Authors: Robbee Wedow, Yeongmi Jeong, Katherine N. Thompson, Kathryn Fiuza Malerbi, Andrew Brubaker, Monica Weindling, Stanley M. Lo, Jamie Amemiya, Brian M. Donovan
Categories: Article, genetics education, genetic essentialism, randomized controlled trial, multifactorial causation, population thinking, Humane Genomics Literacy Curriculum, HGL Curriculum
Source: Human Genetics and Genomics Advances
Authors: Robbee Wedow, Yeongmi Jeong, Katherine N. Thompson, Kathryn Fiuza Malerbi, Andrew Brubaker, Monica Weindling, Stanley M. Lo, Jamie Amemiya, Brian M. Donovan
Despite advancements in genomics, misconceptions about the extent to which genetics contributes to observable differences across racial groups persist. These misconceptions are often rooted in genetic essentialism, a social-cognitive bias that leads individuals to believe that most complex traits are primarily determined by genetics. This scientifically inaccurate belief overlooks the environmental and social influences on complex human outcomes, reinforcing deterministic views about human diversity. Our study examines how and for whom genetics education can reduce genetic essentialist beliefs using targeted interventions. We use data from a randomized controlled trial collected at a large US West Coast public university in 2023, including 2,061 undergraduate students. Participants were randomly assigned to one of four curriculum-based interventions, ensuring balanced characteristics across conditions. Three interventions were population thinking; multifactorial causation; and a curriculum where we combined both approaches, which we call full Humane Genetics Curriculum. Results are reported relative to a control group that taught students about climate change. Using structural equation modeling, we explore the effectiveness of these interventions with our data. We find that all three interventions reduce genetic essentialist beliefs by decreasing the perception of between-group racial variation and by reducing genetic attributions for complex human traits. We also find that the three intervention curricula are highly effective across sociodemographic group characteristics such as self-reported gender, self-reported race, and cultural/political belief systems. However, the interventions were more effective among students who possessed greater baseline genetics knowledge. Using these findings, we offer evidence-based strategies for curriculum development.
The US public often lacks a general understanding of genomic science, including how genes relate to individual and racial group differences in complex traits.^1^ This may be due in part to the fact that genetics education tends not to focus on the complexity of modern human genomics.^2^ For example, two major findings from the last 20 years of behavioral and statistical genetics research are not reflected in current genetics standards. First, human traits, such as intelligence, athleticism, or depression levels are complex because they are influenced by thousands of gene variants, each with small effects that are dynamically intertwined with a person’s surroundings and experiences. Such traits cannot be reduced to single-gene explanations or polygenic factors alone.^3^ Second, we do not have a precise understanding of the specific genetic factors underlying complex traits.^4^^,^^5^^,^^6^ However, high school and even university genetics education often ignores complex trait biology and instead focuses primarily on classical (i.e., Mendelian) and molecular genetic explanations for simple traits. Some evidence suggests that this educational oversight can place students at risk of developing the inaccurate idea that genetic influences on complex human traits are deterministic and that racial group differences in such traits are also genetically caused.^7^^,^^8^
The belief that genes are the defining characteristics of a person, or any group of people, is known as genetic essentialism.^9^ Individuals who adopt this scientifically inaccurate belief assume that racial or ethnic groups are different from one another primarily because of their genes. This minimizes the integral role of the environment by incorrectly attributing individual-level or group-level trait variation, as well as behaviors and identities, to an unspecified genetic “essence.”^10^ Genetic essentialism is scientifically inaccurate for many reasons. One reason is that genetic variation within human populations is proportionally greater than the genetic variation between such populations. Therefore, genetic essentialism overestimates the contribution of between-group genetic variation to the development of complex human traits. (While this statement is generally true, it is important to remember important caveats. Although the relative amount of within-group neutral genetic variation can provide expectations of within-group genetic variation for a quantitative trait, this depends on the features of the trait under investigation. These values can diverge from one another if a given trait is under divergent selection and/or involved in local adaptation [e.g., skin color is a prime example where within-group variation is typically larger than between-group variation, despite what we know about within-group variation for neutral genetic markers]. Human fixation index values indicate that for a given trait we ought to expect small differences unless or until more details about the evolutionary history of the trait arise.^11^^,^^12^^,^^13^) This inaccuracy leads to a simplistic and baseless view that genes are the primary rationale for explaining group differences in human traits.^14^^,^^15^
More important, a growing body of work shows that introducing high school and undergraduate students to genomic and population genetic concepts that refute essentialist views prevents students from developing stronger genetic essentialist beliefs.^12^ In 2024, for example, Donovan and colleagues used data from two different randomized trials to argue that this approach to genetics curriculum is needed to prevent the development of genetic essentialism in adolescents.^13^ This curriculum, called the Humane Genetics Curriculum (HGC),^6^ builds upon the traditional, Mendelian-based curriculum taught in most US schools in two ways. First, HGC incorporates scientifically accurate population-based thinking about the relative amount of within-group and between-group genetic variation. Second, it helps students understand the multifactorial inheritance of complex human traits, in which trait variation within and between human populations is influenced by a combination of polygenic effects, environmental effects, and gene-environment covariance. Rather than ignoring that phenotypic differences between groups exist or that there are some biological differences between groups, HGC uses scientific evidence to help students disentangle the relative contributions of genes and the environment in regard to trait variation within and between groups.^6^ For example, HGC uses the concept of gene-environment covariance to help students understand that genes are an especially poor explanation for variation across groups, given that genetic differences are strongly confounded by environmental differences (an issue that the intervention notes is also still relevant when explaining variation within groups). Donovan and colleagues found that such instruction caused an increase in students’ overall knowledge about genomics and human genetic variation and a decrease in their genetic essentialist beliefs.^11^^,^^12^^,^^13^ This work takes a critical step in demonstrating that educators can use this updated genetics curriculum as a tool to refute scientific inaccuracies in genetic essentialist views of human populations.
