Children transition from simple associations to explicitly reasoned social learning strategies between age four and eight

To differentiate the use of simple associations from use of explicitly reasoned selective social learning, we can look for age-related changes in children’s behaviour that might signify a switch from one social learning strategy to the other. We presented 4- to 8-year-old children visiting a zoo in Scotland (N = 109) with a task in which the perceptual access of two informants was determined by the differing opacity of two screens of similar visual appearance during a hiding event. Initially success could be achieved by forming an association or inferring a rule based on salient visual (but causally irrelevant) cues. However, following a switch in the scenario, success required explicit reasoning about informants’ potential to provide valuable information based on their perceptual access. Following the switch, older children were more likely to select a knowledgeable informant. This suggests that some younger children who succeeded in the pre-switch trials had inferred rules or formed associations based on superficial, yet salient, visual cues, whereas older children made the link between perceptual access and the potential to inform. This late development and apparent cognitive challenge are consistent with proposals that such capacities are linked to the distinctiveness of human cumulative culture.

Alternative approaches in the pre-switch phase A GLMM was built for selection of the knowledgeable informant in each switch trial with fixed effects of age, whether five or more pre-switch trials were required to reach criterion, the presence of screens, and all the interactions between these variables, with a random intercept of Table S1. Effect of reaching the proficiency criterion on switch trial performance. Table S2. Age-effect of reaching the proficiency criterion on switch trial performance.
participant ID. This model was significantly better than the null equivalent (χ 2 (7) = 25.6, p < .001). A significant main effect of age (p = .004) indicated that older children selected the knowledgeable informant more often than younger children. A significant main effect of the number of trials required to meet criterion (p = .002) suggested that children who met criterion after five trials selected the knowledgeable informant more often than children who required more trials. Again, there was no evidence of an effect of the presence of screens (p = .648) or any interactions between the fixed effects (p ≥ .074). For the full model output see Table S3. One-sample t-tests revealed that children who met criterion in five trials selected the knowledgeable informant significantly above chance, in an average of 4.59 switch trials t(33) = 20.0, p < .001, two-tailed. Children who required more trials to reach criterion also selected the knowledgeable informant significantly more often than chance in an average of 3.56 trials t(17) = 3.24, p = .005, two-tailed. A one-way ANOVA comparing the performance of these groups to children who did not reach criterion revealed a significant difference between the three groups F(2, 106) = 24.15, p < .001 (Fig. 3). Post hoc Tukey HSD tests showed that children who met criterion in five trials (M = 4.59 trials, SD = 0.61, N = 34) selected the knowledgeable informant significantly more often than both children who required more trials (M = 3.56 trials, SD = 1.38, N = 18, p = .006) and children who did not reach criterion (M = 2.89 trials, SD = 1.26, N = 57; p < .001). The difference between the latter two groups was not significant (p = .08). Table S3. Effect of the number of pre-switch trials required to reach criterion on switch trial performance.

Explicit verbal reasoning
We investigated whether verbal reasoning was predicted by age, meeting the proficiency criterion, or switch trial performance by conducting ordinal regressions (using the ordinal package 53 in R). Reasoning responses were submitted to an ordinal regression with fixed effects of age, meeting criterion, and the interaction between these variables. Significant main effects of age (p < .001) and meeting criterion (p < .001; Fig. S1) suggested that older children and children who met criterion provided better reasoned responses than younger children and children who did not meet criterion, respectively. There was no evidence of a significant interaction between age and meeting criterion (p = .305). For the full model output see Table S4.  A further ordinal regression explored the relationship between switch trial performance and verbal reasoning. This model included fixed effects of age, meeting criterion, switch trial performance, and the interactions between these variables. A significant main effect of switch trial performance (p < .001) indicated that children who made more knowledgeable informant selections also provided better reasoned responses (Fig. 4). There were no significant main effects of age (p = .287) or meeting criterion (p = .133). However, a significant interaction between switch trial performance and age (p = .032) was likely, at least in part, the product of older children's greater switch trial performance. For full model output see Table S5.The effect of switch trial performance appeared to be enhanced by age. That is, given the absence of a main effect of age, this interaction suggests that older children who performed well in the switch trials were more likely to provide a better reasoned response than younger children who also performed well (

Reasons for not reaching the proficiency criterion
We also explored the possibility that the 52% of children who did not meet criterion were employing alternative yet consistent strategies regarding informant selections. There were a number of potential strategies that could have interfered with children's formation of an association, or inference of a rule, in the number of trials available to them. For example, children may have preferred a particular informant, they may have alternated their choices between the sides, or they may not have used any strategy at all. A chi-square test of independence showed no evidence of an overall preference for either informant (p = .675) or of any particular age group (years) having a preference for either informant (p = .717). To explore whether children had any preferences for copying or not copying the side they selected on the previous trial we conducted further chi-square tests of independence. These revealed no evidence of preference for either copying or not copying the side selected on the previous trial (p = .895) and no evidence of that any age group preferred one of these choices (p = .533). These results suggest that it is unlikely that children who did not reach criterion were using a consistent alternative strategy for informant selections.
Of the 34 children who did not reach criterion, 12 provided reasoned correct verbal justifications suggesting that had there been more trials in which to reach criterion they may have done so. One child's justification appeared to be drawn from the final two trials without screens ("at the end I knew it could be stripes because spots was turned away and couldn't see it"). Nine children gave reasoned but incorrect verbal justifications. The strategies indicated by these children included references to patterns and turn taking (e.g., "I found a pattern, it switched from Spots to Stripes from Spots to Stripes", "because I was taking turns").