Question asking practice fosters aspects of curiosity in science content in young children

Abstract

Children who are more curious learn more in school, but little is known about how to promote curiosity-driven behaviors. In a preregistered experiment, 103 children (54 boys, 49 girls, ages 5–7 years) were randomly assigned to a condition in which they were encouraged to ask questions, or to listen carefully, during eight one-on-one science lessons over 2 weeks. Children in the question-asking condition valued new science information significantly more than children in the listening condition (Wilcoxon r = 0.23). Children with less background knowledge, as measured by their baseline vocabulary and science achievement, showed greater curiosity and learning benefits from question-asking. These results suggest that practice with question-asking can boost some aspects of curiosity and learning in science domains.

Introduction

Curiosity is a core feature of cognition, driving our attention to new information^1–3. Curiosity is thought to enhance learning via a cycle of prediction errors that result in increased exploration, information gathering, and memory formation⁴. Epistemic curiosity, the intrinsic motivation to learn, has been linked to academic achievement⁵. The association between curiosity and early academic achievement is especially strong for children from low socioeconomic status (SES) backgrounds⁶. Such results raise the important question of whether we can foster curiosity early in development, particularly for children who are struggling in school or who have experienced socioeconomic disadvantage⁷.

We have made considerable progress towards understanding the factors that promote children’s curiosity in the moment. Children prefer to explore novel events⁸, and will explore more when evidence is ambiguous^9–11 or surprising^12,13. Children will explore, persist, and learn more when a knowledgeable adult asks them questions with the goal of teaching (“pedagogical questions”)^14–18. Classroom-based experiments have shown that uncertainty can promote state-like curiosity, with benefits for learning¹⁹.

We know less about how to build curiosity in a sustained way. While young children actively seek out explanations to make sense of the world²⁰, become sophisticated at understanding what and who to ask^21–23, and reject unsatisfactory responses^2,24,25, typical schooling does not reliably increase curiosity, and might even diminish it in some contexts, i.e, when learning information by rote memorization is emphasized^26,27. Such direct instruction may stifle curiosity because it focuses learners on just the cued information²⁸). Inquiry-based curricula are designed to promote learning through exploration, but their impacts on curiosity have not been explored in studies with controlled interventions with younger children (though see one promising intervention in past work²⁹, and another intervention with older children³⁰). Further, because these curricula are holistic and broadly defined, it is difficult to identify what should be prioritized in classrooms^31,32.

In this study, we sought to understand the impact of a potential efficient, scalable teaching method to bolster curiosity and increase learning in science domains. We focused on encouraging children to ask questions, inspired by theories of fostering curiosity in schools^33–36 and the considerable research that demonstrates children’s propensity to ask many, meaningful, information-gathering questions in the preschool years^37,38. Questions are core to two related dimensions of epistemic curiosity^39,40: interest-type, the positive emotion associated with the joy of learning^41–43, and deprivation-type, the desire to resolve the negative experience of lacking information. Recognizing the need for a question might raise deprivation-type curiosity, and resolving with an answer might satiate interest-type. Question asking bolsters cognitive reflection: the ability to monitor information as it is presented and consider how it interacts with existing knowledge to identify gaps^44,45. Questions highlight points of uncertainty, which drive exploration-based curiosity. Furthermore, questions may engage the learner in predicting answers, which can highlight errors to increase surprise and promote encoding^46–48. Questions are also empowering: they give children a tool to actively expand their own knowledge by engaging with others³⁸. Thus, discovering how best to foster question-asking behavior has the potential to improve children’s learning via broader increases in intrinsic motivation.

We tested whether encouraging children to ask questions influenced their curiosity-driven behaviors or learning in science domains. We designed a novel teaching paradigm that is both naturalistic, in contrast to many simplified laboratory tasks, and well-controlled, in contrast to the classroom. We focused on teaching children semantic information because of evidence of tight links between curiosity and hippocampal-dependent learning⁴. We selected science information, specifically, because of a broader literature on links between curiosity and science performance^22,44, the link between inquiry-related curiosity and science engagement^49,50, and the role of a teacher’s response to inquiry-based deviations on a science task⁵¹.

