Abstract
Self-to-prototype matching is a strategy of mental comparisons between the self-concept and the typical or “representative” member of a group to make some judgment. Such a process might contribute to interest in pursuing a science career and, relatedly, women’s underrepresentation in physical science, technology, engineering, and mathematics (pSTEM) fields. Across four studies, we measured self–scientist discrepancies on communal, agentic, and scientific dimensions, and assessed participants’ interest in a science career. The most consistent predictor of science interest was the discrepancy between self and scientist on the scientific dimension (e.g., intelligent, meticulous). Study 4 established that students with larger self–scientist discrepancies also had less accurate perceptions of students pursuing science, and that inaccuracy was related to lower science interest. Thus, students with lower science interest do not just perceive scientists differently from themselves but also erroneously. Discrepancy and inaccuracy together explained a significant portion of the gender gap in pSTEM interest.
Prototype matching is a strategy whereby decisions—from the mundane to the profound—are influenced by overlap in one’s self-concept with impressions of a prototypical other. According to this work, when making a complex social decision such as where to live next school year (Niedenthal, Cantor, & Kihlstrom, 1985) or whether to begin smoking as an adolescent (Chassin, Presson, Sherman, Corty, & Olshavsky, 1981), people compare themselves with a prototype of the typical person who would choose each possible option, and choose the outcome that produces the least discrepancy between the self and the prototype. Social actors seem particularly likely to engage in prototype matching strategies in the course of “trying on” various occupation possibilities and deciding which career option to pursue. In the present research, we were particularly interested in using the prototype matching framework to better understand who chooses to pursue physical science, technology, engineering, and mathematics (pSTEM)–related careers. While numerous social psychological factors have been explored as possible contributors to the gender gap in pSTEM fields (for reviews, see Blickenstaff, 2005; Ceci & Williams, 2011; Halpern et al., 2007; Spelke, 2005), almost no work explores the congruence between one’s self-concept and concept of a scientist—in particular on relevant trait dimensions—as a factor in women’s underrepresentation in pSTEM. Just as students can have harmful stereotypes about women’s abilities in pSTEM, they could also have harmful stereotypes about what a scientist is like, leading them to view scientists as very discrepant from the self and contributing to decreased interest in pSTEM (Cheryan & Plaut, 2010). The goals of the present research were to (a) examine perceived self–other discrepancies on trait dimensions relevant to predicting interest in pursuing a career in science; (b) ask for which specific trait dimensions do such discrepancies matter; (c) assess whether these discrepancies account in part for the gender gap in interest in pursuing a science career; and, finally, (d) begin to explore whether such discrepancies covary with inaccurate (e.g., Most scientists are singularly minded nerds) or accurate perceptions of what scientists are like.
Stereotypic Perceptions of Scientists
Exercises in which children draw a scientist (the Draw-A-Scientist Test) have shown that their perceptions of scientists are highly stereotypical—a bearded, bespectacled, lab-coat-clad man alone in his laboratory, surrounded by beakers (Chambers, 1983). More recent research indicates that these stereotypes are still the dominant perception (or prototype) of what a scientist is like (Fralick, Kearn, Thompson, & Lyons, 2009). Furthermore, these stereotypes persist past childhood—college undergraduates who are asked to draw a scientist produce images that are similar in stereotypic content to those of younger children (Rahm & Charbonneau, 1997). One limitation of this body of research is its idiosyncratic focus on drawings of scientists as a measure of stereotypic perceptions. While it is clear that students both young and old visually perceive scientists in a fairly uniform, stereotypic way that is likely different from their perceptions of themselves, it has yet to be determined where exactly those differences lie with respect to the content of students’ perceptions. 1
Of particular interest in the current work was the question of whether some trait dimensions matter more than others. The communal versus agentic distinction has a longstanding history in the field of psychology (Bakan, 1966). As they are traditionally defined, communality concerns itself with others’ needs and union with others, and is seen as feminine, while agency concerns itself with assertiveness and mastery, and is stereotypically perceived as masculine (Eagly & Steffen, 1984). The Stereotype Content Model makes a similar distinction between the dimensions of warmth (tolerant, warm, good natured) and competence (confident, independent, competitive; Fiske, Cuddy, Glick, & Xu, 2002). Diekman and colleagues (Diekman, Brown, Johnston, & Clark, 2010; Diekman, Clark, Johnston, Brown, & Steinberg, 2011) have described the communal and agentic dimensions with respect to goals, identifying communal goals as, for example, helping others or serving the community, and agentic goals as financial reward, status, or self-promotion. They argue that STEM-related careers are perceived by male and female students alike as low in affording communal goal fulfillment, that communal goals are of particular personal importance for female students, and that as a result, women express lower interest in pursuing STEM fields. Carli, Alawa, Lee, Zhao, and Kim (2016) reported that stereotypes of men and scientists share more in common (both are seen as highly agentic) than those of women and scientists, because women are seen as high on communality, whereas scientists are viewed as particularly low on this dimension.
We suggest that in addition to these two dimensions, it is useful to consider a third, more nuanced dimension specifically targeting trait attributes associated with the stereotype of scientists (Banchefsky & Park, 2015). While this cluster of traits shares some overlap with communality and agency, it contains attributes uniquely associated with scientists. Thus, while scientists may be seen as awkward and solitary (low on communality) or competent and independent (high on agency), they are also seen as logical, curious, and precise, which are not reflected in either the communal or the agentic dimensions. Furthermore, many of the traits that traditionally make up the agentic dimension, such as aggressive, dominant, or competitive, do not align with stereotypes of scientists as nerdy and socially awkward. Thus, a key goal of the present research was to examine the role of prototype matching processes along both the well-established communal and agentic dimensions, as well as a more focused scientific dimension. We anticipated that communality (based on previous research) and the scientific dimension (given its relevance) should both matter in terms of self–scientist discrepancies, but that agency as defined by self-confidence and dominance would not relate to interest in pursuing pSTEM.
Self-to-Prototype Matching Strategies in Selection of a pSTEM Career
The basic idea that perceiving greater similarity between oneself and someone in a pSTEM field would produce greater interest in pSTEM has found support across a variety of studies. These studies are frequently grounded in a theoretical approach focused on role models and one’s sense of belonging (Cheryan, Master, & Meltzoff, 2015; Dasgupta, 2011). For instance, role models for women in pSTEM are more effective if they are similar to the self in terms of group membership (e.g., gender; Stout, Dasgupta, Hunsinger, & McManus, 2011), and an individual is more likely to serve as a role model when described as being similar to the participant, although to date, this has always been in abstract terms (e.g., “You are quite similar to the women leaders you read about earlier”; Asgari, Dasgupta, & Stout, 2012). In addition, women feel less similar to role models who embody pSTEM stereotypical hobbies, movie, magazine, and clothing preferences (Cheryan, Siy, Vichayapai, Drury, & Kim, 2011). Stereotypical environmental cues have been shown to reduce women’s pSTEM interest by reducing their ambient belonging and sense of overall similarity to those in a pSTEM field (Cheryan, Plaut, Davies, & Steele, 2009). While this work addresses related ideas about similarity to others and pSTEM interest, these studies do not examine comparisons of the self with the perceived prototypical group member either in general or on specific trait dimensions. In contrast, the present research directly asks about correspondence between perceptions of the self and of the prototypical scientist to examine the role of this in expressed interest in STEM.
