Young children expect and approve of teachers’ gender stereotypes about academic interest Authors: Marianna Y Zhang, Mika Asaba, Julia A Leonard Published: 2026 ABSTRACT AbstractGender stereotypes, such as the belief that boys are more interested than girls in engineering, contribute to gender disparities. Do children enter school already expecting teachers to hold these stereotypes, even without evidence that the teacher is biased? And do they view it as acceptable for teachers to act on such stereotypes? In Experiment 1 (1a: n = 145, 1b: n = 207), 5- to 7-year-old U.S. children predicted that teachers meeting their students for the first time would assign engineering activities to boys and reading activities to girls, even when children knew that the students liked both activities equally. In Experiment 2 (2a: n = 90; 2b: n = 132), children approved of teachers assigning engineering activities to boys and reading activities to girls, and rated these assignments more favorably than assignments counter to gender-interest stereotypes. Together, these findings show that children enter school expecting teachers to hold gender stereotypes about academic interests and view acting on these beliefs as acceptable, highlighting a potential early obstacle to educational equity.Subject terms: Human behaviour, Education Introduction From a young age, children in the United States are exposed to harmful social stereotypes about gender that contribute to inequalities in education and the workforce1–3. Gender stereotypes involve widely-held yet overly broad generalizations about each gender, such as the belief that men are better suited to science and engineering than women4. However, children’s awareness and evaluation of others’ stereotypes, even those that differ from their own beliefs, remains an open question. Here we ask, do young children expect teachers to hold gender stereotypes? And do they approve of teachers acting on such beliefs? Educational contexts provide a particularly powerful setting for exploring these questions, as teachers exert a strong influence on students’ educational beliefs and behaviors5,6. As a result, the mere expectation and approval of teachers’ biases may limit children’s learning and decision-making in the classroom, thereby reinforcing cycles of bias. Prior literature has yielded mixed results as to whether children expect adults to hold gender stereotypes. On the one hand, children believe their parents would disapprove of counter-stereotypical play (e.g., a girl playing with trucks)7, suggesting that children expect their parents to have gender stereotypes. On the other hand, children do not expect their teachers to believe that math performance varies by gender8,9. Yet other work has found that children have an ingroup bias, for example, believing that people think the child’s own gender is best at math10. Discrepant findings regarding children’s expectations of adults’ gender stereotypes may arise from two methodological choices. First, expectations of others’ stereotypes have typically been assessed by explicitly asking children about the beliefs of specific adults in their lives (e.g., their own parents or teachers) or of unspecified groups, such as “people” or “grownups”, rather than of novel individuals. As a result, children may respond by reasoning about particular adults they know, who may have divergent beliefs about gender stereotypes, leading to mixed responses. This methodological choice also leaves open the question of whether children hold a general assumption that adults more broadly hold gender stereotypes, or not. Second, prior work has not accounted for children’s own beliefs, raising the question of whether children are simply relying on their stereotypes to form expectations of others’ beliefs10. This design choice also raises the critical question of whether children can represent an adult as holding stereotypes, separately from the content of their own beliefs. We theorize that, when asked about novel individuals, rather than specific adults or unspecified groups, young children will show a generalized expectation that adults hold gender stereotypes. Specifically, given that gender stereotypes are widely-held beliefs among adults, such beliefs may “leak out” in behavior observable to children. For example, on average, parents themselves believe that boys are more interested in STEM than girls3, and in turn, engage in more challenging conversations about science and provide more STEM materials and activities to their sons than to their daughters3,11. Children may notice such patterns and form a generalized expectation that adults believe boys are more interested than girls in STEM. Additionally, research on the theory of mind indicates that by age 5, children understand that others can hold beliefs that differ from their own12. This capacity raises the possibility that children may also represent adults as holding beliefs in gender stereotypes, even when they do not share those beliefs themselves. Children’s expectations of gender stereotypes could have either positive or negative downstream consequences. On the one hand, if children can represent adults’ stereotypes as distinct from their own beliefs, they may also recognize those stereotypes to be inaccurate, which could enable them to challenge such beliefs13. Furthermore, prior work shows that young children negatively evaluate social exclusion based on gender14,15, suggesting that they may disapprove of adults acting on stereotyped beliefs when those practices result in exclusion of one gender from equal opportunities. On the other hand, due to children’s descriptive-to-prescriptive reasoning16,17, expecting adults to hold gender stereotypes may lead children to view stereotype-based practices as acceptable and justifiable. Consistent with this possibility, children tend to negatively evaluate gender nonconforming behavior18, suggesting that they may similarly approve of practices based on gender stereotypes. Such validation may legitimize these practices and reinforce the gender stereotypes that fuel them. Here, we investigate children’s expectations (Experiment 1) and evaluations (Experiment 2) of teachers’ gender stereotypes. We specifically focus on gender stereotypes about interest in engineering vs. reading, given that these activities are educationally relevant and stereotyped early, with interest in engineering stereotyped in favor of boys19 and interest in reading stereotyped in favor of girls20. To this end, we focus on children ages 5–7 in the United States, capturing the earliest ages at which children have been documented to hold gender stereotypes about engineering interest19 to see if variation in children’s own stereotypes relate to their expectations of others’ stereotypes. In Experiment 1, 5–7-year-old children predicted whether a novel teacher would give an engineering or a story activity to a boy or a girl student. Critically, participants always knew that the student in question was more interested in the counter-stereotypical activity (e.g., a girl student preferred engineering; Experiment 1a), or was interested in both activities equally (Experiment 1b), while the teacher was either knowledgeable or ignorant to the student’s true interests. This design thus