Abstract
Grounded in expectancy-value and stereotype threat theories, this four-year longitudinal study examined associations between changes in stereotype threat and motivation (self-efficacy, task values, and perceived costs) among 425 undergraduates from racial/ethnic groups typically underrepresented in science, technology, engineering, and mathematics (STEM). Growth analyses indicated that students’ stereotype threat and perceived cost of studying science increased during college, whereas science self-efficacy, intrinsic value, and attainment value declined. Parallel growth analyses suggested that higher initial stereotype threat related to a faster decline in attainment value and faster increase in perceived costs throughout college. Higher initial levels and a steeper increase in stereotype threat related to lower STEM GPA. Higher initial levels and a slower decline in motivation variables related to higher STEM GPA and more completed STEM courses. These findings provide empirical evidence for the relations between stereotype threat and motivation among underrepresented minority students during a key developmental period.
Keywords: stereotype threat, expectancy-value theory, underrepresented minorities, motivation, STEM persistence
Students in science, technology, engineering, and mathematics (STEM) disciplines report declines in achievement motivation over time, and such declines are negatively associated with performance and persistence in STEM (e.g., Eccles et al., 1989; Robinson et al., 2019a; Watt, 2004; Wigfield & Eccles, 1992). However, this prior research has primarily considered majority European-American samples (e.g., Kosovich et al., 2017; Robinson et al., 2019a; Watt, 2004) and there is a lack of research with samples of students who identify with racial and ethnic groups traditionally underrepresented in STEM. Due to the persistent opportunity gaps resulting from inequitable social and educational practices, students from ethnic underrepresented minoritized groups (URM) continue to report lower achievement and persistence in STEM compared to their ethnic majority counterparts (e.g., National Science Foundation, 2018; Riegle-Crumb et al., 2019). In fact, research indicates that compared to 58% of European-American students, only 34% of African-American and 43% of Latinx students who declare a STEM major actually persist in earning their STEM degrees (Riegle-Crumb et al., 2019). Moreover, African-American and Latinx students exhibit higher rates of switching from STEM to a different major (40% and 37% respectively) compared to White students (29%; Riegle-Crumb et al., 2019).
A growing body of literature has focused on stereotype threat – anxiety related to confirming the negative stereotypes about one’s group (e.g. one’s ethnic group or gender) – as a key factor for understanding opportunity gaps among different racial and ethnic groups (e.g., Fischer, 2010; Steele & Aronson, 1995; Spencer et al., 2016; Woodcock et al., 2012). Numerous lab-based experimental and cross-sectional naturalistic studies have indicated that stereotype threat negatively relates to URM students’ achievement motivation, enrollment choices, grades, and persistence intentions (e.g., Corra & Lovaglia, 2012; Forbes & Schmader, 2010; Thoman et al., 2013; Woodcock et al., 2012). However, research is lacking on how perceptions of stereotype threat develop in authentic STEM settings and shape URM students’ motivation trajectories, achievement, and persistence over the course of their college careers. Given that both stereotype threat and motivation are highly contextual and dynamic (Kaplan et al., 2018; Steele, 1997; Murphy & Taylor, 2012; Wigfield & Cambria, 2010), it is important to examine these constructs and key outcomes longitudinally in the contexts in which they develop.
Accordingly, we assessed URM students’ perceptions of ethnic stereotype threat and motivation in STEM settings across four years of college. We grounded our study in stereotype threat theory (Steele & Aronson, 1995) and situated expectancy-value theory (Eccles et el., 1983; Eccles & Wigfield, 2020) to understand how perceptions of ethnic stereotype threat and motivation (academic self-efficacy, values, and perceived costs) develop together and relate to one another. Additionally, we examined whether ethnic stereotype threat and motivational trajectories over four years of college were related to STEM achievement and course completion at the end of college.
An Overview of Situated Expectancy-Value Theory
Eccles and colleague’s situated expectancy-value framework (Eccles et al., 1983; Eccles & Wigfield, 2020) provides important insights into the persistence and achievement behaviors of students in a variety of academic domains (e.g., Anderson & Ward, 2014; Durik et al., 2006; Robinson et al., 2019a; Wang et al., 2015; Wigfield et al., 2008). This theory posits that students’ success expectancies and perceptions of ability on a task are central in determining the types of tasks with which students engage, the quality of their engagement with the task, and whether or njot they persist in the task. Moreover, individuals tend to choose and persist in tasks that they enjoy (intrinsic value), view as useful (utility value), and see as connected to their identity (attainment value), and avoid tasks they perceive as costly (cost; Eccles et al., 1983; Eccles & Wigfield, 2002; Wigfield & Eccles, 1992, 2000).
Research with students of various age groups demonstrates that success expectancies and ability beliefs are strong predictors of achievement outcomes, whereas value beliefs are more strongly associated with enrollment and persistence outcomes (e.g., Bong, 2001; Denissen et al., 2007; Eccles et al., 1983; Perez et al., 2014; Stevens et al., 2007; Wigfield et al., 2016). In STEM disciplines in particular, studies with middle school and college students suggest that low competence beliefs in math and science are associated with lower grades in these disciplines (Durik et al., 2006; Eccles et al., 1989; Marsh et al., 2013; Perez et al., 2014). On the other hand, values and cost perceptions for math and science are predictive of students’ aspirations to pursue STEM careers and persistence intentions in STEM among middle school students (e.g., Joyce & Farenga, 2000), young adolescents (Meece et al., 1990), and undergraduates (Perez et al., 2014). Although most of this research has been conducted with majority European-American samples, the few studies with a focus on URM students have yielded similar results. For instance, African-American and Latinx ninth-graders who reported higher science attainment, intrinsic, and utility value reported higher intentions to persist in STEM fields (Andersen & Ward, 2014). Considering the important role of expectancy and task value beliefs in students’ STEM achievement and persistence, understanding how these beliefs develop in authentic STEM contexts and the factors that shape these motivational trajectories is essential (Eccles & Wigfield, 2020).
Trajectories of Expectancies, Values, and Costs and Their Relations to Achievement and Persistence in STEM
Following Eccles and colleagues’ conceptualization of expectancies and values (Eccles et al., 1983; Eccles & Wigfield, 2002; Wigfield & Eccles, 1992, 2000), a growing body of research has focused on understanding how these beliefs develop among students at various academic levels and in various domains. Prior longitudinal research has demonstrated that expectancy-value beliefs generally decrease over time (Robinson et al., 2019a;; Fredricks & Eccles, 2002; Jacobs et al., 2002; Kosovich et al., 2017). For example, longitudinal studies with primary and secondary school students have revealed that students’ math ability perceptions and math intrinsic and utility values decreased over 3–4 years whereas their math difficulty and effort perceptions increased (Chouinard & Roy, 2008; Jacobs et al., 2002; Watt, 2004, 2008).
Similar findings were found with undergraduates. For instance, in a semester-long study with psychology undergraduates, Kosovich and colleagues (2017) found that students started the semester with relatively high utility value and expectancies for success for their psychology class, but both beliefs significantly declined as the semester progressed. Relatedly, research by Robinson et al., (2019a) demonstrated that undergraduate engineering students’ expectancies and values declined over the first two years of college while their perceived costs increased. This research also revealed that changes in motivational trajectories were associated with STEM retention and grades.
Whereas the literature cited above suggests that expectancy-value motivational beliefs decline over time, it is less clear whether these beliefs follow a consistent downward trajectory over the course of the college career or if these beliefs change more dynamically as a function of contextual factors and demonstrate non-linear trends of development. In the current study, we assessed motivation over four time points, which made examination of non-linear change possible. Although, non-linear trajectories are best assessed with six or more waves of data (Xitao & Xiaotao, 2005), measuring URM students’ beliefs yearly over their entire four-year college career provided initial insights into the dynamic nature of motivation and stereotype threat.
Importantly, whereas existing research sheds light on how expectancy-value motivational beliefs develop over time and relate to students’ achievement and persistence, most studies have included student samples that are a majority European- and Asian-American, two ethnic groups that are well represented in STEM fields. Studies comparing ethnic minority and majority students on expectancy-value motivational constructs suggest that URM and non-URM students demonstrate differential levels of expectancies and values (, Robinson et al., 2019a; Robinson et al., 2019b; Robinson, Perez, Nuttall, Roseth, & Linnenbrink-Garcia, 2018; Rosenzweig & Wigfield, 2017). For instance, Rosenzweig and Wigfield (2017) reported that African-American middle school students, compared to their European-American peers, had less adaptive patterns of expectancy-value beliefs. Moreover, a study with STEM undergraduates found that URMs, compared to non-URMs, were less likely to belong to a motivational profile with high expectancies and values and low cost (Robinson et al., 2019b). In contrast, other research indicated that STEM undergraduate URMs compared to non-URMs started their first year with higher expectancy beliefs and interest value and lower cost perceptions with no significant differences between the two groups in the rates of change in these beliefs (Robinson et al., 2019a). These differences between URM and non-URMs on expectancy-value motivation beliefs (Robinson et al., 2019a; Robinson et al., 2019b; Rosenzweig & Wigfield, 2017), the findings illustrating the relations between these beliefs with students’ STEM persistence and achievement (e.g., Eccles et al., 1989; Marsh et al., 2013; Perez et al., 2014), and the underrepresentation of URM students in STEM (National Science Foundation, 2018) all highlight the importance of understanding URM students’ motivational trajectories in STEM and the factors that influence these trajectories.
Cultural Stereotypes and their Role in the Development of Expectancies and Values
A key hypothesis in the situated expectancy-value model is that expectations for success and values are shaped within individuals’ social milieu (Eccles & Wigfield, 2020). Eccles and her colleagues (Eccles et al., 1990; Eccles & Wigfield, 2002) argue that individuals develop success expectancies and ability beliefs based on expectations from important socializers such as parents, peers, and instructors. In turn, socializers’ opinions about individuals’ competencies are assumed to be shaped by the cultural stereotypes that are prevalent about persons of different social groups. In academic environments generally, and in STEM settings particularly, these cultural stereotypes target academic competence of women and ethnic/racial URM students. Thus, important socializing agents may form negative stereotype-based evaluations about female and racial URM students (e.g., women are not as good at math as men; or Black students are underachievers) and convey these stereotypes to students in STEM academic environments. Being exposed to these stereotyped-based evaluations can in turn negatively impact the development of these students’ ability beliefs, diminish their expectations for success, and eventually hurt their achievement and persistence (Wigfield & Eccles, 2002).
Additionally, being a member of a racial or ethnic minoritized group and the target of cultural stereotypes has implications for one’s development of values. When stereotypes are salient in academic domains, persons of stereotyped groups may fear that their academic performance, if poor, could confirm the competence-deficit stereotypes that exist about the people of their group (Steele & Aronson, 1995; Steele, 1997). Such stereotype threat has deleterious effects on students’ motivation and performance particularly for those whose identities and self-worth are closely tied to success in a particular academic domain (Aronson et al., 1999; Major & Schmader, 1998; Steele, 1997; Steele et al., 2002). For these students, the damage caused by potential academic failure is so overwhelming that to protect their self-esteem, students may gradually disidentify from the academic domain (e.g., Spencer et al., 2016; Steele, 1997). Such disidentification eventually leads to devaluing achievement in the domain, increasing the likelihood of withdrawal from the academic domain (Major & Schmader, 1998; Schmader et al., 2001; Steele, 1997; Woodcock et al., 2012). Thus, differences in motivation between students who identify with ethnic minoritized groups and those who identify with majority groups may be reflective of racial and ethnic stereotypes experienced by students from underrepresented groups in STEM.
Empirical studies provide evidence for the negative relations between stereotype threat and expectancy-value beliefs. For instance, a few experimental studies demonstrated that URM college students for whom ethnic stereotype threat was activated had significantly lower performance expectancies on academic tests and performed more poorly on those tests compared to URM students for whom stereotype threat was not activated (Cadinu et al., 2003; Steele & Aronson, 1995). These findings align with results of correlational research with URM and female students in a variety of educational levels and domains including STEM. These studies suggest that students who experienced race or gender identity threats in academic environments had lower expectancies to achieve favorable educational outcomes, lower academic ability self-concept, lower values for the domain, and they also performed more poorly academically and demonstrated reduced intentions to remain in their domains (Eccles et al., 2006; Irving & Hudley, 2005; Plante et al., 2013; Smith et al., 2015). Whereas this literature sheds light on the associations between stereotype threat and some expectancy-value beliefs, perceptions of cost have been left out of these investigations. Stereotype threat is found to be directly related to lower effort investment (Keller, 2002; Stone, 2002). Thus, it is likely that students who experience more stereotype threat find investing time and effort in stereotyped tasks costlier.
Further, there is a lack of research on how stereotype threat and expectancy-value beliefs develop together in academic contexts. A three-year study with URM students revealed that stereotype threat had negative effects on identification with science (similar to attainment value), which in turn negatively related to pursuing a scientific career (Woodcock et al., 2012). Although longitudinal in nature, this study did not provide understandings of how these constructs developed over time. URM students’ perceived stereotype threat could change over time due to the various stereotypic experiences that they may face during their academic career (Aronson & Good, 2002; Steele, 1997). Nevertheless, research examining longitudinal changes in stereotype threat is scarce. We are aware of two studies that examined changes in constructs closely tied to stereotype threat. In both studies, one examining URM undergraduates’ self-reported racial discrimination (Del Toro & Hughes, 2020) and the other investigating STEM students’ self-reported stereotype bias (Cromley et al., 2013), the researchers reported that these perceptions increased over time. In one study, this increase predicted lower likelihood of on-time graduation and lower overall grades (Del Toro & Hughes, 2020); whereas in the other study, the relations between the growth of stereotype threat and retention in STEM yielded inconsistent results across time (Cromley et al., 2013).