A small body of research has investigated questions more closely related to the prototype matching strategy as a factor in pursuing a pSTEM career. However, none of this work quite asks the question of interest here. Specifically, previous studies asked about comparisons with science teachers, or with a student whose favorite subject was science, both of which are only indirectly relevant to comparisons with a scientist in a pSTEM career. Moreover, these studies were conducted with German and/or Dutch high school students, a distinctly different sample than American college students (Hannover & Kessels, 2004; Kessels & Taconis, 2012; Rommes, Overbeek, Scholte, Engels, & De Kemp, 2007; Taconis & Kessels, 2009). Lee (1998) used scientists as the comparison target but assessed prototype matching on a set of global evaluative (e.g., good-bad) content-free semantic differential scales rather than on specific trait dimensions.
Perhaps most directly relevant to the current work is a study by Cheryan and Plaut (2010) who examined the role of perceived similarity to people in a given field in predicting interest in pursuing that field. Their findings—that women’s lower interest in computer science was mediated by their perceptions of themselves as less similar to computer science majors—are consistent with our argument regarding the importance of self–scientist discrepancies. Their approach differed, however, in that they used a single item that measured global perceived similarity (“How similar are you to computer science majors?”). The present work goes beyond this by examining the content of those perceptions, asking which dimensions (communal, agentic, or scientific) are critical in driving the relationship between greater discrepancy and lower interest.
Also, critical to the current work are questions regarding the accuracy of these perceptions. If self–scientist discrepancies predict decreased interest in pursuing STEM fields and these perceptions are inaccurate, this suggests a very different intervention strategy than if they are accurate perceptions. To the extent that large self–scientist discrepancies predict lowered interest in pursuing a science career, if those perceptions of scientists are inaccurate (e.g., believing that most scientists are workaholic nerds), then interest in pursuing a science career might be achieved by correcting these misperceptions. But if perceptions of scientists are in fact accurate (e.g., They have a math ability I genuinely lack), then it makes no sense to try to change this (accurate) perception as a means for increasing interest in pursuing a science career. If students perceive scientists accurately, then a large self–scientist discrepancy simply reflects a genuine mismatch between the student and that career—trying to change their perceptions to reduce self–scientist discrepancies would only set them up for failure if they chose to pursue STEM. We suspected that the former outcome (inaccurate perceptions) was more likely given stereotyping processes, particularly in light of evidence that scientists as a group are seen in stereotypic terms (see above, for example, Fralick et al., 2009). In other words, if students’ perceptions of scientists are based on inaccurate stereotypes, then these perceptions are likely to differ greatly from their self-perceptions, and so a larger self–scientist discrepancy should be seen among those students with more inaccurate perceptions. To date, none of the existing research assesses perceived discrepancies, the accuracy of perceptions, and the relationship between these to more fully characterize how these social perception processes might contribute to the gender gap in pSTEM pursuits.
Goals of the Current Research
In the current research, our primary goal was to use a prototype matching framework to ask whether discrepancies in self- and scientist perceptions are related to interest in pursuing a career in science. Specifically, the studies were designed to directly ask about interest in pursuing a science career, to investigate prototype matching as a strategy among U.S. college students, and perhaps most importantly, to test the effects of self–scientist discrepancies not just in general but specifically along three dimensions selected for their relevance to the judgment outcome (scientific attributes), or their prominence in the existing social psychological literature (communality and agency). The plan was to test which of the trait dimension discrepancies when examined individually predicts interest, as well as which matters when controlling for the others. Moreover, to increase the scope of the work, in addition to trait judgments, we explored discrepancies on behaviors (e.g., help the community) and values (e.g., social justice) related to the three dimensions.
A second major goal of the research was to begin to assess the accuracy of students’ perceptions of students pursuing science, and how that relates to self–scientist discrepancies. That is, the work explores not only whether students perceive people in science fields differently from how they perceive themselves but also whether the perceptions of others are accurate. Overall, we proposed the following hypotheses:
We view Hypothesis 3 as particularly important given the existing emphasis in the social psychological literature on communality perceptions as a deterrent to selecting a STEM career path (e.g., Carli et al., 2016; Diekman, Brown, et al., 2010; Diekman, Clark, et al., 2011). While we expect to see that communality discrepancies do matter, when pitted against the more directly relevant dimension of scientific attributes, we anticipated that the latter would matter most, suggesting that interventions targeting scientific perceptions might ultimately be more effective than trying to convince students that STEM careers fulfill communal goal needs.
With respect to Hypothesis 5, it could be the case that inaccuracy arises from mere unfamiliarity with scientists as a group (with no consistent tendency to view them as, for example, particularly low on communality) rather than stereotypes about them, and in that case we would expect to see no significant relationship between discrepancy and inaccuracy. However, as we do think that inaccuracy in perceptions of scientists is in part due to stereotypic perceptions of scientists, we hypothesize that greater inaccuracy will relate to greater discrepancy.
While we hypothesize that students whose self-concepts are more similar to their concept of a typical scientist will report greater interest in science (and vice versa), we make no strong claims about the causal direction of this relationship between discrepancy and interest. It could be the case, as we suggest, that students use the magnitude of the discrepancy between their self and scientist concepts to inform their judgment of how interested they are in science. However, it is of course equally plausible that students first form their interest in science, adjust their self and scientist concepts through self-stereotyping and self-anchoring processes, and show a reduced self–scientist discrepancy. It is not our aim to argue that only one of these processes is occurring; there are likely to be multiple directions of influence at play here, and the correlational nature of these studies does not allow us to definitively ascertain the primacy of either one of these causal directions. In addition, we do not make strong claims about the process by which this increased similarity occurs with respect to self- versus other-perceptions. The hypothesized relationships between discrepancy and interest are certainly a product of a bidirectional process, wherein students who are interested in science see scientists as a valuable future ingroup and shift both their self-concepts to match their scientist concepts through self-stereotyping, and their scientist concepts to match their self-concepts through self-anchoring; we are not trying to argue that only one of these is likely to occur.
The goals of this project, then, were (a) to use prototype matching as a framework for systematically identifying across three dimensions (communality, agency, and scientific ability) and three attribute types (traits, behaviors, and values) which matters most in terms of predicting science interest; (b) to ask whether the magnitude of self–scientist discrepancies accounts in part for gender differences in expressed interest in pursuing a science career; and (c) to investigate the accuracy of perceptions of scientists, and how this accuracy relates to self–scientist discrepancies and science interest. The first goal (self–scientist discrepancies) was addressed in all four studies; the third goal (accuracy of perceptions) was addressed in Study 4. Studies 1 to 3 asked about interest in pursuing “a science career,” which did not show differences by gender. Study 4 asked specifically about interest in pSTEM, and here evidence of a gender gap was obtained; thus, Study 4 provided the best test of whether self–scientist discrepancies account for some portion of the gender gap in pSTEM pursuit (the second goal). Choice of a pSTEM academic major was used in Studies 1 to 3 as a proxy for interest in pursuing pSTEM to provide a replication test with a behavioral measure.