allows us to pit children’s knowledge of students’ interests against their beliefs of what a teacher might presume based on gender. Importantly, this design overcomes the two key limitations of prior work. First, this paradigm assesses children’s generalized expectations by eliciting inferences about novel teachers rather than familiar or unspecified adults. Second, by varying whether participants and the teachers have access to the same information, this paradigm tests whether children can represent teachers’ stereotypical beliefs separately from their own. In Experiment 2, we tested whether children approve of teachers acting on gender stereotypes in a classroom, thus reinforcing cycles of bias. Specifically, in Experiment 2a, b, children were asked to evaluate the behavior of teachers who assign activities to students based on gender stereotypes (i.e., only boys to engineering activities and only girls to story activities), egalitarianism (i.e., assigns activities equally to both genders) or counter to stereotypes (i.e., only girls to engineering activities and only boys to story activities; Experiment 2b only). Given the pervasiveness of adult gender stereotypes in children’s environments, our primary preregistered prediction in Experiment 1 was that children would expect teachers to hold gender stereotypes. Specifically, we predicted that children would expect the teacher to choose activities based on gender stereotypes when the teacher was ignorant, but expect the teacher to choose activities aligned with students’ interests when the teacher was knowledgeable. In addition, after observing a strong trend in pilot data, we also predicted that girl participants would show these effects more strongly than boy participants, though this pattern did not occur in Experiment 1a, and thus was dropped in Experiment 1b. In Experiment 2, we did not preregister directional predictions based on the strength of competing hypotheses for both positive and negative evaluations of stereotype-enacting behavior. To assess if children’s own interests guided their evaluations, we asked children about their own interests in engineering and reading. Based on results from Experiment 2a, in Experiment 2b we predicted that children with counter-stereotypical interests (e.g., girls who prefer engineering) might be less approving of educational assignments that perpetuate gender stereotypes, compared to egalitarian assignments. Finally, we also measured children’s own stereotypes across experiments. Based on prior work19, we predicted that children would hold gender stereotypes about engineering and reading interest (confirmatory in Experiment 2a, b). We also planned to explore whether these beliefs predicted children’s expectations and evaluations of others’ stereotypes. In particular, we were interested in exploring whether children who do not themselves endorse stereotypes still expect and approve of others’ stereotypes. Such a result would suggest that children’s expectations and evaluations may independently contribute to gender disparities in educational achievement, without the need for children to develop stereotypes themselves. Methods All studies reported received approval from the institutional review board at Yale University (HSC #2000031150). In all studies, informed consent was provided by participating families, assent was provided by participating children, and participants’ families received a $5 gift card for their participation. Example videos of the procedure in each study are available on OSF: https://osf.io/kr5sm/. Experiment 1 Experiment 1 investigated whether 5–7-year-old children expect teachers to rely on gender stereotypes when assigning activities to students. In Experiment 1a, we asked children to predict what activities teachers who were ignorant versus knowledgeable about their students’ interests would assign to boy and girl students who held counter-stereotypical interests. Importantly, expecting the ignorant teacher to select a stereotypical activity requires children to suppress their own knowledge of the student’s true, counter-stereotypical preference, which may have added additional demands on children’s mental state reasoning capacities. Although children in this age range can succeed on standard false-belief tasks, coordinating these competing pieces of information may nonetheless still make the prediction task more challenging21,22. To reduce these demands, in Experiment 1b, we focused on children’s predictions about an ignorant teacher assigning activities to students who were interested in both engineering and stories. This design choice allowed children to reason about the teacher’s beliefs without having to inhibit knowledge of a counter-stereotypical preference. Experiment 1a was run first, and preregistered on OSF on Apr 27, 2023: https://osf.io/3j2wa/. In Experiment 1b, given uncertainty about the youngest age at which children might show stereotype-consistent expectations, we preregistered and tested each age group sequentially, starting with 7-year-olds, then 6-year-olds, and finally, 5-year-olds. The preregistrations are identical except for the age of the participants collected: https://osf.io/68z54/(7-year-olds, September 6, 2023), https://osf.io/w84r9/(6-year-olds, November 1, 2023), and https://osf.io/wbxhn/(5-year-olds, February 21, 2024). There were no major deviations from the preregistrations. Participants Experiment 1a recruited 145 5–7-year-olds in the United States (Mage: 6.34 years, SDage: 0.83, range: 5.01–7.97) on Children Helping Science23. As this was the initial study, we did not conduct a power analysis. Parents reported participants’ gender (n = 75 boys, n = 70 girls) and their family’s race/ethnicity (n = 84 White, n = 25 Asian, n = 7 Hispanic or Latine, n = 3 Black, n = 3 Native Hawaiian or other Pacific Islander, and n = 23 multiracial or other categories). Parental education ranged from 12 to 20 years (M = 18.08, SD = 2.23; missing data from 2 parents), and parental median income was $100,000 (M = $100,708, SD = $48,114; missing data from 12 parents). An additional 5 participants were recruited but excluded for living outside the US. Experiment 1b recruited 207 5–7-year-olds in the United States on Children Helping Science. As mentioned, we preregistered and ran the 7-year-olds first, then 6-year-olds, then 5-year-olds: n = 81 5-year-old participants (Mage: 5.49, SDage: 0.29, range: 5.00–5.99), n = 51 6-year-old participants (Mage: 6.45 years, SDage: 0.31, range: 6.08–6.98), n = 75 7-year-old participants (Mage: 7.47 years, SDage: 0.29, range: 7.06–7.97). After observing the 7-year-old data, the effect was larger than we expected, so we preregistered a smaller sample size for 6-year-olds. However, after observing the 6-year-old data, we returned to the more conservative effect size estimate when selecting the 5-year-old sample (see Supplementary Methods for Experiment 1b for the power analysis). Parents reported participants’ gender (n = 106 girls, n = 100 boys, n = 1 other) and their family’s race/ethnicity (n = 116 White, n = 25 Asian, n = 22 Hispanic or Latine, n = 14 Black, n = 1 Middle Eastern or North African, and n = 30 multiracial or other categories). Parental education ranged from 12 to 20 years (M = 18.04, SD = 2.36; missing