Summer learning loss

Why do identical school holidays leave some children further ahead and others further behind?

The holiday is the most equal thing a school does. Every child in the class stops on the same day and returns on the same day; the school gives nobody anything in between. If treatment is identical, any difference in September must have been made by something else — which means the summer is the one period of the year that reports, cleanly, on what happens when school is not happening.

That is a strange and useful property. Term time confounds everything: home and school act at once and you cannot tell which did what. The holiday switches one of them off. So what does the switched-off period actually show?

Start by sizing the thing school is competing with, because the arithmetic is more startling than any finding.

Take a school year of a hundred and eighty days at six and a half hours of instruction — a US-style calendar; England's is nearer a hundred and ninety days. That is 1,170 hours. A child's waking year, at fifteen hours a day, is 365 times 15, or 5,475 hours. So instruction is 1,170 of 5,475: about twenty-one per cent of waking life. Roughly four-fifths of a child's year is not school, and about 1,155 of those hours — eleven weeks at fifteen waking hours a day — fall in one continuous block.

Now the inference. If children differ in what their out-of-school hours supply, and school hours supply everyone the same, then school is a compressor: the period when everybody is treated alike. The rest of the year is where variation lives, and the summer is that variation running uninterrupted. Entwisle and Alexander named this the faucet theory from their Baltimore cohort study: the resource tap is on for everyone during term, and in summer it runs only in some homes.

That model makes a sharp, checkable prediction, and it is different from the one people usually attribute to it. It does not say poorer children forget more. It says rates diverge in summer and converge in term — so a gap that grows should grow disproportionately in the summer months, and the school year should, if anything, close it. Which invites the obvious challenge: how would you measure a rate that changes with the season?

You would need a test administered at the true start and true end of each term, on a scale where a point means the same amount of learning at every level. Both requirements are harder than they sound, and this is where the literature gets into difficulty.

The famous synthesis is Cooper and colleagues' 1996 meta-analysis, which reported an average summer loss on the order of a month of grade-equivalent learning, larger in mathematics than reading, and larger for lower-income children in reading. That result is quoted everywhere. It is also more fragile than its fame implies, and the reasons are worth naming rather than waving at.

First, autumn tests are not sat on the first day of term and spring tests are not sat on the last, so weeks of instruction get silently counted as "summer". Second, the vertical scales that let you subtract a spring score from an autumn one are model-dependent, and different scaling choices can turn a widening gap into a stable one. Von Hippel and Hamrock re-examined seasonal data with these problems in view and found summer gap-widening far less consistent than the classic account — in some datasets gaps widened during the school year instead.

So what should we actually believe? That the arithmetic of exposure is solid: school is a fifth of waking time and cannot be the main source of variation between children. That the faucet logic — equal treatment compresses, unequal non-treatment spreads — follows from that arithmetic and is hard to argue with. And that the specific empirical claim, "gaps widen measurably every summer by about a month", is weaker than its circulation suggests and is partly an artefact of when tests are sat and how they are scaled.

vocabulary-gap-widening.md is the close neighbour, and the fixed point of difference is where the inequality is located. There the gap widens under identical exposure, because growth is multiplicative and equal proportional rates stretch an absolute difference on their own. Here exposure itself is unequal, and only during a defined period, which is why the seasonal design exists at all. The two mechanisms are additive, not rivals: if summer supplies unequal input to a process that already compounds, the compounding does the rest of the work in term.

One misreading to head off. "Summer learning loss" invites a picture of forgetting — knowledge decaying in the heat. Decay is real for facts held without use, but it is not what the seasonal comparison measures. The comparison is of rates: a child who does no reading in July has not necessarily lost words, but has not gained the words eleven weeks could have supplied, and against a peer who did gain them the effect on a gap is identical. The distinction decides the remedy: forgetting calls for revision, missing input calls for input.

• Whether year-round calendars, which break the eleven weeks into shorter blocks, change gaps or merely redistribute them.
• How vertical test scaling is constructed, and why a change in that construction can reverse a seasonal finding.

Every term the collection defines is gathered in the glossary.