While we know that HGC works to improve genetics knowledge (GK) and reduce genetic essentialist beliefs, we have little insight into which genetic concepts are responsible for the reduction in essentialism. The HGC focuses on two key genetic variation (within and between racial groups) and sources of causation (genetic vs. environmental attributions for complex traits). Prior studies demonstrate that when these concepts are taught in tandem, it causes a reduction in essentialist beliefs.^13^ However, it is unclear whether any such reductions in essentialism are facilitated by developing a better understanding of human genetic variation (i.e., within-group vs. between-group variation) or multifactorial inheritance of complex traits (i.e., genetic vs. environmental attributions). Understanding how the acquisition of genetic concepts relates to changes in essentialist beliefs can provide insight into how to design a better human genetics curriculum. This is the focus of our study.
Relatedly, learning is not simply a passive transmission of knowledge. Decades of cognitive psychology research and science education research demonstrates that learners actively construct their knowledge through an interplay between their prior knowledge, interests, and identity.^16^ For the HGC to be successfully implemented and tailored toward all students, we must therefore understand how the prior beliefs and identities of students influence how they make sense of the ideas on the HGC. HGC may work better or worse for students of different backgrounds, knowledge, and cultural values because of a variety of social-psychological mechanisms.^12^
First, self-reported gender may bias intervention efficacy of the HGC because of socially desirable reporting. Social desirability bias (i.e., the tendency of individuals to answer questions in a way they believe is socially acceptable rather than providing their true beliefs) can be stronger in women than in men, especially when it comes to issues of racial equality.^17^^,^^18^ Men more often attribute racial disparities in outcomes like income to genetic differences.^19^ Therefore, women may be less likely to self-report their essentialist beliefs. Socially desirable reporting could therefore bias the effect of the HGC on genetic essentialism.
Second, the effectiveness of the HGC could differ across racial groups. For instance, white individuals are known to use essentialist ideas when rationalizing the exclusion of racial outgroups, but they reject such ideas if they are used to rationalize the discrimination of racial ingroup members.^20^ Therefore, if white individuals perceive a threat to their ingroup identity as they make sense of the ideas in the HGC, it could introduce a bias into any estimation of the effectiveness of the HGC intervention.
Third, the HGC may be more successful depending on a person’s political affiliation. When White US citizens were exposed to messages about the genetic similarity of racial groups, those with more liberal beliefs about how society should be structured interpreted this as consistent with an environmental explanation for racial inequality, whereas those with more hierarchical beliefs who were self-identified conservatives tended to interpret the same information as consistent with a genetic explanation for racial inequality.^21^ Relatedly, individuals who think that society should be organized hierarchically are also more likely to possess views about the homogeneity of same-race individuals and the discreteness of different racial groups.^22^ Thus, the efficacy of the HGC in changing essentialist beliefs could be greater among individuals with high levels of hierarchical beliefs since they tend to have more inaccurate views about patterns of human genetic variation and thus more potential to dramatically revise these views to be more anti-essentialist.
Fourth, research suggests that more prior knowledge of genomics facilitates greater changes in students’ essentialist views in response to humane genomics instruction because of an expertise effect.^12^ For instance, students with a larger understanding of genomics show greater comprehension of HGC learning materials and are thus more likely to develop the idea that between-group genetic variation is proportionally low (relative to within-group variation) in humans. Once developed, this idea appears to facilitate a greater reduction in their genetic essentialist beliefs about race. However, we do not know whether this mechanism is driven by students’ developing an understanding of genetic variation within groups versus an understanding of genetic variation between groups.
In this study, we build on the work of Donovan et al.^13^ by assessing four learning mechanisms that could facilitate a conceptual change in a student’s belief in genetic essentialism. These involve learning about (1) within-group genetic variation, (2) between-group genetic variation, (3) genetic attributions for complex traits, and (4) environmental attributions for complex traits. We explore how these mediation mechanisms interact with different aspects of a learner’s identity (e.g., their self-identified gender, race, cultural/political views, GK) to alter treatment effect heterogeneity of the HGC. To estimate these effects, we use data that were generated via a randomized controlled trial (RCT) conducted at a large US West Coast public university. This RCT delivered the HGC intervention conditions and control online to undergraduate students recruited through biology courses in the School of Biological Sciences (N = 1,060) and the Department of Psychology (N = 1,001). We use structural equation modeling (SEM) to assess our hypotheses and build a better understanding of the contextual factors that enable undergraduate students to change their belief in genetic essentialism.