data from 2 parents), and parental median income was $90,000 (M = $91,721, SD = $46,072). An additional 9 participants were recruited but excluded for failing the comprehension check questions (n = 2) or living outside the US (n = 2 5-year-olds; n = 1 6-year-old; n = 4 7-year-olds). Stimuli Each test trial showed images of an adult as the teacher and a child as the student. Teachers and students were depicted with pictures showing Black and White men and women, and Black and White boys and girls, respectively. Participants were shown test trials (four trials in Experiment 1a, two trials in Experiment 1b) with unique adult-child pairs (two women, two men; two girls, two boys; normed on age and emotional expressions). These images were overlaid on a cartoon classroom that showed a door and two computer activities (one with an engineering icon, and one with a story icon); see Fig. 1. In the stereotype trials, simple line drawings of a boy and a girl were presented, with a blue thumbs-up and a red thumbs-down icon as response options. Example of Experiment 1a Teacher Ignorant and Teacher Knowledgeable trials. This example shows a girl student (photographs of teacher and student replaced by silhouettes for publication; see OSF for actual stimuli). First, the student revealed their counter-stereotypical interest when the teacher is either absent (Teacher Ignorant) or present (Teacher Knowledgeable). Then, participants were told that the teacher chose a game for the student based on what they think the student likes. As our prediction measure, participants predicted which game the teacher chose: the engineering game or the story game. All participants underwent four trials that crossed the gender of the student (girl or boy) and the teacher’s knowledge of their interest (knowledgeable or ignorant). Experiment 1b was identical, except that the students said that they liked both the engineering and story games, and the teacher was always ignorant. Procedure Children participated remotely with their parents or legal guardians on Children Helping Science23, an asynchronous platform for recruiting and presenting self-paced, pre-recorded, narrated stimuli to children. Experiment 1a Participants were introduced to “Sunny School”, where teachers were meeting their students for the first time. The classrooms had two types of games: engineering games (where “students create or build things, like machines or buildings”; description drawn from Master et al.19) and story games (where “students make up or act out stories”; we used the term “story”, rather than “reading”, to parallel the active nature of engineering), each with their own icon representing the game (see Fig. 1). Participants completed several comprehension check questions to confirm their understanding of the cover story and two activities (see Supplementary Methods for Experiment 1a for all check questions and exclusion criteria). Participants then underwent 4 test trials that crossed the teacher’s knowledge of the student’s interest (Teacher Knowledgeable vs. Teacher Ignorant) with the gender of the student (boy vs. girl). In each trial, participants were introduced to a student who stated which game they preferred (e.g., “here is Nathan. Nathan is looking at the engineering game and the story game. Nathan says, “I really like stories!”). Notably, in Experiment 1a, these preferences were always counter-stereotypical; boy students preferred stories, while girl students preferred engineering. In the Teacher Knowledgeable trials, the student’s teacher entered the room before the student stated their interest, such that the teacher also heard the student’s interest. In the Teacher Ignorant trials, the teacher entered after the student stated their interest, such that the teacher did not hear it. Next, participants were told that the teacher had picked out a game for the student that they think the student will like more, with a pop-out image shown on screen, reminding the participant of the teacher’s presence or absence at the start of the trial. Additional comprehension checks ensured participants understood both the student’s preference and the teacher’s knowledge state. At the end of each trial, participants answered our critical prediction question: which game the teacher chose for the student (“which game do you think the teacher chose for Kelly?”; response options were to click on either the engineering or story icon). Trials were blocked by teacher knowledge (order counterbalanced), with student gender, student race (white vs. black), and teacher gender (male vs. female) counterbalanced within each block. Finally, we solicited participants’ gender interest stereotypes by asking how much boys and girls each like engineering and stories. Based on child-friendly gender stereotype questions from ref. 19, participants were shown a cartoon boy or girl along with the engineering or stories icon and were asked in a two-step format: “do most girls (boys) like engineering (stories) or do most girls (boys) not like engineering (stories)?”, then “how much do they (not) like it? (not) like a little, (not) like, or really (not) like?” Participants responded by clicking thumbs-up or thumbs-down icons, and responses were converted to a 6-point scale (“really not like” to “really like”). Question order (boys or girls) was counterbalanced across participants. Experiment 1b In Experiment 1b, the procedure was identical, except students were interested in both engineering and stories, and we ran the Teacher Ignorant trials only. Statistics All analyses were performed in R version 4.3.224. To test our main hypothesis in Experiment 1a, we preregistered a mixed-effect logistic regression (using the lme4 package25) to model participants’ predictions of which game the teacher assigned to each student (engineering = 1, or stories = 0), as a function of student gender, teacher knowledge, and their interaction, with random intercepts for participants to account for the within-subjects design. To unpack the predicted interaction in Experiment 1a between student gender and teacher knowledge, we preregistered separate mixed-effects logistic regressions within the teacher-knowledgeable trials and within the teacher-ignorant trials, predicting participants’ predictions as a function of student gender, with random intercepts for participants. In Experiment 1b, we preregistered a similar mixed-effect logistic regression for each age group with only the student gender term and random intercepts, as only the Teacher Ignorant trials were run, and age groups were tested one group at a time. In both experiments, we ran the main mixed-effects logistic regression separately for boy versus girl participants (confirmatory in Experiment 1a, exploratory in Experiment 1b). To see whether participants made selective predictions for each student's gender, we ran exploratory binomial tests comparing responses to chance (50%). To test our secondary gender hypothesis in Experiment 1a, we preregistered a mixed-effect logistic regression modeling participants’ predictions of which game the teacher assigned to each student (engineering = 1, or stories = 0) in the Teacher Ignorant trials, as a function of participant gender, student gender, and their interaction, with random slopes (for