Fewer Children, Better Lessons?

AFew claims about schooling win agreement as quickly as the belief that children learn more when there are fewer of them in the room. Parents will pay for it, ministers promise it, and teachers asked what they would buy with an extra pound name it more often than anything else. The evidence, however, has proved unusually hard to assemble, because classes are not filled at random. Schools that keep their classes small tend to serve families who could have bought the same advantage elsewhere, so any gain may belong to the household rather than to the room. Worse, in ordinary schools free to organise as they wish, the smallest groups often hold the pupils who are already struggling, since head teachers use those groups for extra support; comparisons that ignore this can make small classes appear to do harm. Only one design removes the choosing altogether, and that is allocation by lottery.

BThe Kesterfield experiment, which ran from 1987 to 1991 in seventy-nine primary schools, did exactly that. Some 6,300 children entering their first year of school were allocated by lot either to a class of about fifteen or to one of about twenty-three, and the same procedure was applied to the adults: no head teacher chose which colleague taught the smaller room. Pupils remained in the condition to which they had been assigned for four years. By the end of the first year the children in the smaller classes were ahead in reading by 0.18 of a standard deviation, a lead equivalent to about three months of additional progress, with a similar though slightly narrower gap in mathematics. The pattern of the following years is the part most often left out of the retelling. Almost all of the advantage was in place by the end of the second year; over the remaining two years the gap between the two groups neither widened nor closed.

CMarguerite Devlin, an education economist at the Corbin Institute, has followed the Kesterfield cohort into adulthood. In the 2004 follow-up, entry to university among the small-class children exceeded that of their peers by 2.7 percentage points, and that average conceals two very different groups: for children who had been eligible for free school meals the margin was 4.9 points, against 0.9 points for the rest. Devlin states a boundary far more often than those who quote her repeat it. Every one of these gains belongs to children placed in a small class in their first year of school; a smaller group admitted to the trial in its third year showed no measurable gain at all, and she has written twice that the finding should be described as an effect of small classes early, not of small classes. She attributes the gain partly to a fall in interruptions; how much of a teacher's attention each child received was never recorded.

DHenrik Osgood, a fiscal analyst at the Faulkner School of Public Policy, does not dispute the trial result and says so at the outset. His objection is that an experiment and a policy are different objects. Taking classes from twenty-three down to fifteen requires roughly fifty per cent more classroom teachers, and those teachers must come from somewhere. When the jurisdiction of Ferrand capped every primary class at twenty in 1996, the share of its primary teachers without a full qualification rose from 2 per cent to 12 per cent within two years, and the least qualified recruits were not spread evenly: two thirds of them took posts in the districts with the highest poverty. The measure that reaches the disadvantaged child first in a trial, he argues, reaches that child last in a rollout. Put to him directly, he concedes two things: the Ferrand decline cannot be separated from a hiring boom that was drawing graduates out of teaching in the same years, and his cost figure prices every additional post at the average salary, which he agrees overstates it, since new entrants are paid less.

EGovernments have not waited for the argument to end. Since 2010 at least eleven jurisdictions have announced class-size targets, and the announcements travel a good deal further than the results. Much of the confusion is arithmetical. The pupil-teacher ratio and the average class size are different measures, and a ministry that counts every specialist, librarian and part-time coach alongside those who teach a class of their own can report a ratio of fourteen while the ordinary lesson holds twenty-seven. Figures produced by the two methods are then set side by side as though they described one thing. Neither Devlin nor Osgood defends the practice. Both name it as the most misleading habit in the field, and both have argued that any figure of this kind ought to be published together with the rule that generated it. It is the rare point on which the two agree without qualification.

FWhat is emerging is a division of labour rather than a verdict. In the earliest years, before children read independently, small classes have no rival, because the teacher must hear every child read aloud in turn, a task whose cost rises in direct proportion to the number of children present. Where the aim is broader, such as raising attainment across a whole secondary curriculum, sustained coaching of the teachers already employed delivers more for each unit spent, and neither researcher disputes it. Two questions remain beyond the reach of the design that settled the first one. Nobody knows what teachers do differently in a smaller room, since classroom observation formed no part of the Kesterfield protocol. And a policy applied everywhere changes who chooses to enter teaching at all, an effect that a trial confined to seventy-nine schools cannot see, however cleanly its pupils were allocated.