Every year, students attend a variety of classes where they interact with their professors and peers, take tests, and receive achievement feedback. All of these factors can potentially expose URM students to experiences of ethnic discrimination and stereotyping, which over time could have additive effects on students’ perceptions of ethnic stereotype threat (Steele, 1997). Motivational beliefs are highly contextual (Eccles & Wigfield, 2020; Kaplan et al., 2018) and thus can dynamically fluctuate as students’ exposures to discrimination and stereotype threat experiences accumulate. Therefore, it is important that students’ motivational beliefs be examined in parallel with their self-reported experiences of stereotype threat over time and in context.
Current Study
Results from existing research has demonstrated the effects of stereotype threat on some expectancy-value beliefs (e.g., performance expectancies and values for the domain; Cadinu et al., 2003; Smith et al., 2015); however, more research is needed to understand how these beliefs develop over time in URM students and in authentic STEM settings. Importantly, in this study we investigated the associations between the growth of motivation and ethnic stereotype threat with STEM students from underrepresented ethnic groups across four years of college. First, we examined how URM students’ ethnic stereotype threat and expectancy-value beliefs (i.e., academic self-efficacy, task values, and perceived costs) changed over time. We assessed students’ academic self-efficacy as a proxy for their success expectancies in STEM. Although there is little research on longitudinal changes in ethnic stereotype threat with students who identify with ethnic minoritized groups, based on the two studies that indicated students’ perceptions of discrimination and bias increase over time (Cromley et al., 2013; Del Toro & Hughes, 2020), we expected that URM students’ self-reported ethnic stereotype threat would increase throughout college. Also, based on the prior research among diverse, although majority European- and Asian-American samples (e.g., Jacobs et al., 2002; Kosovich et al., 2017; Robinson et al., 2019a), we expected that academic self-efficacy and task values would decrease but cost perceptions would increase throughout college for our URM sample.
Second, we examined how each motivational construct developed in parallel with stereotype threat. Building on prior studies demonstrating the effects of stereotype threat on motivational beliefs (Cadinu et al., 2003; Plante et al., 2013; Smith et al., 2015; Woodcock et al., 2012), we hypothesized that initial levels of ethnic stereotype threat would be negatively associated with initial levels of self-efficacy and values and positively associated with the initial levels of perceived costs. Due to the dearth of prior longitudinal research examining the associations between the growth of stereotype threat and growth of expectancy-value beliefs, we did not make a firm hypothesis about those relations. However, we expected that if associations were found, they would be in the direction that is suggested by the existing experimental and correlational research (Cadinu et al., 2003; Plante et al., 2013; Smith et al., 2015; Woodcock et al., 2012). That is, we expected that a faster increase in ethnic stereotype threat would be associated with faster decreases in self-efficacy and values and faster increases in perceived costs.
Third, we examined whether trajectories of URM students’ ethnic stereotype threat and motivational beliefs were associated with STEM grade point average (GPA) and STEM course completion. STEM GPA provides an indicator of students’ achievement in their STEM courses, which has real-world consequences (e.g., for graduation and admissions to graduate school). STEM course completion provides a behavioral indicator of students’ STEM choices (students’ decisions to take STEM courses) and persistence (students’ successful completion of STEM courses). Based on existing evidence to suggest that stereotype threat predicts academic outcomes among students in various disciplines (e.g., Gonzales et al., 2002; Nguyen & Ryan, 2008; Plante et al., 2013; Steele & Aronson, 1995; Woodcock et al., 2012), we hypothesized that higher levels of initial ethnic stereotype threat and increases in ethnic stereotype threat over time would be associated with lower graduating STEM GPA and fewer completed STEM courses. Based on prior empirical work (e.g., Kosovich et al., 2017; Robinson et al., 2019a; Watt 2004, 2008), we expected that lower initial levels and sharper declines in academic self-efficacy and values would be related to lower STEM GPA and fewer completed STEM courses; these relations would be the opposite for the initial levels and growth of perceived costs (higher levels and sharper increase of perceived cost would be associated with lower outcomes).
Method
Participants and Procedure
We collected data from four cohorts of students, following each cohort for four years, as part of a larger intervention study examining the effects of an early-college summer enrichment program on STEM motivation and persistence. For all cohorts, we started collecting data in students’ first year of college and collected follow-up surveys once each year through all four years of college. Students who received the intervention were excluded from the sample to eliminate the potential confounding effects of the intervention on motivation trajectories.1 This study was conducted at a highly selective private university in the southeastern United States. The study was approved by the Institutional Review Board at the last author’s former (IRB number: 2017–1129) and current (IRB number: STUDY00004417) institutions.
Participants were initially recruited from introductory gateway chemistry courses designed for students intending to pursue STEM majors. Mid-way through their first-year fall-semester chemistry class, participants completed a paper-pencil baseline survey. Approximately 73% of all students enrolled in the chemistry courses completed the baseline survey across all cohorts. This study included 425 URM students who completed the baseline (T1) survey. All 425 URM students were then invited to take the follow-up surveys in the next three years (see missing data analyses below for information about response rates and patterns of missingness). The sample included 58.4% female students with 44.7% African-American, 28.8% Latinx, 25.0% multiracial URM, and 1.4% Native American students. Participants were compensated $10 for their participation in each of the surveys.
Measures
Survey measures included a demographics questionnaire (baseline only), an ethnic stereotype threat scale, a science academic self-efficacy scale, and measures of science values and costs collected at each time point. All survey items were assessed using a 5-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree). We created composite scores for each of the variables to include in our latent growth curve analyses (see below) by calculating the average of the relevant items for each variable at each time-point. All items are presented in the Appendix. Finally, we calculated students’ STEM GPA and the number of completed STEM courses at graduation using data collected from institutional records.
Ethnic Stereotype Threat
Students responded to eight items about ethnic stereotype threat, which were developed by Steele, James, and Barnett (2002; e.g., “If you ask a simple question, people will think it is because of your ethnicity”). The internal consistency coefficients were high across four time points (αs = .91 - .94).
Academic Self-Efficacy
Students responded to five items about their academic self-efficacy in science. Items were adapted from Midgley et al. (2000; e.g., “Even if the work in science is hard, I can learn it”). The internal consistency coefficients were high across four time points (αs = .86 - .93).
Task Values
Students responded to items about their perceptions of value associated with engaging in science tasks using items adapted from Conley (2012). Three different types of science task value were measured including intrinsic value (5 items, e.g., “I enjoy doing science”), attainment value (4 items, e.g., “Being good in science is an important part of who I am”), and utility value (3 items, e.g., “Science will be useful for me later in life”), and their internal consistency coefficients were high across four time points (intrinsic: αs = .92 - .94; attainment: αs = .80 - .85; utility: αs = .83 - .86).
Perceived Effort Cost
Students responded to three items adapted from Perez et al. (2014), which assessed their perceptions of cost associated with the time and effort required in science (effort cost; e.g., “Studying science requires more effort than I’m willing to put in”).2 The internal consistency coefficients were acceptable across four time points (αs = .73 - .79).
STEM Outcomes
Students included in this study are from racial and ethnic groups that are underrepresented across STEM disciplines and we were, therefore, interested in participants’ achievement and persistence in STEM broadly. Further, science motivation, particularly for students in gateway science courses, has important implications for students’ achievement and persistence across STEM disciplines (Dai & Cromley, 2014). Thus, we used students’ cumulative GPA in all graded STEM-related courses during their four years of college, obtained from institutional records, as an indicator of students’ academic achievement in STEM. Students’ number of completed STEM courses was also obtained from institutional records and used as an indicator of STEM persistence.
Analytic Plan
Descriptive statistics and correlations were performed using SPSS version 26; confirmatory factor analyses and latent growth curve analyses were conducted using Mplus version 7.3 (Muthén & Muthén, 1998–2014). Model fit was determined via the comparative fit index (CFI; values ≥ .90), Tucker-Lewis Index (TLI; values >.90), standardized root mean squared residuals (SRMR; values <.08), and the root mean square error of approximation (RMSEA < .08; Hu & Bentler, 1999). For growth analyses, a combination of CFI, TLI, and RMSEA were employed as these fit indices have demonstrated good potential in evaluating the fit of growth curve models (Wu et al., 2009). For confirmatory factor analyses, a combination of CFI, TLI, and SRMR were used as SRMR demonstrates higher power compared to RMSEA in rejecting the non-close fit models particularly when the sample size is not very large (Shi et al., 2020). When comparing two models to see if the constructs are invariant across time, we used changes in CFI less than or equal to .01 as criterion for retaining the more parsimonious model (Chen, 2007; Cheung & Rensvold, 2002). For selecting the model that best captures the growth in constructs over time, we used the χ2 goodness of fit difference test (Muthén & Khoo, 1998).
Latent Growth Curve Models
We tested a series of latent growth curve models using the maximum-likelihood (ML) estimator for each construct including stereotype threat, self-efficacy, values, and effort cost. To examine the nature of change in each construct over time, we first tested unconditional intercept-only (non-growth) followed by linear growth and quadratic growth models. We set the linear slope parameters to 0, 1.3, 2.3, and 3.3 to match the time intervals between assessments. We squared these for the quadratic growth parameters (0, 1.69, 5.29, and 10.89). Next, to examine whether growth in stereotype threat was associated with growth in the other motivational constructs, we tested four parallel process models (Byrne, 2012; Muthén & Muthén, 1998–2017). Each model included stereotype threat with one of the four motivation variables since a single model examining growth trajectories with all variables together would be too complex. Finally, in each parallel process model we added the two outcomes (STEM GPA and number of completed STEM courses) to examine associations between the growth parameters and the outcomes.
Results
Preliminary Analyses
Descriptive Statistics and Correlations
Bivariate correlations and descriptive statistics for study variables are presented in Table 1. Examination of the means indicated that intrinsic, attainment, and utility value as well as academic self-efficacy were relatively high in the first year (Ms = 3.94 – 4.42) but appeared to slightly decrease over time. Academic self-efficacy appeared to decrease in the first three years, then increase slightly. Effort cost and ethnic stereotype threat on the other hand, were low to medium at the first year (Ms = 2.11 and 2.67, respectively), but seemed to slightly increase from the first year to the second year. Both beliefs seemed to stabilize after the second year.
Table 1.
