Most of the time, when the government spends money, it doesn’t make that money back. That’s often fine — we don’t have a military or Medicare, say, because we think they’ll make the government money. We have them because they provide services that we’ve decided the government should provide.
But in a few relatively rare cases, government spending can actually pay for itself. And in those cases, spending can be more than a good idea. It can be a good idea with no real drawbacks.
Those are the kinds of free lunches that economists Nathaniel Hendren and Ben Sprung-Keyser (both of Harvard) think they’ve found in a new working paper. They looked through past economics papers studying 133 policy changes, from the Perry Preschool Project in 1962 to a 2016 experiment trying to increase college enrollment in Michigan.
Using only papers with solid enough methodologies to determine the actual results of those programs, Hendren and Sprung-Keyser estimated, for each of them, a number they call the marginal value of public funds: that is, the value of the benefit to recipients, divided by the cost to the government. MVPFs above 1 mean that the benefits to recipients are greater than the costs to government — either because the benefits are greater than anticipated or because the costs are lower.
And in several cases that Hendren and Sprung-Keyser examined, the MVPF was actually infinite — that is, the programs actually saved the government money over the long run, so their benefits are available for free.
“The findings are driven by the cost side,” Sprung-Keyser told me. “For well-targeted education investment and health, the government recoups much of its investment in increased tax revenue” and decreased health expenditures later on.
That’s a big deal. It suggests that if taxpayers decide, in the short run, to invest heavily in children’s health and education, they’ll actively save money decades in the future, money that could be used for other priorities or even for tax cuts. It’s in voters’ self-interest to try to make kids better off.
Finding programs that pay for themselves
So what programs produced these kinds of free lunches? Mostly, the researchers found, programs for kids.
Expanding Medicaid coverage to pregnant women and their children, for instance, had a precisely estimated infinite payoff. The health gains for the kids meant lower Medicaid and other government health bills later on in life; the government strictly saved money.
Education spending also sometimes has infinite returns. Contrary to some economists who’ve argued the highest returns in education come from pre-K and other early interventions, Hendren and Sprung-Keyser find that some higher ed interventions (like Pell Grants and related state funding for public universities in Texas) and K-12 programs (like programs to equalize funding between majority-nonwhite and majority-white schools) have infinite returns.
These estimates are often imprecise. In some cases, the confidence interval ranges from negative infinity to positive infinity: literally any effect size is possible, and the researchers can’t reject the possibility that the program either paid for itself or was actively destructive. But occasionally (as in the cases cited above), the confidence interval is positive infinity to positive infinity, meaning the evidence is pretty good that the intervention pays for itself.
Programs that cost more than initially thought
The researchers also find some negative estimates. Because they depress labor supply, programs like unemployment insurance, basic incomes (like the Alaska Permanent Fund dividend), and disability insurance cost the government more than $1 per $1 of benefits they provide to recipients. That doesn’t mean those programs are bad — just that they’re programs we have to pay for regularly, which are subject to normal financing requirements.
That’s what makes the positive findings so fascinating to me — they suggest that some programs face no long-run spending constraint at all, because they pay for themselves. They just seem hard to find.
One confounding thing about the paper is that many seemingly similar programs have very different estimated returns. A study of Pell Grants in Texas found precisely estimated infinite returns; a similar study on Pell Grants in Tennessee found finite and possibly negative returns. One national study of K-12 spending equalization found infinite returns; another looking at Michigan did not.
Sprung-Keyser told me that for this reason, they like to emphasize averages of different programs’ effects, not just individual programs’ evaluations. Overall, programs to help young adults enter and pay for college had, they estimated, an infinite payoff; the lower bound of the confidence interval was an MVPF of 4.5, meaning beneficiaries got $4.50 of benefit for every $1 the government spent. Either way, it looks pretty good. By contrast, programs that targeted parents of college students, like income tax deductions for tuition payments, got poor marks. Health insurance for children generally had an infinite estimated return and a low-end estimate of $25 in benefits for every $1 spent.
Still, I’d love to learn more about what, say, made the situation in Texas so different from Tennessee, or how Tennessee schools could change to attain the infinite returns of Texas public schools. Sprung-Keyser suggests that in that particular case, the Texas program might look better because we have better data. “The Texas Pell paper observes medium- and long-run earnings,” he told me, whereas in Tennessee, they only learned about high school and college enrollment and graduation rates, and “use a projection to find the impact on earnings there. In some ways, the Texas paper gives us a more comprehensive measure of the earnings impact.”
That would be incredibly promising if true, because it suggests that Pell Grants in general are incredibly high-impact, and more modest estimates are the result of measurement error. More generally, Sprung-Keyser says, the paper suggests we need bigger-scale interventions with better data to calculate more precise estimates of whether programs pay for themselves, or at least have a high return. “There’s a substantial value to switching to administrative data,” for instance, because of the much higher precision it allows compared to data collected from surveys.
While the paper generally concludes that cash transfers have minimal or negative returns compared to investments in health and education, I’d love to know if child allowances are different, for the same reason health and education investments in children are different. Sprung-Keyser says either a high or low result seems possible; we just need more research.
In the meantime, one lesson I take from the paper is that Medicare for All Children would be a pretty no-brainer step toward universal health care, if the US wants to move in that direction. Unlike normal Medicare, which ensures expensive older people, Medicare for kids would insure youngsters, who are usually quite inexpensive to cover in the short run. Despite that low cost, providing them with free access to preventive care seems to yield really massive benefits over the long run.
Those long-run benefits probably won’t convince legislators who won’t be in office long enough to brag about them. But they suggest that farsighted states should be spending more on kids’ health insurance than they are now.
Sign up for the Future Perfect newsletter. Twice a week, you’ll get a roundup of ideas and solutions for tackling our biggest challenges: improving public health, decreasing human and animal suffering, easing catastrophic risks, and — to put it simply — getting better at doing good.