General Method
The same basic procedure was used across Studies 1 to 4, which differed in small ways that will be noted for each study. Participants completed a survey either in a laboratory or online. The survey asked all participants to rate themselves and the typical scientist on 8-point scales reflecting traits, behaviors, and values that were structured around three dimensions: communal, agentic, and scientific (see Table 1 for a list of all traits, behaviors, and values for each dimension). The specific traits, behaviors, and values were drawn from those used by previous researchers (primarily Eagly & Steffen, 1984, with some from Diekman et al., 2010) and our own theoretical considerations (particularly for the scientific dimension). The trait ratings were bipolar, such that the two ends of each scale were labeled, respectively, with the trait and its semantic opposite (e.g., To what extent is the typical scientist illogical/logical). Behavior and value ratings were unipolar (e.g., To what extent is each behavior or value characteristic of the typical scientist). Ratings of the self were always given before ratings of scientists. Within each target, ratings were given such that traits were first, behaviors were second, and values were last. Within each rating type, the specific traits, behaviors, and values were presented in a random order for each participant. Before making self and scientist judgments, participants also provided their interest in a science career by responding to three items on a 1 to 7 scale: “How interested are you in pursuing a career in science?” (not at all . . . extremely), “I would find a career in science enjoyable” (strongly disagree . . . strongly agree), and “I feel like I would fit in well in a science career” (strongly disagree . . . strongly agree). Finally, they indicated their gender and academic major. If they were still undecided, students indicated which academic major they would choose if they had to choose today. With the exception of 19 participants in Study 3, all participants across Studies 1 to 4 were drawn from an Introduction to Psychology course. As such, our samples are comprised of students early in their college career (more than 80% were either freshmen or sophomores), and incorporate a broad representation of academic majors. The breadth of academic majors among participants in our sample allows us to assess how self–other discrepancy relates to why pSTEM students choose pSTEM, as well as why non-pSTEM students do not choose pSTEM.
Communal, Agentic, and Scientific Traits, Behaviors, and Values.
Signed self–scientist discrepancy scores were calculated by taking the difference between self and scientist ratings on each item and averaging these across items within each of the content dimensions (i.e., communality, agency, scientific), scored in a direction such that more stereotypic ratings of scientists would result in more positive discrepancy scores. That is, for communal ratings, discrepancies were scored as self–scientist, so that viewing scientists as less communal (in accordance with the stereotype) would result in a positive discrepancy score. Both agentic and scientific ratings were scored as scientist–self, so that viewing scientists as more agentic or scientific (in accordance with the stereotype) would result in a positive discrepancy score. The final discrepancy score was in the same unit as the scale participants used for each item. For example, a communal self–scientist discrepancy score of 2 corresponds to a participant’s self-rating on communal items being 2 points higher on the 8-point scale on average than the scientist rating on those same communal items, as judged by that participant. Discrepancy scores were calculated for each dimension and attribute type combination (e.g., communal trait, agentic value, etc.), resulting in nine discrepancy scores. These scores were also averaged across attribute types, within a dimension, to obtain three dimension-level discrepancy scores (i.e., communal trait, behavior, and value discrepancy scores were averaged to obtain a dimension-level communal discrepancy score).
The primary effect of interest was whether self–scientist discrepancy would predict interest in a science career. We were uncertain of the exact effect size to expect, so in Study 1, we collected data from enough participants to achieve at least 80% power for a small-to-medium effect (
Study 1
Method
A total of 237 undergraduate students (117 male, 120 female) participated in exchange for partial course credit. Participants completed a survey either in a laboratory in groups of up to 10 or online (118 participants completed the survey online).
Study 1 differed from the general procedure, in that participants provided ratings not just of self and scientist but also of the typical early childhood educator and the typical person working in the participant’s respective anticipated future job. Participants also constructed histograms on a subset of attributes to measure the perceived variability of all three career targets (see Park & Judd, 1990). These measures did not result in meaningful differences among the targets and so are reported only in the supplemental materials. Finally, in the first study, values and behaviors were included as exploratory judgments. Because of the prominence of communality and agency in past research it was easy to identify items for these, but values and behaviors were not included in the scientific dimension. Thus, only for traits can all three dimensions be compared. For this reason, the behavior and value judgments for communality and agency in Study 1 are reported only in the supplemental materials; in Study 2, we include and report them for all three dimensions.
Results
Discrepancy and interest
Self–scientist discrepancies were calculated as described above. Self-ratings (communal trait α = .64, agentic trait α = .68, scientific trait α = .55), scientist judgments (communal trait α = .64, agentic trait α = .71, scientific trait α = .61), and self–scientist discrepancies (communal trait discrepancy α = .67, agentic trait discrepancy α = .69, scientific trait discrepancy α = .60) showed acceptable reliability considering the small number of items involved (five for each dimension; reliability would be increased if the five items were simple synonyms of each other, but we deliberately chose traits to broadly cover each measured construct). Women had larger self–scientist discrepancies than men on communal traits, t(233) = 3.39, p < .001, and scientific traits, t(233) = 3.58, p < .001, and marginally greater discrepancies on agentic traits, t(233) = 1.66, p = .098.
These self–scientist discrepancies scores were used to predict expressed science interest (the average of the three interest items, α = .95), controlling for subject gender. 2 The simple relations of each attribute to science interest were first evaluated with separate regression analyses, each with two predictors: participant gender and the parameter in question (e.g., self–scientist discrepancy on communal traits). Next, the relative impact of the dimensions was assessed with a simultaneous regression model containing participant gender and discrepancies on all three attribute dimensions. Results from these analyses are presented in Table 2. While the simple relationships for both communal and scientific discrepancies were significant (supporting Hypotheses 1 and 2), the only significant predictor of science interest in the simultaneous model was self–scientist discrepancy on scientific traits (in support of Hypothesis 3): the effect size for which was 3 times as large as that for communality. 3 The more that a student perceived himself or herself as less scientific than the typical scientist with respect to traits such as logical, intelligent, and meticulous, the lower their reported interest in a science career—over and above their self–scientist discrepancies on communal and agentic traits. 4
Study 1 Regression Coefficients Predicting Expressed Science Interest.
Note. Simple relationship models contained only two predictors: participant gender and the parameter in question. Simultaneous relationship models contained participant gender and all listed parameters as predictors. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered.
p < .10. *p < .05. **p < .01. ***p < .001.
