Everyone loves early-childhood education. It provides free day care, which helps working parents, who vote. Boosters say that it can also reduce inequality, help disadvantaged kids get ahead, and . who knows, probably cure shingles. What's not to like? If we had strong evidence that early-childhood education made a big difference in the lives of needy kids, I'd be for it. Even hard-bitten libertarians can acknowledge that children are a special case.
But I'd like better evidence. A lot of the advocacy ends up being based on small studies that showed big results, while larger and better-designed studies are more equivocal. In fact, two recent studies -- one from Quebec, and one from Tennessee-- suggest that the long-term results might actually be slightly negative.
I wouldn't bet too much money on those negative results actually being true, of course; for one thing, I wouldn't bet money on the results of any single study being true, and also, the negative effects were relatively small. But when the largest and best-designed studies consistently show results that range from small benefits to small negative results, it starts to look less likely that early-childhood interventions deliver large benefits, and more likely that they don't do much of anything. Even if they make no difference, statistical noise would make some studies show a small benefit, others a small loss. But that would all be random variation around a mean benefit of zero.
In other words, I think it's probably not true that sending kids to preschool made them do a tiny bit worse than the kids who didn't go, which is what a crude reading of the Tennessee study would suggest. On the other hand, I think it's quite likely to be true that, as the Tennessee study shows, kids who went to preschool showed up to kindergarten better prepared than peers who didn't. But these results faded over time so that by third grade, the kids who had been to preschool had no advantage over their peers. This mirrors the result of other studies on large-scale preschool programs: There's an immediate benefit that fades, leaving little long-term impact, at least on academic performance.
There are two obvious rejoinders to this, and James Heckman, the leading academic advocate of early-childhood education, made both of them in his response to the Tennessee study. The first is that in this field even randomized controlled studies often aren't. Parents are more likely to drop out of the treatment group (which requires them to do something), then the control group, where all they have to do is . whatever it was they were going to do anyway. And the second is that the quality of programs can vary. Tennessee's early- childhood program may not do any good, but that doesn't mean that a better program wouldn't.
I quail to quarrel with James Heckman, who is about a million times smarter than me and has forgotten more about education policy than I will ever learn. But I am still a skeptic about the kinds of mass programs for early childhood that seem to be gaining currency among Democratic policymakers.
It's absolutely true that Tennessee failed to get parental consent for many of the children initially chosen for the treatment group, and had trouble arranging evaluation for the controls in a timely manner (because they were not all conveniently gathered at the preschool). This could well have biased their results. On the other hand, these problems plague other studies that are widely cited, such as Perry and Abecedarian (and indeed, Heckman has done good work correcting some of the problems with Perry). This is simply an inherent problem with studying early-childhood education: It requires the cooperation of parents, and the sort of parents who don't cooperate may be systematically different from the parents who do, in ways that bias most studies you try to do.
It's also true that early-childhood education probably varies in quality. After a discouraging study, defenders fairly say: "You studied a low-quality program, but we don't want to implement a low-quality program. We want to implement a good one!" Sure, but presumably that's what the people behind the studied program wanted to. Why was it not good? I suspect it's not a coincidence that programs derided as "low quality" tend to be very large. In other words that, whatever the benefits of small intensive programs designed by top-flight researchers, that is not the sort of program you get when you kick a vast government bureaucracy into gear.
Introducing a high-quality preschool program requires overcoming a series of obstacles: the political process, budget constraints, pushback from existing bureaucratic stakeholders who are somewhere between disinterested in and threatened by your new plan, recruitment and managerial control of thousands of workers delivering services to millions of children. At each obstacle, your program can die, or it can be altered to solve some immediate problem, at the expense of long-term effectiveness. So it seems quite possible that even if high- quality programs work, they are difficult-to-impossible to implement on any substantial scale.
This doesn't mean I fault the advocates of high-quality programs for wanting them; they sincerely believe that if they could get one through the process, it would do immensely good things for the world. But we also need to keep an eye on the potential costs. What are the odds that we will end up with a high-quality program, versus the odds that we will end up with a low-quality program that spends a lot of money to do no good? What are the odds that such a low-quality program will develop entrenched stakeholders, including the program's employees, who will block any attempt to implement a better program in the future? Given the experience of Head Start, the odds of failure seem higher to me than the odds of success.
That's why the correct approach is to start small, study what you've done, and scale up gradually. Start with a group of small towns. If those show promise, try the programs in bigger towns. Then counties. Then big cities, or groups of counties, doing randomized controlled trials at every stage to see whether your program is still better than doing nothing at all. Have a 20- year comprehensive plan to see if we can create an early- childhood-education program that works, with money already committed, but the next round disbursed only when the previous round is showing positive results.
Unfortunately, while it's easy to imagine how the research should be done, it's hard to imagine any politician proposing a program like that: no immediate benefits for constituents, and a strong possibility that the research will come back saying "nope, we can't seem to make this work." Far better for politicians to claim we know it will work, fire up the government spending machine, and hope like heck that they're right.
Megan McArdle is a Bloomberg View columnist who writes on economics, business and public policy.