student gender) and intercepts per participant to account for the within-subjects design. To follow up on this model, we preregistered separate mixed-effects logistic regressions for boy versus girl participants, modeling boys’ versus girls’ predictions of which game the teacher assigned to each student (engineering = 1, or stories = 0) in the Teacher Ignorant trials, as a function of student gender, and their interaction, with random slopes (for student gender) and intercepts per participant. We also explored participants’ own stereotypes about interests, i.e., what participants believe girls versus boys are interested in. For these analyses, we collapsed across the age groups in Experiment 1b, and ran an exploratory mixed-effects linear regression predicting participants’ interest ratings as a function of the target gender (boy or girl), game domain (engineering or stories), their interaction, and participants’ age (exact), with random intercepts by participant. To explore whether the strength of participants’ stereotypes predicted their expectations, we formalized stereotype strength as two difference scores for each participant, an engineering “stereotype score” and a story “stereotype score”, indicating how much they believed boys like engineering more than stories, and how much they believed girls like stories more than engineering. We entered the relevant stereotype score into a logistic regression to predict the participant’s teacher prediction in the Teacher Ignorant trials for the boy and girl trials separately. All models converged without warnings. Stereotype ratings were not normally distributed, as observed using Shapiro–Wilk tests; linear models were used to estimate mean differences and are robust to non-normality with large sample sizes. The statistical significance of parameter estimates in all models was assessed using Wald tests. Following convention, the alpha value for all tests was set to 0.05. Experiment 2 In Experiment 2, we ask whether 5–7-year-old children approve of teachers acting on gender stereotypes. Specifically, in Experiment 2a, we asked participants whether they approved of a teacher assigning activities in a gender-stereotypical manner (i.e., assigning boys to an engineering activity and girls to a reading activity), or a teacher assigning activities in an egalitarian manner (i.e., assigning boys and girls to each). In Experiment 2b, we exacerbated the severity of teachers’ behavior by making these assignments long-lasting: participants learned that teachers assign students to engineering and story “clubs”, where students play engineering and story games for the entire year. Additionally, to address alternative explanations for approval, such as a general preference for gender-segregated or homogeneous groups26, in Experiment 2b we added a teacher who assigned clubs in a counter-stereotypical manner (i.e., assigning girls to an engineering club and boys to a reading club). Both Experiment 2a, b were preregistered on OSF: https://osf.io/ukypg/(Experiment 2a, July 12, 2024), https://osf.io/ev4xz/(Experiment 2b, November 7, 2024). There were no major deviations from the preregistrations. Participants Experiment 2a recruited 90 5–7-year-olds in the United States (M: 6.30 years, SD: 0.83, range: 5.01–7.97) via online advertisements (e.g., Facebook ads). We did not conduct a power analysis. Parents reported participants’ gender (n = 46 boys, n = 44 girls) and race/ethnicity (n = 52 White, n = 16 Asian, n = 3 Hispanic, Latino, or Spanish origin, n = 2 Black, n = 2 Native Hawaiian or other Pacific Islander, and n = 15 other categories). Parental education ranged from 12 to 20 years of education (M = 18.23, SD = 2.01); parental income was not collected. An additional 8 participants were recruited but excluded on the basis of preregistered criteria, such as failing check questions (n = 2), not completing the study (n = 5), or having an autism diagnosis (n = 1). Experiment 2b recruited 132 5–7-year-olds in the United States (M: 6.4 years, SD: 0.89, range: 5.01–7.97) on Children Helping Science. We did not conduct a power analysis. Parents reported participants’ gender (n = 66 girls, n = 66 boys) and their family’s race/ethnicity (n = 82 White, n = 18 Asian, n = 7 Hispanic, Latino, or Spanish origin, n = 2 Middle Eastern or North African, n = 1 Black, and n = 22 other categories). Parental education ranged from 12 to 20 years (M = 18.17, SD = 2.27), and parental median income was $100,000 (M = $104,545, SD = $46,005; missing data from 12 parents). An additional 46 participants were recruited but excluded due to preregistered criteria of failing check questions (n = 46). However, note that the main results do not change when including these 46 participants (see Supplementary Fig. 4 for details). Stimuli As we did not find differences by teacher gender in Experiment 1 (see Supplementary Analyses 5 and 10), each test trial showed an image of an adult white woman as a teacher and eight children (four girls, four boys; four Black and four White) as students (Fig. 2). These images were overlaid on the cartoon classroom with two computer games, one with an engineering icon and one with a story icon (same icons as in Experiment 1). In the stereotype trials, the same images (e.g., thumbs up and down icons) as in Experiment 1 were used. Experiment 2 stimuli. In Experiment 2a, all participants saw a Stereotypical Teacher trial (teacher assigned only boys to engineering and only girls to stories) and an Egalitarian Teacher trial (teacher assigned boys and girls equally to the games) (left). In Experiment 2b, all participants also saw a Counter-Stereotypical Teacher trial (right; teacher assigned only girls to engineering and only boys to stories). Each trial showed a unique teacher and eight unique students (four boys, four girls, normed on age and emotional expressions; photographs of boy and girl student faces stylized as purple and yellow silhouettes, respectively, for publication; see OSF for actual stimuli). For each trial, participants rated how much they liked what the teachers did. Procedure Experiment 2a Participants were tested in a video call on Zoom. Participants were told that it was the first day of school and teachers were meeting their students for the first time, and they would say whether they liked what the teachers did or whether they did not like what the teachers did. We confirmed participants were comfortable using the full range of the scale to evaluate teachers’ behavior by asking participants to rate a teacher who assigned “fun” activities and a teacher who assigned “not fun” activities to their students; participants rated these behaviors positively and negatively, respectively (see Supplementary Methods for Experiment 2a for this measure and Supplementary Analysis 11 for results). Next, participants were introduced to story games and engineering games, with the same definitions as in Experiment 1, and told that the games were the same amount of fun. In each trial, they were shown a teacher (“Annie” or “Kelly”) and the eight students in her classroom (four boys, four girls) (teacher order, teacher photos, and child photos were counterbalanced across participants). Participants were told that the teacher assigned a game based on what she thinks the