We focused on children in the early grades (Kindergarten through 2nd) because of evidence that science knowledge disparities early in life persist or compound across children’s formal education⁵². The paradigm consisted of engaging science lessons based on curriculum content recommendations for Kindergarten, the youngest students in this experiment, from the United States’ Next Generation Science Standards (e.g., K-LS1:All animals need food in order to live and grow; https://www.nextgenscience.org/kindergarten-topics-model). 103 children ages 5–7 years-old participated in eight one-on-one lessons over 2 weeks. Children were randomly assigned to one of two conditions. In one condition, they were encouraged to ask questions. In the other condition, they were encouraged to listen carefully. We tested for group differences in learning and on a suite of curiosity-related measures adapted for remote administration in children.

We chose measures that allowed us to explore different aspects of curiosity. We refer to these as “curiosity measures” throughout for simplicity. However, note that these individual tasks measure different behavioral outcomes that may relate to curiosity but do not necessarily directly measure curiosity (e.g. see work by Ainley⁵³ on how curiosity and interest are intertwined on behavioral measures). Curiosity is a multifaceted drive that likely impacts a variety of behaviors such as fluency in generating questions, valuation of new information, seeking out information via exploration, and potentially persistence during this search process⁵⁴. We chose a question-asking measure to test whether the question-asking intervention caused children to generate more questions when prompted (inspired by Kemler Nelson et al., 2004)⁵⁵. To investigate whether training shifted the intrinsic value that children place on new science information, we adapted a willingness-to-pay task, which is a well respected measure of curiosity in adults and non-human primates⁵⁶. An exploration measure allowed us to examine whether children would show greater curiosity to explore a cued item with deeper focus, or whether they would show greater interest in exploring more broadly beyond the cued item¹⁷. A persistence measure allowed us to investigate whether children would continue to search for differences in a matching task (inspired by Rholes et al., 1980)⁵⁷. We also tested whether the groups differed in how much they learned.

Finally, we explored whether the efficacy of question asking depended on children’s academic skills, i.e., do children with strong academic performance get even stronger when encouraged to ask questions (the “Matthew Effect”) or do children with weaker academic skills benefit more from question asking? We also asked whether children’s experience of socioeconomic disadvantage moderated the impact of question asking because of the particularly high need to generate strategies to better support these children in the classroom. Further, socioeconomic disadvantage and other stressors have been previously linked to reduced curiosity^58–61 and creates risk for poor learning outcomes^62,63. In sum, we examined whether question asking practice impacts curiosity and learning, and if so, who benefits most from this pedagogical approach.

Methods

Open practices statement

The preregistrations for this study can be accessed at https://aspredicted.org/blind.php?x = 7CD_HK9 and https://aspredicted.org/blind.php?x = ZN2_4Y6. All stimuli, questionnaire data, behavioral data, and analysis code are publicly available at https://osf.io/zsj9h/?view_only=72707fec61b449ebb80a5159809ac86b. Lesson videos and transcripts are available by request with evidence of completed human subjects training.

Participants

This study was approved by the Institutional Review Board at the University of Pennsylvania. The target sample size was 53 per condition (106 total) based on an anticipated medium effect size (d = 0.55). Based on the cognitive training literature, we expected the directly-practiced skill to have the biggest effect size (question asking), and transfer tasks to have medium to small effect sizes (willingness-to-pay, exploration, persistence)⁶⁴. Parents provided written consent. Data were collected from September 2020 through September 2021.

Parents were recruited through Facebook ads about a Virtual Science Camp. Because the study was virtual, enrollment was open to all families in North America who spoke fluent English. 112 children began the data collection process. One child was excluded for being too young, 3 children were excluded for researcher mistakes during the baseline test, one child was excluded for participating in an excessively distracting environment, and one child dropped out. Post-tests were not conducted for these 6 children. Three children were excluded after post-test data were collected because parents answered the diagnosis question differently once children were enrolled than they did during screening (reported ADHD, anxiety, or severe processing disorder). The final sample included 52 children in the question-asking condition and 51 children in the listening condition.

Parent-reported questionnaires

Parents reported their total annual income and education level, as well as the education level of their partner if applicable (99% of parents reported the education level of their partner). Demographics for both groups are shown in Table 1. Groups did not differ on demographics except race. Despite random assignment, the question asking group included more White children than the listening group (F(4, 98) = 5.07, p = 0.03). Parents completed the Child Behavior Checklist (CBCL)⁶⁵ to ensure that children did not have clinically significant internalizing or externalizing symptoms (children completed the version appropriate for their age). They also completed the Life Events Scale for Young Children⁶⁶, the Adverse Childhood Experiences (ACEs) questionnaire⁶⁷, and the 10-item Perceived Stress Scale⁶⁸. They reported on their children’s baseline trait-like curiosity with the I- and D-Type Epistemic Curiosity Scale⁴⁰, which captures children’s curiosity along two dimensions: Interest-type curiosity and Deprivation-type curiosity. We did not readminister this questionnaire at the end of the study because we did not expect parents to update their impressions of their children’s typical behaviors after only 2 weeks.