We first tested whether we could pool the two samples of undergraduate psychology and biology students (Table 1). We regressed undergraduate major, each of the baseline covariates, and demographic variables on each of the three curriculum conditions, using logistic or linear regression depending on the outcome. We pooled our two samples as the three treatment conditions were not significantly different across the two undergraduate majors (Table 2). Moreover, the intervention groups were equivalent at baseline in the proportion of students of self-identified genders (χ^2^(3) = 0.563, p = 0.905), self-identified races (χ^2^(3) = 4.807, p = 0.187), political affiliation groups (F(3,1,815) = 0.311, p = 0.818), levels of baseline GK (F(3,2,057) = 1.554, p = 0.199) and levels of cultural theory of risk (CTR) (F(3,2,057) = 1.658, p = 0.174).Table 1Descriptive statistics for baseline predictors and demographic characteristicsBiology (N = 1,060)****Psychology (N = 1,001)****ProportionSelf-identified gender (N = 2,004) Female0.6970.737 Male0.3030.263Self-identified race (N = 2,061) Non-White0.8780.875 White0.1220.125Political affiliation (N = 1,819) Liberal0.5640.588 Non-liberal0.2370.230 Haven’t thought about it much0.1990.182**Mean (SD)**Cultural theory of risk (N = 2,061)4.839 (0.873)4.847 (0.934)Genetics knowledge (N = 2,061)5.708 (1.993)5.609 (2.059)Table 2Baseline equivalence(1)(2)(3)(4)(5)(6)MaleWhiteMajorPolitical affiliationGenetics knowledgeCultural theory of riskFHGC−0.068 (0.140)0.026 (0.129)0.011 (0.125)−0.050 (0.052)0.079 (0.096)0.173 (0.099)MC−0.082 (0.140)−0.183 (0.128)−0.004 (0.125)−0.030 (0.053)−0.131 (0.100)0.080 (0.100)PT−0.096 (0.140)0.081 (0.130)−0.012 (0.125)−0.027 (0.053)−0.051 (0.099)0.203 (0.103)aIntercept−0.864 (0.098)b0.537 (0.091)b0.058 (0.088)1.642 (0.038)b1.318 (0.068)b1.684 (0.069)bN2,0042,0612,0611,8192,0612,061R^2^–––0.0010.0020.002Estimation methodLogitLogitLogitOLSOLSOLSTest statistic0.5634.8070.0350.3111.5541.658Test p value0.9050.1870.9980.8180.1990.174Standard errors of the mean are in parentheses. FHGC, Full Humane Genetics Curriculum; MC, multifactorial causation; OLS, ordinary least squares; PT, population thinking.ap < 0.05.bp < 0.001.
Students who participated in the HGC showed a significantly lower endorsement of genetic essentialism compared with students who received the climate change control intervention (Table 3). This suggests that by enhancing students’ genomic literacy, HGC can reduce beliefs in genetic essentialism. The effect size on genetic essentialism for the full HGC (FHGC) was greater than both the multifactorial causation (MC) and population thinking (PT) curricula. This difference was significant for the MC but not for the PT curriculum, as indicated by no overlap between the MC effect estimate and the confidence intervals of the FHGC and vice versa.Table 3The effects of all interventions on genetic essentialismGenetic essentialismFull Humane Genetics Curriculum−0.443c (–0.626 to –0.260)Multifactorial causation curriculum−0.174a (–0.340 to –0.008)Population thinking curriculum−0.274b (–0.450 to –0.097)Intercept−1.363c (–1.481 to –1.245)N2,061R^2^0.01295% confidence intervals in parentheses.ap < 0.05.bp < 0.01.cp < 0.001.
We tested the indirect effects from each treatment condition on genetic essentialism via four within-group variation, between-group variation, genetic attribution, and environmental attribution. We first hypothesized that the FHGC treatment intervention would reduce genetic essentialist beliefs through all four mechanisms. Consistent with this prediction, we found significant indirect effects through all four mechanisms for the FHGC intervention (Table 4). Across between-group, genetic attribution, and environmental attribution mechanisms, the indirect effects for the FHGC were stronger than that of the PT and MC interventions, although this difference was significant only for genetic and environmental attribution. The FHGC was the only intervention to show significant indirect effects of within-group perception (β = −0.0305, 95% confidence interval [CI] = −0.0681 to −0.0058]). We demonstrate, as predicted, that the FHGC worked through all four mechanisms.Table 4Indirect effects of three treatment conditions on genetic essentialismFull humane geneticsMultifactorial causationPopulation thinkingIndirect effectsBC 95% CI lowerBC 95% CI upperIndirect effectsBC 95% CI lowerBC 95% CI upperIndirect effectsBC 95% CI lowerBC 95% CI upperWithin-group variation−0.0305−0.0681−0.00580.0255−0.00530.0629−0.0387−0.08960.0071Between-group variation−0.2763−0.3603−0.2064−0.1528−0.2254−0.0955−0.2464−0.3285−0.1782Genetic causation−0.2410−0.3401−0.1606−0.1149−0.1798−0.0616−0.1201−0.1830−0.0696Environmental causation−0.0846−0.1403−0.0411−0.0375−0.0741−0.00770.01720.00190.0472Observations1,033––1,029––1,029––Bonferroni-adjusted 95% confidence intervals (BC 95% CI) are reported.
Next, we assessed whether these effects resulted from the domain specificity of the knowledge taught through the intervention. We hypothesized that indirect effects through within- and between-group variation would be stronger in response to the PT intervention relative to the MC intervention. This is because the PT intervention was designed to facilitate a reduction in genetic essentialism by teaching students that genetic variation between groups is relatively low and variation within groups is high. We expected an opposite pattern of results for the MC intervention relative to the PT intervention, whereby indirect effects via genetic and environmental attributions would be strongest in response to the MC intervention. This is because the MC curriculum materials were designed to demonstrate that variation in complex traits is weakly associated with genetic factors and more strongly with environmental factors.
