Education in science, technology, engineering, and mathematics (STEM) helps both individuals and society. It provides benefits to individuals by sharpening their critical thinking skills, problem solving, and understanding of the physical world; it benefits society by providing groundwork for innovation and invention. The current demand for STEM workers exceeds the supply. The United States would need to increase the number of undergraduates receiving STEM degrees by 33% to keep up with projected demand for STEM professionals (Holdren & Lander, 2012;Xue & Larson, 2015). This suggests that society would benefit by improving STEM education to inspire more students to enter these fields (National Research Council, 2011). Many methods for improving STEM education have been proposed, such as promoting the quality and quantity of STEM learning in both formal and informal environments (National Research Council, 2009;State of Computer Science Education, 2019). In this integrative review, we take a different but complementary perspective and focus on motivational and identity factors, which strongly predict academic outcomes. By "academic outcomes," we include both academic choices and achievement.
Improving students' motivation in STEM is important as a way of altering developmental trajectories starting from an early age. We define motivation from a social-cognitive perspective (Dweck & Leggett, 1988) as a pattern of responses that energizes students toward particular activities. These involve cognitions (beliefs, goals, attributions, values), behaviors (task persistence, effort), and affective responses (interest). Together, these motivational factors can push students to take maximal advantage of STEM learning opportunities and impact students' academic outcomes, including performance and participation (Wigfield et al., 2015). For example, the best predictor of college students' choice to major in STEM (for both males and females) is interest in STEM before college, compared to other variables such as prior STEM courses, GPA in STEM courses, earning potential, or the influence of friends, parents, and teachers (Maltese & Tai, 2011). Overall motivation in math and science shows a particular decrease during adolescence compared to other fields (Wigfield et al., 2015). Between fourth and eighth grade, the number of U.S. children reporting positive attitudes about math and science drops from about 71% to about 48%, and a significant gender gap emerges (Martin, Mullis, & Foy, 2008;Mullis, Martin, & Foy, 2008;TIMSS 2007Assessment, 2009).
Although some students may initially approach STEM with more interest and enthusiasm than others, motivation is malleable. We offer a conceptualization that integrates educational, developmental, and social-psychological research in a novel way and organizes recent research on boosting students' motivation in STEM. Our theoretical model describes the mechanism by which students who face negative stereotypes, such as women in STEM, integrate those stereotypes into their selfrepresentations, with negative consequences for their interest and academic similar patterns of underrepresentation throughout the academic pipeline from high school AP exams to higher education degrees in computer science, engineering, and physics (Cheryan et al., 2017). Computer science is not the same as math, biology, or chemistry. Any full explanations for women's underrepresentation in STEM must also take into account the large variations among STEM fields (Ceci, Ginther, Kahn, & Williams, 2014).
Our review highlights one important source of variation in underrepresentation among STEM fields-stereotypes. Previous explanations for women's underrepresentation have pointed to the math-based nature of STEM fields (e.g., Wang et al., 2013), but this does not explain why women are proportionally more likely to major in math than computer science. We argue that one contributor is that there are differences in the cultural stereotypes about particular STEM fields (Leslie, Cimpian, Meyer, & Freeland, 2015;Master, Cheryan, Moscatelli, et al., 2017). By cultural stereotypes, we mean stereotypes held in a particular society that transcend beliefs within an individual, and are expressed through patterns within many aspects of that culture, including physical objects, media representations, social interactions, and language use (Markus & Kitayama, 2010). These stereotypes can act as gatekeepers that drive girls away from STEM fields such as computer science and engineering (Cheryan, Master, & Meltzoff, 2015). Although limited research has directly connected variation in stereotypes about STEM fields to variation in women's participation in STEM, we take care in this review to specify the field(s) in each study. Due to the limited number of studies, we cannot yet make clear comparisons across fields or draw firm inferences about variation among STEM fields, so we suggest more direct comparisons across STEM fields as a future direction.
We first present our STEreotypes, Motivation, and Outcomes (STEMO) developmental model, and then analyze and synthesize work on four key topics: (a) STEM-gender stereotypes, (b) self-representations affected by stereotypes, including a "sense of belonging" in STEM, (c) interventions that can help draw more students to STEM, chiefly by reducing the impact of stereotypes and enhancing students' sense of belonging, and (d) recommendations for targeted new theorydriven research that can use the STEMO developmental model to specifically address pressing questions in educational theory and practice (thus closing the loop between developmental science, social-psychological studies, and education science). We aim to provide a coherent framework that psychologists can use to better understand recent social-psychological interventions to reduce gender gaps in STEM. Because of space limitations, we postpone discussion of other important issues that contribute to gender gaps, such as societal and institutional structures that have historically contributed to inequality, in part because these have been comprehensively reviewed in other articles (e.g., Ceci et al., 2009;Lee, 2017;Nasir & Hand, 2006). We have aimed to include a comprehensive set of research articles, but note that this is not a complete review of the vast literature on stereotypes, self-representations, and motivation in STEM. We focused our review primarily on research relevant to our new theoretical model. This paper differs from previous reviews on STEM stereotypes in three specific ways by: (a) emphasizing the importance of sense of belonging; (b) acknowledging variation between STEM fields as part of our conceptualization rather than glossing over differences; and (c) considering developmental issues and the motivation of students from preschool through college. To our knowledge, no other review has united these perspectives.
Regarding the sense of belonging, our model is compatible with other theories of student motivation that focus on the link between self-concepts and academic outcomes or factors that make students want to engage with a domain (e.g., Eccles, 2011;Oyserman & Lewis, 2017; for a review, see Wigfield et al., 2006), but also emphasizes the importance of a sense of belonging (see also Eddy & Brownell, 2016;Fiske, 2010;Lewis, Stout, Pollock, Finkelstein, & Ito, 2016). Much research on sense of belonging and STEM has been conducted within social psychology (e.g., Walton & Cohen, 2007), and we aim to bring together this line of research with an expectancy-value perspective. Expectancy-value theory highlights how students' expectations of success and valuing of an academic domain influence their academic choices and behaviors, and has been a dominant focus in research on academic motivation (for a review, see Eccles, 2011). We focus on a subset of the variables typically included within expectancy-value theory to highlight the theoretical pathway by which stereotypes affect students' academic outcomes, and show how sense of belonging fits into this pathway. Although sense of belonging is not explicitly mentioned as part of the expectancy-value framework (Eccles, 2011), it is highly relevant for women and underrepresented minorities' motivation in STEM fields. Pervasive negative stereotypes and common social and environmental cues can signal to members of these groups that they do not belong, which reduces their motivation to pursue these fields (Good, Rattan, & Dweck, 2012;Lewis et al., 2016;Master, Cheryan, & Meltzoff, 2016a;Murphy, Steele, & Gross, 2007;Walton & Cohen, 2007).
Regarding variation in STEM fields, in this review we do not lump all STEM fields together because we see important differences among them. We specify the particular STEM fields used in each research study (e.g., math vs. engineering) throughout this review, and highlight recent research that suggests that differences in gender participation across STEM fields align with differences in stereotypes about each field. The unique cultural stereotypes about different STEM fields reduce many students' sense of belonging in those fields and make social identity particularly meaningful for STEM motivation. By incorporating stereotypes into our developmental model, we offer a way for future researchers to continue to organize important differences between STEM fields.
