Causal Effects of Parent Number Talk on Preschoolers’ Number Knowledge

100 children (52 female), mean age 3.10 years (SD 0.43, range 2.54-4.36) and their parents

Two- to four-year-olds and their parents, stratified at randomization by baseline number-knower level (pre-knowers, 1-2-knowers, 3-4-knowers).

Child Development 91(6). Individually randomized three-arm trial in which the INSTRUCTIONAL CONTENT, not the instructor, is manipulated: parents read purpose-built picture books with their own child daily for four weeks, and the books differed only in whether they contained small number words (1-3), large number words (4-6), or non-numerical adjectives. Dose achieved was substantial and logged — a mean of 29.74 readings across 17.39 sessions. Outcome was the Give-N task (place N toy fish in a pond), a well-established laboratory measure of cardinal number knowledge but a researcher-administered, treatment-aligned one, scored as knower-level change from pre- to posttest. n = 100 with subgroup cells of 49 and 26; the authors themselves flag the subgroup analyses as needing cautious interpretation.

Parents delivered the instruction and it worked — but only when the content was pitched at the child's current level, which is the finding that generalizes. Children given Small Number (1-3) books gained 0.97 knower-levels against 0.23 for the non-numerical control, t(95) = 3.32, p = .001. Children given Large Number (4-6) books gained no more than controls overall, t(95) = 0.68, p = .500. The interaction is a clean expertise/readiness effect: among 1-2-knowers, Small Number books helped (1.13 vs 0.15 gain, t(44) = 2.51, p = .016) and Large Number books did nothing (t(44) = 0.08, p = .935); among 3-4-knowers, who had already mastered the small numbers, BOTH Small Number (t(21) = 2.85, p = .010) and Large Number books (t(21) = 2.73, p = .013) helped. The parent was a perfectly adequate delivery vehicle for a scripted, correctly-levelled input; the variable that decided the outcome was the match between content and child, not the instructor. What the study does not show is transfer or durability — no standardized maths outcome, no follow-up.

Low for the contrast, and this is the design's value: randomization was within family-and-child, stratified on baseline knower level, so the parent, the home, and the shared genes are held constant across arms and only the book content varies. The absolute gains are from a volunteer sample of parents willing to read daily for a month and should not be read as population averages.

No independent replication located. The readiness-matching pattern is a home-delivery instance of the expertise-reversal / prior-knowledge boundary this archive already records for worked examples and guidance (Kalyuga; van Gog & Sweller). The authors flag the small cells (n = 26 for 3-4-knowers), the unmeasured number talk outside the books, unknown generalizability to commercial books, and the absence of any long-term maths outcome.