We found mixed results (Figure 1; Table 4). First, there were no indirect effects via within-group variation for the PT or MC interventions. Second, the indirect effects via between-group variation were stronger in response to the PT intervention as compared to the MC intervention. Third, the indirect effects via environmental attributions were stronger in response to the MC intervention compared to the PT intervention. Fourth, the indirect effects via genetic attributions did not differ across both MC and PT interventions. These findings suggest that the PT and MC interventions had some domain-specific and also domain-general effects. In general, these two different concepts reduced essentialist beliefs by causing learners to believe that complex human traits are weakly influenced by genes. However, instruction about MC facilitated a reduction in genetic essentialism via a greater increase (relative to the PT intervention) in environmental attributions Similarly, the PT intervention had a more precise domain-specific effect on genetic essentialist beliefs via changes to learners’ perceptions of between-group genetic variation.Figure 1The indirect effects of three treatment conditions on genetic essentialismControl condition is indicated by the dashed red line. Each estimate represents a difference from the control group, denoted by the dashed red line at zero. Bonferroni-adjusted 95% confidence intervals are reported.
As the FHGC operated through all four mechanisms and was most effective in reducing genetic essentialism, we assessed whether its effectiveness varied based on the characteristics of the students. We hypothesized that the full curriculum would reduce beliefs in genetic essentialism differently across types of students. That is, indirect effects (the A×B path in Figure 2C) would vary by self-identified gender, self-identified race, egalitarian views of how society should be structured (CTR), and baseline GK.Figure 2Three statistical models analyzed in the present studyThree statiscial models analyzed in the present study. Panel A indicates our linear regression models, Panel B indicates our mediation models, and Panel C indicates our moderated mediation models.
We found no evidence of moderation in the indirect effects via gender, race, or egalitarian views. There was no moderation on the A×B path. Thus, the full curriculum worked equally well through all four mechanisms regardless of these student characteristics. There were no statistically significant differences in the indirect effects between men and women (Figure 3A; Table S1), White and non-White racial groups (Figure 3B; Table S2), and students with high and low CTR (Figure 3C; Table S3). We report robustness checks for a different cutoff for high and low levels of CTR in Note S1 (Figure S1; Table S5) and find similar results.Figure 3Moderated mediation resultsModerated mediation figures showing indirect effects of the Full Humane Genetics Curriculum for gender (A), race (B), cultural theory of risk (C), and baseline genetics knowledge (D). Each estimate represents a difference from the control group, denoted by the dashed red line at zero. Bonferroni-adjusted 95% confidence intervals are reported.
For prior GK, however, we show that the full curriculum had a greater impact in reducing essentialist beliefs through the genetic attribution mechanism among students with higher prior GK compared with those with lower baseline GK (Figure 3D; Table S4). We found significant moderated mediation, indicating that the indirect effect (A×B path) of the full curriculum on genetic essentialism operating through genetic attribution varies across levels of GK. GK did not significantly moderate any of the other within-group variation, between-group variation, and environmental attribution (Figure 3).
We then separately tested whether prior GK moderated the effect of the full curriculum on the genetic attribution mechanism (the A path) or the effect of the genetic attribution mechanism on genetic essentialism (the B path). We found no indirect effect due to moderation on the A path or the B path separately (Table S7), suggesting that the moderation occurs at the level of the overall indirect effect, rather than through moderation of the individual pathways. This suggests that teaching students about the limited contribution of genetics to complex human traits leads to a reduction in essentialism more so for those who started with higher baseline GK. We report robustness checks for a different cutoff for high and low levels of GK in Note S1 (Figure S2; Table S6) and find similar results.
The effects of the FHGC suggest that genetic essentialist beliefs can be reduced by altering (1) how students perceive patterns of human genetic variation and (2) how students attribute complex human trait variation to patterns of human genetic variation. The relative effects of the MC and PT conditions further suggest that genetic essentialist beliefs are reduced through two distinct but related psychological (1) causal reasoning (represented by the MC condition results) and (2) social categorization (represented by the PT condition results). The social categorization mechanism involves how learners perceive between-group and within-group genetic variation. The causal reasoning mechanism, in contrast, involves the extent to which learners attribute complex trait variation to genetic (i.e., genetic attribution) versus environmental variation (i.e., environmental attribution).
Both mechanisms are related, because each appears to reduce essentialism by affecting the extent to which students attribute complex traits to genes (see Figure 1, genetic attribution indirect effects). However, each mechanism is also distinct, because each operates in a domain-specific manner. For example, the social categorization mechanism (relative to the causal reasoning mechanism) appears to reduce belief in genetic essentialism by causing a greater reduction in students’ perceptions of the amount of genetic variation that exists between/across groups (see Figure 1, between-group variation indirect effects). Meanwhile, the causal reasoning mechanism (relative to the social categorization mechanism) appears to reduce genetic essentialist beliefs by causing a greater increase in students’ environmental attributions for complex traits (see Figure 1, indirect effects of MC vs. PT condition for environmental attributions).
Each psychological mechanism also has unique educational constraints that may affect its power to regulate genetic essentialist beliefs. For example, our results suggest that reductions in genetic essentialism are mediated more strongly by students’ perceptions of between-group genetic variation than their perceptions of within-group genetic variation (see within-group and between-group variation results of FHGC in Figure 1). Previous studies have found that learners have difficulty conceptualizing within-group variation.^23^ Other studies have found that when generic noun phrases (the phrase “lions roar” is a generic noun phrase, whereas “some lions roar” is not) are used to describe collections of individuals, they tend to cause people to think of the individuals within those groups as more homogenous.^24^ Therefore, when learners read sentences that include generic noun phrases such as “There is a lot of genetic variation within Black people,” it may attenuate their ability to develop an accurate perception of within-group variation. These two factors, generic noun phrases and cognitive difficulty, may explain why the social categorization mechanism appears to operate more strongly through perceptions of between-group variation.