Discrepancy and pSTEM major status
One impetus for this project was to better understand women’s underrepresentation in science. Notably, however, no gender difference in reported science interest was obtained in this study, t(235) = −0.59, p = .56. This is consistent with national data where, when science is defined broadly (i.e., including the life sciences), no difference in gender representation emerges (National Science Foundation, Division of Science Resources Statistics, 2015; see Tables 2-8). Because participants were asked simply to indicate their interest in a “science career,” they were free to broadly define such a career. To explore the role of prototype matching in the gender gap within pSTEM specifically, subsequent analyses were conducted using pSTEM major status as the outcome of interest. All participants indicated their current academic major (or, for undecided participants, the major they would choose if they had to choose today), and these majors were coded, such that +1 indicated a pSTEM major and 0 indicated a non-pSTEM major. A major was classified as pSTEM if it was a science, technology, engineering, or mathematics major, excluding the life and social sciences (e.g., biology or psychology). A list of all academic majors and their pSTEM coding is provided in the supplemental materials. Of 65 possible majors, 35 (54%) were represented in Study 1. Forty-eight participants (20%) had not yet chosen a major, and 36 (15%) were classified as having a pSTEM major. Men were more likely than women to have or intend a pSTEM major in our sample, χ²(235) = −79.36, p = .002, providing a gender-disparate outcome variable to predict with the self–scientist discrepancies.
Study 1 Logistic Regression Coefficients Predicting pSTEM Major Status.
Note. Simple relationship models contained only two predictors: participant gender and the parameter in question. Simultaneous relationship models contained participant gender and all listed parameters as predictors. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered. pSTEM = physical science, technology, engineering, and mathematics.
p < .10. *p < .05. **p < .01. ***p < .001.
Study 2 Regression Coefficients Predicting Expressed Science Interest.
Note. Gender was not included in these analyses as all participants identified as the same gender (female). Simple relationship models contained only two predictors: condition (the variability manipulation) and the parameter in question. Simultaneous relationship models contained condition and all listed parameters as predictors. All continuous predictors were mean centered.
p < .10. *p < .05. **p < .01. ***p < .001.
Study 3 Regression Coefficients Predicting Expressed Science Interest and pSTEM Major Status.
Note. Simple relationship models contained only two predictors: participant gender and the parameter in question. Simultaneous relationship models contained participant gender and all listed parameters as predictors. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered. Coefficients for the model predicting pSTEM major status are in log odds from a logistic regression. pSTEM = physical science, technology, engineering, and mathematics.
p < .10. *p < .05. **p < .01. ***p < .001.
Survey Items Answered by Fall 2012 Physics 1110 Students.
Note. For the communal and agentic goal items, participants were instructed as follows: “Think about the career you plan to obtain from your major, then rate the likelihood you will be able to meet each of the following goals in that career. (1 = very unlikely, 5 = very likely).”
Study 4 Regression Coefficients Predicting Expressed pSTEM Interest.
Note. Simple relationship models contained only two predictors: participant gender and the parameter in question. Simultaneous relationship models contained participant gender and all listed parameters as predictors. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered. pSTEM = physical science, technology, engineering, and mathematics.
p < .10. *p < .05. **p < .01. ***p < .001.
Mean Inaccuracy Tendencies.
Note. Asterisks reflect significant differences in a t test comparing the difference between the actual and estimated means to 0.
p < .10. *p < .05. **p < .01. ***p < .001.
The previous models using communal, agentic, and scientific trait discrepancies were rerun, predicting pSTEM major status in a logistic regression, treating the predictors both individually and simultaneously (see Table 3). Greater self–scientist discrepancies on communal and scientific traits were significantly related to a lower likelihood of being in a pSTEM major in both the simple relationships and the simultaneous model. The sign for agentic traits changed direction in the simple versus simultaneous models, such that, on its own, larger discrepancies were associated with a (nonsignificant) lower likelihood of being in a pSTEM major, but when controlling for discrepancies on communal and scientific traits, a larger agentic trait discrepancy marginally positively predicted being in a pSTEM major (suggesting a possible suppression effect). Of note, the gender difference in pSTEM major status remained significant even after accounting for discrepancy scores, but, consistent with Hypothesis 4, controlling for the three trait discrepancies reduced the gender gap in pSTEM major status by 21%.
Discussion
Study 1 demonstrated that self–other discrepancy is a useful tool with which to investigate interest in a science career. Students who see themselves as very different from the typical scientist report lower interest in a science career. Discrepancies with respect to both communality (seeing the self as more communal than the typical scientist) and scientific-ness (seeing the self as less scientific than the typical scientist) were important, although the effect sizes for scientific discrepancies were of a larger absolute magnitude than those for communality (see Table 2). Moreover, although there was no gender difference in interest in pursuing science in general, when we narrowed the focus to students who are or express an interest in pursuing a pSTEM major, a significant gender difference emerged that was partially accounted for by these self–scientist discrepancy scores. Given that we found evidence of particular importance for the scientific dimension, we wanted to include it not just among trait ratings but also in behavior and value ratings, to test whether there is something uniquely important about scientific traits or whether the dimension would be broadly important across behaviors and values as well. Study 2 addressed this issue and provided a replication of the effects obtained in Study 1.
Study 2
Method
One hundred one female undergraduates participated in exchange for partial course credit. Participants completed the survey in a laboratory on campus in groups of up to 10. Study 2 differed from the general procedure in a few ways. The study design included a manipulation to increase the perceived variability of scientists, and only women were included in the sample as we anticipated that the variability manipulation might be most effective for them (i.e., recognizing that there are many different ways to be a scientist might especially increase women’s interest in pursuing science). Although a manipulation check demonstrated the effectiveness of the variability manipulation itself, the predicted “downstream” consequences of greater interest in pursuing science in the high variability condition were not obtained (all ps > .3); thus, details and analyses of the manipulation are reported only in the supplemental materials. The analyses reported here control for condition to ensure that this manipulation did not inadvertently alter any of the effects. As in Study 1, participants provided ratings of the perceived variability of scientists, and these are reported in the supplemental materials.
Results
Discrepancy and interest
Given the inclusion of the scientific dimension among all attribute types, we were able to run models looking at the independent contribution of all three dimensions among each attribute type (see Table 4). That is, in addition to testing the simple relationships between each type of discrepancy and interest (α = .94 for the three interest items), the simultaneous relationships across the three dimensions were tested for traits, behaviors, and values. Looking just at trait discrepancies, both communal and scientific trait discrepancies showed a significant negative relationship with interest. This was true for both the simple relationships and the simultaneous relationships. Those who perceived themselves as more communal than scientists, and also those who perceived scientists as more scientific than them, expressed lower levels of interest in a science career (again supporting Hypotheses 1 and 2). Looking just at behaviors, only the scientific behavior discrepancy was significantly negatively related to interest, again both in the simple relationships and in the simultaneous model (supporting Hypotheses 2 and 3). Finally, among values, none of the discrepancies were significantly predictive of science interest, although there was a marginal negative relationship between communal value discrepancy and science interest in the simple and simultaneous models. Both for parsimony and to obtain a more robust measure of the three dimensions, we averaged the discrepancy scores across attribute types within each dimension (see Table 4). At the dimension level, self-judgments (communal dimension α = .84, agentic dimension α = .79, scientific dimension α = .79), scientist judgments (communal dimension α = .85, agentic dimension α = .76, scientific dimension α = .83), and self–scientist discrepancies (communal discrepancy α = .82, agentic discrepancy α = .71, scientific discrepancy α = .72) all showed high reliability. Combining traits, behaviors, and values, we again see significant negative relationships with interest for both communal and scientific discrepancies, both in isolation and in the simultaneous model. Although the effect size for scientific discrepancy was larger than that for communality (consistent with Hypothesis 3), the difference was not that large in magnitude (the effect size for scientific discrepancy was 1.2 times that of communality).