Descriptive Statistics and Pearson Correlations
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
| ||||||||||||
| 1. Stereotype threat T1 | — | |||||||||||
| 2. Stereotype threat T2 | .59** | — | ||||||||||
| 3. Stereotype threat T3 | .60** | .77** | — | |||||||||
| 4. Stereotype threat T4 | .46** | .68** | .68** | — | ||||||||
| 5. Academic self-efficacy T1 | .002 | −.01 | −.14† | −.002 | — | |||||||
| 6. Academic self-efficacy T2 | .01 | −.06 | −.07 | .04 | .45** | — | ||||||
| 7. Academic self-efficacy T3 | −.14 | −.19* | −.10 | .01 | .42** | .65** | — | |||||
| 8. Academic self-efficacy T4 | −.002 | −.02 | −.16† | −.02 | .34** | .61** | .60** | — | ||||
| 9. Intrinsic value T1 | −.06 | −.05 | −.20* | −.02 | .40** | .26** | .20* | .25** | — | |||
| 10. Intrinsic value T2 | −.09 | −.19* | −.22* | −.20* | .17* | .45** | .31** | .24* | .64** | — | ||
| 11. Intrinsic value T3 | −.24** | −.15 | −.23** | −.13 | .16* | .32** | .36** | .24* | .58** | .74** | — | |
| 12. Intrinsic value T4 | −.03 | −.17† | −.28** | −.08 | .21** | .35** | .28** | .56** | .52** | .70** | .72** | — |
| 13. Attainment value T1 | −.003 | .006 | −.08 | −.07 | .41** | .28** | .16* | .22** | .48** | .39** | .43** | .38** |
| 14. Attainment value T2 | −.07 | −.05 | −.09 | −.13 | .18* | .35** | .17† | .26** | .46** | .64** | .61** | .55** |
| 15. Attainment value T3 | −.17* | −.10 | −.14† | −.12 | .18* | .24* | .42** | .26** | .50** | .56** | .72** | .62** |
| 16. Attainment value T4 | −.05 | −.20* | −.20* | −.13 | .14† | .30** | .26** | .48** | .46** | .59** | .64** | .77** |
| 17. Utility value T1 | .05 | .09 | −.02 | .04 | .29** | .14† | .17* | .28** | .47** | .34** | .36** | .36** |
| 18. Utility value T2 | .004 | −.10 | −.11 | −.12 | .20** | .43** | .27** | .30** | .46** | .66** | .54** | .53** |
| 19. Utility value T3 | −.09 | −.20* | −.19* | −.09 | .13† | .27** | .39** | .45** | .43** | .58** | .72** | .64** |
| 20. Utility value T4 | .05 | −.05 | −.19† | −.05 | .11 | .39** | .30** | .57** | .29** | .53** | .51** | .72** |
| 21. Effort cost T1 | .08 | .08 | .19* | −.10 | −.37** | −.31** | −.35** | −.21** | −.43** | −.35** | −.25** | −.25** |
| 22. Effort cost T2 | 13† | .24** | .11 | .05 | −.12 | −.29** | −.29** | −.07 | −.21** | −.38** | −.30** | −.19† |
| 23. Effort cost T3 | 13† | .27** | .21** | .19* | −.09 | −.25** | −.36** | −.35** | −.19* | −.40** | −.41** | −.51** |
| 24. Effort cost T4 | .16* | .17† | .23* | .05 | −.05 | −.31** | −.19* | −.22** | −.16* | −.41** | −.33** | −.38** |
| 25. STEM GPA | −23** | −.21** | −.33** | −.26** | .02 | .20** | .22** | 15† | .13* | .24** | .25** | .19* |
| 26. STEM Courses | −.12* | −.16* | −.20** | −.08 | .11* | .23** | .20* | .26** | .27** | .33** | .46** | .43** |
|
| ||||||||||||
| M | 2.67 | 2.98 | 2.99 | 3.01 | 4.09 | 3.90 | 3.81 | 3.90 | 4.15 | 4.08 | 4.12 | 4.08 |
| SD | .88 | .97 | 1.01 | .98 | .64 | .72 | .75 | .80 | .71 | .76 | .77 | .78 |
| Cronbach’s α | .91 | .93 | .94 | .93 | .86 | .91 | .88 | .93 | .92 | .93 | .94 | .94 |
| 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
| ||||||||||||||
| 13. Attainment value T1 | — | |||||||||||||
| 14. Attainment value T2 | .55** | — | ||||||||||||
| 15. Attainment value T3 | .56** | .67** | — | |||||||||||
| 16. Attainment value T4 | .54** | .69** | .81** | — | ||||||||||
| 17. Utility value T1 | .50** | .19** | .36** | .33** | — | |||||||||
| 18. Utility value T2 | .46** | .69** | .53** | .60** | .38** | — | ||||||||
| 19. Utility value T3 | .38** | .54** | .67** | .69** | .44** | .68** | — | |||||||
| 20. Utility value T4 | .33** | .40** | .51** | .71** | .32** | .58** | .75** | — | ||||||
| 21. Effort cost T1 | −.23** | −.24** | −.29** | −.22** | −.46** | −.42** | −.30** | −.18* | — | |||||
| 22. Effort cost T2 | −.14† | −.25** | −.26** | −.20* | −.22** | −.44** | −.41** | −.26* | .49** | — | ||||
| 23. Effort cost T3 | −.10 | −.20* | −.34** | −.41** | −.24** | −.34** | −.56** | −.49** | .35** | .66** | — | |||
| 24. Effort cost T4 | −.08 | −.24* | −.25** | −.29** | −.14† | −.39** | −.39** | −.43** | .34** | .48** | .59** | — | ||
| 25. STEM GPA | .09 | .18* | .13 | .18* | .04 | .17* | .24** | .22** | −.07 | −.26** | −.26** | −.30** | — | |
| 26. STEM Courses | .25** | .47** | .37** | .51** | .10 | .41** | .43** | .31** | −.18** | −.24** | −.22** | −.30** | .45** | — |
|
| ||||||||||||||
| M | 3.94 | 3.83 | 3.75 | 3.76 | 4.42 | 4.24 | 4.12 | 4.18 | 2.11 | 2.43 | 2.49 | 2.45 | 2.91 | 16.27 |
| SD | .69 | .79 | .82 | .89 | .61 | .71 | .88 | .77 | .83 | .99 | .97 | .98 | .62 | 7.48 |
| Cronbach’s α | .80 | .84 | .85 | .85 | .83 | .84 | .86 | .85 | .73 | .77 | .78 | .79 | — | — |
Note. T1 = Time 1, T2 = Time 2, T3 = Time 3, T4 = Time 4, STEM Courses = number of completed STEM courses. STEM GPA was measured on a 4-point scale. Number of completed STEM courses ranged from 2 to 33.
p ≤ .10
p < .05
p < .001
Missing Data Analyses
We used full information maximum likelihood (FIML) estimation to handle missing data (Davey & Savla, 2010). Data were collected from 425 URM students who completed the baseline (T1) survey.3 All 425 students who had completed Time 1 surveys were invited to participate in each follow-up survey regardless of their participation in prior surveys.4 Of the original baseline sample of 425 students, 44% completed the second year (T2) survey, 39% completed the third year (T3) survey, and 41% completed the fourth year (T4) survey. Of the 425 students included in this study, 62.8% completed surveys on at least two time-points, 39.3% completed surveys on at least three time-points, and 20.4% completed surveys at all four time-points. For those who completed surveys, item-level missing data was negligible at each wave of data collection (range from 2.4% - 7.0%, across items and time-points). When calculating composite mean scores for stereotype threat and motivational variables, we only included participants who completed at least 2/3 of the items for each scale. Implementing this approach did not further eliminate any participants as the students who had completed each survey rarely missed any individual items.
We used Little’s MCAR test (Little, 1988) to examine the missing data mechanism. A significant χ2 result would provide evidence that missingness did not occur completely at random; however, results of Little’s MCAR test were not significant χ2 (463) = 457.30, p = .566. We conducted further analyses to explore whether missingness could be accounted for by other variables in the study. To assess whether participants with missing data were comparable to those with complete data, we compared the two groups on key study variables at Time 1, outcomes, and demographic variables. Results of a multivariate analysis of variance (MANOVA) indicated that participants with any missing data at Time 2, Time 3, or Time 4 did not significantly differ from participants with complete data on Time 1 intrinsic value, attainment value, utility value, effort cost, self-efficacy, and stereotype threat Wilks’ λ (6, 317) = .98, p = .223, η2= .03. Moreover, there were no significant differences between the two groups on STEM achievement and number of completed STEM courses at the end of the semester: Wilks’ λ (2, 360) = .99, p = .188, η2= .01. However, the two groups differed significantly with regard to gender. Those with missing data were more likely to be male χ2 (1) = 9.10, p = .003. Therefore, we included gender as a missing data correlate in our growth models using the auxiliary (m) command in Mplus (Asparouhov & Muthén, 2008; Graham, 2003). Since we identified a correlate of missingness, we assumed data were missing at random rather than missing completely at random, which meets the assumption of FIML estimation (Enders, 2010). Further, including missing data correlates in the model reduces bias in FIML estimation (Enders, 2008; Graham, 2003).
Confirmatory Factor Analyses
To ensure that the expectancy-value variables (self-efficacy, intrinsic value, attainment value, utility value, and cost) are perceived as distinct constructs by participants, we performed confirmatory factor analyses (CFA) at each time-point, including the latent variables of all of the expectancy-value constructs in the same model. Results indicated that the data fit the model well at T1, χ2 (179) = 459.55, p<.001, CFI = .94, TLI = .94, SRMR = .05; at T2 χ2 (179) = 390.97, CFI = .92, TLI = .90, SRMR = .06; at T3, χ2 (179) = 309.28, CFI = .95, TLI = .94, SRMR = .06; and at T4, χ2 (179) = 329.52, CFI = .95, TLI = .94, SRMR = .05.
Additionally, we conducted CFAs on the ethnic stereotype threat measure at each time point to assess the construct validity of this measure. Results indicated poor fit for stereotype threat at all four time-points (CFIs < .84, TLIs <.77, and SRMRs > .06). An examination of the 8-item stereotype threat measure revealed that different pairs of items tapped into common aspects of stereotype threat. For example, two items highlighted the threat to students’ ability (“Some people believe that you have less ability,” and “If you are not better than average, people will assume that you are limited”). Two items tapped students’ concerns about how professors view and treat them (“Professors expect you to do poorly,” and “Professors are less likely to encourage you”). Moreover, another pair highlighted students’ concerns about how their academic performance will be viewed related to their ethnicity (“If you ask a simple question, people will think it is because of your ethnicity,” and “If you do poorly on a test, people will assume that it is because of your ethnicity”). Modification indices aligned with the substantive interpretation of the similarity of these items, suggesting that the residual variances for these pairs of items should be correlated. Correlating the residuals for these three pairs at all 4 time-points resulted in acceptable fit at T1: χ2 (17) = 110.97, CFI = .96, TLI = .93, and SRMR = .05, T2: χ2 (17) = 42.55, CFI = .98, TLI = .96, SRMR = .04, T3: χ2 (17) = 78.32, CFI = .94, TLI = .01, SRMR = .05, and T4: χ2 (17) = 51.45, CFI = .97, TLI = .95, and SRMR = .04.
Longitudinal Confirmatory Factor Analyses
We conducted longitudinal confirmatory factor analyses on stereotype threat, self-efficacy, interest value, attainment value, utility value, and effort cost to examine measurement invariance across time (Meredith, 1993; Vandenberg & Lance, 2000). For each construct, we tested configural invariance (i.e., equivalent patterns of fixed and free factor loadings), weak invariance (i.e., equivalent factor loadings), and strong invariance (i.e., equivalent item intercepts). Measurement invariance was inferred if the change in CFI was ≤ 0.01 (Chen, 2007). We were able to establish weak and strong measurement invariance for stereotype threat, academic self-efficacy, intrinsic value, attainment value, and cost, but not for utility value (see Table 2). Since longitudinal measurement invariance is an assumption of latent growth curve analysis, we were unable to examine growth in utility value. Therefore, we dropped utility value from further analyses.
Table 2.
Fit Statistics for Longitudinal Confirmatory Factor Analysis of Stereotype Threat and Motivation Variables
| Measurement Invariance | χ2 | df | CFI | ΔCFI | TLI | SRMR |
|---|---|---|---|---|---|---|
|
| ||||||
| Stereotype threat1 | ||||||
| Configurai | 822.73 | 399 | .931 | − | .914 | .066 |
| Weak | 853.82 | 420 | .929 | −.002 | .916 | .072 |
| Strong | 900.54 | 441 | .925 | −.004 | .916 | .075 |
| Academic self-efficacy | ||||||
| Configurai | 208.44 | 134 | .973 | − | .962 | .051 |
| Weak | 233.76 | 146 | .968 | −.005 | .958 | .070 |
| Strong | 261.11 | 158 | .962 | −.006 | .955 | .077 |
| Intrinsic value | ||||||
| Configurai | 193.25 | 134 | .985 | − | .979 | .043 |
| Weak | 221.32 | 146 | .981 | −.004 | .975 | .063 |
| Strong | 257.46 | 158 | .975 | −.006 | .970 | .066 |
| Attainment value | ||||||
| Configural | 117.14 | 74 | .978 | − | .964 | .048 |
| Weak | 126.89 | 83 | .978 | 0 | .968 | .060 |
| Strong | 147.46 | 92 | .972 | −.006 | .963 | .063 |
| Utility value2 | ||||||
| Configurai | 45.66 | 30 | .990 | − | .977 | .044 |
| Weak | 70.78 | 36 | .977 | −.013 | .958 | .100 |
| Effort cost | ||||||
| Configural | 25.56 | 30 | 1.00 | − | 1.01 | .040 |
| Weak | 42.87 | 36 | .991 | −.009 | .984 | .057 |
| Strong | 55.96 | 42 | .982 | −.009 | .972 | .053 |
Note. CFI = comparative fit index; TLI = Tucker_Lewis index; RMSEA = root-mean-square error of approximation. Measurement invariance was inferred if the change in CFI was ≤ 0.01 (Chen, 2007).
Configural, weak, and strong measurement invariance were established after correlating three sets of item residual variances.
We were not able to establish measurement invariance for the 3-item utility value variable therefore this measure was dropped from the analysis.
Growth Trajectories of Stereotype Threat and Motivational Variables
The first aim of the study involved examining the change in ethnic stereotype threat and the motivational variables over four years. Using the composite observed variables at each time point, we fit three latent growth curve models including intercept-only, linear slope, and quadratic slope models.
The intercept-only models for all of the variables, except for intrinsic value, demonstrated poor model fit, suggesting that these variables may be changing significantly over time. All linear slope models, including the linear slope model for intrinsic value, showed a statistically significant improvement over the intercept-only models. The intercepts and linear slopes for ethnic stereotype threat and all of the motivational variables were statistically significant, indicating that ethnic stereotype threat and all of the motivational variables changed, on average, over time. Next, quadratic models for ethnic stereotype threat, academic self-efficacy, and cost fit significantly better than the linear models, all with excellent fit (see Table 3 for the fit statistics). The intercept, linear, and quadratic slopes were all statistically significant for these three models. This suggests that URM students’ ethnic stereotype threat, self-efficacy, and effort cost showed varying rates of change over time; whereas, their intrinsic and attainment values tended to change linearly over time. See Figure 1 for the model-implied growth trajectories.
Table 3.
