The inherence bias in preschoolers' explanations for achievement differences: replication and extension

The study investigates whether preschool children show an inherence bias when explaining achievement differences—tending to attribute outcomes to inherent, person-based factors (e.g., intelligence, personality, effort) rather than extrinsic, contextual factors (e.g., teacher behavior, family background, access to resources). This bias is a generalization of the correspondence bias and has been linked to essentialist thinking and acceptance of social inequalities. Cognitive mechanisms (attentional focus on actors, memory accessibility, and working-memory constraints) and Western cultural norms emphasizing individual agency may promote inherent explanations. Because such explanations can legitimize status disparities and potentially harm the self-concepts and self-efficacy of marginalized children, the authors aimed to replicate and extend prior findings with a larger sample, improved methodology, and broader achievement contexts. Study 1 tested replication of inherence-biased explanations for classroom oral participation and examined competence/warmth judgments and ancillary hypotheses about social-class-linked participation perceptions. Study 2 extended to explanations of math performance differences, including both doing better and worse than peers, and elicited explanations about the self versus fictional peers.

Prior work documents an inherence bias in explanations across domains (social conventions, history, language, natural phenomena) and ties it to psychological essentialism. The correspondence bias literature shows a tendency to infer dispositions from behavior. Cultural psychology suggests Western contexts emphasize individual agency, amplifying inherent explanations. Developmental research indicates children often use inherent traits to explain group differences, which in turn increases perceived fairness and acceptance of inequality. Educational research shows early social-class disparities in classroom participation, with working-class children contributing less and for shorter durations in whole-class discussions. Earlier work (Goudeau et al.) found preschoolers explained participation differences via inherent factors and evaluated higher contributors as more competent and warm, potentially legitimizing social-class-based achievement gaps.

Two studies with French preschoolers in Grande-Section (kindergarten-equivalent) were conducted individually in quiet rooms; sessions were audio-recorded and transcribed. Ethics approval was obtained (CER-TP n°2021-10-01). Data, materials, and code are available on OSF. Study 1 (preregistered): N=306 (142 girls, 164 boys; Mage=5.6, range 4.9–6.6) from 27 classrooms in a largely ethnically homogeneous region. Social class was proxied by the Social Position Index (mean 109, SD 26.1); median split yielded working-class (N=150) vs middle/upper-class (N=156). Power sensitivity indicated 80% power for χ² ω ≥ 0.16 and one-sample t d ≥ 0.16. Children heard two hypothetical scenarios about whole-class discussions: (1) frequency (teacher calls on target more often) and (2) duration (target talks longer). Order and protagonist gender were counterbalanced. Measures: (a) Open-ended explanations coded by two independent raters into inherent, extrinsic, incoherent/irrelevant, or no explanation; high inter-rater reliability (Scenario 1: 96% agreement, κ=0.92; Scenario 2: 94% agreement, κ=0.86; excluding no-response: Scenario 1 κ=0.82; Scenario 2 κ=0.52). (b) Competence and warmth ratings on 1–5 scales with explicit midpoint (3): intelligence, academic achievement (competence), niceness, and teacher’s liking (warmth). Three control items probed response biases: “is mean/bad,” “is stupid,” and “likes strawberries.” Order of seven items was counterbalanced. Additional ancillary measures: children’s perceived similarity to protagonist (self-comparison, 1–5) and identification of a classmate similar to the protagonist; classmates’ social class determined via median split. Analyses used R (v4.0.3; jmv v2.3.4). Study 2 (extension): N=304 (144 girls, 160 boys; Mage=5.7, range 5.05–6.7) from 25 classrooms in the same region (Social Position Index M=104, SD=29.2). Stimuli presented with PsychoPy (2021.2.3). Children provided explanations for two math scenarios about themselves: (1) downward comparison (child does better than a classmate) and (2) upward comparison (child does worse). Fixed order (success first) minimized negative affect. Open-ended responses were coded independently for inherent and extrinsic content using non–mutually exclusive codes (to accommodate longer, more complex answers); high inter-rater reliability: Downward-inherent 99% (κ=0.94), Downward-extrinsic 97% (κ=0.90), Upward-inherent 96% (κ=0.84), Upward-extrinsic 95% (κ=0.84). Non-response rates were higher than Study 1 (41.1% downward; 37.2% upward), possibly due to limited classroom comparative feedback practices. Analyses used McNemar’s tests to compare inherent vs extrinsic mention rates; examined social class and gender predictors.