Discrepancy and pSTEM major status
Replicating the analysis from Study 1, using pSTEM major status as the outcome, a logistic regression using the three dimension-level discrepancies as predictors revealed a significant negative relationship between communal dimension discrepancy and pSTEM major status (b = −0.76, p = .04). Thirty-one different majors (48% of the 65 possible) were represented among participants in Study 2. Twenty-one participants (21%) had not yet chosen a major, and 12 (12%) were classified as having a pSTEM major. Across the entire university, women comprise only 26% of students in pSTEM majors, contributing to the small number observed in this sample. Given the small number of participants with a pSTEM major in this study, this analysis should be interpreted with caution. 5
Discussion
Study 2 replicated our previous finding that self–scientist discrepancies are related to lower expressed interest in a career in science. It expands beyond Study 1 by finding similar relationships between larger discrepancies with respect to the scientific dimension and lower science interest for scientific behaviors and (directionally) for values, in addition to replicating the effect from Study 1 for traits. Study 3 replicated the findings of Study 2 (examining traits, behaviors, and values) with both male and female participants.
Study 3
Method
A total of 247 undergraduates (161 female, 86 male) participated. Two hundred twenty-eight (146 female, 82 male) participants completed the survey in a laboratory in groups of up to 10 in exchange for partial course credit; 19 participants completed the survey online in exchange for entry in a raffle with a US$25 prize. Participants provided trait, behavior, and value ratings of the self and the typical scientist. 6
Results
Discrepancy and interest
We replicated the self–scientist discrepancy analyses from Studies 1 and 2. At the dimension level, self-judgments (communal dimension α = .84, agentic dimension α = .80, scientific dimension α = .85), scientist judgments (communal dimension α = .85, agentic dimension α = .78, scientific dimension α = .85), and self–scientist discrepancies (communal discrepancy α = .77, agentic discrepancy α = .68, scientific discrepancy α = .77) all showed acceptable reliability. As in Study 1, women had greater self–scientist discrepancies than men on the communal, t(243) = 4.97, p < .0001, scientific, t(243) = 2.00, p = .047, and agentic, t(243) = 3.74, p = .0002, dimensions (agentic was marginally significant in Study 1; recall that Study 2 only included women, precluding these comparisons). Importantly, and consistent with Hypothesis 3, discrepancy on the scientific dimension (collapsing across trait, behavior, and value judgments) was most strongly negatively related to interest in pursuing a science career (see Table 5, α = .95 for the three interest items). This was true in both the simple and simultaneous models, where the effect size for scientific discrepancy was 6.5 times that of communal discrepancy, and 4 times that of agentic discrepancy. Communal discrepancy was marginally predictive of science interest in the simultaneous model, and, as in Study 2, discrepancies on the agentic dimension were positively related to science interest, and in this study, significantly so. The latter effect indicates that students with strong negative discrepancy scores on agency (seeing themselves as much more agentic—aggressive, powerful, dominant—than scientists) were also the least interested in pursuing a science career.
Discrepancy and pSTEM major status
Forty-four different majors (68% of the 65 possible) were represented among participants in Study 3. Forty-eight participants (19%) had not yet chosen a major, and 46 (19%) were classified as having a pSTEM major. In logistic regressions predicting pSTEM major status, only scientific discrepancy was significantly related to the likelihood of being in a pSTEM major, both across the simple relationships and in the simultaneous model (see Table 5). Students who saw themselves as more different from scientists on science-related attributes were less likely to be in a pSTEM academic major. There was an effect of gender, such that men were significantly more likely to be in a pSTEM major than women. As in Study 1, and consistent with Hypothesis 4, controlling for the three discrepancy scores reduced the magnitude of this gender gap by 17%.
Discussion
The discrepancy analyses clearly replicated Studies 1 and 2, showing that large self–scientist discrepancy perceptions, particularly for the scientific dimension, predict lower interest in pursuing a science career and a lower likelihood of being in a pSTEM major. Moreover, accounting for these discrepancy scores partially reduces the gender gap in selection of a pSTEM major.
This might suggest that one way to increase interest in pSTEM fields is by changing perceptions of scientists to reduce the size of these self–scientist discrepancies. However, it could be the case that students generally perceive scientists accurately, and that the variance among students that leads to differences in self–scientist discrepancies comes merely from differences in their self-perceptions (which may also be accurate). Having demonstrated that the way students perceive scientists (in relation to the way that they perceive themselves) is important in predicting their science interest, our next goal was to test the accuracy of these perceptions, and assess whether accuracy is related to self–scientist discrepancies and science interest. We hypothesize that students have stereotypic perceptions of scientists that lead them to view themselves as very different from the typical scientist (larger discrepancy). We also believe that these stereotypic perceptions of scientists are inaccurate. If this is the case, then self–scientist discrepancy should be related not just to lower interest but also to less accurate perceptions of scientists. Similarly, inaccurate stereotype-driven perceptions of scientists should be related to lower interest as well.
One of the greatest difficulties in asking about the accuracy of social perceptions is identifying an appropriate criterion against which to compare perceivers’ judgments (Judd & Park, 1993). Specifically, it is critical that the criterion be based either on a random (and hence representative) sample of the target population, or that the entire population is measured on the same items for which perceivers’ perceptions will be assessed. Study 4 used responses to a survey administered to all students (n = 787) enrolled in a calculus-based physics course in the fall of 2012, collected as part of a different project. Participants in Study 4 were asked to predict the responses of this population (all students enrolled in the calculus-based physics course in the fall of 2012) on the specific survey items. This course is a degree requirement for nearly all pSTEM majors on campus, meaning that the accuracy of participants’ perceptions of these students clearly reflects the accuracy of their perceptions of pSTEM students on campus more broadly. While the target group in this dataset was physics students rather than actual scientists, and perceptions needed to be restricted to judgments of the particular survey items asked of these students, the advantage in using responses to these survey items is that they provide an appropriate and relevant criterion (e.g., endorsement of communal and agentic goals from a specifiable population of students pursuing a pSTEM degree) against which perceivers’ judgments could be compared.
Study 4
Method
Two hundred twenty-nine undergraduates (139 women, 90 men) participated in exchange for partial course credit in a laboratory in groups of up to 10. Participants first provided estimates of the average responses of students in a calculus-based physics course at the University of Colorado Boulder. Specifically, they were told the following: In the Fall semester of 2012, we surveyed 787 students at the University of Colorado Boulder who were enrolled in the course Physics 1110; Physics 1110 is a calculus-based required course for most science and engineering majors at CU. These students were asked to respond to a variety of questions about themselves on a 1-5 point scale, where 1 indicated “Strongly Disagree” and 5 indicated “Strongly Agree” . . . for each statement below, please indicate what you think the average level of agreement with that statement was among the Physics 1110 students that we surveyed . . .