Table 1.

Demographics. Data on race were missing for two participants (one in each condition)

	Condition
Variable	Listening (n = 51)	Question-Asking (n = 52)
Age (years)	M = 6.30 (SD = 0.70)	M = 6.47 (SD = 0.71)
Gender	24 girls, 27 boys, 0 non-binary	25 girls, 27 boys, 0 non-binary
Race
White	30 (59%)	40 (77%)
Multiracial or Other	12 (24%)	9 (17%)
Asian	3 (6%)	2 (4%)
Black	3 (6%)	0 (0%)
American Indian	2 (4%)	0 (0%)
Ethnicity
Not Latine/Latinx	47 (92%)	48 (92%)
Latine/Latinx	3 (6%)	2 (4%)
Parent Education (years)
Completed college	39 (76%)	39 (75%)
Did not complete college	12 (24%)	13 (25%)
Annual Household Income (US dollars)	Mdn = $88 K (SD = $56 K)	Mdn = $88 K (SD = $55 K)
Parent Report of Children’s Epistemic Curiosity
Deprivation-type	2.74 (0.59)	2.90 (.65)
Interest-type	3.42 (0.48)	3.58 (0.48)

Data on ethnicity were missing for three participants (one in the Listening condition, two in the Question-Asking condition).

Pre-training assessments

Pre-training assessments were conducted on the first day of the study (Monday of Week 1). Over video call, children completed standardized tests of knowledge and cognitive skills. Children completed the Science subtest from the Woodcock-Johnson IV Tests of Achievement (WJ IV ACH⁶⁹, and the Vocabulary, Matrix Reasoning, and Digit Span subtests from the Wechsler Intelligence Scales for Children (WISC-V⁷⁰). Scores on pre-training assessments by condition are shown in Supplemental Table 1.

Training

The learning paradigm took place over 2 weeks by video call (Tuesday-Friday of Week 1, Monday-Thursday of Week 2) with each child meeting one-on-one with the experimenter for each session. Thus, there were ~800 recorded sessions. Each lesson lasted 20–45 min and consisted of three parts: a story from the book We Build Our Homes: Small Stories of Incredible Animal Architects by Laura Knowles, a video from Mystery Science (an online science education resource), and an activity (Table 2).

Table 2.

Science lessons

Theme	Story (3–8 min)	Science Video (5–20 min)	Activity (5–8 min)
Protecting offspring	Tailorbirds build a nest out of leaves to protect their chicks.	Mother ducks scare off predators and teach babies to swim.	Building a nest (sticks, yarn, paper).
Safety	Gopher tortoises dig tunnels for themselves and other animals to hide in.	Praying mantises look fierce, gophers hide in holes, snails use shells for protection.	Acting out safety behaviors.
Animal-plant systems	Honeybees collect nectar from flowers to make honey and wax.	Bees collect nectar from flowers, enzymes thicken nectar.	Making honey (cornstarch and water).
Camouflage	Caddisfly larvae decorate themselves with pebbles.	Tigers hide in grass, katydids look like leaves, owls blend in with tree bark.	Finding paper moths on a cardboard tree.
Homes	Aardvarks dig burrows.	Squirrels make nests in trees.	Acting out building homes: wood mouse, mallard duck.
Hibernation	Polar bears give birth and hibernate in dens in the winter.	Bears, skunks, and bats put their bodies on pause.	Guessing game: how do animals prepare to hibernate?
Food	Woodpeckers use their beaks to drill holes to find grubs.	Quails scratch and peck ground, raccoons catch fish.	Guessing game: where do animals look for food?
Modifying environments	Beavers build dams.	Woodpeckers make holes in trees and other animals move in.	Building a dam (clay, rocks, sticks).

Each science lesson consisted of three different sections: a story from the book We Build Our Homes: Small Stories of Incredible Animal Architects by Laura Knowles, a video from Mystery Science, an online science education resource, and an interactive activity.

