Dweck’s Social-Cognitive Model of Achievement Motivation in Science

Abstract

Dweck’s social-cognitive model has long been used as a basis for achievement motivation research. However, few studies have examined the comprehensive model with interactions between perceived ability and achievement goals, and even fewer studies have focused on this model in a science academic context. With a sample of undergraduates (n = 1,036), the relations among mindsets, science academic self-efficacy, achievement goals, and achievement-related outcomes in science were examined. Fixed mindset related to performance goals. Growth mindset related to mastery goals and the number of courses completed. There was a significant indirect effect of growth mindset on interest value via mastery goals. Contrary to Dweck’s model, the relation of performance goals to outcomes did not vary as a function of science academic self-efficacy. The findings provide empirical evidence for a more nuanced understanding of Dweck’s model. They provide practical insights for how to support undergraduate students who are pursuing science-related career.

1. Introduction

Dweck’s social-cognitive model, often also called mindset theory, provides a useful lens to understanding motivational processes of individuals’ achievement-related behaviors through implicit theories of intelligence, perceived ability, and achievement goals (Dweck, 1986; Dweck & Leggett, 1988; see Dweck & Yeager, 2019). Researchers have investigated the relations among these three constructs to explain and predict individuals’ behavioral patterns in achievement settings (Cho et al., 2019; Gonzalez-DeHass et al., 2023; Yu & McLellan, 2020). However, few studies have examined the comprehensive model representing the structural relations among all three constructs, and even fewer studies have focused on this model in a science academic context. Studying these processes for undergraduates in science is important given the prevalence or shared societal views of fixed mindsets in science, technology, engineering, and mathematics (STEM) fields relative to other fields, such as liberal arts (Miller, 2015). The long-lasting problem of high attrition rates in a science major (National Science Foundation, 2011), often called the ‘leaky pipeline,’ needs to be considered in light of recent evidence indicating that students perceive STEM-related majors as “brilliance-required” fields (Leslie et al., 2015). This means students might tend to believe that STEM-related ability is determined by innate and stable factors within individuals (i.e., fixed mindset) rather than malleable and changeable factors (i.e., growth mindset), which could further lead some students to leave such a challenging field.

Thus, investigating Dweck’s model among science undergraduates is important for better understanding attrition from science majors, a critical issue related to gaps in the U.S. STEM workforce (National Academies of Sciences, Engineering, and Medicine, 2016), based on the underlying social-cognitive processes of achievement motivation and because it provides a useful context for studying these psychological processes in a high-challenge, failure-prone context. As such, the current study investigated the relations among college students’ mindsets (also called implicit theories of intelligence), science motivation, and science persistence. This investigation can also inform theory such that testing the full structural relations among mindsets, perceived ability, and achievement goals can provide a nuanced understanding of unique roles of these three different constructs in predicting achievement-related behaviors. In doing so, we aimed to provide new empirical evidence toward a more complete portrait of these motivational processes as well as insights into how to support college students’ interest and persistence in science.

2.1. Dweck’s Social-Cognitive Model

In Dweck’s social-cognitive model of achievement motivation (Dweck, 1986; Dweck & Leggett, 1988; see Figure 1), also known as mindset theory (Dweck & Yeager, 2019), individuals’ beliefs about whether intelligence or basic ability is stable (i.e., fixed mindset, or entity theory) or malleable and can grow (i.e., growth mindset, or incremental theory) impact on their responses to challenges or setbacks. In Dweck’s (1986) original theorizing, two types of mindsets and two types of achievement goals were introduced, with different mindsets presumed to result in different types of achievement goals. Dweck proposed that students holding fixed mindset would endorse performance goals, in which they seek to gain favorable judgments of their competence or avoid negative judgments of their competence. In contrast, students with growth mindset would endorse learning goals, in which they seek to increase their competence to understand or master new knowledge or skills.

Figure 1. Mindset, Achievement Goals, and Behavior Patterns Based on Dweck’s Social-Cognitive Model.

Note. This theoretical model was slightly modified from the articles by Dweck (1986) and Dweck and Leggett (1988).