student might like. Notably, participants were not given any information about what the students in fact liked. Participants were asked a series of check questions to check their understanding of the cover story (see Supplementary Methods for Experiment 2a for wording and exclusion criteria). In the Stereotypical Teacher trial, the teacher assigned the four boys to the engineering game and the four girls to the story game. In the Egalitarian Teacher trial, the teacher assigned two girls and two boys to the engineering game and two girls and two boys to the story game. After seeing each teacher make assignments, participants were asked the critical evaluation question: whether they liked or did not like what the teacher did (two-step format: “do you like what the teacher did, or did you not like what the teacher did?”, then “do you kind of (not) like it, do you (not) like it, or do you really (not) like it?”, with ratings converted to 6-point scale from “really not like” to “really like”), as well as an open-ended explanation for that rating to increase engagement (see Supplementary Analysis 18 for participants’ explanations). Thus, each participant provided two evaluative ratings-one for the Stereotypical Teacher’s behavior and one for the Egalitarian Teacher’s behavior. Finally, participants were asked for their preference of which of the two classrooms they would rather be in, the Stereotypical Teacher’s or the Egalitarian Teacher’s (see Supplementary Analysis 13 for results of this measure). To measure participants’ own stereotypes, participants completed the same four stereotype trials as in Experiment 1 (in counterbalanced order). Finally, to measure participants’ own preferences, participants were asked which game they would prefer to play: the engineering game or the story game (see Supplementary Analysis 14 for results from this measure). Experiment 2b The procedure of Experiment 2b was identical to Experiment 2a except for the following changes. First, to eliminate concerns about experimenter influence in Experiment 1, the study was conducted on children helping science in an asynchronous, recorded, self-paced format, with only their parent or guardian present. Second, participants learned that teachers assign students to engineering and story “clubs”, where students play engineering and story games for the entire year. As before, participants answered a series of check questions to confirm their understanding of the cover story (see Supplementary Methods for Experiment 2b for more details). Third, in the test trials, we added a Counter-Stereotypical Teacher who assigned all the boys to the story club and all the girls to the engineering club. Thus, each participant provided three ratings (of the behaviors of the Stereotypical Teacher, Egalitarian Teacher, and Counter-Stereotypical Teacher). Statistics All analyses were performed in R version 4.3.2. To assess whether participants’ evaluations (1–6) of the Stereotypical Teacher’s practices were significantly negative or positive, in both Experiment 2a, b, we preregistered two-tailed one-sample t-tests comparing ratings to the midpoint of the scale (3.5), and used similar tests to explore participants’ ratings of the other teachers. To compare evaluations of different teacher practices, in both Experiment 2a, b, we preregistered a mixed-effect linear regression (using the lme4 package25) predicting participants’ ratings (1–6), as a function of the teacher behavior and participant age (exact), with random slopes for teacher behavior by participant and random intercepts by participant. To assess if participants’ own interests predicted their evaluations (exploratory in Experiment 2a, preregistered in Experiment 2b), we ran a mixed-effect linear regression predicting participants’ ratings, as a function of teacher behavior, participants’ own interests (0 = stereotypical, 1 = counter-stereotypical), and their interaction, as well as participant age (exact), with random slopes for teacher behavior by participant and random intercepts by participant. To follow up, we ran separate analyses (same model structure as the analysis reported above, without the participant interest variable) for participants with stereotypical interests and participants with counter-stereotypical interests (preregistered in Experiment 2b). To confirm that participants themselves held stereotypes (preregistered in both Experiment 2a, b), we ran mixed-effect linear regressions predicting stereotype ratings, as a function of target gender (girl or boy), task domain (engineering or stories), their interaction, and age (exact), with random slopes and intercepts for the interaction by participant. In Experiment 2b, we also preregistered separate mixed-effect linear regressions for engineering and stories separately, predicting their stereotype ratings as a function of target gender and age, with random slopes for target gender by participant. To explore if participants’ own stereotypes predicted their evaluations, we calculated the same boy and girl stereotype scores as in Experiment 1 for Experiment 2a, b, and summed them for each participant. We then ran exploratory linear regressions predicting each participant’s teacher rating by their combined stereotype score. All reported models converged without warnings. For the teacher ratings, we observed deviations from normality using Shapiro–Wilk tests; however, one-sample t-tests are robust to non-normality with large sample sizes, and thus we report parametric tests here. For completeness, nonparametric Wilcoxon signed-rank tests showed the same pattern of results (see Supplementary Analyses 12 and 20). The statistical significance of parameter estimates in these models was assessed using Wald tests, with alpha at 0.05. Stereotype ratings were not normally distributed, as observed using Shapiro–Wilk tests; linear models were used to estimate mean differences and are robust to non-normality with large sample sizes. To follow-up on nonsignificant results, we also ran exploratory Bayesian t-tests using the ttestBF function from the BayesFactor package, which uses a default Cauchy prior on standardized effect size (δ) with scale r = 0.707. Following convention, we interpreted a BF10 > = 3 as weak evidence and > = 10 as strong evidence in favor of a difference, and a BF10 < = 0.33 as weak evidence and < = 0.10 as strong evidence in favor of the null hypothesis. Results Children expect teachers to hold gender stereotypes In Experiment 1, we found that 5–7-year-old children expect teachers to rely on gender stereotypes about interest when assigning activities to students. As predicted, in Experiment 1a, we found a statistically significant interaction between student gender and teacher knowledge (b = 3.27, p < 0.001, OR = 27.33, 95% confidence intervals = [12.85, 58.13]), showing that participants’ predictions of what the teachers chose for boy versus girl students were sensitive to the teacher’s knowledge of the student’s interests (see Fig. 3). To unpack this interaction, we looked at each teacher knowledge trial in Experiment 1a separately. As predicted, in the Teacher Knowledgeable trials, participants were more likely to predict that the teacher chose the engineering game for girl students than boy students (b = 