Children were randomly assigned to one of two conditions: question asking or listening. Children completed a similar number of lessons across the two conditions (question-asking: 7.69 lessons, listening: 7.72 lessons). Children needed to complete at least six lessons to be included in the experiment. Before the first lesson the child was told the purpose of the lessons: “I want you to get really good at [asking questions/listening carefully]. [Question asking/careful listening] is super important because [asking questions/listening] is an amazing way to learn about new, cool things! [Question asking/listening closely] is a great skill for you to use in school, and I want us to have a goal of becoming really good [question askers/listeners] so that we can be some of the best learners in the future!” This goal was reaffirmed at the start of each subsequent lesson.

In the question-asking condition, teachers modeled question asking throughout the lesson and children were encouraged to ask questions themselves. Questions that children asked were followed by praise and an answer from the teacher. The listening condition followed the same prompting schedule, and included praise for listening carefully or if the child was having trouble listening, a reminder to do so. If a child in the listening condition asked a question, the teacher would give a short answer (with no explicit praise), or redirect the child with a reminder to listen carefully. Such interruptions were relatively rare. See OSF materials, linked below, for detailed response directives to children’s questions and comments for researchers.

At the end of each lesson, children were praised for their question asking or listening, and given a short summary of the lesson, including a fact that would be included in the post-training science test. Finally, they were reminded to ask questions or listen carefully next time.

To track children’s progress during training, researchers transcribed the questions children asked during each lesson from recorded videos of the lessons. The transcribed questions were marked as relevant to the learning context, e.g., “Why do aardvarks help so many animals?”, or irrelevant, e.g., “Did you see my ninja costume?” Twenty percent of lessons were double coded. Inter-rater reliability was calculated for the total number of relevant questions per lesson and was very high (82% exact agreement, intra-class correlation = 0.99).

Post-training assessments

Post-training assessments were completed on the final day (Friday of Week 2). Post-tests were completed by a new researcher, who was masked to the child’s assigned condition.

Materials for all assessments are available on OSF: https://osf.io/zsj9h/?view_only=72707fec61b449ebb80a5159809ac86b.

Willingness-to-pay

This task was developed following methods by Brosseau-Liard (2014)⁷¹ where children trade-off stickers against information sources (see also work by Rowles & Mills⁷²). The willingness-to-pay score is the minimum number of stickers that a child was willing to give up to watch a preselected science video. A greater number of stickers indicated greater curiosity about the video. This task allows us to measure general curiosity about science in terms of the child’s perceived “reward” for information.

Novel animal questions

Children were shown a picture of a novel animal, a drongo. Children were given 3 min to ask as many questions as they had. In order to test whether question-asking practice additionally shapes the content of children’s questions, we coded their novel animal questions based on four properties: causal structure (e.g., “why”, “what for”, “how does”), functional content (i.e., questions about animals’ ecological niches), perceptual content, and lesson-related questions.

Exploration

Children’s exploration was measured with the Apothecary Task. This task is used to quantify playful exploration behaviors similar to past studies of novel toy exploration, via their discovery of unknown options, objects, or functions as it trades off against a pedagogical cue to focus exploration on the cued target²⁸. Such a measure allowed us to examine whether curiosity to the target function is increased (leading to more focused attention on a cued item relative to other options) or whether curiosity to broader possibilities is increased (leading to more exploratory focus on the non-cued items). Children were shown 16 envelopes (four envelopes of four colors) and asked: “How might the things in the blue envelopes go together? Can you think about that?” Then, the child was invited to explore the envelopes for up to 6 min, or until they notified the experimenter that they were finished playing. We preregistered that we would examine the number of envelopes opened and the amount of time spent playing with toys, as well as the percent of total playtime with the contents of the blue envelopes. Because of ceiling effects on the number of envelopes opened and total play time, we only report percent of total playtime with the contents of the blue envelopes. Additionally, following standards in past work²⁸, we include an exploratory analysis focusing on the number of blue envelopes opened in the first minute, to get a measure of how children trade-off between the directed function and exploration.

Persistence

Children were shown two nearly-identical pictures. Children were instructed to find as many differences as they could between the two pictures, to tell the experimenter each time they found a difference, and to let the experimenter know when they were all done looking. The main outcome measure was the time spent looking for differences between the images.

Science learning

Children were asked 16 questions about the animal stories they had learned about in their eight science lessons (two per lesson). Accuracy is reported as percent correct.

Statistical approach

Question asking during training (question-asking condition)

We tested whether children in the question-asking condition increased in the number of relevant questions they asked over the course of training in two ways. First we compared the number of relevant questions asked in the first training session with the number of questions asked in the last training session attended by the child, using a 1-tailed paired t-test. Second, we computed the slope of questions asked with a Poisson mixed model (estimated using ML and Nelder-Mead optimizer) in R. The model was fit with day and lesson order as fixed effects, allowing each participant an intercept and a random slope for day for a maximally-specified model. Lessons were excluded from question coding if there were technical problems (10 lessons across 8 children). Children were excluded from analyses of slope of question-asking improvement if fewer than five lessons had usable data (n = 1).