The items on the survey included some that were directly relevant to the communal, agentic, and scientific dimensions (e.g., communal and agentic goals; see Table 6). The survey also included items that asked about psychologically meaningful perceptions such as the utility value of science, how much students felt a sense of belonging in science, stress related to studying physics, and self-efficacy regarding physics performance, and hence were also included in the accuracy assessment. 7 In addition, as a more objective criterion that would not be subject to shifting standards (Biernat & Manis, 1994), participants were asked to estimate the average SAT math score and high school grade point average (GPA) of the physics students. Students who completed the fall 2012 survey were asked for permission to access these institutional records, and the criterion was based on those students who consented to have these data provided (n = 758). Participants were asked to estimate responses to the survey items for the entire class, and then separately for just the male and just the female students in the class (the latter two were counterbalanced).
Following estimates of responses to the survey items, participants were asked to rate the same attributes as in Studies 1 to 3 (Table 1). In a slight departure from the general procedure, after rating the self, participants rated the “typical person pursuing a pSTEM career” rather than the “typical scientist,” so that the target for the self–scientist discrepancy score better corresponded to the target for which the accuracy judgments were made. In addition, participants rated their personal interest in a pSTEM career, using the same items as Studies 1 to 3 but with “pSTEM” in place of “science” and defining pSTEM for participants as including “physical sciences such as physics, chemistry, or astronomy, as well as technology, engineering, and mathematics.” This change allowed the target of the discrepancy scores to match the level of specificity for the interest questions. It also narrowed the definition of a science career to pSTEM, increasing the likelihood that a gender difference in expressed science interest (not present in Studies 1-3) would emerge in Study 4. Restricting the interest questions to pSTEM better reflects the existing gender gap that this research was intended, in part, to investigate.
Results
Discrepancy and interest
Consistent with Studies 1 and 3, women had greater self-pSTEM discrepancies than men on the communal, t(225) = 3.26, p = .0013, agentic, t(225) = 6.13, p < .0001, and scientific dimensions, t(225) = 3.00, p = .003. At the dimension level, self-judgments (communal dimension α = .87, agentic dimension α = .82, scientific dimension α = .85), scientist judgments (communal dimension α = .87, agentic dimension α = .85, scientific dimension α = .91), and self–scientist discrepancies (communal discrepancy α = .77, agentic discrepancy α = .70, scientific discrepancy α = .77) were all reliable. Again, the relationship seen in Studies 1 to 3 between scientific discrepancy scores and interest (α = .94 for the three interest items) was replicated (see Table 7). In both the simple relationships and the simultaneous model, the scientific dimension discrepancy was significantly negatively related to pSTEM interest, supporting Hypothesis 2. Communal dimension discrepancy scores were not related to interest in pursuing a pSTEM career, either when considered alone or in conjunction with the other dimensions. As in Studies 2 and 3, there was evidence of a suppression relationship for agentic discrepancies, such that in the simultaneous model that controlled for both communal and scientific discrepancies, agentic discrepancy was significantly positively predictive of interest in a pSTEM career. The magnitude of the effect size for scientific discrepancy scores was larger than either of the other dimensions (by a factor of at least 7.5), supporting Hypothesis 3. Also, of note, males expressed higher interest in pursuing a pSTEM career than females, even after controlling for the discrepancy scores (see the simultaneous relationships model in Table 7). Together, the discrepancy scores accounted for 17% of the gender gap in expressed pSTEM interest.
Accuracy assessment
The next set of analyses tested participants’ accuracy in their perceptions of the pSTEM students. While participants provided estimates for the students overall, just the male students, and just the female students, the patterns of relationships were largely the same across these three target groups. The most robust relationships were seen for the judgments of the students overall, and so the inaccuracy analyses here use that set of judgments. We assessed the accuracy of these judgments in two ways (see Judd & Park, 1993): First, we calculated inaccuracy scores—deviations of the participant’s estimate from the true criterion value for each item, and then averaged across items (see below for more details). Second, a sensitivity correlation was computed by comparing the participant’s estimate and the criterion value for each item, calculating a correlation between these across all items. A higher correlation indicates greater accuracy in perceptions. Sensitivity correlations ask a somewhat different question than the inaccuracy scores. Rather than measuring how well the perceiver accurately gauges the group’s standing on a given item, it measures the degree to which the perceiver accurately orders the importance or endorsement of the items by the group. One can appropriately order the items by degree of endorsement (e.g., judging that a greater percentage of scientists endorse self-direction than serving humanity as goals) even if one’s specific estimates for those judgments are “inaccurate” in the sense that they differ numerically from the exact criterion value. Perhaps most importantly, sensitivity correlations are relatively uninfluenced by scale usage differences across judges.
The actual mean responses of the students in the physics course for each construct appear in Table 8, as do the mean perceptions of participants in Study 4 when asked to predict these responses. Overall, participants were relatively inaccurate in their perceptions, with the estimated mean significantly deviating from students’ actual responses on five of the six constructs.
Discrepancy and subjective inaccuracy
The primary question of Study 4 was whether self-pSTEM discrepancies would be related to inaccurate perceptions of those pursuing a pSTEM career. We examined this by calculating inaccuracy scores and sensitivity correlations across the full set of items that constituted the constructs in Table 8. A composite inaccuracy score was computed as the arithmetic difference between a participant’s estimate for a given item and the criterion value on that item, scored such that higher numbers indicated more stereotypic perceptions; these (signed) individual item discrepancies were then averaged to arrive at a single inaccuracy score for each participant. Stereotypic direction was defined consensually, using the average estimates in Table 8. For example, on average participants underestimated the extent to which the physics students endorsed communal goals, indicating a stereotypic perception that physics students endorse communal goals less than they actually did. For each communal item, then, the inaccuracy score was the actual mean rating of the physics students on that item minus the participant’s estimate, reflecting the degree to which the participant underestimated endorsement of this communal goal. Consider a given subject, Sarah, who rated endorsement of “serving humanity” as 3.7, whereas the actual mean rating across students in physics was 4.2. The extent to which Sarah underestimated “serving humanity” as a goal would be 0.5 (4.2-3.7). These inaccuracies were then averaged across all items, such that larger discrepancy scores indicated more stereotypic perceptions of the physics students. For one construct, stress, participants’ mean rating did not differ from the criterion (see Table 8), so the inaccuracy value was scored in what we judged to be a stereotype-consistent direction—overestimating the physics students’ stress: participant’s estimate—actual aggregate rating of stress. The same items used to create the inaccuracy score composite were used to calculate sensitivity correlations for each participant by simply correlating the participant’s estimate with the aggregate rating from the physics students, across the 19 items in Table 6.