Regarding including preschool and early developmental phases, we differ from other valuable reviews (e.g., Ceci & Williams, 2011;Rosenzweig & Wigfield, 2016) by considering students from preschool through college. In recent years studies have emerged showing that starting as early as preschool and early elementary school, social identities and group memberships can influence students' STEM motivation (Master, Cheryan, & Meltzoff, 2017). This is part of the foundation of human development (Meltzoff, 2007). Examining STEM motivation starting at such a young age is important, because the recruitment of young women into a variety of STEM fields is a larger problem than retaining women in these fields (Miller & Wai, 2015). In addition, interventions designed to address recruitment or spark interest in STEM may be most effective if they start early (Maltese & Tai, 2010;Master, Cheryan, Moscatelli, et al., 2017;Musu-Gillette, Wigfield, Harring, & Eccles, 2015), with the potential for cascading effects across the lifespan.
The chief goal for this paper is to synthesize evidence that cultural stereotypes have an effect on students throughout the lifespan by affecting their selfrepresentations, which in turn drive many outcomes in STEM. Figure 1 shows our developmental model (in solid lines) and how theory-based interventions fit into this conceptualization (in dashed lines).
This conceptualization, which we call the STEreotypes, Motivation, and Outcomes (STEMO) developmental model, is different from previous models in several ways that we articulate below. However, it builds upon expectancy-value theory (Eccles, 2011) by integrating it with social-psychological research on the importance of social identities for academic motivation (Cohen & Garcia, 2008;Master, Cheryan, & Meltzoff, 2016b) and developmental research that has shown the effects of societal stereotypes in children as early as preschool (Hilliard & Liben, 2010;Patterson & Bigler, 2006;Rhodes, Leslie, & Tworek, 2012). Our developmental model is not intended to displace expectancy-value theory, but provides a way to complement it by incorporating newer research that emphasizes sense of belonging as a particularly central self-representation for students who face negative stereotypes about their social identities. We also note that this model currently focuses on individuals' social beliefs, attitudes, and representations, and does not (yet) explicitly incorporate the many ways in which the social context can affect all of these elements.
As shown in Figure 1, the STEMO developmental model highlights how social factors (e.g., pervasive stereotypes about one's own social identities, such as gender) influence students' interest and academic outcomes in STEM through selfrepresentations. When students have a social identity that is the target of negative stereotypes in a domain such as a STEM field, they may experience more negative self-representations in that domain, such as reduced identification, reduced belief in their ability to succeed, and a lower sense of belonging. These negative selfrepresentations lead to less interest and poorer academic outcomes in STEM (e.g., choosing not to enroll in optional STEM courses). Not all arrows in Figure 1 are depicted as bi-directional to simplify the illustration. However, we acknowledge that each element in the figure may affect the others as these processes influence others and play out continuously over time (e.g., Jacobs, Lanza, Osgood, Eccles, & Wigfield, 2002;Marsh & Martin, 2011). In this model, students encounter stereotypes about social groups (e.g., STEM-gender stereotypes that women do not fit or have as much ability in STEM as men). When those stereotypes are relevant to students ' own social identity (e.g., gender), this affects their self-representations in STEM, including less identification, reduced ability beliefs, and a lower sense of belonging. This can impair their interest and academic outcomes (e.g., participation) We also emphasize the role of developmental processes within this model in several ways. First, we consider these processes to begin to emerge with children's first experiences in school, during preschool and early elementary school, or perhaps even earlier. As soon as children learn about social categories, they begin to use that information to organize their expectations of the social world (Hilliard & Liben, 2010;Master, Markman, & Dweck, 2012;Patterson & Bigler, 2006). Second, we use this model as a framework to ask developmental questions: when do children begin to endorse these stereotypes, and when do stereotypes begin to affect their self-representations? How does this process unfold over time as children gain new social and academic experiences? Current research suggests that stereotypes are already linked to self-representations by early elementary school (Cvencek, Meltzoff, & Greenwald, 2011;Master, Cheryan, Moscatelli, et al., 2017). Finally, once we have a better understanding of how this process emerges developmentally, the model can be used in the design of future interventions to help make predictions about which interventions will be effective at which ages.
The STEMO model has three valuable characteristics. First, it can be applied to any stereotyped social identity. Similar processes to those proposed here would apply for members of other groups who face negative stereotypes in STEM, such as certain racial/ethnic groups. In this paper, we focus on gender to illustrate our model, due to the large amount of research on this social identity, and we return to the issue of other stereotyped groups at the end of this review. Second, we explicitly emphasize developmental processes and change over time in this model. It can be applied to students from preschool through higher education. Thus, the model is directed at the roots and earliest manifestations of the role of stereotypes, even before formal schooling begins. Although these dynamic processes are often overlooked outside of developmental psychology, they have important practical implications, such as insights into the optimal timing for different types of interventions. Third, a unique feature of the STEMO developmental model is that it also serves as a basis for designing and classifying interventions that effectively buffer against the effects of stereotypes and enhance women's self-representations in STEM, as shown by the dashed lines in Figure 1.
When combined with research into the development of stereotypes and selfrepresentations, the STEMO developmental model can help make predictions about which interventions will be successful at what ages, and the mechanisms by which they work. In a separate section of this paper, we will use the model to discuss three types of evidence-based interventions.
The STEMO model is related to Eccles and colleagues' seminal expectancy-value theory with several differences, including the emphasis on sense of belonging.
Another notable difference is in our use of "interest" as a motivational outcome influenced by self-representations such as ability beliefs (our superordinate term that includes expectations of success). Expectancy-value theory models typically use "interest/enjoyment/intrinsic value" as a predictor of behavioral outcomes such as achievement-related choices and performance, in parallel with ability beliefs (Eccles, 2011). In contrast, our model uses interest as an outcome variable, with ability beliefs as one of its predictors. We do this for three reasons.
First, ability beliefs are more likely to predict interest over time than the reverse (Wigfield & Eccles, 2002), which provides empirical support for our approach.
Second, interest has distinct motivational properties, including the predisposition to re-engage with a domain over time (Hidi, 2006;Renninger & Hidi, 2011;Sansone, 2009). This makes it particularly valuable to consider as an outcome for long-term STEM pathways. Third, many studies have found empirically that interest more strongly predicts academic choices than ability beliefs do (Barron & Hulleman, 2015;Wigfield et al., 2015). The STEMO model is also conceptually related to a large body of literature on the relation between group stereotypes and self-representations. This is a bi-directional relation, in which stereotypes of the group influence beliefs about the self at the same time that beliefs about the self influence perceptions of the group. This distinction has been referred to as self-stereotyping vs. self-anchoring, or deduction-to-the-self vs. induction-to-the-group, in social psychology (Latrofa, Vaes, Cadinu, & Carnaghi, 2010;van Veelen, Otten, Cadinu, & Hansen, 2016); as the attitudinal pathway versus the personal pathway in developmental psychology (Liben & Bigler, 2002); and is theorized to be balanced in a three-way relation among self, group, and academic field (e.g., math) in balanced identity theory (Cvencek et al., 2011;Cvencek, Meltzoff, & Kapur, 2014;Patterson & Bigler, 2018).