3.19, p < 0.001, OR = 23.92, 95% CI [13.08, 43.73]), since the teacher knew the students preferred the counter-stereotypical games. In the Teacher Ignorant trials, however, there is no evidence that participants made different teacher predictions for girl versus boy students (b = −0.08, p = 0.72, OR = 0.88, 95% CI = [0.56, 1.38]). Predictions of teacher game choice from Experiment 1a, b. Bars depict the proportion of participants a (Experiment 1a: of 145 participants, b Experiment 1b: of 207 participants) predicting that the teacher chose the engineering game (rather than the story game) for boy students (purple bars) or for girl students (yellow bars). Error bars show 95% CIs; dashed lines indicate chance. Experiment 1b results are collapsed across 5–7-year-olds (see Supplementary Fig. 1 for separate plots for each age group). To explore whether participants in Experiment 1a Teacher Ignorant trials were simply responding randomly or being systematic with their responses, we looked at participant-level patterns. The majority of participants either predicted that the teacher would select the stereotypical game for each student (35.2%) or predicted that the teacher would choose the game that reflected students’ actual (counter-stereotypical) interests (33.1%), while few participants predicted the same game for both students (both story, 17.9%; both engineering, 13.8%). This pattern of results is different from what would be expected by chance (Chi-square vs. 25% for each pattern: X2(3) = 19.99, p < 0.001, V = 0.21), suggesting that participants expected the teacher to either assign students games based on gender, or to give the students the games the student actually preferred, rather than guessing randomly on each trial. Given the large proportion of participants who expected an ignorant teacher to give students what they really liked, we suspected that participants may have had difficulty overriding the students’ true interests, perhaps because they wanted students to receive the game they preferred. Experiment 1b tested this possibility by stipulating that the students like both games equally, so students should be happy receiving either game. Since Experiment 1a established that participants do attend to the teacher’s knowledge, Experiment 1b included only the Teacher Ignorant trials. As predicted, all age groups ages 5 through 7 were significantly more likely to predict that the teacher would give the engineering game to the boy student than to the girl student (5-year-olds: b = −0.70, p = 0.029, OR = 0.50, 95% CI = [0.27, 0.93]; 6-year-olds: b = −1.57, p < 0.001, OR = 0.21, 95% CI = [0.09, 0.48]; 7-year-olds: b = −1.51, p < 0.001, OR = 0.22, 95% CI = [0.11, 0.44]) (see Supplementary Fig. 1). Specifically, 6- and 7-year-olds predicted that the teacher would choose the engineering game for a boy (6-year-olds: 70.6% engineering, p = 0.005, h = 0.43, 95% CI [57.0%, 81.3%; 7-year-olds: 70.7%, p < 0.001, h = 0.43, 95% CI [59.6%, 79.8%]) and the story game for a girl (6-year-olds: 66.7% story, p =0.024, h = 0.34, 95% CI [53.0%, 78.0%]; 7-year-olds: 65.3%, p = 0.011, h = 0.31, 95% CI [54.1%, 75.1%]). However, we did not find that 5-year-olds made selective predictions about what the teacher would choose for a boy (58.0% engineering, p = 0.182, h = 0.16, 95% CI [47.2%, 68.2%]) or for a girl (59.3% story, p = 0.119, h = 0.19, 95% CI [48.4%, 69.3%]; see Supplementary Fig. 1). Note that we did not observe that boys and girls made different predictions in Experiment 1a (see Supplementary Analysis 1) or 1b (see Supplementary Analysis 6). In sum, participants in Experiment 1b predicted that an ignorant teacher would assign games to students based on gender stereotypes. Children approve of educational assignments that enact gender stereotypes Experiment 1 showed that 5–7-year-old children expect teachers to hold gender stereotypes about interest, but do children actually approve of educational assignments based on such stereotypes? In Experiment 2a, b, we found that children do approve of such assignments, just as positively as egalitarian assignments, and more so than counter-stereotypical assignments. Participants endorsed gender-stereotypical educational assignments in both Experiment 2a (t(89) = 6.46, p < 0.001, mean = 4.57, 95% CI [4.24, 4.89], d = 0.68) and 2b (t(131) = 8.32, p < 0.001, mean = 4.66, 95% CI [4.38, 4.93], d = 0.72; see Fig. 4). Participants also endorsed egalitarian assignments (2a: t(89) = 8.23, p < 0.001, mean = 4.69, 95% CI [4.40, 4.98], d = 0.87; 2b: t(131) = 7.29, p < 0.001, mean = 4.56, 95% CI [4.27, 4.85], d = 0.63) and counter-stereotypical assignments (2b: t(131) = 3.44, p < 0.001, mean = 4.05, 95% CI [3.73, 4.36], d = 0.63). Participants gave equally positive ratings to the stereotypical and egalitarian assignments (2a: assignment type: b = −0.02, p = 0.913, 95% CI [−0.44, 0.40], paired t-test: BF10 = 0.137 (±0.11%); 2b: b = −0.09, p = 0.619, 95% CI [−0.49, 0.29], paired t-test: BF10 = 0.109 (±0.16%)). However, exploratory analyses in Experiment 2b revealed that participants were significantly less approving of counter-stereotypical assignments compared to the stereotypical (b = −0.61, p = 0.002, 95% CI [−1.00, −0.22]) or egalitarian assignments (b = −0.52, p = 0.009, 95% CI [−0.90, −0.13]), suggesting that children are relatively less approving of educational practices that act against gender stereotypes. Teacher assignment evaluations from Experiment 2a (n = 90) and 2b (n = 132). Larger points represent mean ratings for each assignment, and smaller points represent individual ratings. Participants rated the teachers' assignments on a scale from 1 (really not like) to 6 (really like). The dashed line shows the midpoint of the rating scale (3.5). Error bars show 95% CIs. Children’s own interests do not predict their evaluations We asked participants at the end of Experiment 2a, b whether they would prefer to play the engineering game or the story game to assess whether children’s own interests guided their evaluation of assignment practices (see Supplementary Analyses 14 and 21 for results from this measure). In an exploratory finding from Experiment 2a, we found an interaction between teacher assignment and participants’ own interest (b = −1.34, p = 0.006, 95% CI [−2.30, −0.38]). Specifically, participants with counter-stereotypical interests were more approving of the egalitarian assignment than the stereotypical assignment (b = −1.04, p = 0.007, 95% CI [−1.80, −0.28]), but participants with stereotypical interests did not show evidence of a difference in approval (b = 0.29, p = 0.254, 95% CI [−0.21, 0.79]). This exploratory result suggested that participants’ evaluations may come from an egocentric projection of whether their personal interests would be satisfied by the teacher’s practices. Consequently, in Experiment 2b, we preregistered that the minority of participants who had counter-stereotypical interests (i.e., girls who preferred engineering, boys who preferred reading) would be less approving of the stereotypical assignment compared to the egalitarian assignment. However, unlike in Experiment 2a, we did not detect a significant interaction between assignment type and participants’ own interest (egalitarian vs. stereotypical assignment and participant