demonstration of a genuine causal effect of the cash payments is all the more notable for that."

The result was written down in advance. It was never corrected.

Two things about that number should be held apart. The first is that it was predicted before anyone looked. DunedinPACE was entered on the trial's public registration as a secondary outcome before the age-four data were collected, and the team posted a detailed analysis plan after collection but before the methylation data existed. The paper states that the hypotheses and analyses are reported exactly as preregistered. Secondary here means the trial was not sized around this measure. It does not mean anyone went looking afterward.

The second is that the P value stands uncorrected. The trial's prespecified epigenetic tests come as a set of four: two indices, each run in the children and in the mothers. The correction the team planned covers a different pair (the supplementary clocks it expected to stay flat), not these. Spread across the four main tests, the simplest adjustment would ask for a P below about 0.0125, and 0.037 does not reach it. The range around the effect almost touches zero, the trial's size was fixed by its cognitive and behavioral outcomes, and no separate power calculation was made for the epigenetic ones at all.

Nothing else about these children moved with it.

The measure that shifted is, so far, attached to nothing else in the same children. The team checked it against brain activity in every frequency band measured, against a laboratory test of executive function, against vocabulary, against body mass index and against how mothers rated their child's overall health. None of them tracked DunedinPACE. The authors read that as consistent with an old idea in developmental biology: that early conditions mark the body before they show up in behavior. It also leaves nothing visible in a four-year-old for the result to be anchored to.

How to read a clock of this kind in a small child is itself unsettled. DunedinPACE was built on adult blood and applied here to a preschooler's saliva, a stretch the authors describe and defend at length. They also point to an open disagreement among researchers about what epigenetic aging is in a body that is still growing, and how the two processes are related.

The second marker ran the wrong way.

A second index was preregistered alongside the first. Epigenetic-g is a methylation score trained to predict how adults perform on cognitive tests, and the prediction here was that more cash would raise it. The estimates came out lower instead. In the version the team prefers, the drop cannot be told apart from zero; in the alternative version it can (−0.19 of a standard deviation, P = 0.016), and even there it fails one of the study's checks for who dropped out. The authors call the finding inconclusive, decline to build anything on it, and write that it is possible the cash gifts "will be associated with worse cognitive outcomes later in development."

The 777 mothers showed no sign of an effect on anything: not on pace of aging, not on Epigenetic-g, not on the supplementary clocks, and not after weighting for who returned and who did not. That sits beside earlier results from the same trial. There was no evidence the extra money improved housing quality, moved families to better neighborhoods or reduced hardship and food insecurity.

The payments themselves have ended; the last one went out in September 2025, after the children's sixth birthday. Whether a difference measured at four survives that, and whether it predicts anything about a life, is what the next waves of the study exist to answer. "It remains to be seen whether these differences persist as the children grow up," the senior authors, Kimberly Noble of Teachers College, Columbia University, and K. Paige Harden of the University of Texas at Austin, said in the same announcement. The summarized data and the analysis code are public, at ICPSR and on GitHub, so anyone who wants to check this can start from the same files the authors used.