Our STEMO model focuses on one pathway in this theorized three-way relation: how group stereotypes exert an influence on self-representations. We do so for several reasons. First, this pathway is supported by empirical developmental data. Although data is limited, it appears that endorsement of stereotypes about the group precedes the development of gender differences in self-concepts (Cvencek et al., 2011;Eccles, 2009; see also Eccles, Wigfield, Harold, & Blumenfeld, 1993;Liben & Bigler, 2002). Second, research suggests that this pathway is especially strong when a group has lower status, such as women in STEM (Latrofa et al., 2010). Third, we do so for pragmatic reasons. We are focused on explaining motivation at the individual level, rather than explaining the formation of beliefs about the social group. Thus, we focus on how beliefs about the group influence beliefs about the self, rather than the other way around. This direction of causality is also more relevant for interventions targeting girls' low interest in STEM. If stereotypes have a causal impact on interest, then interventions can focus on girls who endorse cultural stereotypes in hopes of changing their stereotypes to boost their motivation in STEM.
There are prevalent cultural stereotypes in Western countries associating boys and men with many STEM fields (Master, Cheryan, & Meltzoff, 2014). We will refer to these stereotypes collectively as STEM-gender stereotypes. These stereotypes are transmitted to students by parents, teachers, peers, and media (Cheryan et al., 2015;Cvencek, Kapur, & Meltzoff, 2015;Cvencek et al., 2011;Gunderson, Ramirez, Levine, & Beilock, 2012;Keller, 2001;Leaper, Farkas, & Brown, 2012;Rhodes & Leslie, 2017;Riegle-Crumb & Morton, 2017). According to our STEMO model (Figure 1), as children learn about cultural stereotypes, those stereotypes begin to shape the development of their self-representations.
Our review distinguishes between two types of STEM-gender stereotypes (Cheryan et al., 2015;Master & Meltzoff, 2016;Wynn & Correll, 2017): stereotypes about who likes or is associated with a field (which we will refer to as stereotypes about interest; also referred to as stereotypes about cultural fit) and stereotypes about who is good at or has superior ability in a field (which we will refer to as stereotypes about ability). These two types of stereotypes show different developmental patterns and may have different relations to STEM selfrepresentations and interest (Master, Meltzoff, & Cheryan, 2020). The combination of both types of stereotypes about STEM fields contributes to the underrepresentation of women in these fields by acting as a gatekeeper that keeps women from entering. Women may worry both that they do not fit the image of a STEM person and that they do not have the ability to succeed in STEM. In distinguishing between these two types, our evaluation of the influence of stereotypes is much broader than the influence of stereotype threat research, which focuses on how ability stereotypes impact women's performance in STEM.
Stereotype threat itself may be insufficient to account for gender gaps in participation (Stoet & Geary, 2012), and represents only one small piece of a larger mechanism by which stereotypes influence STEM participation.
Stereotypes about interest. Stereotypes about interest encompass beliefs about which social groups typically enjoy or are interested in STEM (as well as stereotypes about who is associated with STEM fields, although these may be slightly different conceptually). Interest stereotypes have been measured using both direct (e.g., self-report) and indirect methods. Indirect methods include implicit measures, such as the Child Implicit Association Test (e.g., Cvencek et al., 2011), which assesses children's implicit reactions that boys are linked with math and girls with reading. This test is built on the principle that people respond more quickly to associations that are congruent with their implicit stereotypes and biases (e.g., math = boys and reading = girls) than to associations that are incongruent with those stereotypes (reading = boys and math = girls). In terms of math, North American children as early as second grade already register the implicit stereotype that math = boys (Cvencek et al., 2011), and Chilean boys report this implicit stereotype in kindergarten (del Río, Strasser, Cvencek, Susperreguy, & Meltzoff, 2019). Another indirect method, the Draw-a-Scientist test, has found that children become more likely to draw a male scientist than a female scientist as they get older, with girls more likely than chance to draw a male scientist beginning around age 10 (Miller, Nolla, Eagly, & Uttal, 2018). In terms of technology, 6-year-old North American children explicitly report that boys will enjoy a programming or robotics activity more than girls (Master, Cheryan, Moscatelli, et al., 2017), and 6 th graders are more likely to depict a computer expert as male than female (Mercier, Barron, & O'Connor, 2006). In terms of science, by fourth grade (the youngest age tested so far) European and North American girls and boys associate boys with science, and expect boys to enjoy science more than girls (Archer et al., 2012;Farenga & Joyce, 1999). Thus, endorsement of cultural stereotypes about interest appears to emerge during elementary school across multiple STEM fields.
Most previous research has focused on stereotypes about whether boys or girls are "better at math" (often conceptualized as having more fixed or innate ability). Very little research has examined children's stereotypes about ability in technology, engineering, or computer science. The existing findings suggest a developmental shift during elementary school, although findings differ based on nationality and the exact methods and wording used to measure stereotypes. When explicitly asked which group is "better" at math or science, in the youngest ages tested (from kindergarten to second grade), North American and some European children report either that the genders are equal in ability (Hargreaves, Homer, & Swinnerton, 2008;J. Steele, 2003) or show a bias that their own-gender group is better (Kurtz-Costes, Rowley, Harris-Britt, & Woods, 2008;Master, Cheryan, Moscatelli, et al., 2017;Passolunghi, Rueda Ferreira, & Tomasetto, 2014), although this can vary by culture (Lummis & Stevenson, 1990).
Research suggests that some European and Chinese girls do not endorse the stereotype that boys are better at math until fourth grade or later (Martinot & Désert, 2007;Muzzatti & Agnoli, 2007;Passolunghi et al., 2014;Zhao, Zhang, Alterman, Zhang, & Yu, 2018; see also del Río & Strasser, 2013), although North American middle school students sometimes report counter-stereotypes that girls are better than boys at math and science (Evans, Copping, Rowley, & Kurtz-Costes, 2011;Plante, Théoret, & Favreau, 2009;Rowley, Kurtz-Costes, Mistry, & Feagans, 2007). In terms of technology, North American children begin to believe that boys are better than girls at computer science and engineering by age six (Master, Cheryan, Moscatelli, et al., 2017). Thus, endorsement of cultural stereotypes about ability appears to emerge later and less consistently across fields than stereotypes about interest.
Children's career aspirations begin to form during elementary school, which is also the period during which children become more familiar with stereotypes involving both interest and ability. The choice of whether students intend to pursue science careers in particular often occurs during middle school (Tai, Liu, Maltese, & Fan, 2006). This is also an important time period for the long-term trajectories of students' expectations of success and valuing of different fields (Jacobs et al., 2002). Thus, as girls begin to form and express these early career choices, they are already aware that boys are associated with and widely believed to be better than girls at many STEM fields. Endorsement of these stereotypes is also correlated with lower interest in STEM among girls and higher interest in STEM among boys (Blazěv, Karabegović, Burusǐć, & Selimbegović, 2017;Master, Cheryan, Moscatelli, et al., 2017;Plante, De la Sablonnière, Aronson, & Théorêt, 2013; see also Weisgram, Bigler, & Liben, 2010).
Stereotypes may also influence students externally through the people around them, who socialize girls and boys differently when it comes to STEM (Eccles, 2009(Eccles, , 2015; for a review, see Gunderson et al., 2012). Parents and teachers may hold math-and science-gender stereotypes and attitudes that influence their expectations for and interactions with children (e.g., Beilock, Gunderson, Ramirez, & Levine, 2010;Gunderson et al., 2012), which then influence children's selfconcepts (del Río et al., 2019;Friedrich, Flunger, Nagengast, Jonkmann, & Trautwein, 2015).