Fit Statistics for the Intercept-Only, Linear, and Quadratic Models
| Model | χ2 | df | Δχ2 | Δdf | CFI | TLI | RMSEA |
|---|---|---|---|---|---|---|---|
|
| |||||||
| Stereotype threat | |||||||
| Intercept-only | 89.89** | 11 | − | − | .75 | .86 | .13 |
| Linear | 28.40** | 8 | 61.49** | 3 | .93 | .95 | .08 |
| Quadratic | 5.29 | 4 | 23.11** | 4 | 1.00 | .99 | .03 |
| Academic self-efficacy | |||||||
| Intercept-only | 70.45** | 11 | − | − | .66 | .82 | .11 |
| Linear | 19.01** | 8 | 51.44** | 3 | .94 | .95 | .06 |
| Quadratic | 2.28 | 4 | 16.73* | 4 | 1.00 | 1.00 | .00 |
| Intrinsic value | |||||||
| Intercept-only | 40.52** | 11 | − | − | .91 | .95 | .08 |
| Linear | 15.91* | 8 | 24.61** | 3 | .98 | .98 | .05 |
| Quadratic1 | 14.32* | 7 | 1.59 | 1 | .98 | .98 | .05 |
| Attainment value | |||||||
| Intercept-only | 75.07** | 11 | − | − | .80 | .89 | .12 |
| Linear | 18.07* | 8 | 57.00** | 3 | .97 | .98 | .05 |
| Quadratic1 | 15.79* | 7 | 2.28 | 1 | .97 | .98 | .05 |
| Effort cost2 | |||||||
| Intercept-only | 100.59** | 11 | − | − | .41 | .68 | .16 |
| Linear | 42.95** | 8 | 57.64** | 3 | .77 | .83 | .11 |
| Quadratic | 5.19 | 4 | 37.76** | 4 | .99 | .99 | .03 |
Note. Final growth models are in bold. Residuals for observed variables are set to be equal across time. The number of free parameters in all intercept-only and linear models were 3 and 6, respectively. All quadratic models had 10 free parameters except for the intrinsic and attainment value quadratic models, which had 7 free parameters.
The variance for the quadratic slope was fixed to 0 to avoid the non-positive definite error caused by this parameter.
Cohort 1 is excluded from the analysis since they did not have baseline data on effort cost.
p < .05
p < .001
Figure 1. Growth Trajectories of Stereotype threat and Motivation Variables.
Note. Stereotype threat, academic self-efficacy, and effort cost changed quadratically; whereas, intrinsic and attainment values changed linearly over four years.
Table 4 displays the model parameters. On average, URM students’ science academic self-efficacy and values were moderate to high in their first semester (M latent intercepts = 3.94 to 4.15) and showed small but statistically significant decreases over time (M latent linear slopes = −0.24 to −0.04). Further, the quadratic slope for academic self-efficacy (M latent quadratic slope= 0.06), indicated that students’ beliefs decreased until the end of third year and then slightly increased in their fourth year. Ethnic stereotype threat and effort cost were low in the first semester (M latent intercept = 2.67 and 2.11, respectively) and significantly increased over time (M latent linear slope = 0.27 and 0.49, respectively). Further, the quadratic slopes (M quadratic slopes= −0.05 and −0.12 for ethnic stereotype threat and effort cost, respectively) indicated that both variables increased from the first year to the second year and then stabilized.
Table 4.
Unstandardized Parameters for the Growth Models
| Model | Means (M) | Variances (S2) | Correlations | |||||
|---|---|---|---|---|---|---|---|---|
|
|
|
|
||||||
| Intercept | Linear | Quadratic | Intercept | Linear | Quadratic | Intercept | Linear | |
|
| ||||||||
| Stereotype threat | ||||||||
| Intercept-only | 2.80** | − | − | .52** | − | − | − | − |
| Linear | 2.69** | .12** | − | .52** | .03** | −.02 | − | |
| Quadratic | 2.67** | .27** | −0.05** | .53** | .18 | .01 | −.48 | −.90** |
| Academic self-efficacy | ||||||||
| Intercept-only | 3.99** | − | − | .22** | − | − | − | − |
| Linear | 4.07** | −.08** | − | .18** | .02* | .21 | − | |
| Quadratic | 4.10** | −.24** | 0.06** | .21** | .15 | .01 | <.01 | −.92** |
| Intrinsic value | ||||||||
| Intercept-only | 4.12** | − | − | .33** | − | − | − | − |
| Linear | 4.15** | −.04* | − | .34** | .02** | − | −.06 | − |
| Attainment value | ||||||||
| Intercept-only | 3.87** | − | − | .32** | − | − | − | − |
| Linear | 3.94** | −.07** | − | .27** | .02** | − | .26 | − |
| Effort cost1 | ||||||||
| Intercept-only | 2.32** | − | − | .36** | − | − | − | − |
| Linear | 2.16** | .15** | − | .28** | .02 | .35 | − | |
| Quadratic | 2.11** | 49** | −.12** | .40** | .53** | .04** | .00 | −.96** |
Note. Final growth models are in bold. Residuals for observed variables are set to be equal across time. We reported unstandardized results for the growth parameter estimates and standardized results for correlations between the intercepts, and linear and quadratic slopes.
Cohort 1 is excluded from the analysis since they did not have baseline data on effort cost.
p < .05
p < .01
Although model fit improved by adding quadratic slopes and the quadratic slopes were statistically significant, it is important to be cautious when interpreting quadratic change based on only four waves of data. This is particularly the case for ethnic stereotype threat and academic self-efficacy as the linear-slope model fit was acceptable in these models. The addition of the quadratic slope may have led to overfitting of the models for these two variables. We discuss this issue further in the limitations section.
Parallel Process Growth Models
We used parallel process growth models to examine whether changes in the motivation variables were related to change in ethnic stereotype threat over time. We fit four models; each model included one of the motivation variables along with ethnic stereotype threat. Although the individual growth analyses suggested non-linear growth patterns for some of the variables, we only tested the relations among the linear growth trajectories of the variables in the parallel process models in order to minimize collinearity issues and maximize interpretability of the models.5
The model fit was good for all four parallel process models. See Table 5 for the parallel process model parameters and model fit results. Results indicated that the intercept of ethnic stereotype threat was negatively correlated with the linear slope of attainment value (r = −.28, p = .041), suggesting that higher initial levels of ethnic stereotype threat were associated with a faster decline in attainment value over four years of college. On the other hand, the intercept of ethnic stereotype threat was positively associated with the slope of effort cost (r = .38, p = .030). This correlation indicated that higher levels of ethnic stereotype threat in the first year were associated with steeper increases in perceived effort cost. The slope of ethnic stereotype threat was not associated with the intercept or slope of any of the motivational variables.
Table 5.
Parameter Estimates and Their Correlations for Parallel Process Models
| χ2 (df) | CFI | RMSEA | M | SE | Correlations | |||
|---|---|---|---|---|---|---|---|---|
| 1 | 2 | 3 | ||||||
|
| ||||||||
| Academic self-efficacy | 33.60 (26) | .98 | .03 | |||||
| 1. Stereotype threat intercept | 2.67** | .04 | ||||||
| 2. Stereotype threat linear | 0.27** | .06 | −.02 | |||||
| 3. Self-efficacy intercept | 4.10** | .03 | <.00 | .09 | ||||
| 4. Self-efficacy linear | −0.25** | .05 | −.07 | −.14 .12 | ||||
| Intrinsic value | 51.35** (27) | .96 | .05 | |||||
| 1. Stereotype threat intercept | 2.67** | .04 | ||||||
| 2. Stereotype threat linear | 0.27** | .06 | −.03 | |||||
| 3. Intrinsic intercept | 4.15** | .03 | −.12† | −.04 | ||||
| 4. Intrinsic linear | −0.04** | .02 | −.21 | .06 | −.06 | |||
| Attainment value | 43.71* (27) | .97 | .04 | |||||
| 1. Stereotype threat intercept | 2.67** | .04 | ||||||
| 2. Stereotype threat linear | 0.27** | .06 | −.03 | |||||
| 3. Attainment intercept | 3.94** | .03 | <.01 | −.09 | ||||
| 4. Attainment linear | −0.07** | .02 | −.28* | .03 .25 | ||||
| Effort Cost1 | 63.70** (26) | .91 | .07 | |||||
| 1. Stereotype threat intercept | 2.68** | .05 | ||||||
| 2. Stereotype threat linear | .28** | .06 | −.05 | |||||
| 3. Effort cost intercept | 2.11** | .05 | .17† | −.06 | ||||
| 4. Effort cost linear | .47** | .07 | .38* | −.01 .22 | ||||
Note. Intercept and slope estimates are unstandardized. Significant coefficients are in bold. The quadratic slope coefficients for stereotype threat, academic self-efficacy, and effort cost were included in the model due to improved model fit; however, we fixed the variance of these quadratic slopes to zero to avoid estimation issues. Only the relations between the linear trajectories are tested.
The number of free parameters in academic self-efficacy and effort cost models were 18 and the number of free parameters in intrinsic and attainment value models were 17.
Cohort 1 is excluded from the analysis since they did not have baseline data on effort cost.
p≤.10
p < .05
p < .001
Relations of Growth Trajectories to STEM Achievement and Persistence
To examine whether trajectories of stereotype threat and motivation predicted STEM achievement and persistence, we examined the relations among the intercepts and slopes of stereotype threat and each of the motivation variables with STEM GPA and number of completed STEM courses (see Figure 2 for a parallel process model predicting STEM outcomes). Model fit was good for all four conditional parallel process models. Results for the parallel process models with outcomes, including fit indices, are displayed in Table 6.
Figure 2.
Parallel Process Model of Motivation and Stereotype Threat and the Relations to STEM Outcomes
Note. Stereotype = Stereotype Threat. Stereotype threat and motivation measured at times 1 through time 4 were used to calculate the intercept and linear slope of the growth parameters and their covariances. Path coefficients are represented by single-headed arrows. Double-headed arrows indicate covariances. The (0, 1.3, 2.3, and 3.3) intercepts at Time1 through Time 4 represent the time intervals between each measurement. Quadratic slopes were estimated for stereotype threat, academic self-efficacy, and effort cost, but are not represented in the figure for clarity.
Table 6.
Parameters for the Parallel Process models of Stereotype Threat and Motivation Variables Predicting STEM Achievement and Course Completion
| STEM Achievement | STEM Course Completion | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Predictors | χ2 (df) | CFI | RMSEA | b | β | R 2 | b | β | R 2 |
|
| |||||||||
| Academic self-efficacy | 46.75 (34) | .98 | .03 | .20** | .15* | ||||
| 1. Stereotype threat intercept | −.22** | −.26 | −1.43* | −.14 | |||||
| 2. Stereotype threat linear | −.68 | −.19 | 2.65 | .06 | |||||
| 3. Self-efficacy intercept | .06 | .04 | 2.56 | .15 | |||||
| 4. Self-efficacy linear | 1.221† | .27 | 16.05† | .30 | |||||
| Intrinsic value | 60.48** (35) | .97 | .04 | .18** | .30** | ||||
| 1. Stereotype threat intercept | −.20** | −.23 | −.39 | −.04 | |||||
| 2. Stereotype threat linear | −.82* | −.22 | 1.18 | .03 | |||||
| 3. Intrinsic intercept | .15* | .14 | 4.56** | .35 | |||||
| 4. Intrinsic linear | .881 | .19 | 23.86** | .42 | |||||
| Attainment value | 57.50** (35) | .97 | .04 | .17** | .34** | ||||
| 1. Stereotype threat intercept | −.20** | −.23 | −.41 | −.04 | |||||
| 2. Stereotype threat linear | −.821 | −.22 | 2.48 | .06 | |||||
| 3. Attainment intercept | .06 | .05 | 3.54* | .25 | |||||
| 4. Attainment linear | .82 | .19 | 24.32** | .46 | |||||
| Effort Cost 1 | 71.22* (34) | .93 | .06 | .36* | .18* | ||||
| 1. Stereotype threat intercept | −.04 | −.04 | .08 | .01 | |||||
| 2. Stereotype threat linear | −.96* | −.30 | −1.21 | −.03 | |||||
| 3. Effort cost intercept | −.06 | −.05 | −2.85† | −.21 | |||||
| 4. Effort cost linear | −1.761 | −.49 | −14.60† | −.33 | |||||
Note. b = unstandardized coefficients; β = standardized coefficients. Significant coefficients are in bold. Significance of the relations between growth parameters and STEM outcomes was determined based on p values for the unstandardized coefficients. The quadratic slope coefficients for stereotype threat, academic self-efficacy, and effort cost were included in the model due to improved model fit; however, we fixed the variance of these quadratic slopes to zero to avoid estimation issues.
The number of free parameters in academic self-efficacy and effort cost models were 31 and the number of free parameters in intrinsic and attainment value models were 30.
Cohort 1 is excluded from the analysis since they did not have baseline data on effort cost.
p ≤ .10;
p < .05;
p < .01
STEM Achievement
With the exception of the effort cost-ethnic stereotype threat model, the ethnic stereotype threat intercepts were significantly related to STEM achievement (βs between −.26 and −.23, ps ≤ .003 in the self-efficacy, intrinsic, and attainment value models) such that URM students who had lower perceptions of ethnic stereotype threat in the first year graduated with higher STEM GPAs. Furthermore, there was a significant relationship between the intrinsic value intercept and STEM GPA (β = .14, p = .027) indicating that higher levels of intrinsic value in the first year in college was related to higher STEM GPA at the end of college, controlling for the intercept and slope of ethnic stereotype threat. Additionally, the linear slope of ethnic stereotype threat was significantly related to STEM achievement (β = −.22, p = .033 in the intrinsic value model; β = −.30, p = .020 in the effort cost model). Higher ethnic stereotype threat slopes (steeper rates of increase) were associated with lower cumulative STEM GPAs at graduation, controlling for the intercept and slope of intrinsic value and effort cost.