interest: b = 0.08, p = 0.854, 95% CI [−0.75, 0.91], BF10 = 0.52 (±2.06%); counter-stereotypical vs. stereotypical and participant interest: b = 0.47, p = 0.267, 95% CI [−0.36, 1.30], BF10 = 0.63 (±2.34%); see Supplementary Analysis 22 for more details). Thus, children do not seem to use their own interests to guide their evaluations of educational assignments. Children hold gender stereotypes, but these do not predict their expectations and evaluations of others’ gender stereotypes Across all experiments, we found that 5–7-year-old children hold gender stereotypes that associate boys with interest in engineering and girls with interest in reading. In Experiment 1 (exploratory) and Experiment 2 (preregistered), we found that participants held gender stereotypes about interest in engineering and stories, as indicated by a significant interaction between the target gender and domain on participants’ estimates of each gender’s interest in engineering vs. reading (1a: b = 2.62, p < 0.001, 95% CI [2.10, 3.14]; 1b: b = 2.60, p < 0.001, 95% CI [2.16, 3.05]; 2a: b = 3.43, p < 0.001, 95% CI [2.80, 4.06]; 2b: b = 2.98, p < 0.001, 95% CI [2.43, 3.54]) (see Fig. 5). In both experiments, participants rated boys as more interested in engineering than girls (1a: b = −1.81, p < 0.001, 95% CI [−2.17, −1.46]; 1b: b = −1.51, p < 0.001, 95% CI [−1.83, −1.19]; 2a: b = 1.80, p < 0.001, 95% CI [1.36, 2.25]; 2b: b = 1.91, p < 0.001, 95% CI [1.52, 2.30]). Participants also rated girls as more interested in stories than boys (1a: b = 0.86, p < 0.001, 95% CI [0.41, 1.18]; 1b: b = 1.09, p < 0.001, 95% CI [0.77, 1.41]; 2a: b = 1.63, p < 0.001, 95% CI [1.18, 2.08]; 2b: b = 1.08, p < 0.001, 95% CI [0.68, 1.47]) (see Fig. 5). Such results confirm that children possess gender stereotypes about engineering and reading interest, consistent with prior work on such stereotypes19,20. Children’s own stereotypes for all experiments. Participants (Experiment 1a: n = 145, Experiment 2b: n = 207, Experiment 2a: n = 90, Experiment 2b: n = 132) provided four responses of how much most boys (in purple) and girls (in yellow) like engineering/stories, from 1 (really not like) to 6 (really like). Large points represent mean responses, and small points are individual responses. Error bars are 95% CIs, and dashed lines indicate chance. Although participants as a group endorsed both stereotypes, there was variability in participant-level responses (see Fig. 3). We leveraged this variability to explore how participants’ own stereotypes relate to their expectations and evaluations of others’ stereotypes. In Experiment 1, we explored whether participants who hold stronger stereotypes also hold stronger expectations of others’ gender stereotypes. However, we found that participants’ gender stereotypes about a game were not related to their likelihood of predicting an ignorant teacher to assign that game to a boy (1a: b = 0.01, p = 0.90, 95% CI [−0.14, 0.13], BF10 = 0.07 (±1.75%); 1b: b = 0.10, p = 0.138, 95% CI [−0.03, 0.22], BF10 = 0.19 (±3.36%)) or to a girl (1a: b = −0.08, p = 0.22, 95% CI [−0.21, 0.05] BF10 = 0.14 (±2.63%); 1b: b = 0.04, p = 0.462, 95% CI [−0.07, 0.16], BF10 = 0.08 ± 2.02%). In other words, we found that children do not use their own stereotypes as the basis for their belief that others hold stereotypes. If children’s own stereotypes are unrelated to their expectations of others’ stereotypes, it is possible that even participants who do not personally hold gender stereotypes may nonetheless expect others to hold gender stereotypes. Indeed, participants in Experiment 1b who did not endorse either engineering or story gender stereotypes (n = 49/205; about 25% of participants; Mage = 6.00, SDage = 0.73, range = 5.01–7.80) nonetheless predicted that the teacher would assign an engineering game to a boy more than a girl (b = −1.18, p = 0.005, 95% CI [−2.00, −0.36]; see Supplementary Fig. 2 and Supplementary Analysis 9 for more details). Thus, even the minority of participants who do not hold gender stereotypes themselves still assume that others hold such stereotypes, demonstrating robust stereotype expectation. In Experiment 2, we explored whether participants’ own stereotypes predicted their evaluations of teachers’ assignments. Participants’ combined stereotype scores did not predict their ratings of stereotypical assignments (2a: b = 0.04, p = 0.448, 95% CI [−0.06, 0.14]; 2b: b = 0.04, p = 0.302, 95% CI [−0.03, 0.11]), but did marginally or weakly predict their ratings of egalitarian assignments (2a: b = −0.08, p = 0.071, 95% CI [−0.16, 0.01]; 2b: b = −0.09, p = 0.017, 95% CI [−0.16, −0.02]), and did predict their ratings of the counter-stereotypical assignment (2b: b = −0.09, p = 0.020, 95% CI [−0.17, −0.02]). Overall, participants who had stronger gender stereotypes were also slightly less approving of the egalitarian and counter-stereotypical assignments, although these weak, exploratory effects should be treated with caution. Discussion Children in the United States grow up in an environment where adults frequently hold harmful gender stereotypes1,3. Here, we find that by the time children are just 5–7-year-olds, they have already developed accurate, yet troubling, expectations that teachers hold gender stereotypes about academic interests. Children held these expectations even if they know these stereotypes are not accurate in our scenarios, and even if they do not personally believe in such gender stereotypes. Strikingly, children also approve of educational assignments that reinforce gender stereotypes about academic interests (e.g., assigning boys to an engineering club and girls to a story club), suggesting that such stereotypes are not only expected but also socially validated at an early age. Our findings help adjudicate between prior competing accounts to establish that children expect adults to hold gender stereotypes. We found that children predict that a novel teacher will assign a gender-stereotypical activity (engineering to boys, reading to girls), even if the child knows the student is equally interested in both (Experiment 1b). When the student was more interested in the counter-stereotypical activity (Experiment 1a), children showed weaker stereotype-based predictions, likely due to additional task demands rather than an absence of stereotypical expectations. By focusing on novel adults, our design supports the conclusion that children generally expect teachers to hold stereotypes. Thus, mixed results in prior studies asking children what “most adults” or “most grownups” think8 may arise because children reasoned about particular, familiar adults, some of whom may reject gender stereotypes. Moreover, our findings do not support the account that children’s own stereotypes inform their awareness of others’ stereotypes27, since even children who did not endorse gender stereotypes themselves nonetheless predicted that the teacher would assign activities based on gender stereotypes. Our results reveal that children not only expect teachers to hold gender stereotypes about academic interests (Experiment 1), but also approve of teachers acting on those stereotypes (Experiment 2). In Experiment 2, children approved of a teacher assigning activities to students based on gender stereotypes (i.e., boys to engineering