Fewer than 20 children in a class is an expensive luxury. Kids don't learn more in small classes

When children go back to school after the summer break, they will often find fewer than 20 classmates waiting for them. Classes of 15 are common, even in cities. I myself treated class size as an important criterion when choosing a school for our four children: it makes intuitive sense that in a small class the teacher has more time for each child. Dozens of scientific studies have examined this intuition. The result? The number of children in a class has a negligible effect on their education.

Not all the studies agree. There are 62 of them, and together they report 1,767 different estimates. Moreover, published studies need not reflect the actual scientific findings impartially: because of publication bias, results that are interesting or consistent with the prevailing theory get published more often. The studies also differ in methodological quality.

The statistical method for summarizing the results of studies on the same topic is called meta-analysis. It corrects the literature for publication bias and takes differences in quality into account. It originally comes from medicine, but it can also be used, for instance, to compute the optimal rate of inflation or the ideal tax on CO2 emissions, or to assess the usefulness of investing in hedge funds.

Our research group specializes in meta-analysis and helps organize an annual workshop attended by experts from around the world. Besides developing the method, we aim for research that can be put directly to use in practice. The latest meta-analysis is unambiguous: changes in class size within the range of 15 to 35 pupils do not affect academic results, not even for minorities and children from poor families.

How do we judge which of the 62 studies are of good quality? We do not have to do much judging. The studies differ from one another in 42 respects (different data and methods), but a negligible effect of class size is the result common to all approaches. With a single exception: the experiment in the state of Tennessee, where in the 1980s elementary schools received $28 million to make some of their classes smaller.

But Tennessee was not a proper experiment of the kind Julie Chytilová and Michal Bauer of Charles University publish in Nature. The Tennessee experiment was not randomized. So there is no way to verify whether children and teachers were assigned to classes at random. More proactive parents may have pushed to get their children into the smaller classes. Principals may have handed out the smaller classes to better teachers as a reward.

A controlled experiment is unrealistic here. Many studies therefore use a clever trick, one of the things for which a Nobel Prize in economics was awarded: in many places, if 60 first-graders enroll, the school opens two classes. If 61 enroll, it opens three. Under ideal conditions this gives us a natural experiment in which the split into two classes (60 pupils) or three classes (61 pupils) does not depend on the abilities of the children or the teachers.

The method is called regression discontinuity, and new studies by Nobel laureate Angrist show the same thing as the meta-analysis. The other methods that try to tackle the "correlation is not causation" problem, instrumental variables and regression analysis with control variables, give the same results too: class size matters precious little for children's education. Why? The studies disagree, but apparently it is not easy to adapt teaching effectively to a smaller class.

Class size does matter to parents and teachers, however. Reducing class sizes is so popular a policy that since 2010 alone at least 17 countries and regions around the world have adopted it. This September, for instance, class sizes in New York will shrink. The justification: "Studies show that children learn faster in smaller classes." Behold, evidence-based policy in its postmodern form.

Smaller classes not only fail to help, they do harm in two areas. First, they drain money that could otherwise go toward teachers' pay. Cut class size from 30 children to 20 and you need 50 percent more teachers. The budgetary equivalent is a 50 percent pay raise. Second, when you need that many new teachers, a smaller share of them will be people who truly have their heart in the job. Fewer children get to experience a top teacher.

The bottom line: we do not need smaller classes and more teachers. We need the best teachers teaching larger classes and being paid top salaries. Let us start by raising the minimum permitted average number of children per class in a school from the current 17 (!) to at least 20. And let us give principals the money saved to pay bonuses to the best teachers.