Parents may provide more or fewer opportunities for their children to participate in math and science activities, depending on their beliefs about gender-appropriate activities (Jacobs, Davis-Kean, Bleeker, Eccles, & Malanchuk, 2005). In one study, parents who endorsed gender stereotypes were also more likely to give intrusive, uninvited assistance on math homework to girls, which led to lower math ability beliefs for those girls (Bhanot & Jovanovic, 2005). Parents explain less about scientific concepts to girls than boys (Crowley, Callanan, Tenenbaum, & Allen, 2001), and underestimate their daughters' interest in science books (Ford, Brickhouse, Lottero-Perdue, & Kittleson, 2006). On average, parents also believe that science is less interesting and more difficult for daughters compared to sons, and their beliefs are correlated with their children's science interest and beliefs about their ability to learn science (Tenenbaum & Leaper, 2003). Parents are also influenced by the gendered packaging of mechanical toys when they guide their children in play, encouraging more building for "boys' toys" and more narrative play for "girls' toys" (Coyle & Liben, 2020).
Potential ways that teachers' stereotypes affect their students may include explicit verbal statements about gender, differential treatment (e.g., interacting more with boys than girls in math class; Becker, 1981), or modeling their own math anxiety (Beilock et al., 2010;Gunderson et al., 2012). Elementary school math teachers' gender biases have long-lasting effects, including the number of advanced math and science courses that students take in high school (Lavy & Sand, 2018). Even well-intentioned language such as "Girls are as good as boys at math" can unintentionally emphasize the talent of boys and reinforce stereotypes (Chestnut & Markman, 2018). Although not specific to STEM, teachers may give more negative academic feedback to girls in general (compared to negative behavioral feedback to boys), which has a negative impact on girls' ability beliefs (Dweck, Davidson, Nelson, & Enna, 1978). Thus, stereotypes may cause parents and teachers to treat girls and boys differently when it comes to STEM, which may affect children's opportunities and attitudes.
As students become aware that others hold these stereotypes, they may begin to experience social identity threat. "Social identity threat" refers to the concern about being viewed negatively due to social group membership (C. Steele, Spencer, & Aronson, 2002). For members of negatively stereotyped groups (e.g., women in STEM), a situation in which that group identity is highlighted raises the possibility that others will make assumptions about them based on their membership in that group. Certain cues in the environment can confirm this possibility and create a sense of social identity threat. These cues include the social identities of other people in that environment, such as a male-dominated group of students or teachers (Master et al., 2014); the social identities of other people in a potential environment (Murphy et al., 2007); and the design of the physical classroom environment (Master et al., 2016a). For social identity threat, students' awareness of stereotypes may be more influential than their endorsement of stereotypes (Cvencek, Nasir, O'Connor, Wischnia, & Meltzoff, 2015;Huguet & Régner, 2009), although students' self-representations may be shaped more strongly by their endorsement (Master, Cheryan, & Meltzoff, 2019;Owens & Massey, 2011).
Social identity threat is a broad category that includes stereotype threat, the concern about being seen through the lens of a negative stereotype about one's group (C. Steele, 1997). Many studies have shown that making group membership salient (e.g., checking a box for gender) leads to decreased performance for members of stereotyped groups (for a meta-analysis, see Nguyen & Ryan, 2008), especially those who identify strongly with that domain (Kahalon, Shnabel, & Becker, 2018;Schmader, 2002). Stereotype threat affects children and early adolescents as well as adults (Flore & Wicherts, 2015). Although most stereotype threat research has examined performance as the outcome, some research has examined effects on motivation (Murphy et al., 2007;Thoman, Smith, Brown, Chase, & Lee, 2013). Negative effects of stereotype threat on performance and motivation can create a negative feedback loop that harms students' selfrepresentations, which leads to further negative effects on performance and motivation (Cohen, Garcia, Purdie-Vaughns, Apfel, & Brzustoski, 2009;Cohen & Sherman, 2014). In the next section, we discuss the important role of students' self-representations in greater detail.
Stereotypes can also influence students by changing how they view themselves (Cvencek et al., 2011;Cvencek et al., 2014;Tobin et al., 2010). According to our STEMO developmental model, when students with a stereotyped social identity start to endorse these stereotypes, they may develop representations of the self that are aligned with those stereotypes. In turn, those representations shape their developing interest in STEM fields and influence their STEM outcomes. Again, we intend for the term self-representations (also known as self-perceptions; Banchefsky, Lewis, & Ito, 2019) to include self-concepts and self-relevant motivational beliefs. We focus on three specific self-representations: identification, ability beliefs, and sense of belonging in STEM. We focus only on these three specific self-representations due to space limitations and due to their strong and established relations to STEM-gender stereotypes (Eddy & Brownell, 2016), but other beliefs and values are also important, such as the perceived costs (Perez, Cromley, & Kaplan, 2014) or utility value associated with STEM fields (Rozek, Hyde, Svoboda, Hulleman, & Harackiewicz, 2015). Stereotypes have also been linked directly to academic outcomes, such as participation in science. For example, one study found that gender differences in college students' plans to pursue a career in science or the humanities were driven by students' implicit associations between science and gender (Lane, Goh, & Driver-Linn, 2012). Across 66 countries, sciencegender stereotypes are correlated with women's lower participation in science at the college and professional levels (Miller, Eagly, & Linn, 2015), highlighting the importance of gender stereotypes for STEM outcomes cross-culturally. Relatedly, when students feel more similar to the prototypical student in a field, they report greater interest in that field themselves (Kessels, 2005;Kessels, Heyder, Latsch, & Hannover, 2014).
We view the literature as supporting our STEMO model if: (a) correlational and/or experimental studies link that self-representation to interest and academic outcomes in STEM, (b) gender differences related to STEM are evident for that selfrepresentation (to confirm that gender moderates the effects of stereotypes), and (c) correlational and/or experimental studies link stereotyping to that selfrepresentation. We review each type of evidence below.
Links to STEM outcomes. Identification involves both linking the self to a domain or social group (e.g., "I'm a math person") and valuing that domain or group (e.g., "math is important to me"). It is highly similar to "attainment value" as described in expectancy-value theory, when applied to a domain such as math (Wang & Degol, 2013). Identification can predict students' interest and academic outcomes (Osborne & Jones, 2011). Women's science, engineering, and physics identification is significantly correlated with their career intentions and persistence in science/engineering/physics in college (Hazari, Sonnert, Sadler, & Shanahan, 2010;Jones, Ruff, & Paretti, 2013;Smith, Brown, Thoman, & Deemer, 2015;Stout, Dasgupta, Hunsinger, & McManus, 2011;Young, Rudman, Buettner, & McLean, 2013). Girls' sense that "being a scientist" is incompatible with their gender identity can decrease their interest in STEM (Lei, Green, Leslie, & Rhodes, 2019). For this reason, describing science with identity language ("Let's be scientists!") instead of action language ("Let's do science!") decreased girls' persistence in science games (Rhodes, Leslie, Yee, & Saunders, 2019).