Relatedly, when it comes to causal reasoning, reductions in genetic essentialism were mediated more strongly by learners’ genetic attributions for complex traits than their environmental attributions. Previous studies have found that individuals tend to prioritize causal factors that are believed to be specific, direct, and stable across contexts.^25^ Causal factors that are specific have a one-to-one relationship with their effect; they cause only one outcome. Direct factors do not operate through other factors via mediation. Stable factors do not operate in conjunction with other factors via moderation. Several studies of the curriculum and teaching methods involved in high school genetics education demonstrate that the gene is often conceptualized for students as a specific, direct, and stable cause of traits.^26^ For instance, many students learn about “genes for” traits, and textbooks rarely discuss how the effects of genes depend on the environment. The only factors that mediate the effects of genes in the genetics curriculum are transcription and translation factors, and the curriculum rarely discusses how the environment can epigenetically moderate gene expression.^2^^,^^26^ Therefore, students may be educationally and cognitively predisposed to explain trait variation via genes. If so, then this may explain why reductions in genetic essentialism in our study were facilitated more strongly by reductions in genetic attributions than increases in environmental attributions—students are more predisposed to explain traits in a genetic versus environmental manner.
While we contend that the cognitive and educational mechanisms described above are the most probable explanation for our results, we are nevertheless aware of the causal inference issues with mediation models,^27^^,^^28^^,^^29^especially when it comes to omitted variable bias and the directional dependence of results. To address these concerns, we performed a number of robustness tests. In Note S1, we find that our results are robust after controlling for the covariates undergraduate major, gender, race, measures of CTR, GK, and political affiliation (Table S8). We next performed a serial correlation test (Table S9), and the results provide reasonable evidence that there are no unmeasured confounders influencing both the mediators and the genetic essentialism outcome. Importantly, we are also concerned about directional dependence, in that we want to be certain that the mediators affect genetic essentialism, and not the other way around. In Tables S10–S12, we report the results of a residual distribution test, a robust Breusch-Pagan test of heteroscedasticity, and a non-linear correlation test. The results provided evidence that the mediators affect genetic essentialism, instead of genetic essentialism affecting the mediators. In light of these results, we remain reasonably confident that our results do not suggest alternative hypotheses that would nullify the explanations described in the previous paragraphs. Nevertheless, additional studies are needed to test the validity of the proposed interpretations for our findings. If our interpretations are correct, then they suggest two mechanisms for reducing the prevalence of genetic essentialist beliefs about race. One involves teaching about human population genetics, and the other involves teaching about the multifactorial inheritance of complex human traits.
That said, the conceptual change of strongly held beliefs often depends on a learner’s prior knowledge, motivations, and identities.^16^ However, we found no difference in indirect effects across self-reported gender and race or across different cultural values/political belief systems. These null results tentatively suggest that HGC education may be an effective intervention for reducing essentialist beliefs regardless of variation in learner identity and motivation. However, this assertion is more of a hypothesis than a generalizable claim since our sample was obtained from a single university setting and it could be biased by potential power issues (i.e., an overrepresentation of women in the sample, as well as an underrepresentation of some minority groups and conservative viewpoints).
At the same time, we did find that HGC education was more effective for students with a higher level of baseline knowledge about genetics, including knowledge of Mendelian and molecular genetics, as well as polygenic and multifactorial inheritance. Our results therefore suggest that a learner’s prior knowledge about genetics is more important than their ingroup identities or political motivations when it comes to the conceptual change of essentialist beliefs that can be triggered by an HGC via the mediation mechanisms described above. Our findings thus suggest a learning progression for reducing genetic essentialism.
The literature on learning progressions suggests that there is an optimal sequential order for teaching scientific concepts to students, progressing from the least to most cognitively difficult concepts.^30^^,^^31^^,^^32^ While learning progressions have limitations (e.g., the assumption that learning is linear and not recursive or dependent on identities or experiences), our results do suggest an optimal conceptual sequence for learning about race and genetics in a university setting. Specifically, our study suggests students should first learn about Mendelian inheritance and its molecular basis in the central dogma and then move beyond Mendel to learn about polygenic inheritance. Next, students should learn about the MC of complex traits. Armed with this knowledge, the stage is set to help students understand the concepts involved in HGC education—namely that genetic variation is proportionally larger within human populations than between them, and thus complex trait variation within racial groups is more plausibly related to genetic inheritance than is trait variation between such groups since racial groups experience different social environments, on average, in the United States. These are the core ideas that an HGC educator can teach students to refute genetic essentialist beliefs about race. Our study demonstrates how this approach may work for learners of varying identities, political motivations, and prior knowledge.
This study was approved by the University of California, San Diego institutional review board under project number 800880.
Our RCT data were collected in 2023 and includes undergraduate students (N = 2,061) recruited through biology courses in a school of biological sciences (N = 1,060) and through a department of psychology subject pool (N = 1,001) at a large US West Coast public university. Students provided informed consent to participate, and the study procedure was approved by the institutional review board of the university where our intervention was given.
For students in our psychology sample, our intervention was one of several online studies students could choose to take part in for course research credit. For our biology sample, the intervention was made available to instructors who then offered it to students for course extra credit. While our psychology sample has previously been used as a replication dataset by Donovan et al.,^13^ the biology sample data have never been used in a published study.