Dweck’s social-cognitive model further posits that achievement goals in turn lead to differential patterns of achievement-related behaviors (Dweck, 1986; Dweck & Leggett, 1988). Learning goals are hypothesized to result in an adaptive or mastery-oriented pattern of behaviors, including high levels of interest, performance, and persistence, regardless of whether difficulties are faced, and seeking of challenge. The outcome patterns of performance goals, however, are differentially hypothesized depending on the level of perceived ability. Performance goals would have a positive impact on achievement outcomes (similar to learning goals), when students’ perceived ability is high. When students’ perceived ability is low, conversely, performance goals would lead to a maladaptive or helpless pattern of behavior, including low levels of interest, performance, and persistence, in the face of setbacks. It is worth noting that these different behavior patterns are expected as responses to challenges or obstacles. Thus, mindset theory is not primarily concerned with academic outcomes in general; rather, it should be associated with academic outcomes, especially among individuals confronting challenges (Yeager & Dweck, 2020).

As such, Dweck’s original social-cognitive model included these two types of goals, learning goals (which have also been referred to as mastery goals) and performance goals. In subsequent work, performance goals were further differentiated into performance-approach goals and performance-avoidance goals (Harackiewicz et al., 1998). Thus, the trichotomous achievement goal model consists of mastery goals (focusing on developing competence), performance-approach goals (focusing on demonstrating competence), and performance-avoidance goals (focusing on avoiding showing incompetence). Later, the approach-avoidance dimension was also applied to mastery goals in the 2 × 2 model, introducing mastery-avoidance goals (Elliot & McGregor, 2001). In the current study, we did not include mastery-avoidance goals as these goals were not presumed in Dweck’s model. Additionally, we did not differentiate between performance-approach and performance-avoidance goals but instead used these two goals as combined performance goal latent variable. While this approach of combining performance-approach and -avoidance goals into a single construct does not align with more modern approaches, it is consistent with Dweck’s original theorizing (see also our factor analysis in Measures) and, importantly, allowed us to test Dweck’s original theorizing that perceived ability would moderate the effect of performance goals.

Also of note, performance goals have been further differentiated into distinct forms of goals (Grant & Dweck, 2003; Hulleman et al., 2010), including normative goals (desire to outperform others or avoid underperforming others), appearance or ability goals (desire to show competence or avoid showing incompetence), and outcome goals (desire to get a good grade). We assessed performance goals as appearance goals because appearance goals are better aligned with the conceptions of performance goals suggested by Dweck’s social-cognitive model (Dweck, 1986; Dweck & Leggett, 1988).

2.2. Relations of Mindsets to Academic Outcomes

Prior research has examined the relations of mindsets to various motivational and achievement outcomes. In general, prior research found that endorsing a growth mindset was beneficial for a range of outcomes, such as academic motivation (Cho et al., 2019; Haimovitz et al., 2011; Lee et al., 2022), educational retention (Dai & Cromley, 2014), and achievement (McCutchen et al., 2016; Paunesku et al., 2015; Yeager et al., 2014; Yu & McLellan, 2020). Focusing specifically on achievement goals, one meta-analytic study showed that, compared to fixed mindset, growth mindset was positively related to mastery goals and negatively related to both performance-approach and -avoidance goals, with a more negative relation with performance-avoidance goals (Burnette et al., 2013). The same patterns were replicated in a laboratory experimental study such that university students adopted higher levels of mastery goals and lower levels of performance-avoidance goals in the incremental condition than in the entity condition (Dinger & Dickhäuser, 2013).

Mindsets are also associated with self-efficacy, educational retention, and academic achievement. Children’s fixed mindset was negatively correlated with self-efficacy and behavioral and cognitive engagement in reading (Lee et al., 2022), whereas undergraduate students in an incremental intervention showed higher entrepreneurial self-efficacy and task persistence in their class project (Burnette et al., 2020). For educational retention, low initial level and negative slope of growth mindset in biology was associated with leaving STEM majors (Dai & Cromley, 2014). Regarding achievement, students holding high growth mindset tended to achieve higher course grades (Lee et al., 2022; Yeager et al., 2014; Yu & Mclellan, 2020) or show slower decreases in grades over time than those with high fixed mindset (Dai & Cromley, 2014; McCutchen et al., 2016). Meta-analytic studies (Macnamara & Burgoyne, 2023; Sisk et al., 2018), however, found only a weak relation between growth mindset and achievement, questioning the predictability of mindset for outcomes. Empirical evidence often reveals such weaker mindset associations than theoretically expected (Li & Bates, 2020; Lou & Li, 2023; Yeager & Dweck, 2020). Consequently, more data may be needed if important mediating or moderating variables have been overlooked (Tipton et al., 2023).