STEM Course Completion
We found that the intercepts of intrinsic value (β = .35, p < .001) and attainment value (β = .25, p = .024) were positively related to STEM course completion. URM students who started college with higher levels of intrinsic and attainment value completed more STEM courses before graduation, controlling for the intercept and slope of ethnic stereotype threat. Moreover, the intercept of stereotype threat negatively related to STEM course completion; however, this relation was only significant in the academic self-efficacy model (β = −.14, p = .043). URM students who had lower levels of stereotype threat early in college completed more STEM courses, controlling for the intercept and slope of academic self-efficacy. Additionally, the linear slopes of intrinsic value (β = .42, p < .001) and attainment value (β = .46, p = .001) positively related to STEM course completion, controlling for the intercept and slope of ethnic stereotype threat. Students with higher intrinsic value and attainment value slopes (i.e., slower rates of decrease) completed significantly more STEM courses before graduation.
Discussion
In this study, we investigated the growth of ethnic stereotype threat and science motivation with undergraduate students from underrepresented racial/ethnic groups in STEM over four years of college. We also examined whether changes in ethnic stereotype threat were associated with changes in science motivation. Finally, we examined whether growth trajectories of ethnic stereotype threat and motivation beliefs were related to STEM academic outcomes at graduation (STEM GPA and STEM course completion). Results indicated significant changes in ethnic stereotype threat and motivation over four years of college, relations between changes in ethnic stereotype threat and some motivation variables, and relations among the growth of ethnic stereotype threat and motivation and STEM outcomes. We discuss these findings in further detail below.
Growth of Motivational Beliefs
On average, the URM students in our sample started their college careers with relatively high levels of academic self-efficacy and values but these beliefs declined slightly over time. On the other hand, effort cost perceptions were relatively low at the beginning of college but increased over four years. These findings are consistent with studies that include mainly ethnic majority samples, which suggest that students’ STEM motivation declines over time (e.g., Kosovich et al., 2017, Robinson et al., 2019a; Watt, 2004). One explanation for these trends is that self-efficacy, values, and costs are affected by students’ achievement in STEM courses (e.g., Perez et al., 2014). It is possible that these high-achieving students experienced declines in motivation after experiencing difficult STEM courses at a highly competitive institution.
Interestingly, although intrinsic and attainment value decreased linearly over time, results suggested academic self-efficacy and effort cost changed more dynamically over time. The steady and relatively slow average decreases in intrinsic and attainment value reflect theoretical and empirical research that suggests these two values may be more connected to one’s identity and thus change more slowly in response to contextual influences (Eccles, 2009; Robinson et al., 2019a). On the other hand, academic self-efficacy demonstrated somewhat steep average decreases in the first three years and then increased more slowly from the third to the fourth year. We observed the opposite pattern with perceived effort cost such that these beliefs increased somewhat steeply from year one to year two and then stabilized after year two. This may reflect URM students who persist in the major start to feel more competent as they gain more experience in the major, accumulate mastery experiences, and get closer to graduation.
Additionally, it is common that the difficult gateway STEM courses (e.g. organic chemistry and calculus) are taken during the first two or three years of the college. Therefore, it is possible that as students approached the fourth year, putting difficult “weed-out” courses behind them, their self-efficacy improved. Indeed, making it through these difficult STEM courses may in itself boost students’ feelings of confidence in their science ability. This hypothesis is consistent with social-cognitive theory (Bandura, 1997), which highlights that previous success experiences are important sources of self-efficacy. Such mastery experiences are especially powerful when individuals overcome obstacles and succeed despite difficult challenges. In terms of effort cost, the initial steeper increase from year one to year two may reflect students’ adjusting to the workload of college-level STEM courses in the first year. However, as they completed the difficult courses (i.e., mastery experiences) and moved closer to graduation, their self-efficacy increased, and they began to perceive time and effort expenditure as less costly.
Because we assessed beliefs four times over four years, we were able to model quadratic change in the variables, which revealed differences in trajectories among various motivational beliefs. Overall, our results suggested that with some of the motivational variables, changes were more dramatic during the first year of college. These findings are in line with the literature that suggests freshman year is perhaps the most sensitive and critical time for STEM students’ motivation and persistence decisions (Hernandez et al., 2013; Perez et al., 2014). Relatedly, many interventions to sustain STEM students’ motivation and persistence target these students during the first two years of college (e.g., Cromley et al., 2017; Harackiewicz et al., 2014; Hulleman et al., 2010). Aligned with the goal of these prior studies, our findings also suggest that motivational interventions may be best implemented during the first years of college when some adaptive motivational beliefs decline more sharply. The timeliness of these interventions may be particularly fruitful for retaining students who would otherwise drop out before graduation as our findings suggest that for students who made it through the fourth year, the declines in some of the motivational beliefs gradually leveled off. The somewhat dynamic trends of change in self-efficacy and perceived cost compared to the steady trends of change in intrinsic and attainment value may signal that self-efficacy and perceived cost are more influenced by contextual experiences. Thus, it may be that self-efficacy and perceived cost are more malleable to change even in the absence of interventions, but intrinsic and attainment value do not change course unless perhaps targeted directly by a motivation-promoting intervention. Whereas the quadratic growth patterns suggest dynamic change in some of the variables, it should be noted that future research should include more waves of data to test for quadratic change (see Limitation section below). Thus, although our results suggest quadratic patterns of change over four years, we interpret these results cautiously.
Growth of Ethnic Stereotype Threat
We found that students started their college career with moderate perceptions of ethnic stereotype threat, but these beliefs increased significantly over time, particularly from the first year to the second year. The increase in self-reported ethnic stereotype threat implies that students were exposed to stereotype-threatening experiences early in college. This speculation is consistent with the theoretical proposition (Steele, 1997) that suggests exposure to identity-related threats has long-term effects on individuals’ vulnerability to stereotypes. This finding is also in line with similar existing studies that indicated students’ racial stereotype bias (Cromley et al., 2013) and self-reported racial discrimination (Del Toro & Hughes, 2020) increased over time. This study is distinct from this prior research, however, in that we model how ethnic stereotype threat changes over URM students’ entire college career.
Researchers argue that there are various minority-status-related stressors that pose challenges to academic adjustment of racial minoritized students during freshman year (e.g., Hurtado et al., 1996; Smedley et al., 1993). These stressors include racial experiences that undermine students’ academic competence and belonging to the university (Smedley et al., 1993). Additionally, the various characteristics of the university can affect the adjustment of racial minoritized students during the freshman year (e.g., Hurtado, 1992; Chickering & Reisser, 1993). For instance, college selectivity could signal high expectations from minoritized students and impose performance-related pressures on them (Hurtado, 1992). Many selective universities, including the university in which our study was conducted, are racially imbalanced in favor of European- and Asian-American students. Research indicates that URM students who attend predominantly White institutions (PWI), frequently experience racial stereotyping and discrimination which negatively affects their psychological and educational adjustment (e.g., Harper, 2015; Harwood et al., 2012). The racial climate in addition to the high expectations relayed by the selective institution may have increased students’ stereotype threat-inducing experiences during this transition year by (1) increasing minoritized students’ exposure to negative race-based judgements from the majority privileged peers and professors, and (2) increasing pressure to perform well and succeed, which could in turn heighten apprehension about performing poorly and confirming the negative stereotypes about one’s racial group.
The Relations Between Ethnic Stereotype Threat and Motivation Trajectories
The results indicated thathigher initial levels of ethnic stereotype threat were related to faster declines in attainment value and faster increases in perceived cost. Results with regards to attainment value–the centrality of a task to one’s identity–align with the literature on the effects of stereotype threat on domain identification (Steele, 1997; Osborne & Walker, 2006; Woodcock et al., 2012). According to previous studies, continuous exposure to ethnic stereotype threat results in a gradual disidentification from the stereotyped domain (Woodcock et al., 2012). This phenomenon happens when the expectation of poor performance and confirming the negative racial stereotypes cause racial minoritized students to put psychological distance between their identity and the domain and gradually devalue the domain. Similarly, our findings suggested that students who had higher self-reported ethnic stereotype threat early in college also reported higher rates of decrease in attainment value for science, which may indicate that science became less central to students’ identity for those who experienced higher levels of ethnic stereotype threat in their first semester of college.
Regarding perceived effort cost, results suggested that students with higher self-reported ethnic stereotype threat in the first semester of college found it increasingly costly to invest time and effort on academic science tasks. The finding of the relations between ethnic stereotype threat and effort cost is a novel contribution to the literature. Although there is prior research on relations between stereotype threat and values (e.g., Smith et al., 2015), there is no research, to our knowledge, on relations between stereotype threat and perceived cost. Such relations are important to understand as perceived costs are an important determinant of the overall value of a task (Eccles et al., 1983; Eccles; Eccles & Wigfield, 2020) and the overall value of a task is central to persistence in the task. This finding may be explained by existing research that suggests stereotype threat is associated with effort investment such that individuals who anticipate experiencing discrimination are more likely to withdraw effort and engage in self-handicapping behavior (Keller, 2002; Stone, 2002). Individuals may perceive that investing effort in a task is particularly costly when they expect that the outcome would nevertheless be unfavorable due to discrimination and stereotypes.
Finally, although the difficulty of a task may be perceived as a value (i.e., an indicator of task importance) when the task is identity-congruent (Oyserman & Destin, 2010), task-difficulty may be perceived as costly when the task is non-identity congruent. Thus, it is possible that students who experience more ethnic stereotype threat start to deidentify with science (as indicated by decreases in attainment value), which leads to feelings that the effort required for success in science is costlier. Although this possibility was not explicitly tested in this study, prior research supports the connections between attainment value and effort cost (Perez et al., 2019)
Overall, our findings suggest that URM students’ self-reported experience of ethnic stereotype threat early in college plays an important role in the development of their attainment value and effort cost beliefs over the next four years. This finding underscores the importance of students’ early experiences with the culture of their college and the various race-related experiences they face in the first year of college. Universities should strive to promote diversity and inclusion in STEM fields as research shows that being underrepresented in such fields leads to higher experiences with stereotype threat (Inzlicht & BenZeev, 2000; Murphy et al., 2007), which may also partially explain students’ increase in ethnic stereotype threat in this study. Diversity-related supports implemented by colleges including programs that are designed to facilitate positive race-related experiences such as forming URM-specific peer-groups, providing URM students with mentors, and engaging faculty and administrators in diversity-related and implicit bias trainings may significantly benefit URM students (e.g., Chang et al., 2016; Toretsky et al., 2018). Research that has examined the relations between positive identity-related experiences and students’ socioemotional and academic adjustment suggests that such experiences positively relate to marginalized students’ motivation, well-being, and belonging on college campuses (Mendoza-Denton et al., 2002). Promoting a welcoming climate for URM students and decreasing their experiences with racism and stereotypes during the first year of college could help sustain students’ motivation in STEM fields.
Our results did not indicate any significant associations between ethnic stereotype threat and academic self-efficacy. This result was unexpected given findings in existing theory and empirical research regarding the negative relations between stereotype threat and outcome expectancies (Cadinu et al, 2003; Eccles et al., 1990; Irving & Hudley, 2008; Steele & Aronson, 1995). In these studies, the researchers reported that students who had higher self-reported ethnic stereotype threat predicted their performance to be poorer. Studies indicate that outcome expectancies (belief about one’s performance leading to a successful outcome) and self-efficacy expectancies (belief about one’s ability to successfully perform) are independent constructs and have unique effects on behavior (Maddux et al., 1982; Shell et al., 1989). Thus, it is possible that outcome expectancies and self-efficacy beliefs may be affected by stereotype threat differently. Students who have higher ethnic stereotype threat may expect lower grades because they anticipate discrimination and unfair evaluation and not because they do not believe in their abilities to learn and complete assignments. Future research could investigate these differences further as they could have important implications for interventions designed to mitigate the negative effects of ethnic stereotype threat on motivation and academic outcomes.
Prediction of STEM Outcomes
Motivational Beliefs
We found that after controlling for ethnic stereotype threat, only the growth parameters of intrinsic value significantly related to STEM achievement. URM students earned higher grades if they had higher initial levels of intrinsic value. However, other motivation variables were not related to STEM achievement, controlling for ethnic stereotype threat. These findings suggest that, for URM students, change in ethnic stereotype threat was typically more important in their STEM achievement than motivation (we discuss the associations between ethnic stereotype threat and STEM outcomes below). Regarding STEM course completion, the story was different. We found that after controlling for ethnic stereotype threat, the intercepts and linear slopes for intrinsic value and attainment value were related to course completion. That is, URM students completed more STEM courses if they had higher initial levels of intrinsic and attainment values and if these motivational beliefs decreased more slowly. These results make sense given theory and empirical research that suggest that motivational beliefs, and particularly task values, are stronger predictors of students’ choice behaviors (i.e. persistence) than performance (i.e. achievement; Eccles, 1984, 2005; Perez et al., 2014; Stevens et al., 2007). Therefore, after controlling for ethnic stereotype threat, these motivational beliefs still significantly related to STEM course completion.