Given that both the inaccuracy scores and sensitivity correlations included items relevant to all three dimensions (communal, agentic, scientific), as well as items not uniquely related to any of these, an “overall discrepancy” score was also calculated, averaging discrepancy across the three dimensions, to obtain a parallel level of generality. Supporting Hypothesis 5, greater inaccuracy in the direction of more stereotypic perceptions of Physics 1110 students was related to greater discrepancy between how the self and the typical person pursuing a pSTEM career were rated (see Table 9). Similarly, these students with larger discrepancy scores also demonstrated smaller sensitivity correlations, demonstrating less ability to accurately order these items from least to most highly endorsed. Thus, participants who perceived a greater disparity between their self-concept and their concept of the typical person pursuing pSTEM were also more likely to have inaccurate judgments of students pursuing pSTEM.
Study 4 Regression Coefficients Predicting Overall Discrepancy Scores With Inaccuracy.
Note. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered.
p < .10. *p < .05. **p < .01. ***p < .001.
These results indicate that self–other discrepancy is related to lower pSTEM interest, and also that this discrepancy is related to less accurate perceptions of those others. We can now ask the question of whether, like discrepancy, inaccuracy is also related to lower pSTEM interest. We expect that because inaccuracy is in part driven by stereotypic perceptions of those pursuing pSTEM, it will show a similarly negative relationship with pSTEM interest. However, one could also imagine that inaccuracy would be unrelated to pSTEM interest if inaccuracy is instead derived from mere inexperience with scientists as a group. We find that as expected, and similarly to discrepancy, greater inaccuracy was related to significantly lower pSTEM interest (see Table 10), both for inaccuracy scores and for sensitivity correlations. Women reported significantly lower interest in pSTEM than men even after accounting for inaccuracy.
Study 4 Regression Coefficients Predicting Expressed pSTEM Interest With Accuracy.
Note. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered. Sensitivity correlations were z-transformed before analysis. pSTEM = physical science, technology, engineering, and mathematics.
p < .10. *p < .05. **p < .01. ***p < .001.
Discrepancy and objective inaccuracy
Relations of inaccuracy to discrepancies and interests were also examined for the objective judgments of physics students—high school GPA and math SAT scores. In this analysis, scientific dimension discrepancies were used because they include items specifically related to intelligence, and thus map more closely onto GPA and SAT than overall discrepancy. For both GPA and SAT, inaccuracy was scored directionally as a tendency toward overestimating, in line with the stereotype that people who pursue pSTEM are highly intelligent. Participants’ estimated SAT score did not significantly differ from the physics students’ actual average math SAT score, t(228) = −0.63, p = .53; however, participants tended to underestimate physics students’ high school GPA—the actual mean was 3.63, while the average estimate across all participants was 3.53, t(228) = −5.36, p < .0001. For GPA, greater inaccuracy was related to larger self–scientist discrepancy scores and marginally to lower interest (see Tables 11 and 12). Notably, women had significantly greater inaccuracy in their judgments of GPA than men. For SAT judgments, the only significant effect was an interaction of gender with SAT inaccuracy predicting scientific discrepancy, such that women with greater overestimation in their judgments of the student’s SAT scores were particularly prone to greater scientific discrepancies. There was a weak trend, such that greater SAT inaccuracy was directionally related to lower interest (p = .15).
Study 4 Regression Coefficients Predicting Scientific Discrepancy Scores With Objective Accuracy.
Note. Gender was contrast coded, such that female = +1 and male = −1, and all continuous predictors were mean centered. GPA = grade point average.
p < .10. *p < .05. **p < .01. ***p < .001.
Study 4 Regression Coefficients Predicting Expressed pSTEM Interest With Objective Accuracy.
Note. Gender was contrast coded such that female = +1 and male = −1, and all continuous predictors were mean centered. Sensitivity correlations were z-transformed before analysis. pSTEM = physical science, technology, engineering, and mathematics; GPA = grade point average.
p < .10. *p < .05. **p < .01. ***p < .001.
Mediation of gender gap in pSTEM interest
Part of the motivation for this project was to attempt to better understand the gender gap in pSTEM. The final analytical step was to test for partial mediation of the gender difference in pSTEM interest in Study 4 (the only study with a continuous measure of interest on which there was a significant gender difference). Multiple mediation models were run testing overall discrepancy, inaccuracy scores from the questionnaire items in Table 6 (termed “subjective” inaccuracy), and a standardized composite of GPA and SAT inaccuracy (“objective” inaccuracy), bootstrapped 10,000 times. This model was tested twice: once using inaccuracy scores from the subjective judgments as the criterion (Figure 1) and once using sensitivity correlations (also from the subjective judgments; Figure 2). Using inaccuracy scores as the subjective inaccuracy variable, the mediators together comprised a significant total indirect effect (b = −0.151, p = .04, 95% confidence interval [CI] = [−0.292, −0.01]). The total indirect effect accounted for 20% of the gender effect on pSTEM interest; however, the relationship between gender and pSTEM interest was still significant after accounting for self–other discrepancy and inaccuracy. Among the individual mediators, only overall discrepancy had a significant indirect effect (b = −0.15, p = .01, 95% CI = [−0.261, −0.035]). Using sensitivity correlations as the subjective inaccuracy variable, the mediators as a set again had a significant total indirect effect (b = −0.163, p = .02, 95% CI = [−0.297, −0.029]) and accounted for 24% of the gender effect on pSTEM interest. Again, the relationship between gender and pSTEM interest was significant even after accounting for the mediators, and again only overall discrepancy had a significant individual indirect effect (b = −0.148, p = .012, 95% CI = [−0.263, −0.32]). 8
Mediation model attempting to explain the gender difference in pSTEM interest in Study 4 with self–other discrepancy, subjective inaccuracy scores, and objective inaccuracy scores.
Mediation model attempting to explain the gender difference in pSTEM interest in Study 4 with self–other discrepancy, subjective sensitivity correlations, and objective inaccuracy scores.
Discussion
Study 4 showed that students with larger discrepancies between judgments of the self and of the typical person pursuing a pSTEM career were also less accurate in their judgments of students pursuing a pSTEM career. Furthermore, this inaccuracy was related to diminished interest in pSTEM. The multiple mediation model indicated that the total effect of discrepancy and inaccuracy accounted for part of the gender gap in pSTEM interest, with discrepancy emerging as a significant individual mediator. In sum, students have stereotypic, inaccurate perceptions of those pursuing pSTEM that covary with a greater discrepancy between perceptions of the self and the typical person who pursues a pSTEM career, and a decreased interest in a pSTEM career. Thus, the results of Study 4 indicate that self–scientist discrepancies tend not to reflect accurate perceptions of what those who are pursuing pSTEM careers are like.
General Discussion
Across four studies, students’ discrepancy between their concept of themselves and their concept of a typical scientist was significantly related to their interest in a science career—the greater the self–scientist discrepancy, the lower the student’s reported interest in pursuing a career in science. This was particularly the case with respect to the scientific dimension of these self and scientist perceptions—including attributes such as being logical, intelligent, and meticulous. The scientific discrepancy was consistently predictive of lower science interest over and above discrepancies on the communal and agentic dimensions. These results suggest that one reason that students choose not to pursue pSTEM could be that they do not see themselves as “scientific enough”—not as intelligent, logical, or meticulous as they perceive the typical scientist to be. This discrepancy may push students away from pSTEM and toward fields where the self–other discrepancy is not as great.