The RCT consisted of four curriculum conditions, including three treatment groups and one control (1) PT, whereby students are taught that genetic variation within groups is proportionally greater than between groups; (2) MC, whereby students are taught that complex traits are caused by both genes and the environment and that group differences in traits are better explained by environmental factors; (3) FHGC, which combines both PT and MC; and (4) a climate change control curriculum, which teaches students about patterns in temperature and precipitation that vary within and between geographic regions over time and how these patterns are attributed to human behavior.
Students learned the contents in the climate curriculum in the same way and by using the same pedagogical practices as the material in the genetics curriculum was presented to ensure comparability across treatment and control groups (i.e., online video modules). Climate change was chosen as our control condition topic to control for the effects of learning about widely debated complex scientific concepts and to control for the learning difficulties that students have when it comes to reasoning about variability and causal inference (see Donovan et al.^11^^,^^12^^,^^13^ for more elaboration on the validity of the learning condition as a control). Therefore, this RCT tested the effects of each component of the curriculum independently (PT and MC) and when combined (FHGC), compared with a control condition (climate change).
Undergraduates were randomized into one of the four computerized learning conditions, each with equal probability of 0.25 for assignment into each condition (N = 514 in PT, N = 514 in MC, N = 518 in FHGC, and N = 515 in the control condition). Unlike cluster-randomized designs, the individual randomization into each of our four conditions ensures an even distribution of both observed and unobserved characteristics across the four experimental conditions,^13^ which eliminates the need to adjust standard errors of the treatment effect for within-group correlations across classrooms. Therefore, we use ordinary least squares (OLS) multivariate regression to generate efficient and unbiased estimates of the treatment effects (see Weiss et al.^33^ for additional arguments in support of this estimation method). The climate change control curriculum is used as the reference group across all analyses.
The RCT was conducted in the same way across all conditions. First, students read the information about informed consent online and agreed to be a part of the study. Second, we assessed baseline characteristics of the students. Third, students watched several video clips and answered quick, basic knowledge check questions after each video. Fourth, the students completed post-test assessment questions, and we collected demographic information from the students. In total, the intervention lasted about 90 min for each student. We provide an example of a video clip that closely matches the style and content of a clip used in our intervention as Video S1.
Video S1. An example video clip that closely matches the style and content of a lesson from our Humane Genetics Curriculum (HGC) online intervention materialsSome images in the movie are used/adapted with permission from Donovan et al. 2019. Creative Commons license CC-BY-NC-ND.
Our PT condition and its respective online module was aimed at increasing a scientifically accurate understanding that genetic variation is proportionally greater within than between groups. Our MC condition and its respective online module was aimed at increasing a scientifically accurate understanding that complex human outcomes are caused by both genes and the environment but also that group differences in these outcomes are better explained by environmental causes. Our FHGC condition and its respective online module was aimed at increasing a scientifically accurate understanding for all of these.
We use the Rasch measurement model to construct our dependent and baseline-measured moderator and post-test-measured mediator variables (described below). This transforms raw scores on survey items to infer a latent construct on an interval scale. Unlike traditional item response theory (IRT) methods that fit models to data, Rasch modeling fits data to models, allowing a researcher to characterize how items might perform for comparing student ability to item difficulty.^34^ The Rasch model uses the intuition that a person who is scoring better overall on a series of questionnaire items should also be getting more of the harder questions correct (i.e., when the test scale is a knowledge assessment). A comparison of expected and observed scores can therefore indicate when a valid instrument in a latent construct is created, and individuals who are likely guessing on questions (i.e., students who are performing poorly overall, but are getting more difficult questions correct) can be identified and temporarily removed from the Rasch modeling method to improve estimates of item difficulty. Then, these students are added back into the analysis to yield person-measures of the latent constructs using unbiased item difficulties.
Each construct was created using WINSTEPS 5.2.3.^35^ When question responses used a Likert scale (i.e., our genetic essentialism questions or our cultural theory of risk questions; see below), or a slider bar option (i.e., our mediator variables; see below), a Rasch partial credit model was applied.^36^ This is a two-parameter logistic (2PL) IRT model with latent regression. When the question response used multiple response options with only one correct answer (i.e., our GK questions; see below), a Rasch one-parameter logistic (1PL) model was applied.^37^
To assess our Rasch items, we examined (1) reliability, or how reproducible the items and students are along a continuum, which allows us to distinguish between high and low levels of ability in students and difficulty in items; (2) item and person fit using unstandardized mean square fit values, allowing us to gauge data-to-model fit; and (3) dimensionality using principal-component analysis of residuals, ensuring that our measures are valid by testing unidimensionality.
The descriptive statistics for the baseline predictors assessed are shown in Table 1, split by biology and psychology students. Descriptively, the two subsamples present similarly. Both reflect the racial and gender demographics for our large US West Coast public university when the data were collected in 2023, and students have a relatively high belief in an egalitarian organization of society.
At the end of the computerized module for each of our four conditions, we assessed baseline sample demographic characteristics to be used as moderator variables in our analyses. We measured students’ self-identified gender (0 = female and 1 = male), self-identified race (0 = non-White and 1 = White), self-identified political affiliation (0 = moderate, conservative or “haven’t thought about it much,” and 1 = liberal), and undergraduate major (1 = students recruited from biology courses in a school of biological sciences and 2 = students recruited from a department of psychology subject pool). Note that not all but most students recruited through biology courses had biology or closely related majors (75%), and most students recruited through the psychology subject pool had psychology or closely related majors (70%).