Interestingly, there were no significant relations between the effort cost growth parameters and the STEM outcomes after controlling for the ethnic stereotype threat growth parameters. The ethnic stereotype threat intercept was also not related to the STEM outcomes in this model even though such relations were found in other models. The growth parameters of ethnic stereotype threat and effort cost correlated more highly than those of stereotype threat and other motivation variables. Further research is needed to understand these findings, but it is possible that there was no unique variance in the intercepts and slopes of effort cost and ethnic stereotype threat left to explain the STEM outcomes. This finding may have implications for theory in that ethnic stereotype threat may be partially experienced as perceived cost. Another explanation may simply be that the sample size in the effort cost model was smaller than the sample size in the other models. As was reported in the Method section, Cohort 1 students were not included in the analyses that involved effort cost. Indeed, the marginal though non-significant findings for the relations between effort cost growth parameters and STEM outcomes may attest to this possibility. Similar analyses with a larger sample size may yield significant findings for the relations between effort cost growth parameters and STEM outcomes.
These findings have important implications for practice. The results imply that pre-college supports designed to boost URM students’ interest in STEM fields may improve achievement for URM students, even for those with lower levels of motivational beliefs early in college. Further, interventions implemented during the four years of college that would slow down or reverse the downward trajectory of motivation might facilitate URM students’ positive development of science identity and interest in STEM fields, which could in-turn increase their persistence in their STEM majors.
Ethnic Stereotype Threat
We found that the ethnic stereotype threat growth parameters were related to STEM achievement after controlling for the effects of the motivation growth parameters. Although extant research (e.g., Good et al., 2008; Spencer et al., 2016; Woodcock et al., 2012) has demonstrated the effects of stereotype threat on grades, our study further demonstrates that ethnic stereotype threat can explain additional variance in achievement beyond the variance explained by motivational beliefs (although the exact relationship patterns varied depending on which motivational variables were included in the same model as stereotype threat). Our findings suggested that both the intercept and slope of ethnic stereotype threat related to cumulative STEM achievement. That is, students who started college with higher ethnic stereotype threat, and those with faster rates of increase in ethnic stereotype threat, received lower grades in their STEM courses. Our findings with regards to the growth of ethnic stereotype threat and academic outcomes are novel. We believe that the increase in ethnic stereotype threat in the first year and its lack of decrease in the following years may reflect a chronic effect. This notion is aligned with Steele’s (1997) argument that continuous experiences with stereotype threat chronically affects students’ identification with the domain and results in negative long-term outcomes. Relatedly we found that early formation of ethnic stereotype threat beliefs related to a decrease in attainment value (which is related to domain identification). Moreover, early self-reported ethnic stereotype threat as well as the increase in these beliefs over the years related to lower achievement in the long-term.
With regards to STEM persistence, results indicated that after controlling for values and effort cost, ethnic stereotype threat growth parameters did not significantly relate to STEM course completion; however, after controlling for self-efficacy, the ethnic stereotype threat intercept significantly related to this outcome. This finding can be explained by the theoretical expectation and empirical research that suggest expectancies are stronger predictors of achievement whereas values are stronger predictors of students’ choice behaviors such as course enrollment and persistence (Eccles, 1984, 2005; Perez et al., 2014; Stevens et al., 2007). Relatedly, our data suggested that in the models that included task values (intrinsic value, attainment value, and effort cost), the majority of variance in STEM course-completion was explained by task values, which likely resulted in the non-significant relations between stereotype threat and STEM course completion. On the other hand, in the models with academic self-efficacy, the majority of variance in STEM course completion was likely explained by stereotype threat, which resulted in null findings for the relation between self-efficacy and STEM course completion.
Similar to the findings for the motivation variables, these findings have important implications. Results suggest URM students’ early experiences with stereotypic judgements or racially-biased treatment may have long-term effects on their academic performance. Additionally, depending on a variety of factors such as the frequency of exposures to discrimination during their college years, our findings demonstrated that self-reported ethnic stereotype threat may change, and such changes are connected to academic performance. Thus, institutions that seek to support ethnically minoritized students should consider implementing interventions that reduce negative stereotype-based experiences early in college and also provide students with support through all four years of college to mitigate increases in ethnic stereotype threat. This can be achieved by facilitating positive interactions among members of the minoritized groups (Abrams et al., 2006), providing access to successful role models from the minoritized groups (Marx & Goff, 2005), and educating the members of the majority group about their implicit biases (Jackson et al., 2014). Identity-safe environments could, in turn, help students develop more adaptive motivational beliefs, which could result in improved achievement and persistence outcomes.
Limitations and Future Directions
Although participants were reasonably compensated with monetary incentives for survey completion each year, attrition in our sample was high after our first survey. The large amount of missing data, although comparable to other multi-year studies (e.g., Jiang et al., 2020; Guo et al., 2015), could lead to biased parameter estimates, which could result in making inaccurate interpretations about the findings. We attempted to reduce potential bias due to missing data by including gender as a missing data correlate in our models since gender was related to missingness and incorporating missing data correlates into the model reduces bias in FIML estimation (Asparouhov & Muthén, 2008). Importantly, our diagnostic tests indicated that missingness was not related to any of our key variables assessed at year one and outcomes at the end of college. Even so, it will be important to find ways to retain more students who identify with underrepresented groups in STEM in future longitudinal studies. Additionally, our sample size might have been insufficient to detect some of the smaller effects, particularly for the conditional parallel process growth models with outcomes. However, even in those models, we still detected effects that were consistent with theory and prior research. Nevertheless, it is important to replicate these findings with a larger sample size to ensure the reliability of the results with sufficient power.
Another limitation involves the estimation of quadratic slopes with four time points. Although four time-points make estimation of quadratic slopes possible (Diallo, Morin, & Parker, 2014), some researchers recommend using at least six waves of data to avoid convergence issues caused by estimating quadratic slopes (Xitao & Xiaotao, 2005). Indeed, in our parallel process analyses, we faced model estimation issues that were likely due to the inclusion of the quadratic slope term. To remedy this error, we fixed the variance of the quadratic slopes to zero, which prevented us from examining the associations between the quadratic slopes and other parameters. Measurement of the study variables at six time-points in future research could potentially help circumvent this problem.
Moreover, whereas our approach of measuring motivation and stereotype threat once per year was well-suited to our research questions, the growth of stereotype threat and motivation is likely more dynamic and the relations among these beliefs are more complex than a yearly survey can capture. Therefore, further short-term longitudinal research with more frequent assessments would complement these findings by unraveling potentially reciprocal, dynamic, and complex trends of change.
Related to understanding potential reciprocal effects, it is likely that students’ early achievement in STEM courses may have affected their later motivation which, in turn, may have impacted their graduating STEM GPA. This assumption is in line with situated expectancy-value theory that posits students’ early achievement experience shapes their later success expectancies and values (Eccles & Wigfield, 2020). Existing empirical research also supports this theoretical expectation. Perez and colleagues (Perez et al., 2014) reported that achievement in the middle of the semester positively related to competence beliefs and values later in the semester which, in turn, positively predicted achievement at the end of the semester. Future long-term longitudinal research could examine the directionality of these relations in greater detail using cross-lagged panel models.
Finally, the results do not provide insights into the causal relations between ethnic stereotype threat and motivation. In other words, although we demonstrated that the initial level of ethnic stereotype threat was related to the changes in some of the motivation variables, we cannot make the conclusion that early perceptions of stereotype threat caused changes in motivation. Our findings also did not provide any indication of the factors that shape the growth of ethnic stereotype threat. We found that ethnic stereotype threat increases over time; however, we did not find any significant relations between the slope of ethnic stereotype threat and the growth parameters of motivation variables. We suspect that, consistent with theory (Steele, 1997), continuous exposure to discrimination and stereotypes might have led to an increase in ethnic stereotype threat over time. However, since we did not measure experiences with discrimination, further research is needed to examine this possibility.
Conclusion
Promoting the achievement and persistence of underrepresented minoritized students in STEM fields is an urgent priority for researchers, educators, and policymakers. The current study highlighted the socio-cultural and psychological factors that contribute to URM students’ persistence and achievement in STEM. Our study revealed that URM students’ positive motivational beliefs declined over four years of college while negative motivational beliefs increased (i.e., perceived effort cost). Sharper declines as well as lower initial levels of some of the motivational beliefs predicted lower achievement and persistence of URM students in STEM fields. We also found that ethnic stereotype threat was related to students’ motivational decline such that those who started college with higher perceptions of threat reported a faster decline in attainment value and steeper increase in effort cost perceptions. Findings also revealed that ethnic stereotype threat increased non-linearly over time and negatively influenced URM students’ achievement in STEM. This study has important implications for theory as it demonstrates how ethnic stereotype threat as a sociocultural factor is associated longitudinally with expectancy-value beliefs. There are also practical implications of the study for interventions seeking to enhance the motivation of URMs in STEM. Our findings imply that diversity-related supports should be implemented before or during the first year of college to buffer the negative effects of ethnic stereotype threat on motivation development. Moreover, continuous diversity-related support and motivation interventions during college could facilitate development of adaptive motivation and promote students’ achievement and persistence outcomes at the end of the college career.
Highlights.
Perceptions of stereotype threat increased over four years of college
Self-efficacy, intrinsic, and attainment values declined but cost increased over time
Early perceptions of stereotype threat related to a faster decline in motivation
Growth of stereotype threat and motivation related to grades and course completion
Acknowledgments
This work was supported by the National Institutes of Health under award number R01GM094534. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Appendix
Stereotype Threat (Steele et al., 2002)
How often do you feel that because of your ethnicity…
Some people believe that you have less ability.
If you are not better than average, people will assume that you are limited.
Professors expect you to do poorly.
Professors are less likely to encourage you.
You are not fully accepted or included by your program.
If you ask a simple question, people will think it is because of your ethnicity.
If you do poorly on a test, people will assume that it is because of your ethnicity.
People of your ethnicity face unfair evaluations because of their ethnicity.
Academic Self-Efficacy (Midgley et al., 2000)
I am certain I can master the skills taught in science classes.
I am certain I can figure out how to do the most difficult class work in science.
I can do almost all the work in science classes if I don’t give up.
Even if the work in science is hard, I can learn it.
I can do even the hardest work in science, if I try.
Task Values (Conley, 2012)
Intrinsic value
I enjoy the subject of science.
I enjoy doing science.
Science is exciting to me.
I am fascinated by science.
I like science.
Attainment value
It is important for me to be a person who reasons scientifically.
It is important for me to be someone who is good at solving problems that involve science.
Being someone who is good at science is important to me.
Being good in science is an important part of who I am.
Utility value
Science concepts are valuable because they will help me in the future.
Science will be useful for me later in life.
Being good in science will be important for my future (like when I get a job or go to graduate school).
Effort Cost (Perez et al., 2014)
When I think about the hard work needed to be successful in science, I am not sure that studying science is going to be worth it in the end.
Studying science requires more effort than I’m willing to put in.
Considering what I want to do with my life, studying science is just not worth the effort.
Footnotes
The intervention was a science summer enrichment program designed to support students’ motivation and sense of belonging in science, particularly for students who are underrepresented in STEM. Variables used in this study were targeted by the intervention, which is why we excluded students who participated in the intervention.
We did not start measuring effort cost until the second year of the larger project; therefore, we removed cohort 1 (N = 90) from the analyses with effort cost. We measured effort cost in all four years for all other cohorts.
For effort cost, the number of students with baseline data was 335 because we did not start assessing effort cost until the second year of the project.
Forty-seven students completed the Time 3 survey after not completing the Time 2 survey. Additionally, 57 students completed the T4 survey after not completing the Time 3 survey. There were 34 students who returned to complete the T4 survey after not completing either the Time 2 or Time 3 surveys.
The reasoning for this decision was twofold: (1) the very high correlations between the linear and quadratic slopes of each variable (see Table 4) led to a non-positive definite covariance psi matrix; (2) interpreting the association between two quadratic slopes or the association between a quadratic slope and a linear slope is complex. Whereas the first aim of the study was to model the developmental trends of stereotype threat and the motivation variables (including linear and quadratic trends), linear parallel process models aligned with the second and third aims of the study, which focused more on the interrelations between the trajectories of stereotype threat and motivation variables and less on making distinctions between different patterns of change.