Main effects of condition

We preregistered one-tailed independent sample t-tests for willingness-to-pay, novel animal questions, and exploration, with the prediction that we would see higher performance in the question-asking condition. We preregistered two-tailed t-tests for science learning and persistence because we thought we could see greater performance in either condition. After collecting the data, we found that all measures significantly deviated from normal distributions: science learning (W = 0.97, p = 0.03); willingness-to-pay (W = 0.76, p < 0.001); novel animal questions (W = 0.89, p < 0.001); exploration (W = 0.80, p < 0.001); persistence (W = 0.91, p < 0.001). Further, because the groups differed by racial/ethnic composition and key outcomes were related to age even in the narrow age range in which we recruited⁷³, we ran quantile regressions testing for effects of group controlling for minoritized racial group (binary White/non-White) and age. We used the qreg package in Stata to run these regressions. Because we evaluated 5 core measures, we corrected for multiple comparisons with a false discovery rate (FDR) correction⁷⁴.

We examined the effect of condition on the content categories of questions about the novel animal, preregistered here: https://aspredicted.org/blind.php?x = ZN2_4Y6. We preregistered one-tailed t-tests for causal structure, functional content, perceptual content, and lesson-related questions. We predicted that children in the question-asking condition would ask more questions in the causal, functional, and lesson-related categories, while children in the listening group would ask more questions in the perceptual category, because practice with paying attention might modulate visual perception. We excluded children who asked 0 questions from the content analysis. We also preregistered repeating these analyses with percent questions, as well as excluding outliers (+/− 3 SD from the mean). After data collection, we found that all question type measures were significantly non-parametric (p < 0.001), so as in the analyses above, we conducted quantile regressions controlling for age and race to compare groups.

Moderation analyses

With quantile regression, we tested whether cognitive abilities (parent report of curiosity, and WISC vocabulary, matrix reasoning, and working memory) and baseline science knowledge (WJ Science) interacted with condition to predict post-test measures. We also tested whether demographic and stress exposure measures interacted with condition to predict post-test measures. These moderation analyses were pre-registered so we share the results for transparency, but note that the sample size is likely too small to be well-powered for interaction analyses.

Ethics statement

This research was approved by University of Pennsylvania IRB #825656.

Results

Developmental trends in learning curiosity measures

Age was positively correlated across conditions with science learning (r_s(101) = 0.36, p < 0.001), willingness-to-pay (r_s(101) = 0.23, p = 0.02) and the number of questions asked in the novel animal task (r_s(101) = 0.26, p = 0.009). Age was not associated with other curiosity measures (see Supplementary Fig. 1).

Associations among curiosity measures

Associations among curiosity measures by condition are shown in Fig. 1. Correlations among behavioral curiosity measures were generally small (r_s = −0.06–0.27). Parent report of I-type and D-type epistemic curiosity were highly correlated, which could be due to shared method variance (r_s(101) = 0.70)⁷⁵.

Fig. 1. Spearman correlations among curiosity measures by condition.

Correlations across both groups are shown in gray text in the top rows. Correlations within the Question-asking condition (QA) are shown in dark blue. Correlations within the Listening (L) condition are shown in light orange. * p < 0.05, **p < 0.01, ***p < 0.001 uncorrected.

Question-asking during lessons

Children in the question-asking condition asked an average of 5.51 relevant questions (SD = 3.68) during their first lesson and 7.43 questions during their last lesson (SD = 5.30), showing a significant increase in question asking between these training sessions (t(40) = 2.36, p = 0.001, Supplementary Fig. 2). This difference was not reflected in a slope analysis (b = 0.01, p = 0.395), suggesting that the increased question asking occurred non-linearly over the sessions. We did not code question asking in the listening condition because question asking was not the target behavior, it occurred rarely, and we did not predict changes over 2 weeks.

Main effects of question-asking

Children in the question-asking condition had a significantly higher willingness-to-pay score than children in the listening condition, indicating that they valued science knowledge more (Fig. 2a, Question-Asking (QA): Mdn = 7, Listening (L): Mdn = 5, b = 2, 95% CI [.76, 3.24], t = 3.19, p = 0.002, p_FDR = 0.010, n = 103).