It is important to note that given the correlational nature of the studies presented here, we cannot make any argument about the causal direction of the relationships seen. This applies both to the nature of the discrepancy itself (whether it is driven more by self or other perceptions) and to the direction of the relationship between discrepancy and interest—while we posit that discrepancy might arise first and that students use the perceived discrepancy to determine their level of interest, it could also be the case that students start with a given level of interest and use that interest to guide their perceptions of the self and others, resulting in a greater or smaller discrepancy.
The stereotype that scientists lack communality—they are cold, work with things rather than people, and are not helpful to others—has been an important focus of interventions that attempt to increase women’s interest in STEM majors and careers. For instance, Diekman and colleagues (2011) have used an intervention that specifically focuses on scientists performing communal behaviors such as helping people and working together. Our results show that while this is not an inappropriate choice—communal discrepancies were often significantly related to lower interest in addition to scientific discrepancies—there is perhaps another aspect of the stereotypic perceptions students have of scientists that has been overlooked and underutilized in these types of interventions: the scientific dimension. Students perceive scientists as hyperbolically “scientific.” These stereotypic perceptions, then, are markedly discrepant from students’ more modest perceptions of themselves, producing a substantial discrepancy between self and scientist which, as we have seen, relates to decreased interest in science. This means that interventions to increase science interest by decreasing the discrepancy between self and scientist could focus on the scientific trait dimension in two possible ways: either by making perceptions of scientists on these traits less extreme (e.g., “Scientists make mistakes sometimes too”) or by boosting students’ perceptions of themselves on the same scientific traits (e.g., “I’m more analytical than I think I am”).
It could be the case that perceptions of scientists are generally consistent across participants, and that the variation that leads to self–scientist discrepancies is all due to differences in self-perceptions that may be completely veridical. In other words, students with large self–scientist discrepancies may simply be accurately perceiving themselves as not cut out for a science career, which certainly does require some amount of scientific-ness. However, Study 4 suggests that this is not the case—participants with greater self–scientist discrepancies also had less accurate perceptions of science students, and both inaccuracy and discrepancy were related to reduced science interest. It was not just that students perceived science students differently from how they perceived themselves; they also perceived these people erroneously. As such, trying to increase the accuracy of these perceptions could be a promising route to increasing students’ science interest and broadening participation in pSTEM. Again, however, causation could flow in both directions. Inaccurate perceptions of people in pSTEM may lead to reduced pSTEM interest, but a student interested in pSTEM may spend more time around others pursuing pSTEM and as a result perceive them more accurately. The students in our samples are early in their college careers (i.e., many have not yet chosen a major, and most are in their first year), and so little exposure to other pSTEM students has occurred. Still, the possibility of increasing pSTEM interest through increased exposure of less interested students to science courses and students, which in turn increases the accuracy of their perceptions, is an intriguing possibility for intervention research. As with discrepancy, we view both patterns of causation as interesting and potentially important.
In the current work, we assessed only accuracy of students’ perceptions of others and not the accuracy of their self-perceptions. It is certainly possible that inaccurate self-perceptions also contribute to larger self–scientist discrepancies. Self-perceptions are also influenced by stereotypes. For example, exposure to the stereotype that women are less rational or intelligent may influence a young woman’s concept of herself with respect to her scientific-ness to align with that stereotype, even if she is in fact quite rational and intelligent. In addition, self-perceptions are influenced by how one is perceived by others; being exposed to teachers or parents who endorse an inaccurate stereotype may also influence one’s self-concept in an inaccurate direction. Biased or inaccurate self-perceptions would obviously contribute to a larger self–other discrepancy just as we have found that biased or inaccurate perceptions of others do.
The findings that discrepancy and inaccuracy together partially mediated the gender gap in pSTEM interest in Study 4, and that women had significantly less accurate perceptions of scientists in Study 3, suggest that even though the relationships of these variables with interest do not differ by gender, women may start off at a particular disadvantage in terms of their perceptions of scientists. Women perceive scientists less accurately and more discrepantly from how they perceive themselves, partially explaining their reduced interest in science. Increasing accuracy and reducing discrepancy may help remedy the gender gap by increasing women’s interest in pSTEM.
The results of these four studies suggest several directions for future research: Further investigation could isolate the specific psychological mechanisms behind the relationship between self–scientist discrepancy and science interest. Some of these mechanisms may involve social identity—for instance, the mismatch created by a large self–scientist discrepancy leads to reduced identification with scientists and subsequently reduced interest. As social identities satisfy fundamental psychological needs (Greenaway, Cruwys, Haslam, & Jetten, 2016), the selection of a career path that “fits” one’s identity is important for personal well-being. Students who do not see science as matching their identity are likely to avoid such careers, even if the perceived mismatch is erroneously based on stereotypes. Other mechanisms may involve the relationship between self- and other-perceptions. Previous literature indicates that self-stereotyping processes, in which attributes of an ingroup are incorporated into the self-concept (e.g., Latrofa, Vaes, Cadinu, & Carnaghi, 2010; Pickett, Bonner, & Coleman, 2002), as well as self-anchoring processes whereby attributes of one’s self-perceptions are incorporated into perceptions of an ingroup (e.g., van Veelen, Otten, & Hansen, 2011), both play a role in how people perceive themselves and members of groups to which they belong (or desirable groups to which they anticipate that they will belong). It is likely that both these processes influence the formation of self–scientist discrepancy and its relationship with science interest. Furthermore, understanding the developmental trajectory that these perceptions of the self and of others take in the process of producing career interest and aspirations (e.g., Bian, Leslie, & Cimpian, 2017; Lauermann, Tsai, & Eccles, 2017) will also be a key piece in developing our understanding of the mechanisms behind these phenomena.
Although these studies demonstrate that self–scientist discrepancies are related to inaccurate perceptions of scientists, the issue of where these discrepancies come from remains to be explored—there are certainly other factors at play besides inaccurate perceptions of scientists. These could include developmental, sociocultural, and economic variables that work together to shape both perceptions of scientists and the self-concept. Future research could try to disentangle some of these additional questions about the origins of and the causal relationships between self-perceptions, scientist perceptions, and pSTEM interest using a longitudinal design, and beginning when students are much younger, ideally grade school. It is worth noting that the primary outcome (expressed interest in a science career) assesses a behavioral intention rather than a behavior per se. Still, in Studies 1 to 3 the pSTEM status of participants’ chosen academic major—which is a behavior—was also used as an outcome variable. The fact that the effects seen with expressed interest as the outcome largely corresponded to the effects seen with pSTEM major status as the outcome suggests that students’ behavioral intentions are related to their actual behavior. One avenue for future research could be to assess self–scientist discrepancy in the context of other behavioral outcomes, such as the number of pSTEM courses a student chooses to take or college graduates’ actual career choices. For now, we view the present studies as an important contribution to the growing body of work aimed at understanding how social perception processes shape who pursues a pSTEM career path.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by National Science Foundation Grant NSF1251590 awarded to Tiffany A. Ito and NSF1551099 awarded to Bernadette Park.
Notes
Supplemental Material
Supplementary material is available online with this article.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