We have no known conflict of interest to disclose.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- Abrams D, Eller A, & Bryant J. (2006). An age apart: The effects of intergenerational contact and stereotype threat on performance and intergroup bias. Psychology and Aging, 21(4), 691–702. 10.1037/0882-7974.21.4.691. [DOI] [PubMed] [Google Scholar]
- Andersen L, & Ward TJ (2014). Expectancy-value models for the STEM persistence plans of ninth grade, high-ability Students: A comparison between Black, Hispanic, and White students. Science Education, 98(2), 216–242. 10.1002/sce.21092. [DOI] [Google Scholar]
- Aronson J, & Good C. (2002). The development and consequences of stereotype vulnerability in adolescents. In Pajares F, & Urdan T. (Eds.), Academic motivation of adolescents (pp. 299–330). Greenwich, CT: Information Age. [Google Scholar]
- Aronson J, Lustina MJ, Good C, Keough K, Steele CM, & Brown J. (1999). When white men can’t do math: Necessary and sufficient factors in stereotype threat. Journal of Experimental Social Psychology, 35(1), 29–46. 10.1006/jesp.1998.1371. [DOI] [Google Scholar]
- Asparouhov T, & Muthén B. (2008). Auxiliary variables predicting missing data. https://www.statmodel.com/download/AuxM2.pdf. [Google Scholar]
- Bandura A. (1997). Self-efficacy: The exercise of control. New York: Freeman. [Google Scholar]
- Bong M. (2001). Between-and within-domain relations of academic motivation among middle and high school students: Self-efficacy, task value, and achievement goals. Journal of Educational Psychology, 93(1), 23–34. 10.1037/0022-0663.93.1.23. [DOI] [Google Scholar]
- Byrne BM (2013). Structural equation modeling with Mplus: Basic concepts, applications, and programming. Routledge. [Google Scholar]
- Cadinu M, Maass A, Frigerio S, Impagliazzo L, & Latinotti S. (2003). Stereotype threat: The effect of expectancy on performance. European Journal of Social Psychology, 33(2), 267–285. 10.1002/ejsp.145. [DOI] [Google Scholar]
- Chang JM, Buonora P, Stevens L, & Kwon C. (2016). Strategies to recruit and retain students in physical sciences and mathematics on a diverse college campus. Journal of College Science Teaching, 45(3), 14–22. 10.2505/4/jcst16_045_03_14. [DOI] [Google Scholar]
- Chen FF (2007). Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling, 14(3), 464–504. 10.1080/10705510701301834. [DOI] [Google Scholar]
- Cheung GW, & Rensvold RB (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9(2), 233–255. 10.1207/S15328007SEM0902_5. [DOI] [Google Scholar]
- Chickering AW, & Reisser L. (1993). Education and Identity. The Jossey-Bass Higher and Adult Education Series. 350 Sansome St., San Francisco, CA 94104: Jossey-Bass Inc., Publishers. [Google Scholar]
- Chouinard R, & Roy N. (2008). Changes in high-school students’ competence beliefs, utility value and achievement goals in mathematics. British Journal of Educational Psychology, 78(1), 31–50. 10.1348/000709907X197993. [DOI] [PubMed] [Google Scholar]
- Conley A. (2012). Patterns of motivation beliefs: Combining achievement goal and expectancy-value perspectives. Journal of Educational Psychology, 104(1), 32–47. 10.1037/a0026042. [DOI] [Google Scholar]
- Corra M, & Lovaglia MJ (2012). Too small to see? African-American underenrollment in advanced high school courses. Journal of Women and Minorities in Science and Engineering, 18(4), 295–314. 10.1615/JWomenMinorScienEng.2013001748. [DOI] [Google Scholar]
- Cromley JG, Kaplan A, & Perez T. (2017). Combined cognitive & motivational supports for STEM learning beta. http://hdl.handle.net/2142/97878.
- Cromley JG, Perez T, Wills TW, Tanaka JC, Horvat EM, & Agbenyega ETB (2013). Changes in race and sex stereotype threat among diverse STEM students: Relation to grades and retention in the majors. Contemporary Educational Psychology, 38(3), 247–258. 10.1016/j.cedpsych.2013.04.003. [DOI] [Google Scholar]
- Dai T, & Cromley JG (2014). Changes in implicit theories of ability in biology and dropout from STEM majors: A latent growth curve approach. Contemporary Educational Psychology, 39(3), 233–247. 10.1016/j.cedpsych.2014.06.003. [DOI] [Google Scholar]
- Davey A, & Savla J. (2010). Statistical power analysis with missing data: A structural equation modeling approach. NY: Rutledge. [Google Scholar]
- Del Toro J, & Hughes D. (2020). Trajectories of discrimination across the college years: Associations with academic, psychological, and physical adjustment outcomes. Journal of Youth and Adolescence, 49(4), 772–789. 10.1007/s10964-019-01147-3. [DOI] [PubMed] [Google Scholar]
- Denissen JJ, Zarrett NR, & Eccles JS (2007). I like to do it, I’m able, and I know I am: Longitudinal couplings between domain-specific achievement, self-concept, and interest. Child Development, 78(2), 430–447. 10.1111/j.1467-8624.2007.01007.x. [DOI] [PubMed] [Google Scholar]
- Diallo TM, Morin AJ, & Parker PD (2014). Statistical power of latent growth curve models to detect quadratic growth. Behavior Research Methods, 46(2), 357–371. 10.3758/s13428-013-0395-1. [DOI] [PubMed] [Google Scholar]
- Durik AM, Vida M, & Eccles JS (2006). Task values and ability beliefs as predictors of high school literacy choices: A developmental analysis. Journal of Educational Psychology, 98(2), 382–393. 10.1037/0022-0663.98.2.382. [DOI] [Google Scholar]
- Eccles J. (2005). Subjective task value and the Eccles et al. model of achievementrelated choices. Handbook of Competence and Motivation, 105–121. [Google Scholar]
- Eccles J. (2009). Who am I and what am I going to do with my life? Personal and collective identities as motivators of action. Educational Psychologist, 44(2), 78–89. 10.1080/00461520902832368. [DOI] [Google Scholar]
- Eccles (Parsons) JS, Adler TF, Futterman R, Goff SB, Kaczala CM, Meece JL, & Midgley C. (1983). Expectancies, values, and academic behaviors. In Spence JT (Ed.), Achievement and achievement motivation (pp. 75–146). San Francisco, CA: W. H. Freeman. [Google Scholar]
- Eccles JS, Elliot AJ, & Dweck CS (2005). Subjective task value and the Eccles et al. model of achievement-related choices. Handbook of competence and motivation (pp. 105–121). New York, NY: Guildford. [Google Scholar]
- Eccles JS, Jacobs JE, & Harold RD (1990). Gender role stereotypes, expectancy effects, and parents’ socialization of gender differences. Journal of Social Issues, 46 (2), 183–201. 10.1111/j.1540-4560.1990.tb01929.x. [DOI] [Google Scholar]
- Eccles JS, & Wigfiled A. (2020). From expectancy-value theory to situated expectancyvalue theory: A developmental, social cognitive, and sociocultural perspective on motivation. Contemporary Educational Psychology. 10.1016/j.cedpsych.2020.101859. [DOI] [Google Scholar]
- Eccles JS, & Wigfield A. (2002). Motivational beliefs, values, and goals. Annual Review of Psychology, 53(1), 109–132. 10.1146/annurev.psych.53.100901.135153. [DOI] [PubMed] [Google Scholar]
- Eccles JS, Wigfield A, Flanagan C, Miller C, Reuman D, & Yee D. (1989). Selfconcepts, domain values, and self-esteem: Relations and changes at early adolescence. Journal of Personality, 57(2), 283–310. 10.1111/j.1467-6494.1989.tb00484.x. [DOI] [PubMed] [Google Scholar]
- Eccles JS, Wong CA, & Peck SC (2006). Ethnicity as a social context for the development of African-American adolescents. Journal of School Psychology, 44(5), 407–426. 10.1016/j.jsp.2006.04.001. [DOI] [Google Scholar]
- Enders CK (2008). A note on the use of missing auxiliary variables in full information maximum likelihood-based structural equation models. Structural Equation Modeling: A Multidisciplinary Journal, 15(3), 434–448. [Google Scholar]
- Enders CK (2010). Applied missing data analysis. Guilford press. [Google Scholar]
- Fischer MJ (2010). A longitudinal examination of the role of stereotype threat and racial climate on college outcomes for minorities at elite institutions. Social Psychology of Education, 13(1), 19–40. 10.1007/s11218-009-9105-3. [DOI] [Google Scholar]
- Forbes CE, & Schmader T. (2010). Retraining attitudes and stereotypes to affect motivation and cognitive capacity under stereotype threat. Journal of Personality and Social Psychology, 99(5), 740–754. 10.1037/a0020971. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fredricks JA, & Eccles JS (2002). Children’s competence and value beliefs from childhood through adolescence: Growth trajectories in two male-sex-typed domains. Developmental Psychology, 38(4), 519–533. 10.1037/0012-1649.38.4.519. [DOI] [PubMed] [Google Scholar]
- Good C, Aronson J, & Harder JA (2008). Problems in the pipeline: Stereotype threat and women’s achievement in high-level math courses. Journal of Applied Developmental Psychology, 29(1), 17–28. 10.1016/j.appdev.2007.10.004. [DOI] [Google Scholar]
- Gonzales PM, Blanton H, & Williams KJ (2002). The effects of stereotype threat and double-minority status on the test performance of Latino women. Personality and Social Psychology Bulletin, 28(5), 659–670. 10.1177/0146167202288010. [DOI] [Google Scholar]
- Graham JW (2003). Adding missing-data-relevant variables to FIML-based structural equation models. Structural Equation Modeling, 10(1), 80–100. [Google Scholar]
- Guo J, Parker PD, Marsh HW, & Morin AJ (2015). Achievement, motivation, and educational choices: A longitudinal study of expectancy and value using a multiplicative perspective. Developmental Psychology, 51(8), 1163. [DOI] [PubMed] [Google Scholar]
- Harackiewicz JM, Canning EA, Tibbetts Y, Giffen CJ, Blair SS, Rouse DI, et al. (2014). Closing the social class achievement gap for first-generation students in undergraduate biology. Journal of Educational Psychology, 106(2), 375–389. 10.1037/a0034679. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Harper SR (2015). Black male college achievers and resistant responses to racist stereotypes at predominantly White colleges and universities. Harvard Educational Review, 85(4), 646–674. 10.17763/0017-8055.85.4.646. [DOI] [Google Scholar]
- Harwood SA, Huntt MB, Mendenhall R, & Lewis JA (2012). Racial microaggressions in the residence halls: Experiences of students of color at a predominantly White university. Journal of Diversity in Higher Education, 5(3), 159–173. 10.1037/a0028956. [DOI] [Google Scholar]
- Hernandez PR, Schultz P, Estrada M, Woodcock A, & Chance RC (2013). Sustaining optimal motivation: A longitudinal analysis of interventions to broaden participation of underrepresented students in STEM. Journal of Educational Psychology, 105(1), 89–107. 10.1037/a0034254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hu LT, & Bentler PM (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. 10.1080/10705519909540118. [DOI] [Google Scholar]
- Hulleman CS, Godes O, Hendricks BL, & Harackiewicz JM (2010). Enhancing interest and performance with a utility value intervention. Journal of Educational Psychology, 102(4), 880–895. 10.1037/a0019506. [DOI] [Google Scholar]
- Hurtado S. (1992). The campus racial climate: Contexts of conflict. The Journal of Higher Education, 63(5), 539–569. 10.1080/00221546.1992.11778388. [DOI] [Google Scholar]
- Hurtado S, Carter DF, & Spuler A. (1996). Latino student transition to college: Assessing difficulties and factors in successful college adjustment. Research in Higher Education, 37(2), 135–157. 10.1007/BF01730113. [DOI] [Google Scholar]
- Inzlicht M, & Ben-Zeev T. (2000). A threatening intellectual environment: Why females are susceptible to experiencing problem-solving deficits in the presence of males. Psychological Science, 11(5), 365–371. 10.1111/1467-9280.00272. [DOI] [PubMed] [Google Scholar]
- Irving MA, & Hudley C. (2008). Cultural identification and academic achievement among African American males. Journal of Advanced Academics, 19(4), 676–698. 10.4219/jaa-2008-833. [DOI] [Google Scholar]
- Jackson SM, Hillard AL, & Schneider TR (2014). Using implicit bias training to improve attitudes toward women in STEM. Social Psychology of Education, 17(3), 419–438. [Google Scholar]
- Jacobs JE, Lanza S, Osgood DW, Eccles JS, & Wigfield A. (2002). Changes in children’s self -competence and values: Gender and domain differences across grades one through twelve. Child Development, 73(2), 509–527. 10.1111/1467-8624.00421. [DOI] [PubMed] [Google Scholar]
- Jiang S, Simpkins SD, & Eccles JS (2020). Individuals’ math and science motivation and their subsequent STEM choices and achievement in high school and college: A longitudinal study of gender and college generation status differences. Developmental Psychology, 56(11), 2137–2151. 10.1037/dev0001110. [DOI] [PubMed] [Google Scholar]
- Joyce BA, & Farenga SJ (2000). Young girls in science: Academic ability, perceptions and future participation in science. Roeper Review, 22(4), 261–262. 10.1080/02783190009554048. [DOI] [Google Scholar]
- Kaplan A, Garner JK, & Brock B. (2019). Identity and Motivation in a Changing World: A Complex Dynamic Systems Perspective, 20. Advances in Motivation and Achievement (pp. 101–127). Emerald Publishing Limited. 10.1108/S0749-742320190000020006. [DOI] [Google Scholar]