Fig. 2. Main effects of question asking.

Data distributions are shown within the question-asking and listening conditions for four measures of curiosity driven behavior: (a) willingness-to-pay (trade stickers) for new information (a novel video), (b) number of questions asked about a novel animal, (c) exploration following a pedagogical cue during play, and (d) persistence on a challenging task. Science learning (% of post-test questions correct) is shown in (e). Box-plots indicate the following: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range. Data points are slightly jittered to prevent overlap.

No condition difference was observed in the novel animal task in which children were prompted to ask questions (Fig. 2b, QA: Mdn = 3, L: Mdn = 3, b = −0.39, 95% CI [−1.51, 0.74], t = −0.68, p = 0.5, p_FDR = 0.500, n = 103). The content of the questions also did not differ across conditions. Children in the question-asking condition did not ask more questions with causal structure (e.g., “why”, “what for”, “how does”), functional content (i.e., questions about animals’ ecological niches), perceptual content (i.e., questions about details in the picture cue), or lesson-related questions (i.e., questions about topics in the curriculum) (all ps > 0.2). There were also no condition differences when normalizing by total number of questions asked or when excluding outliers that were >3 SD from the mean (all ps > 0.1).

In the exploration task, almost all children in both groups opened all the envelopes and played for the entire allotted time, making statistical comparisons of these measures uninformative. Additionally, we found no condition differences for the percent of total time spent playing with the contents of the cued blue envelopes. We examined whether the groups differed in how many of the four cued blue envelopes were opened in the first minute of playtime (following standards to assess first minute of play in related studies^17,28). Controlling for race and age, the listening group opened more cued envelopes than the question asking group (Fig. 2c, QA: Mdn = 0, L: Mdn = 2, b = −1, 95% CI [−1.88,−0.12], t = 2.27, p = 0.02, p_FDR = 0.05).

We found no differences between conditions on the persistence measure (Fig. 2d, QA: Mdn = 110 s, L: Mdn = 132 s, b = −21.73, 95% CI [−64.13, 20.67], t = −1.02, p = 0.31, p_FDR= 0.39, n = 103). We also did not find a significant difference on the science learning measure (Fig. 1e, QA: Mdn = 62.5%, L: Mdn = 50.0%, b = 7.74, 95% CI [−1.93, 17.40], t = 1.59, p = 0.12 p_FDR = 0.20, n = 103).

Moderation effects on post-test measures

We examined whether baseline measures of cognitive skills and demographics moderated the impact of question asking on willingness-to-pay because we observed a main effect on this measure. We found a significant interaction between condition and baseline science knowledge on willingness-to-pay (Fig. 3a, WJ-IV Science x Condition: b = −0.40, 95% CI [−0.70, −0.10], t(97) = −2.65, p = 0.009; main effect of WJ-IV Science: b = 0.40, 95% CI [0.17, 0.63], t(97) = 3.45, p = 0.001). The condition difference in willingness-to-pay was greater for children with lower background science knowledge. We did not find moderations of condition on willingness-to-pay by other cognitive measures, age, sex, or stress exposure (Supplementary Figs. 1 and 3).

Fig. 3. Moderations of curiosity and learning outcomes by baseline cognitive skills.

The Listening group is shown in light orange and the Question-Asking group is shown in dark blue. a Scatterplot of the association between baseline science knowledge and willingness-to-pay, by condition (WJ-IV Science x Condition: b = −0.40, 95% CI [−0.70, −0.10], t(97) = −2.65, p = .009; main effect of WJ-IV Science: b = 0.40, 95% CI [0.17, 0.63], t(97) = 3.45, p = .001). b Scatterplot of the association between WISC vocabulary raw scores and science learning, by condition (WISC Vocabulary x Condition: b = −1.80, 95% CI [−3.40, −0.20], t(97) = −2.23, p = .028; main effect of WISC Vocabulary: b = 3.21, 95% CI [1.84, 4.58], t(97) = 4.65, p < 0.001).

We preregistered moderation analyses of baseline measures on science learning. We found that vocabulary interacted with condition to predict science learning (Fig. 3b, WISC Vocabulary x Condition: b = −1.80, 95% CI [−3.40, −0.20], t(97) = −2.23, p = .028; main effect of WISC Vocabulary: b = 3.21, 95% CI [1.84, 4.58], t(97) = 4.65, p < 0.001). The relationship between final science learning scores and vocabulary was weaker in the question-asking condition, suggesting that this intervention “equalized” children’s learning outcomes. We did not find moderations by other cognitive measures, demographics, or stress exposure (Supplementary Figs. 1 and 3).