Study 1: - Open-ended explanations favored inherent over extrinsic factors. • Scenario 1 (frequency): χ²(1, N=203)=106.45, p<0.001; no effects of child social class (p=0.71) or gender (p=0.93). • Scenario 2 (duration): χ²(1, N=232)=179.38, p<0.001; no effects of social class (p=0.52) or gender (p=0.71). - Competence and warmth evaluations of the high-participation peer were above the midpoint. • Scenario 1 means (1–5 scale, midpoint=3): intelligence M=3.61, t=8.03, p<0.001, d=0.46; academic achievement M=3.58, t=7.93, p<0.001, d=0.45; niceness M=3.59, t=7.85, p<0.001, d=0.45; teacher liking M=3.51, t=7.35, p<0.001, d=0.42. • Control items: “bad” M=2.37, t=−8.45, p<0.001, d=−0.48; “stupid” M=2.46, t=−6.81, p<0.001, d=−0.39; “likes strawberries” M=3.57, t=7.24, p<0.001, d=0.41. • Scenario 2 means: intelligence M=3.27, t=3.37, p<0.001, d=0.19; academic M=3.27, t=3.52, p<0.001, d=0.20; niceness M=3.30, t=3.78, p<0.001, d=0.22; teacher liking M=3.18, t=2.35, p=0.019, d=0.13. Controls: “bad” M=2.52, t=−5.87, p<0.001, d=−0.34; “stupid” M=2.65, t=−4.54, p<0.001, d=−0.26; “likes strawberries” M=3.26, t=3.26, p=0.001, d=0.19. • To address possible halo/response bias (strawberry item), regressions centered at midpoint showed positive intercepts for all competence/warmth judgments in Scenario 1 (ps<0.001) and for three of four in Scenario 2 (ps<0.031); exception: “liked by the teacher” in Scenario 2 (b=0.09, SE=0.07, p=0.24). - Ancillary Hypothesis 3 (self-similarity by social class): not supported. • Scenario 1: working-class M=2.96 (SD 1.14) vs middle/upper M=2.88 (SD 1.15), t=0.57, p=0.57, d=0.07. • Scenario 2: working-class M=2.87 (SD 1.15) vs middle/upper M=2.91 (SD 1.08), t=−0.29, p=0.77, d=−0.03. - Ancillary Hypothesis 4 (peer-nominated classmates’ social class): mixed. • Scenario 1: among 93 nominators (30.4%), working-class classmates less often identified, χ²(1, N=93)=5.68, p=0.017 (37.6% working-class; 62.4% middle/upper). • Scenario 2: among 126 nominators (41.2%), no significant difference, χ²(1, N=126)=0.51, p=0.48 (46.8% working-class; 53.2% middle/upper). Study 2: - Children more often invoked inherent than extrinsic explanations for both success and failure involving themselves. • Downward comparison (did better): inherent n=158 vs extrinsic n=30, McNemar χ²(1)=103.39, p<0.001; social class ns (ps>0.18); girls mentioned more inherent explanations (p=0.032), extrinsic ns (p=0.86). • Upward comparison (did worse): inherent n=161 vs extrinsic n=35, McNemar χ²(1)=94.13, p<0.001; social class ns (ps>0.35); gender ns (ps>0.86). - Overall, the inherence bias generalized across positive/negative outcomes, beyond oral participation to math, and to explanations about the self, using varied question wording.

The studies robustly show that preschoolers prefer inherent explanations over extrinsic ones for achievement differences. This pattern replicated prior findings on oral participation and generalized to math performance and to self-referential explanations for both better- and worse-than-peer outcomes. Children also evaluated high contributors as more competent and warm, and these effects largely persisted after adjusting for response biases. The results support theoretical accounts of an inherence bias in early explanations and illuminate mechanisms by which educational inequalities may be legitimized in children’s minds. Ancillary analyses suggested that while children’s self-perceptions did not differ by social class, peers were more likely to identify middle/upper-class classmates as frequent contributors, aligning with documented participation disparities. These findings underscore how explanatory tendencies can shape perceptions of ability and social warmth, potentially reinforcing class-based disparities.

This work provides robust evidence that preschoolers exhibit an inherence bias when explaining achievement differences, extending prior research with larger samples, improved measures, and broader contexts (math, self-focused outcomes, positive and negative comparisons). By linking this bias to perceptions of competence and warmth, the study contributes to theories of the early construction and legitimization of educational inequalities. Future research should examine the causal consequences of children’s explanations for motivation, self-efficacy, and performance; investigate why children’s self-reports may under-detect participation disparities; and test interventions that prompt structural or extrinsic explanations to mitigate inequity-amplifying effects.

Key limitations include: (1) consequences of explanatory style were not directly tested; (2) potential low-level response or halo biases on evaluation items (e.g., “likes strawberries”), though adjusted analyses largely upheld main conclusions; (3) Study 1 scenarios did not include low-participation cases (addressed in Study 2 with better/worse math scenarios); (4) higher non-response rates in Study 2, possibly due to limited comparative feedback in classrooms; (5) sample drawn from a mostly ethnically homogeneous French region and ethnicity data were not collected due to legal constraints, limiting generalizability; (6) social class operationalized via Social Position Index and median split for some analyses; (7) some inter-rater reliability coefficients (e.g., Scenario 2 coding excluding no-response) were lower though acceptable; (8) vignette-based and self-report measures may not fully capture classroom realities.