- Keller J. (2002). Blatant stereotype threat and women’s math performance: Selfhandicapping as a strategic means to cope with obtrusive negative performance expectations. Sex Roles, 47(3–4), 193–198. 10.1023/A:1021003307511. [DOI] [Google Scholar]
- Kosovich JJ, Flake JK, & Hulleman CS (2017). Short-term motivation trajectories: A parallel process model of expectancy-value. Contemporary Educational Psychology, 49, 130–139. 10.1016/j.cedpsych.2017.01.004. [DOI] [Google Scholar]
- Little RJ (1988). A test of missing completely at random for multivariate data with missing values. Journal of the American Statistical Association, 83(404), 1198–1202. 10.1080/01621459.1988.10478722. [DOI] [Google Scholar]
- Maddux JE, Sherer M, & Rogers RW (1982). Self-efficacy expectancy and outcome expectancy: Their relationship and their effects on behavioral intentions. Cognitive Therapy and Research, 6(2), 207–211. 10.1007/BF01183893. [DOI] [Google Scholar]
- Major B, & Schmader T. (1998). Coping with stigma through psychological disengagement. In Prejudice (pp. 219–241). Academic Press. 10.1016/B978-012679130-3/50045-4. [DOI] [Google Scholar]
- Marsh HW, Abduljabbar AS, Abu-Hilal MM, Morin AJ, Abdelfattah F, Leung KC, et al. (2013). Factorial, convergent, and discriminant validity of TIMSS math and science motivation measures: A comparison of Arab and Anglo-Saxon countries. Journal of Educational Psychology, 105(1), 108–128. 10.1037/a0029907. [DOI] [Google Scholar]
- Marx DM, & Goff PA (2005). Clearing the air: The effect of experimenter race on target’s test performance and subjective experience. British Journal of Social Psychology, 44(4), 645–657. 10.1348/014466604X17948. [DOI] [PubMed] [Google Scholar]
- Meece JL, Wigfield A, & Eccles JS (1990). Predictors of math anxiety and its influence on young adolescents’ course enrollment intentions and performance in mathematics. Journal of Educational Psychology, 82(1), 60–70. 10.1037/0022-0663.82.1.60. [DOI] [Google Scholar]
- Mendoza-Denton R, Downey G, Purdie VJ, Davis A, & Pietrzak J. (2002). Sensitivity to status- based rejection: Implications for African American students’ college experience. Journal of Personality and Social Psychology, 83(4), 896–918. 10.1037//0022-3514.83.4.896. [DOI] [PubMed] [Google Scholar]
- Meredith W. (1993). Measurement invariance, factor analysis and factorial invariance. Psychometrika, 58(4), 525–543. 10.1007/BF02294825. [DOI] [Google Scholar]
- Midgley C, Maehr ML, Hruda LZ, Anderman E, Anderman L, Freeman KE, et al. (2000). Manual for the Patterns of Adaptive Learning Scales (PALS). Ann Arbor, MI: University of Michigan. Retrieved from: http://www.umich.edu/~pals/PALS%202000_V12Word97.pdf. [Google Scholar]
- Muthén BO, & Khoo ST (1998). Longitudinal studies of achievement growth using latent variable modeling. Learning and Individual Differences, 10(2), 73–101. 10.1016/S1041-6080(99)80135-6. [DOI] [Google Scholar]
- Murphy MC, Steele CM, & Gross JJ (2007). Signaling threat: How situational cues affect women in math, science, and engineering settings. Psychological Science, 18 (10), 879–885. 10.1111/j.1467-9280.2007.01995.x. [DOI] [PubMed] [Google Scholar]
- Murphy MC, & Taylor VJ (2012). The role of situational cues in signaling and maintaining stereotype threat. In Inzlicht M, & Schmader T. (Eds.), Stereotype threat: Theory, process, and application (pp. 17–33). Oxford University Press. [Google Scholar]
- Muthén L, & Muthén B. (2012). Mplus user’s guide (version 7). Los Angeles: Author. National Science Foundation. (2018). S&E Indicators 2018. Retrieved September 30, 2019, from https://www.nsf.gov/statistics/2018/nsb20181/report/sections/highereducation-in-science-and-engineering/undergraduate-education-enrollment-anddegrees-in-the-united-states. [Google Scholar]
- Nguyen HHD, & Ryan AM (2008). Does stereotype threat affect test performance of minorities and women? A meta-analysis of experimental evidence. Journal of Applied Psychology, 93(6), 1314–1334. 10.1037/a0012702. [DOI] [PubMed] [Google Scholar]
- Osborne JW, & Walker C. (2006). Stereotype threat, identification with academics, and withdrawal from school: Why the most successful students of colour might be most likely to withdraw. Educational Psychology, 26(4), 563–577. 10.1080/01443410500342518. [DOI] [Google Scholar]
- Oyserman D, & Destin M. (2010). Identity-based motivation: Implications for intervention. The Counseling Psychologist, 38(7), 1001–1043. 10.1177/0011000010374775. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Perez T, Cromley JG, & Kaplan A. (2014). The role of identity development, values, and costs in college STEM retention. Journal of Educational Psychology, 106(1), 315–329. 10.1037/a0034027. [DOI] [Google Scholar]
- Perez T, Wormington SV, Barger MM, Schwartz-Bloom RD, Lee Y. k., & Linnenbrink-Garcia L. (2019). Science expectancy, value, and cost profiles and their proximal and distal relations to undergraduate science, technology, engineering, and math persistence. Science Education, 103(2), 264–286. 10.1002/sce.21490. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Plante I, De la Sablonnière R, Aronson JM, & Théorˆet M. (2013). Gender stereotype endorsement and achievement-related outcomes: The role of competence beliefs and task values. Contemporary Educational Psychology, 38(3), 225–235. 10.1016/j.cedpsych.2013.03.004. [DOI] [Google Scholar]
- Riegle-Crumb C, King B, & Irizarry Y. (2019). Does STEM stand out? Examining racial/ethnic gaps in persistence across postsecondary fields. Educational Researcher, 48(3), 133–144. 10.3102/0013189X19831006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Robinson KA, Lee Y. k., Bovee EA, Perez T, Walton SP, Briedis D, & Linnenbrink-Garcia L. (2019a). Motivation in transition: Development and roles of expectancy, task values, and costs in early college engineering. Journal of Educational Psychology, 111(6), 1081–1102. 10.1037/edu0000331. [DOI] [Google Scholar]
- Robinson KA, Perez T, Carmel JH, & Linnenbrink-Garcia L. (2019b). Science identity development trajectories in a gateway college chemistry course: Predictors and relations to achievement and STEM pursuit. Contemporary Educational Psychology, 56, 180–192. 10.1016/j.cedpsych.2019.01.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Robinson KA, Perez T, Nuttall AK, Roseth CJ, & Linnenbrink-Garcia L. (2018). From science student to scientist: Predictors and outcomes of heterogeneous science identity trajectories in college. Developmental Psychology, 54(10), 1977–1992. 10.1037/dev0000567. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rosenzweig EQ, & Wigfield A. (2017). What if reading is easy but unimportant? How students’ patterns of affirming and undermining motivation for reading information texts predict different reading outcomes. Contemporary Educational Psychology, 48, 133–148. 10.1016/j.cedpsych.2016.09.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schmader T, Major B, & Gramzow RH (2001). Coping with ethnic stereotypes in the academic domain: Perceived injustice and psychological disengagement. Journal of Social Issues, 57(1), 93–111. 10.1111/0022-4537.00203. [DOI] [Google Scholar]
- Shell DF, Murphy CC, & Bruning RH (1989). Self-efficacy and outcome expectancy mechanisms in reading and writing achievement. Journal of Educational Psychology, 81(1), 91–100. 10.1037/0022-0663.81.1.91. [DOI] [Google Scholar]
- Shi D, Maydeu-Olivares A, & Rosseel Y. (2020). Assessing fit in ordinal factor analysis models: SRMR vs. RMSEA. Structural Equation Modeling: A Multidisciplinary Journal, 27(1), 1–15. 10.1080/10705511.2019.1611434. [DOI] [Google Scholar]
- Smedley BD, Myers HF, & Harrell SP (1993). Minority-status stresses and the college adjustment of ethnic minority freshmen. The Journal of Higher Education, 64 (4), 434–452. 10.1080/00221546.1993.11778438. [DOI] [Google Scholar]
- Smith JL, Brown ER, Thoman DB, & Deemer ED (2015). Losing its expected communal value: How stereotype threat undermines women’s identity as research scientists. Social Psychology of Education, 18(3), 443–466. 10.1007/s11218-015-9296-8. [DOI] [Google Scholar]
- Spencer SJ, Logel C, & Davies PG (2016). Stereotype threat. Annual Review of Psychology, 67, 415–437. 10.1146/annurev-psych-073115-103235. [DOI] [PubMed] [Google Scholar]
- Steele CM (1997). A threat in the air: How stereotypes shape intellectual identity and performance. American Psychologist, 52(6), 613–629. 10.1037/0003-066X.52.6.613. [DOI] [PubMed] [Google Scholar]
- Steele CM, & Aronson J. (1995). Stereotype threat and the intellectual test performance of African Americans. Journal of Personality and Social Psychology, 69 (5), 797–811. 10.1037/0022-3514.69.5.797. [DOI] [PubMed] [Google Scholar]
- Steele J, James JB, & Barnett RC (2002). Learning in a man’s world: Examining the perceptions of undergraduate women in male-dominated academic areas. Psychology of Women Quarterly, 26(1), 46–50. 10.1111/1471-6402.00042. [DOI] [Google Scholar]
- Stevens T, Wang K, Olivárez A, & Hamman D. (2007). Use of self-perspectives and their sources to predict the mathematics enrollment intentions of girls and boys. Sex Roles, 56(5–6), 351–363. 10.1007/s11199-006-9180-2. [DOI] [Google Scholar]
- Stone J. (2002). Battling doubt by avoiding practice: The effects of stereotype threat on self-handicapping in white athletes. Personality and Social Psychology Bulletin, 28(12), 1667–1678. 10.1177/014616702237648. [DOI] [Google Scholar]
- Thoman DB, Smith JL, Brown ER, Chase J, & Lee JYK (2013). Beyond performance: A motivational experiences model of stereotype threat. Educational Psychology Review, 25(2), 211–243. 10.1007/s10648-013-9219-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Toretsky C, Mutha S, & Coffman J. (2018). Breaking barriers for underrepresented minorities in the health professions. Retrieved from Healthforce Center at UCSF website https://healthforce.ucsf.edu/publications/breaking-barriers-under represented-minorities-health-professions. [Google Scholar]
- Vandenberg RJ, & Lance CE (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70. 10.1177/109442810031002. [DOI] [Google Scholar]
- Wang MT, Degol J, & Ye F. (2015). Math achievement is important, but task values are critical, too: Examining the intellectual and motivational factors leading to gender disparities in STEM careers. Frontiers in Psychology, 6, 36. 10.3389/fpsyg.2015.00036. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Watt HM (2004). Development of adolescents’ self-perceptions, values, and task perceptions according to gender and domain in 7th-through 11th-grade Australian students. Child Development, 75(5), 1556–1574. 10.1111/j.1467-8624.2004.00757.x. [DOI] [PubMed] [Google Scholar]
- Watt HM (2008). A latent growth curve modeling approach using an accelerated longitudinal design: The ontogeny of boys’ and girls’ talent perceptions and intrinsic values through adolescence. Educational Research and Evaluation, 14(4), 287–304. 10.1080/13803610802249316. [DOI] [Google Scholar]
- Wigfield A. & Cambria J. (2010), Expectancy-value theory: retrospective and prospective. In Urdan T, & Karabenick S. (Ed.) The Decade Ahead: Theoretical Perspectives on Motivation and Achievement (Advances in Motivation and Achievement, Vol. 16 Part A). Emerald Group Publishing Limited, Bingley, pp. 35–70. [Google Scholar]
- Wigfield A, & Eccles JS (1992). The development of achievement task values: A theoretical analysis. Developmental Review, 12(3), 265–310. 10.1016/0273-2297(92)90011-P. [DOI] [Google Scholar]
- Wigfield A, & Eccles JS (2000). Expectancy–value theory of achievement motivation. Contemporary Educational Psychology, 25(1), 68–81. 10.1006/ceps.1999.1015. [DOI] [PubMed] [Google Scholar]
- Wigfield A, Hoa LW, & Klauda SL (2008). The role of achievement values in the regulation of achievement behaviors. Motivation and Self-regulated Learning: Theory, Research, and Applications, 169–195. [Google Scholar]
- Wigfield A, Tonks S, & Klauda SL, (2016). Expectancy-value Theory. In Wentzel KR, & Miele DB (Eds.), Handbook of motivation at school (pp. 55–74). Oxon: Routledge. [Google Scholar]
- Woodcock A, Hernandez PR, Estrada M, & Schultz P. (2012). The consequences of chronic stereotype threat: Domain disidentification and abandonment. Journal of Personality and Social Psychology, 103(4), 635–646. 10.1037/a0029120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wu W, West SG, & Taylor AB (2009). Evaluating model fit for growth curve models: Integration of fit indices from SEM and MLM frameworks. Psychological methods, 14(3), 183–201. 10.1037/a0015858. [DOI] [PubMed] [Google Scholar]
- Xitao F, & Xiaotao F. (2005). Power of latent growth modeling for detecting linear growth: Number of measurements and comparison with other analytic approaches. The Journal of Experimental Education, 73(2), 121–139. 10.3200/JEXE.73.2.121-139. [DOI] [Google Scholar]