Gender differences. Gender differences in identification with STEM emerge surprisingly early. As early as elementary school, U.S. girls identify less strongly with math than do boys, as measured by their performance on a Child Implicit Association Test (Cvencek et al., 2011), and middle school girls report lower science identity than boys (Vincent-Ruz & Schunn, 2018). Adult women also report lower identification than men in physics and math (Hazari et al., 2010;Seyranian et al., 2018;Smith & White, 2001).
Links to stereotypes. Although causal pathways have not been empirically determined, several studies have found that endorsement of STEM-gender stereotypes correlates negatively with girls' identification with STEM. Singaporean and German girls' implicit and explicit math-gender stereotypes are negatively correlated with their math identification in elementary school (Cvencek et al., 2011;Cvencek et al., 2014;Cvencek, Kapur, et al., 2015;Steffens, Jelenec, & Noack, 2010). Similarly, college-aged women's implicit science-gender stereotypes are negatively correlated with their science identification (Cundiff, Vescio, Loken, & Lo, 2013;Young et al., 2013; see also Smith et al., 2015). Women in Silicon Valley tech companies who endorsed interest stereotypes about tech workers also reported lower identification with the tech profession (Wynn & Correll, 2017).
Conversely, the more that college-age women implicitly associate women with math, the stronger their math identification and interest (Gilbert, O'Brien, Garcia, & Marx, 2015). Experiences of stereotype threat may also cause women to disidentify from scientific domains, choosing other domains instead (Smith et al., 2015).
Links to STEM outcomes. We use ability beliefs to refer to students' judgments about their ability to succeed in a particular domain or task, including beliefs about present ability or future level of success, as well as confidence in that domain.
When students approach a STEM task or domain, do they have the sense that they can be successful? We use ability beliefs as an umbrella term to cover several types of beliefs that have been measured in previous research, including "self-efficacy" (beliefs about current ability to succeed in a task or domain), "expectations of success" (beliefs about future success), "perceived competence," "self-confidence," or "academic self-concept" (which can be measured for perceived abilities in a specific academic domain). These types of beliefs are typically highly linked to one another empirically (Eccles & Wigfield, 1995;Eccles, 2009). These beliefs about the self can play an important role in students' interest and academic outcomes (Perez et al., 2014).
Ability beliefs about the self are correlated with course selection and major. One study found that North American girls' math self-concept in both 6 th and 10 th grade predicted the number of math courses they took in high school, and science selfconcept in 10 th grade predicted the number of science courses they took in high school (Simpkins, Davis-Kean, & Eccles, 2006). Women who believed they had the ability to master challenging math tasks in 12 th grade were three times more likely to major in physical science, engineering, math, and computer science in college than women who did not believe they had that ability (Perez-Felkner, Nix, & Thomas, 2017). College students with high computing self-efficacy were more likely to take a computer science class during the following year (Beyer, 2014). A longitudinal study from 12 th grade through the second year of college found that Canadian college students with stable or increasing science self-efficacy were more likely to persist across a variety of science majors compared to students with decreasing self-efficacy (Larose, Ratelle, Guay, Senécal, & Harvey, 2006). Ability beliefs predict STEM interest and outcomes above and beyond prior performance.
One study found that 12 th grade students' math ability self-concepts predicted their participation in a variety of STEM careers at age 29, even controlling for math performance (Eccles & Wang, 2016).
As posited by expectancy-value theory, positive ability beliefs in STEM are a necessary but not sufficient predictor of STEM interest and academic outcomes (Wang & Degol, 2013). Even when girls have the belief that they can succeed in math and science, they may still be uninterested in math and science activities and careers (Jacobs et al., 2005;Weisgram & Bigler, 2007). Ability beliefs also may be more strongly linked to girls' motivation in STEM compared to boys' motivation.
One study found that actual performance and perceived math self-concept predicted Australian and North American female adolescents' career plans for a variety of STEM fields, but not male adolescents' plans (Watt et al., 2017).
Gender differences. A large international dataset with students ages 14-16 found that boys reported higher math self-confidence than girls, with a small effect size of d = .15 (Else-Quest, Hyde, & Linn, 2010). Boys report higher math ability beliefs than girls even when their objective ability is identical (Perez-Felkner et al., 2017).
Several studies with adults have also found that adult women underestimate their scientific capabilities compared to men and hold themselves to a higher standard before applying to math-and science-related college majors or job openings (Ayalon, 2003;Ehrlinger & Dunning, 2003), and this underestimation is linked to their lower interest in math-related fields (Bench, Lench, Liew, Miner, & Flores, 2015). Other studies have also found that boys in the United States, Australia, and Germany have more positive math self-concepts than girls, although competence beliefs decline for both boys and girls across grades K-12 (Nagy et al., 2010; see also Carr, Steiner, Kyser, & Biddlecomb, 2008). Middle school boys in the United States were more likely to report high confidence in their science ability than girls (P.W. Hill et al., 2017). In terms of STEM fields other than math and science, one study found that high school boys in Sweden reported higher self-efficacy than girls in engineering and computer programming, and this gender difference partially mediated the gender difference in interest in these fields (Tellhed, Bäckström, & Björklund, 2017). College women in computer science, physics, and math also report lower self-efficacy than men (Ellis, Fosdick, & Rasmussen, 2016;Lewis et al., 2017).
Links to stereotypes. Girls' lower ability beliefs in STEM are correlated with cultural stereotypes that girls have lower ability in these domains. As early as elementary school, North American and European girls' explicit endorsements of math-gender and programming-gender stereotypes are negatively correlated with their selfperceptions of math ability, programming self-efficacy, and math academic selfconcepts (Master, Cheryan, Moscatelli, et al., 2017;Passolunghi et al., 2014;Plante et al., 2013;Steffens et al., 2010). Similarly, for adults, college women who worried about negative stereotypes reported lower science self-efficacy and lower intentions to major in physics (Deemer, Thoman, Chase, & Smith, 2014;Stout et al., 2011), and female engineers with higher implicit stereotypes linking male = engineering also reported lower self-efficacy in engineering (Block, Hall, Schmader, Inness, & Croft, 2018). Even within a sample of German women majoring in computer science and engineering who had positive academic self-concepts, the more that women endorsed interest and ability stereotypes about STEM, the lower their STEM self-concept (Ertl, Luttenberger, & Paechter, 2017). The more that high school girls endorsed ability stereotypes about math, the more they underestimated their own performance on a high-stakes math exam (Chatard, Guimond, & Selimbegovic, 2007). Similarly, the more that college women in math-related majors endorsed ability stereotypes about math, the lower their confidence in their future math performance (Schmader, Johns, & Barquissau, 2004).
Links to STEM outcomes. Sense of belonging is a self-representation that indicates how much students see themselves as fitting in with those around them (Master et al., 2016a). We argue that sense of belonging plays a particularly important role in students' STEM interest and academic outcomes. Belonging has been described as one of five core social motives (Fiske, 2010), or as one of the psychological needs that facilitate intrinsic motivation (Ryan & Deci, 2016). Students who feel a strong sense of belonging and social connectedness in school show positive changes in motivation and academic achievement over time (Osterman, 2000). Connections to others can also help to trigger greater interest in STEM subjects (Thoman, Sansone, Fraughton, & Pasupathi, 2012).