At the suggestion of an anonymous reviewer, we also examined potential moderation by children’s attention problems as measured by the CBCL. This analysis was not pre-registered. We found some evidence that children with greater attention problems showed greater science learning in the question-asking condition (Supplementary Fig. 4; CBCL attention problems x condition: b = 4.13, 95% CI [0.08, 8.191], t (97) = 2.02, p = 0.046; main effect of CBCL attention problems: b = −2.23, 95% CI [−5.41, 0.94], t(97) = −1.40, p = 0.166). This relationship suggests that this intervention might better support children with attention challenges. However, because this effect of attention on science learning was driven by two outliers, and not preregistered, it should be interpreted with caution.

Discussion

We implemented an intensive science learning paradigm with 5–7 year-old children to test whether question-asking practice led to improvements in curiosity-related behaviors and learning. We found that children in the question-asking condition valued information more than children in the listening condition, demonstrated by a higher willingness-to-pay for new information. The condition effect on willingness-to-pay was larger for children with less baseline science knowledge. Children in the question-asking condition responded differently to a pedagogical cue meant to guide their exploratory play. No significant group differences were found for novel question asking, persistence, or science learning. However, the condition effect on science learning was moderated by baseline vocabulary. Because baseline knowledge moderated the condition effects on both willingness-to-pay and science learning, these results suggest that fostering question asking could be a promising component in interventions designed to support academic development. However, our sample included some demographic differences, so further research is needed to understand whether question-asking interventions could be tailored to address academic disparities in diverse populations.

Practice with question asking caused children to value new science information more. We chose willingness-to-pay because it is a commonly used measure of curiosity in adults, and even non-human primates⁵⁶. Changing how children value new information could be a powerful mechanism for increasing intrinsic motivation, with long-term benefits for science learning. The mechanisms by which asking questions improves willingness-to-pay for information are still unknown. It could be that when children ask questions, they practice surveying their existing knowledge, and notice how much they do not know yet. They could also feel empowered to seek new information, or learn that science is important or interesting, reflecting a greater interest in the content rather than curiosity⁵³. Children who started the intervention with a weaker academic foundation showed greater increases in how much they valued science information following the question asking intervention. Future work is necessary to understand whether these changes in children’s valuation of information will lead them to seek out more science learning opportunities in the future, building up their knowledge and skills.

We found some evidence that question-asking practice changed a novel measure of children’s exploration following a cued prompt. Children in the listening group opened more cued envelopes than children in the question-asking group. We did not predict this pattern of data, so any interpretation is speculative, but it could be that children in the listening condition may have focused on the pedagogical question’s goal in the exploration task because careful listening practice honed this sensitivity. However, we are cautious in over-interpreting this result, as differences between other exploratory measures on this task were not significant.

Children in the question-asking condition learned to ask more questions over the course of training. This finding is consistent with prior work showing that cognitive interventions lead to improvements on the practiced task^76–78. However, we did not find condition effects on the question generation measure, which we initially conceptualized as a near-transfer measure. Generating questions spontaneously because of an intrinsic drive for information is different from generating questions when prompted to do so. The question generation task was inspired by past work, but by prompting children to generate questions, we may have masked the potential effects of our training on spontaneous question asking. Instead, generating many questions over several minutes might require unpracticed skills like verbal fluency or creativity. Indeed, our vocabulary measure was correlated with the number of questions generated (controlling for age and group: t(102) = 2.34, p = 0.02).

We did not observe a significant main effect of question asking on science learning (p = 0.09). However, there was heterogeneity in this measure. For some children, asking too many questions could be distracting or could draw children too far from the lesson content, as can happen through “busybody”-like exploration⁷⁹. For others, asking questions could be helpful. Indeed, we found that question asking may be especially helpful for scaffolding learning for children with smaller vocabularies, or less baseline knowledge. Question asking could therefore be well-positioned to improve learning because questions create opportunities to monitor and reflect on one’s own knowledge gaps.

We conceptualized carefully listening as a kind of treatment-as-usual control group. Often in classroom settings children are reminded to pay attention. However, it could be the case that the careful listening directive negatively impacted children’s learning and curiosity, where-as the question asking had little effect. We cannot discern from this data whether both conditions had a positive effect and the question asking effect was stronger, or whether the careful listening had no effect or even a negative effect compared to question asking. We think it is implausible that both practice at careful listening and question asking would have a negative effect on learning, given the literature in education that suggests such pedagogical prompts can encourage active listening and learning^80,81, however it remains an open question regarding the degree to which students’ question asking bolsters curiosity above baseline. Unfortunately, a pre-post design was not possible with the chosen curiosity measures as the interest and value of the options would diminish with repeated exposure. Future work could explore these open questions with novel measures of curiosity and additional baseline groups.