and girls to stories), even when those activity assignments would be long-lasting (Experiment 2b). Children rated such a practice just as highly as egalitarian practices (i.e., assigning each activity equally between girls and boys). Critically, children were less approving of counter-stereotypical assignments (Experiment 2b), showing that children are specifically sensitive to whether educational assignments align with gender stereotypes, beyond merely producing gender-segregated or homogeneous groups. These findings suggest that children endorse gender-stereotypical practices in the classroom, even if such practices limit students’ educational experiences and opportunities. What drives children’s approval of gender-stereotypical practices? We found that 5–7-year-old children in the United States hold gender stereotypes about engineering and reading interest, supporting prior work establishing that even first-graders hold gender stereotypes about engineering interest19. However, children’s own stereotypes were unrelated to their evaluations (Experiment 2), indicating that they did not endorse educational assignments simply because they believed the teacher shared their own views. Further, we did not find evidence that children’s own interests were related to their evaluations. One possibility is that children may generally defer to teachers as authority figures, thereby granting them a high baseline level of approval. However, this account is insufficient on its own. Children were significantly less approving of a teacher assigning counter-stereotypical activities and were outright critical of teachers assigning “not fun” activities to students (see Supplementary Methods for Experiment 2a for warmup trials). Thus, children’s trust in teachers does not translate into indiscriminate approval. Instead, even against a backdrop of teacher deference, children are sensitive to whether teachers’ practices violate gender stereotypes. We suggest that children’s expectations of others’ stereotypes might provide a basis for their evaluations. In everyday life, children may observe adults acting based on gender stereotypes (e.g., that parents give STEM activities more to boys than to girls), which may lead them to form generalized expectations about how adults, such as teachers, act. Children may then turn these expectations of descriptive norms into prescriptive ones16,17, and as a result, negatively evaluate teachers’ practices when they do not align with gender stereotypes. In this way, children’s sensitivity to recurring patterns of adult behavior may shape both their expectations and evaluations, ultimately reinforcing gender-based inequalities. Children’s expectation and acceptance of adults’ holding gender stereotypes may shape their own learning behaviors and social interactions in meaningful ways. First, prior work suggests that children expect adults to react with disapproval or surprise when children violate gender stereotypes7,28. As a result, if children assume that their teachers believe in gender stereotypes (and receive no evidence to the contrary), they may be motivated to engage in gender-stereotypical activities to avoid negative evaluations, even if they do not hold gender stereotypes themselves. Second, children’s endorsement of practices that reinforce gender stereotypes could guide their peer interactions29,30, such as boys selecting other boys for male-stereotyped activities, which could perpetuate gender divisions in opportunity. In short, our findings show that children enter school expecting and accepting that teachers hold gender stereotypes, which may lead them to validate such practices and conform to stereotyped roles, with cascading effects on gender equity in higher education and the workforce. A critical direction for future research is to understand how children’s expectations of their particular teachers’ gender stereotypes shape their decision-making in the classroom, and importantly, how best to intervene on this process. Limitations Our work has several limitations. First, our sample is predominantly from families with high education and/or income levels in the United States, therefore limiting generalizability to broader socioeconomic or cultural contexts. Second, given that our study only tested children’s reasoning about teachers, future research could test whether children have expectations for other adults, e.g., parents, other gender stereotypes, e.g., gender stereotypes about ability, or other common stereotypes, e.g., race or class-based stereotypes. Third, our work raises questions about whether children make these assumptions of all adults, or if the strength of children’s expectations depends on their own social identities, the stereotype in question, or the social identities of the adult that they are reasoning about. Conclusion We find that children enter school already expecting teachers to act on gender stereotypes and approving of such actions. Importantly, preventing children from developing gender stereotypes may not be sufficient to prevent gender disparities from emerging. That is, children’s mere expectation and approval of others’ gender stereotypes may be enough to shape their choices down gendered paths. Supplementary information Transparent Peer Review File Supplemental Material Acknowledgements We thank Julian Jara-Ettinger and Yarrow Dunham for helpful discussions. We also thank Natalie Masetti, Brandon Carrillo, Zahra Yarali, Christina Norberg, Bethel Asomaning, and Elaine Wang for help with data collection and stimuli creation. This work was supported by an NSF SPRF to M.A., and a Jacobs Foundation Research Fellowship awarded to J.A.L. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author contributions M.A. contributed conceptualization, formal analysis, funding acquisition, methodology, project administration, writing-original draft, writing-review and editing. M.Y.Z. contributed conceptualization, methodology, writing-original draft, writing- review and editing. J.A.L. contributed conceptualization, funding acquisition, methodology, supervision, writing-review and editing. Peer review Peer review information Communications Psychology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary handling editors: Troby Ka-Yan Lui. A peer review file is available. Data availability All data necessary to reproduce the results reported in this work are publicly available on the Open Science Framework (OSF): 10.17605/OSF.IO/KR5SM. Code availability All analysis code necessary to reproduce the results reported in this work are publicly available on the Open Science Framework (OSF): 10.17605/OSF.IO/KR5SM. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Marianna Y. Zhang, Mika Asaba. Supplementary information The online version contains Supplementary material available at 10.1038/s44271-026-00459-y. References Associated Data Supplementary Materials Transparent Peer Review File Supplemental Material Data Availability Statement All data necessary to reproduce the results reported in this work are publicly available on the Open Science Framework (OSF): 10.17605/OSF.IO/KR5SM. All analysis code necessary to reproduce the results reported in this work are publicly available on the Open Science Framework (OSF): 10.17605/OSF.IO/KR5SM.