CTR was assessed at baseline and captures how people interpret danger and trust or distrust institutions.^38^ In part, CTR provides a measure of the extent to which individuals believe society should be structured in an egalitarian fashion or not. Students were asked to evaluate six statements from the egalitarian-hierarchical component of the CTR scale^39^ with six-point Likert scales with responses from “strongly disagree” to “strongly agree.” Some of these statements included “We have gone too far pushing equal rights in this country” or “Our society would be better off if the distribution of wealth was more equal.” An average score on these measures of 3 or below indicates that a student believes society should be organized around social hierarchies, while an average score of 4 or above suggests that students believe in a more egalitarian organization of society. Our final variable was a 2PL Rasch partial credit model,^36^ as described above.
We measured students’ GK at baseline using eight questions, each with five response categories, with only one scientifically correct response each. We provide two example questions in Note S1.
Higher scores on these questions indicate greater knowledge of Mendelian patterns of inheritance, greater knowledge of patterns of genetic variation in human populations, and greater knowledge of multifactorial inheritance. Our final variable was a 1PL Rasch partial credit model.^37^
To test the effects of the three treatment conditions, compared to the control climate change condition, we created four mediating variables, which we call (1) within-group variation, (2) between-group variation, (3) genetic attribution, and (4) environmental attribution.
Students were asked a series of 22 questions. In the first set of 10 questions, students were asked to slide a bar ranging from 0 to 100 to indicate by what percentage individuals of the same group differ from one another in their variable DNA, for Black, White, Asian, and Native American individuals separately. Students were then asked to do the same but this time to indicate by what percentage individuals of different groups differ from one another in their variable DNA, for White versus Black, Asian versus Black, White versus Asian, White versus Native American, Black versus Native American, and Asian versus Native American groups separately. In the second set of 12 questions, students were asked to slide a bar ranging from 0 to 100 to indicate by what percentage individuals differ from one another in terms of their (1) social environment, (2) genes, and (3) personal choices, separately for (i) science ability, (ii) the structure and function of their bodies, (iii) the structure and function of their brains, and (iv) athletic ability, for a total of 12 continuous variables ranging from 0 to 100. The raw scores from these questions were then summed into four mediating variables to indicate the (1) amount of within-group genetic variation, (2) the amount of between-group genetic variation, (3) genetic attributions for within-group variation in complex human traits, and (4) environmental attributions for within-group variation in complex human traits. The 4 raw scores were binned into deciles and then Rasch modeled with a 2PL Rasch partial credit model,^36^ as described above, to arrive at our final analytic mediating variables.
Genetic essentialism was measured after the online module was completed by students. Students were asked to evaluate 12 statements using 6-point Likert scales with response options “strongly disagree,” “disagree,” “somewhat disagree,” “somewhat agree,” “agree,” and “strongly agree.” Some of these statements included “Racial differences in academic ability are caused by genetics” or “Members of one racial group are stronger than members of another racial group because of genetics” or “Genetics causes differences in intelligence.” These psychometrically valid questions were taken from the genetically based racism instrument.^40^ An average score on these measures of 3 or below indicates that a student explicitly (as opposed to implicitly) disagrees with genetic essentialist beliefs, while an average score of 4 or above suggests that a student agrees with genetic essentialist beliefs.
Before testing intervention effects, we assessed whether we were able to pool the two samples (biology N = 1,060, psychology N = 1,001, pooled total sample N = 2,061) to maximize sample size and statistical power to detect the effects of our intervention conditions. While the samples are descriptively similar (Table 1), it is possible that each of the three treatment conditions (PT, MC, and FHGC) can have different effects on genetics essentialism compared with our control condition for gender, race, political affiliation, and major and higher or lower levels of CTR and GK.
As a first step, we use OLS multivariate regression to test the effect of the three treatment conditions on reducing genetic essentialism relative to controls (Equation 1; see also Figure 2A):(Equation 1)Yi=β0+β1Treatmenti+β2Xi+εi
Here, Yi is a measure of an individual i’s belief in genetic essentialism. Treatmenti is the genetics education curriculum that one received (PT, MC, or FHGC). The climate change condition is the control (or reference) group. Xi represents individual characteristics of self-identified gender and race, degree major, self-reported political affiliation, and our measures of GK and CTR.
We hypothesize that receiving the HGC will lead to a greater reduction in students’ belief in genetic essentialism compared with the climate change curriculum. We predicted that the FHGC would have the most substantial impact among the three curricula, as it integrates contents of both the MC and the PT curricula.
As a second step, we test for the mediation effects of within-group variation, between-group variation, genetic attribution, and environmental attribution. We note that we fully recognize the statistical assumptions that weaken the interpretation of mediation analyses, and we are careful to perform a series of critical robustness checks in Note S1, which we also clearly address in the discussion section. We calculate indirect effects of the PT, MC, and FHGC conditions relative to the control group using a two-stage mediation model in an SEM framework. STATA 19^41^ was used for all SEM models. Details of our SEM models are included in the appendix A.
For each of the three treatment conditions, we estimate the indirect effects through each mediator by multiplying coefficients from the A path (a11-34) and the B path (b1-4). Specifically, the A path coefficients represent the effects of each treatment condition on the four mediators, while the B path coefficients capture the effects of each mediator on genetic essentialism. Each of the 12 A paths and 12 B paths are indicated in Figure 2B.