The lack of differences on our persistence, prompted question asking, and science learning measures could raise questions about the true nature of the effects of question asking on curiosity. As per the preregistration, we hypothesized effect sizes that were larger based on prior work and piloting than we found in practice. These differences could have been due to the online nature of testing, duration of our intervention, or temporing from the pandemic sample. Additionally, our recruitment for an online Science Camp may have preselected participants with already higher interest in science topics, which could have resulted in either stronger or weaker results depending on whether a question intervention might be bolstered by initial interest or whether there are potential ceiling effects. We expect that larger effects would probably be revealed if this strategy were deployed over longer time scales, which we hope is an avenue for future work. Additionally, we chose a diversity of measures because our goal was to tap into different aspects of motivation related to curiosity, including potentially more distal effects of persistence and learning. Our null results on those particular measures may reflect true null differences. However, they could also be due to differences in the psychometric validity of the measures. Unfortunately, there are currently not many psychometrically validated tools for measuring these aspects of curiosity, but we hope this project is a first step towards inspiring psychometrically valid measures in this domain, which will be critical for future work.

Despite random assignment, the groups were not matched in racial/ethnic composition. We therefore controlled for race and ethnicity in the core analyses. It is also worth noting that the groups did not differ in the racial/ethnic identities of the research team with whom they interacted. 13 children in the question asking group and 12 children in the listening group were taught by a researcher from an underrepresented racial or ethnic background. 36 children in the question asking group and 36 children in the listening group were tested by a researcher from an underrepresented racial or ethnic background. However, there could have been interactions between researcher and student identities that were difficult to statistically control, particularly in a small sample.

We are cautious in generalizing our results beyond the specific content and delivery of the paradigm and the population included. Question asking might be particularly good for science content because such knowledge may be represented as causal, abstract, and explanatory; the “why” and “how” questions preschoolers tend to generate in these domains likely supports such conceptual development. Such training may not have the same impact in other domains like learning basic math facts or developing reading skills. Further, some features of our intervention design could have introduced unexpected effects. The intervention was delivered remotely at a time when children had a mix of experiences with virtual school. An in-person intervention might have different impacts. Additionally, the post-test was administered by a researcher who was unaware of the child’s condition to minimize potential biases, but this also meant that the absence of rapport could have impacted performance⁸². Finally, our sample included few children from low socioeconomic backgrounds, limiting our ability to detect moderating effects of demographics and stress exposure on intervention outcomes. Further, children in this sample were likely experiencing adversities related to the pandemic, which could have obscured associations between stress and learning. In future work, we will seek a better understanding of which pedagogical strategies bring more equity to the classroom by accounting for variability in learning needs. Future work is also necessary to map out the contexts in which question asking is most effective: does the identity or group membership of a teacher or a student influence question asking dynamics or the efficacy of question-based pedagogy?

The current study serves as proof-of-principle that question asking can improve curiosity-driven behaviors. The intervention effects were relatively small. We chose a constrained condition difference for analytical precision, but a real-world intervention to enhance curiosity might produce bigger effects with multiple active ingredients, as holistic interventions are typically most effective for globally improving cognition^83,84. Encouraging question asking is simple, allowing teachers and parents to use the strategy with minimal training. Indeed, this approach converges with recent suggestions from the education literature that teachers should model questions in the classroom to support curiosity⁸⁵. Further, question asking can be applied to diverse content domains in diverse settings. We hope that by sharing this rich dataset of children’s science learning, future work will develop hypotheses about the mechanisms connecting question asking to greater curiosity and learning. In sum, our findings demonstrate that question asking can improve how children value new information, and opens the door for future investigation: who might benefit most from learning to ask questions?

Author contributions

All authors, A.P., J.C., L.D., S.S., A.K., E.B., and A.M. contributed to this project and have read and approve this manuscript.

Data availability

Data is available at: https://osf.io/zsj9h/?view_only=72707fec61b449ebb80a5159809ac86b.

Code availability

Code for analysis is available at: https://osf.io/zsj9h/?view_only=72707fec61b449ebb80a5159809ac86b.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41539-025-00384-5.