College women's sense of belonging across a variety of STEM fields strongly correlates with their motivation in those fields (Lewis et al., 2017;Smith, Lewis, Hawthorne, & Hodges, 2013). Longitudinal studies of college students majoring in a variety of STEM fields have found that sense of belonging is positively correlated with women's intentions to persist and actual persistence in their major, to a greater extent than men's sense of belonging (Banchefsky et al., 2019;London, Rosenthal, Levy, & Lobel, 2011; see also Good et al., 2012;Stout & Blaney, 2017).
Similarly, women's perceptions that they matched the cultural stereotype of people who succeed in technology (in other words, that they "fit in") was a stronger predictor of intentions to persist in technology companies than their sense that they matched the skill stereotype of people who succeed in technology in a simultaneous regression analysis (Wynn & Correll, 2017). Belonging is also related to academic performance. In one study, the more that college women wrote about social belonging during a values affirmation, the better they performed on a subsequent math test (Shnabel, Purdie-Vaughns, Cook, Garcia, & Cohen, 2013). Although limited research on sense of belonging has been conducted with young children, preschool children showed greater persistence and performance on STEM tasks when they felt socially connected to a group (Master et al., 2017).
Different types of belonging may matter for different groups in STEM. For members of racial/ethnic minority groups, both social belonging (concerns that they may not fit in with others around them due to their numeric minority status) and academic belonging (concerns that other students around them have skills that they lack) may be uncertain. For girls and women in STEM, academic belonging in STEM may be especially salient at times when they are sufficiently numerically represented (Lewis & Hodges, 2015; see also Aday & Schmader, 2019). Important influences that may decrease academic belonging for women in STEM include perceptions that they have fewer skills or that they have to work harder than other students to succeed (Smith et al., 2013;Walton & Cohen, 2007). Both academic and social belonging predict women's intentions to persist in STEM fields (Banchefsky et al., 2019).
Gender differences. Adult women feel a lower sense of social belonging than men in STEM fields such as computer science and physics (Lewis et al., 2017;Stout & Blaney, 2017), and this gender gap can widen over the course of a single introductory college course (Sax et al., 2018). Girls in Western countries feel a lower sense of belonging in STEM fields such as computer science and engineering by high school, which is correlated with lower interest, even controlling for other variables such as expectations of success and utility value (Ito & McPherson, 2018;Master et al., 2016a;Tellhed et al., 2017).
Links to stereotypes. Students' sense of belonging in STEM is directly related to culturally-held STEM-gender stereotypes (Dasgupta, 2011). For example, collegeage women's awareness of stereotypes about ability correlates with lower sense of belonging in a variety of STEM majors (Ahlqvist, London, & Rosenthal, 2013), and female engineers with higher implicit male = engineering stereotypes reported a lower sense of fit with their organization (Block et al., 2018). Similarly, college women who felt less similar to the "prototypical" computer scientist or engineer reported lower interest in computer science and engineering (Ehrlinger et al., 2018).
Experimental studies indicate that changing stereotypical cues in the environment can increase women's sense of belonging and interest in computer science (Cheryan, Meltzoff, & Kim, 2011;Cheryan, Plaut, Davies, & Steele, 2009). A classroom that contained objects fitting a stereotypical image of computer science as "geeky" and masculine made high school girls feel that they would not belong in that classroom (Master et al., 2016a). Girls also reported low interest in taking a computer science course in that classroom. In contrast, a classroom that contained neutral objects (e.g., potted plants) made girls feel a significantly higher sense of belonging, with greater interest in taking that course. In this way, stereotypes can act as a barrier or "gatekeeper" that prevents girls from feeling a sense of belonging in STEM, and thus keeps them out of these fields (Cheryan et al., 2015).
Boys show greater motivation than girls for many STEM fields starting at an early age, but motivation is malleable, and can be changed. Experimentally designed interventions have succeeded in using social influences to boost students' identification (Ramsey, Betz, & Sekaquaptewa, 2013;Stout et al., 2011), ability beliefs (Dennehy & Dasgupta, 2017;Luzzo, Hasper, Albert, Bibby, & Martinelli, 1999;Shin, Levy, & London, 2016), and sense of belonging (Gehlbach et al., 2016;Walton, Logel, Peach, Spencer, & Zanna, 2015; see also Yeager & Walton, 2011).
In this section we highlight three promising intervention approaches. See Figure 1 for how each intervention targets the key components of our proposed STEMO model. These experimental interventions have been able to: (a) broaden stereotypes about STEM and who belongs in STEM, (b) reduce the impact of negative stereotypes through a growth mindset, or (c) strengthen belonging by reducing belonging uncertainty. These interventions are in no way intended to absolve the people within STEM disciplines, classrooms, and workspaces from their responsibility to reduce structural and societal disparities. However, they provide a way for teachers and students to reduce the effect of cultural stereotypes while larger societal issues remain. These three types of interventions create a buffer that protects students' self-representations in STEM despite pervasive societal stereotypes.
Intervention Type 1: Broaden Stereotypes about STEM Broadening cultural stereotypes. If women feel that they do not belong in STEM because of cultural stereotypes, one solution is to change the stereotypes. Although eliminating cultural stereotypes is difficult (J. Steele, 2003), an alternative is to broaden the stereotypes by giving students diverse examples of who does STEM and what STEM involves. For example, changing the objects in a computer science classroom or office environment to be less stereotypically masculine and "geeky" can encourage high school girls to show an interest in that environment (Master et al., 2016a). Showing non-stereotypical role models with diverse personalities and interests can increase women's math identification, as well as their sense of belonging, expectations of success, and interest in computer science (Cheryan, Drury, & Vichayapai, 2013;Stout et al., 2011). Female role models may also reduce high school girls' worries about being negatively stereotyped in a computer science class (Master et al., 2014).
Creating an image of STEM as communal. Another stereotype about STEM is that it involves solitary, non-communal work (Diekman, Clark, Johnston, Brown, & Steinberg, 2011). Studies have found that the more that people value communal goals (which include both altruistic helping and collaborating with others), the less interested they are in science and engineering careers, even controlling for differences in previous experiences and STEM self-efficacy (Diekman, Brown, Johnston, & Clark, 2010). One way that stereotypes about STEM may decrease women's identification with science is by undermining their perception that science is communal (Smith et al., 2015). Changing these stereotypes by helping young women and members of other underrepresented groups see the potential for communal experiences in STEM can boost students' self-representations in STEM (Thoman, Muragishi, & Smith, 2017). Experimentally highlighting communal values can have a causal boost on students' interest in STEM. For example, female college students who learned about a day in the life of a scientist whose daily activities were highly collaborative had more positive attitudes towards science careers compared to students who learned about an independent scientist (Diekman et al., 2011).
Growth mindsets and STEM. One way to reduce the impact of stereotypes is by teaching growth mindsets. A growth mindset is the idea that abilities such as intelligence are malleable and can be changed through effort (Dweck & Master, 2009). This is in contrast to a fixed mindset, the idea that abilities such as intelligence are fixed and unchangeable. Growth mindsets may make students more willing to take on challenging STEM courses. Growth mindsets about intelligence in 6 th grade predicted enrollment in more advanced math courses in middle school (Romero, Master, Paunesku, Dweck, & Gross, 2014), and endorsing growth mindsets about math in 9 th grade predicted intent to take college math courses (Priess-Groban & Hyde, 2017). A belief that abilities are fixed may also be related to women's underrepresentation in STEM fields. Research has shown that the more that people within a field endorse the belief that "natural" talent or genius (often conceived of as biologically fixed) is required for that field, the lower the proportion of women in the field (Leslie et al., 2015). These beliefs are more frequent within fields such as computer science and engineering. Similarly, learning about genetic sex differences increased adolescent girls' belief that science ability is innate, which in turn decreased their future STEM interest (Donovan, Stuhlsatz, Edelson, & Buck Bracey, 2019).