We hypothesize that(1)For our PT treatment condition, the indirect effects for the within-group variation- and between-group variation-mediating variables will be significant and stronger than the indirect effects for the genetic causation- and environmental causation-mediating variables. This is because the PT intervention materials are designed to facilitate a reduction in genetic essentialism by leading students to believe that genetic variation between groups is relatively low, while variation within groups is high.(2)For our MC treatment condition, the indirect effects for the genetic causation- and environmental causation-mediating variables will be significant and stronger than the indirect effects for the within-group variation- and between-group variation-mediating variables. This is because the multifactorial intervention materials are designed to help students view variation in complex human traits as only weakly associated with genetic factors, but more strongly influenced by environmental factors.(3)For the FHGC treatment condition, since this condition targets the mediating mechanisms in both the PT and MC treatments, the effects will be significant and strongest for all four mediating variables in this condition.
Evidence of this last hypothesis would indicate that for reducing beliefs in genetic essentialism, the FHGC treatment is the strongest intervention. This would also indicate that this is the model we would want to move forward into our moderated mediation analyses described below. However, one reason why the FHGC may not be the strongest intervention is because of the complexity of the content. Developing the ability to reason about patterns of genetic variation within and between populations is conceptually analogous to understanding an intraclass correlation coefficient from a linear mixed model. Reasoning about the multiple causes of trait variation within and between groups is therefore a difficult cognitive task. Several studies suggest that debiasing interventions that use scientific information to challenge misinformed beliefs are more likely to backfire as the cognitive complexity of the ideas in the interventions increase.^42^ Therefore, there is reason to believe that the FHGC will be less effective than the PT or MC interventions.
As a third step, we test which moderators (i.e., types or groups of students) have differential effects on the mediation path of the HGC reducing beliefs in genetic essentialism. Details of our moderated mediation SEM models are available in the appendix A.
There are three ways in which social identity factors could moderate the mediation model (Figure 2C). First, we test the moderation of the A×B path (i.e., intervention → mediator → outcome), as individual characteristics could influence how much of the effect of the curriculum on genetic essentialism operates through the four mediating knowledge areas (within-group variation, between-group variation, genetic attributions, and environmental attributions). Therefore, moderation on the A×B path may occur if that pathway or total indirect effect of the curriculum on reducing genetic essentialism works better or worse for people with different social identity factors. Second, if there is evidence of significant moderation on the A×B path, then we test the moderation of the A path (i.e., intervention → mediator), as individual characteristics could influence how students receiving the curriculum content acquire knowledge in the four areas of within-group variation, between-group variation, genetic attribution, and environmental attribution. Therefore, moderation may occur on the A path due to learners of different backgrounds interpreting or interacting with the intervention materials in different ways, leading to varying degrees of change in mediation measures. Third, if we observe significant moderation on the A×B path, we test the moderation of the B path (i.e., mediator → outcome), as characteristics could influence the extent to which an increase in knowledge on within/between variation and genetic/environmental attributions reduces genetic essentialism. Therefore, moderation on the B path may occur if learners internalize the intervention’s effect on mediators similarly, but preexisting social, behavioral, or cognitive biases create differences in learners’ abilities to grow or change their views. Our four moderating social identity factors, as described above, are (1) self-identified gender, (2) self-identified race, (3) our Rasch-modeled measure of CTR, and (4) our Rasch-modeled measure of GK.
We hypothesize the following(1)Indirect effects will vary by self-identified gender (e.g., women vs. men). We predict that this moderation effect will occur on the B path (i.e., the association between the mediator and genetic essentialism), given that social desirability bias for issues of racial equality are often stronger in women than in men.^17^^,^^18^ As a result, women may be less likely to openly endorse essentialist beliefs, which could attenuate the observed effect of the mediators on the outcome, even if their internal beliefs have shifted.(2)Indirect effects will vary by self-identified race (e.g., White vs. non-White individuals). We predict this moderation effect will occur on the B path, given that White individuals tend to endorse essentialist thinking when justifying the exclusion of racial outgroups, but they reject such beliefs when they are used as a rationalization to exclude members of a racial ingroup.^20^ Thus, the path from mediator to genetic essentialism may be weaker for white students than for non-White students.(3)Indirect effects will vary by individuals who hold more or less egalitarian views of how society should be structured, measured through our CTR variable. We predict this moderation effect will occur on the A and/or B paths, given that individuals who think that society should be organized hierarchically are more likely to possess views about the homogeneity of same-race individuals and the discreteness of different racial groups.^22^(4)Intervention effects will vary by individuals with more or less baseline GK, measured through our GK variable. We predict this moderation effect on the A×B path, given that students with higher prior GK are more likely to correctly interpret HGC content and internalize its intended message (A path) and to apply this understanding in a way that reduces essentialist beliefs (B path).^12^
There are restrictions to the availability of the code and dataset for this paper due to concerns about the reidentification of students who participated in this study.
We thank Dr. Donna Ferullo for her incredibly insightful feedback about copyright considerations for the images that appear in this paper. This work was funded by National Science Foundation EDU Core Research (ECR) grant no. 1660985, called “Toward a More Human(e) Genetics Education: Exploring how Knowledge of Genetic Variation and Causation Affects Racial Bias among Adolescents.” We also acknowledge the work of Kristen Hunt, Theresa (Tracie) Wagner, Melissa A. Jones, and Jessica Hathaway, without whose amazing administrative and careful business expertise this project would not be possible. Finally, we acknowledge the hard work of the UCSD Biology Department instructors, who were crucial in administering the study.
R.W. is a research fellow at AnalytiXIN, which is a consortium of health-data organizations, industry partners, and university partners in Indiana, primarily funded through the Lilly Endowment, IU (Indiana University) Health, and Eli Lilly and Company.