Reading a book to a child, what the trials measure

Trials almost always concern a specific protocol, often the dialogic reading described by Grover Whitehurst in the late 1980s, where the adult asks open questions and prompts rather than reading straight through.

So it is not reading in general that is evaluated, but a way of reading, taught to adults through a short training. The distinction bears on how the results are interpreted.

Dialogic reading rests on a few precisely described techniques: inviting the child to complete a sentence, asking what they see, taking up their answer and expanding on it. That precision is what makes it testable, and what rules out extending the results to every form of reading.

Meta-analyses find a positive effect on young children's expressive vocabulary, of moderate size, clearer in preschool children than in older ones.

An older meta-analysis, published in 1995 by Adriana Bus, Marinus van IJzendoorn and Anthony Pellegrini, looked at the frequency of shared reading rather than at a protocol. It found an association with language, established largely on correlational data.

Effects on later reading skills are weaker and less consistent. Several syntheses conclude that the effect on phonological awareness, which strongly predicts entry into reading, is weak or null.

Blinding is impossible, and vocabulary is often measured on the words appearing in the books used, which brings the test close to what was taught and inflates the apparent effect.

The most rigorous trials, measuring vocabulary independent of the materials, find smaller effects. That gap is a textbook case of the problem of outcome measures too close to the intervention.

Several trials find larger effects in families where the practice was initially less frequent, which is expected since there is more room to move. That reading calls for a caution: a larger gain where the starting level is low may also reflect regression to the mean, a statistical phenomenon affecting any group selected on an initial score far from the average.

That result is cited in support of book distribution programmes. Evaluations of those programmes, notably in English-speaking countries, find clearer effects on reported reading frequency than on measured skills.

None of these data licenses presenting shared reading as a condition of language development. Children learn to speak in societies where the practice does not exist, as comparative work recalls.

What the trials establish is narrower: a particular way of interacting around a book raises, on average and moderately, vocabulary measured in the short term.