The effects of stereotypes on self-representations may be particularly strong for those who hold fixed mindsets. Indeed, fixed ability beliefs predict increased concerns about belonging for women in STEM (Bian, Leslie, Murphy, & Cimpian, 2018;Deiglmayr, Stern, & Schubert, 2019). In contrast, the growth mindset can help students realize that social identities will not prevent them from achieving success and that stereotypes are not destiny (Good et al., 2012). High school girls who endorsed a growth mindset about math had higher expectancies of success in math, which predicted higher math achievement (Degol, Wang, Zhang, & Allerton, 2018).
Teachers' and parents' mindsets can also influence students' self-representations and academic outcomes. For example, students from negatively stereotyped racial groups earned higher grades and reported greater motivation in STEM courses when their professors endorsed a growth mindset (Canning, Muenks, Green, & Murphy, 2019). Adults with fixed mindsets can also send messages to students that have a negative impact on their motivation (Park, Gunderson, Tsukuyama, Levine, & Beilock, 2016). Even well-meaning messages (for example, to reassure students struggling in math that "not everyone can be a math person") can inadvertently undermine students' motivation if they reinforce a fixed mindset. This type of "comfort-oriented" feedback sends the message that teachers have low expectations, which undermined students' expectations for success in math in one study (Rattan, Good, & Dweck, 2012).
In one study, college students who were experimentally taught to have a growth mindset about using computers subsequently showed greater self-efficacy in that domain (Martocchio, 1994).
Similarly, when teachers emphasized the mistakes, failures, and struggles of others in science (emphasizing how famous scientists worked hard and learned over time), high school students showed greater interest and learning in science (Hong & Lin-Siegler, 2012). Interventions that have taught a growth mindset have shown benefits to students' motivation and academic outcomes (Blackwell, Trzesniewski, & Dweck, 2007), especially for members of negatively-stereotyped groups (Aronson, Fried, & Good, 2002;Broda et al., 2018) and low-achieving students (Paunesku et al., 2015;Yeager et al., 2019). A growth-mindset intervention in computer science classes significantly increased students' sense of belonging and interest in computer science (Burnette et al., 2020). In terms of women in STEM, hearing a STEM professor communicate the belief that everyone has the potential for growth eliminated the gender gap in STEM majors' sense of belonging in a STEM course (Rattan et al., 2018). An intervention that incorporated growth messages into an invitation to a tutoring program in an introductory STEM course led to more women signing up and attending the program, which then improved their course grades (Covarrubias, Laiduc, & Valle, 2019).
Belonging uncertainty. Stereotypes can make members of stereotyped groups more vigilant for cues about whether their group membership will be relevant. This can also lead to a phenomenon known as "belonging uncertainty" (Walton & Cohen, 2007). When students from negatively stereotyped groups have doubts about their belonging in academic settings, they may show constant alertness for cues (such as environmental cues, Master et al., 2016a) about whether they belong. Their sense of belonging may be contingent on the most recent cues they experience. On good days, they may feel that they belong, but minor incidents (such as being excluded from social events by other students) can threaten their sense of fit in that environment (Cohen & Garcia, 2008). In this way, students who are uncertain of whether they belong may base their current sense of belonging on their most recent experiences in that setting.
Interventions to strengthen sense of belonging. Some interventions have strengthened women's sense of belonging by targeting belonging uncertainty: by framing adversity as normal, something all students experience that gets better over time (Walton & Cohen, 2007). This changes how students interpret adversityas normal and temporary, instead of a sign that the student does not belong there.
An intervention with first-year students in strongly male-dominated engineering majors found that this social-belonging intervention increased women's GPA compared to a control condition, and led women to report more positive experiences and greater confidence that they could succeed in engineering (Walton et al., 2015; see also Ramsey et al., 2013). Similarly, a letter from a female role model that normalized concerns about belonging improved women's course grades in chemistry (Herrmann et al., 2016).
Other interventions have used role models to strengthen women's sense of belonging. An intervention with pre-med majors found that exposure to successful female physicians increased women's sense of belonging, which led to increased interest in a medical career (Rosenthal, Levy, London, Lobel, & Bazile, 2013).
This integrated review has provided a developmental model for how stereotypes and self-representations (including identification, ability beliefs, and sense of belonging) can affect students' motivation in STEM, and also provided summaries of evidence-based interventions that have been effective. However, there are some key gaps in the literature. Gillette et al., 2015). If stereotypes truly have an influence on girls' sense of belonging in STEM, then awareness and endorsement of stereotypes about a STEM field should develop before or concurrently with a lower sense of belonging in that field and predict changes in belonging over time. We also are unaware of existing longitudinal studies examining how girls' sense of belonging in various STEM and non-STEM fields changes over time. Although there are some valuable existing longitudinal data sets, many of these studies (e.g., exploring expectancy-value theory; Jacobs et al., 2002;Musu-Gillette et al., 2015;Nagy et al., 2010) were conducted in the 1980s or 1990s, examined only math-related variables, and did not measure students' sense of belonging.
There is also a strong need for intervention studies that test mechanisms experimentally and then continue to follow students over time to show how long causal effects on behavior may last (e.g., Goyer et al., 2019;Harackiewicz, Rozek, Hulleman, & Hyde, 2012) , 2018b, 2018c). Researchers should also consider how intersectionality might affect students in STEM who belong to multiple stereotyped groups. Although this article has focused primarily on gender, and has only briefly mentioned racial/ethnic minority status, there are many important social identities that intersect with gender in students' experience of stereotypes (e.g., Rogers & Meltzoff, 2017;Turner & Brown, 2007). For example, Black women majoring in STEM report lower sense of belonging in STEM compared to other groups (Rainey, Dancy, Mickelson, Stearns, & Moller, 2018), and may benefit from both academic belonging interventions and social belonging interventions (Lewis & Hodges, 2015). Are belonging interventions most effective for young girls with multiple stereotyped identities, who may worry that they do not belong due to both their gender and racial identities? On the other hand, there is also some evidence that ethnic minority girls are less likely to endorse STEM-gender stereotypes and less likely to be subject to STEM-gender stereotypes from others (Casad, Hale, & Wachs, 2017;O'Brien, Blodorn, Adams, Garcia, & Hammer, 2015;Purdie-Vaughns & Eibach, 2008).
Other variables such as general racial stigma, information about and support for attending higher education, and family influences on self-representations may also be important for understanding the academic experiences of members of these groups (Simpkins, Price, & Garcia, 2015). Much of the existing research on negatively stereotyped groups such as racial minority students is highly compatible with our model and supports the importance of examining these students' sensitivity to stereotypes and sense of belonging (e.g., Gillen-O'Neel & Fuligni, 2013;Thoman et al., 2017). Nevertheless, it remains valuable to explore how psychological constructs are similar or different across social groups that are often underrepresented in research (Hall, Yip, & Zárate, 2016), as well as to recognize students as individuals who participate in cultural communities and practices rather than using their ethnic group as a trait for making generalizations (Gutiérrez & Rogoff, 2003).
There may be other differences between groups of students that impact the effectiveness of interventions. For example, growth mindset interventions are often most effective for struggling students (Paunesku et al., 2015;Yeager et al., 2019). Are growth mindset interventions more effective for girls who worry that their effort is a sign that they do not belong in STEM (Smith et al., 2013)? Are interventions to broaden stereotypes more effective for girls with a tendency to essentialize gender differences (Coleman & Hong, 2008;McPherson, Banchefsky, & Park, 2018)? There may also be meaningful variations in STEM stereotypes across cultural contexts that impact the effectiveness of interventions (e.g., Rattan, Savani, Naidu, & Dweck, 2012). The strength of the stereotype that STEM = boys also varies across countries (Nosek et al., 2009), so interventions to broaden stereotypes may be less effective than other approaches in countries where gender stereotypes are weaker.
Recent findings suggest that researchers should pay attention to variations among STEM fields. Although math abilities and self-representations are important for a variety of STEM fields (Watt et al., 2017), researchers may gain predictive power by examining beliefs specific to each STEM field separately. There may be crucial differences in the stereotypes about different STEM fields (Cheryan et al., 2017). Fields which are believed to require natural ability, such as computer science, engineering, and physics, are also the fields in which women are most underrepresented (Leslie et al., 2015;Meyer, Cimpian, & Leslie, 2015). Even as early as elementary school, computer science and engineering gender stereotypes are stronger than gender stereotypes about math and science (Master, Cheryan, Moscatelli, et al., 2017;Master et al., 2020; see also Leslie et al., 2015). In college-aged students, there are differences even among different types of engineering fields (Walton et al., 2015). All STEM fields cannot be lumped together.
In this review, our ability to directly compare STEM fields was limited by the fact that most studies described here were conducted within a single STEM domain, so we were unable to compare across multiple domains.
Our STEMO model (Fig. 1) would predict that the strength of stereotypes among STEM fields should correspond to more negative self-representations, as well as greater underrepresentation of women. In other words, the STEM fields with the strongest stereotypes should also be the fields in which women report the lowest levels of identification, ability beliefs, sense of belonging, and interest. Although some evidence does support this prediction (Leslie et al., 2015;Master, Cheryan, Moscatelli, et al., 2017), future research could directly quantify the strength of stereotypes across different STEM fields and relate that to girls' and women's selfrepresentations and interest within those fields.
Another question for future research involves the root causes of differences in stereotypes across fields. Why might computer science and engineering have stronger gender stereotypes than math and biology? One source of this variation may be the relative lack of experience that students have with these fields (Cheryan et al., 2017). For example, students have many opportunities to see girls enjoying and succeeding at math, which may help to counter traditional stereotypes. Students may have very few experiences to counter traditional gender stereotypes about computer science. Another source of variation may lie in the masculine cultures of the fields, including the norms and values within these fields and how family-friendly they are perceived to be (Diekman et al., 2011;Weisgram & Diekman, 2017). Computer science and engineering are perceived to have particularly masculine cultures (Cheryan et al., 2015). Finally, it is difficult to tease apart the causal links between stereotypes about gender representation, stereotypes about ability, and stereotypes about interest in each field. Believing that few women participate in engineering may lead students to infer both that women are less interested and that they have lower ability compared to men. Conversely, believing that most women are uninterested in engineering may lead to inferences that they are unlikely to choose that field and that they lack engineering talent. In this way, stereotypes can act as self-fulfilling prophecies. Knowing these stereotypes pushes girls away from these fields, and reinforces the stereotypes for the next generation.
Recommendation 4: Open, Replicable, and Meaningful Science Finally, there is a need for more pre-registered studies and replications in education (see Makel & Plucker, 2014). Pre-registration allows improved transparency (van't Veer & Giner-Sorolla, 2016) and reduces the frequency of the selective reporting of only statistically significant results out of multiple analyses, also known as phacking (Nelson, Simmons, & Simonsohn, 2018). Few of the studies reviewed here were pre-registered, and many of the interventions have not yet been reproduced by independent groups of researchers. Reproducibility and replication are especially important for educational interventions (U.S. Department of Education, 2018a). These interventions typically replace classroom learning opportunities, so it is important to examine them systematically to better understand when and for which students they are effective (Schneider, 2018). We also urge researchers to consider effect sizes and confidence intervals when evaluating the effectiveness of interventions (Cumming, 2013;C. J. Hill, Bloom, Black, & Lipsey, 2008). For example, if a growth mindset intervention leads to an increase of d = .11 in grade points for low-achieving students, how meaningful is such an effect (Yeager et al., 2019)? The average effect size, scalability (which has implications for cumulative impact across large numbers of students), and cost effectiveness of such interventions should all be considered (Kraft, 2019).
We have presented the STEMO developmental model to show the way that social identities and the stereotypes about them can impact students' STEM interest and academic outcomes through self-representations (Figure 1). When students face negative stereotypes about their social identity, those stereotypes can affect their self-representations. Those self-representations, in turn, can affect interest and academic outcomes (such as choice of STEM career or grades in STEM courses). We have argued that self-representations, including identification and ability beliefs, are a key part of this process. In particular, we emphasize the importance of students' sense of belonging for interest and academic outcomes in STEM. By organizing and synthesizing the research in this way, our goal is to draw attention to research on students' sense of belonging for STEM education.
We have also drawn attention to the many ways that stereotypes about students' social identities can impact their STEM outcomes through their beliefs, attitudes, and behavior regarding STEM. We have described both negative pathways (e.g., negative stereotypes about social groups that reduce students' sense of belonging in STEM) and positive pathways (e.g., interventions that broaden beliefs about STEM, use growth mindsets to reduce the impact of stereotypes, and normalize doubts about belonging) between stereotypes and STEM outcomes.
Our STEMO developmental model makes several novel contributions to our understanding of gender gaps in STEM. To help contribute to a more interdisciplinary and nuanced perspective on STEM education, we have made efforts to: (a) emphasize the importance of self-representations such as sense of belonging, (b) distinguish among STEM fields instead of grouping them together, and (c) consider developmental processes across a wide range of ages starting as early as preschool. Although the self-representations discussed here (identification, ability beliefs, and sense of belonging) are quite distinct from each other, our STEMO model highlights how each can be shaped by cultural stereotypes and social identities to change how students view themselves in relation to academic fields. Am I a STEM person? Do I have what it takes to succeed? Do I belong here? Each self-representation also acts in ways to shape students' interest and academic outcomes. By bringing together these disparate constructs, we aim to draw other researchers' attention to the parallels between them.
We also hope that the STEMO model outlined here can help guide educational research from correlational toward more experimental studies showing causal links between constructs. The emphases on stereotypes, social identity, variations among STEM fields, and developmental processes can help guide discussions about future interventions to provide theoretically-informed predictions. When and for whom will interventions be most effective when scaled up among heterogeneous populations of students?
We suggest that by changing messages about "who belongs" and can succeed in STEM, we can help girls, women, and members of other underrepresented groups feel a greater sense of belonging in a variety of STEM fields. In turn, this may encourage them to develop interest, pursue STEM opportunities, and contribute to the next generation of STEM innovators.