Prior studies have further tested mediational models where mindsets predict achievement goals which, in turn, predict outcomes. In one study, college students’ growth mindset had a positive and indirect effect on achievement through mastery goals and performance-approach goals (Chen & Wong, 2015). In another study focusing on upper elementary school students, fixed mindset negatively predicted reading achievement via performance-avoidance goals (Cho et al., 2019). Collectively, fixed mindset generally related to performance goals whereas growth mindset was linked to mastery goals. Additionally, growth mindset was more positively related to self-efficacy, educational retention, and academic achievement than fixed mindset.

2.3. Relations of Achievement Goals and Perceived Ability to Academic Outcomes

2.3.1. Achievement Goals

Different types of achievement goals have distinct associations with academic-related outcomes, such as interest value (i.e., enjoyment from engaging in the task or activity; Eccles & Wigfield, 2020), academic achievement, and choice behavior (e.g., course choices). With respect to interest, meta-analytic studies (Baranik et al., 2010; Huang, 2011; Hulleman et al., 2010; Scherrer et al., 2020) found that mastery-approach goals were more strongly related to interest (rs = .42 to .61) than performance-approach goals (rs = .04 to .17). Performance-avoidance goals exhibited a negative or weak association with interest (rs = −.08 to .09), with the majority of studies showing a negative correlation. Hulleman and colleagues (2010) differentiated performance-approach goals into normative goals and appearance goals, finding a non-significant correlation between normative goals and interest (r = .08). Huang (2011), using a dichotomous goal framework like our study, found that mastery goals (r = .42) were more strongly correlated with interest than performance goals (r = .07).

Regarding academic achievement, most meta-analyses (Hulleman et al., 2010; Van Yperen et al., 2014; Wirthwein et al., 2013) found positive associations for mastery-approach goals (rs = .11 to .13) and performance-approach goals (rs = .06 to .08), but negative associations for performance-avoidance goals (rs = −.14 to −.12). When separating performance-approach goals into normative goals and appearance goals (Hulleman et al., 2010), normative goals were positively associated with academic performance (r = .14), whereas appearance goals were negatively associated (r = −.14). With the dichotomous goal framework (Huang, 2012), mastery goals were positively related to achievement (r = .13), whereas performance goals were not significantly related (r = −.00).

A more limited number of studies have examined the association between achievement goals and choice behavior, but other choice-related outcomes, such as engagement and perseverance, may help to imply this association. For instance, mastery goals positively related to behavioral, cognitive, and emotional engagement with higher-order learning, reflective and integrative learning, and perseverance (Cho et al., 2019; Millet et al., 2021; Yu & McLellan, 2020). Contrarily, performance-avoidance goals were negatively related to these engagement indicators, while the findings for performance-approach goals were more mixed, including positive relations (Yu & McLellan, 2020) and non-significant relations (Cho et al., 2019). In summary, mastery goals are expected to have a positive relation with academic-related outcomes, whereas the relations for performance goals, which encompass both performance-approach and performance-avoidance goals in this study, are expected to be more mixed.

2.3.2. Perceived Ability

Perceived ability, rather than actual ability, has consistently been found to strongly predict individuals’ performance (Klassen & Usher, 2010; Pajares, 1997). According to Dweck’s social-cognitive model, individuals’ behavior patterns derived from achievement goals may differ based on their perception of ability, rather than their actual intelligence (Dweck, 1986; Dweck & Leggett, 1988). In this sense, perceived self-efficacy, often referred to as self-efficacy, can be considered reflective of such perceived ability. Self-efficacy is defined as individuals’ evaluative perceptions of their abilities to perform a certain task (Usher, 2015), and many researchers have used self-efficacy as a proxy for perceived ability given various competence perceptions (e.g., perceived competence, self-efficacy) share conceptual and empirical similarities (e.g., Cho et al., 2018; Cury et al., 2006; Kaplan & Midgley, 1997).

Considering self-efficacy for science may be particularly important given perceptions that the domain of science, similar to other STEM disciplines, is perceived as highly competitive and demanding. Accordingly, self-efficacy plays a pivotal role in shaping how students effectively engage in scientific tasks and manages the academic challenges linked to pursuing a career in science (Perez et al., 2014; Robinson et al., 2018). High science academic self-efficacy is associated with increased science course enrollment, higher scores in science courses, greater science interest, and the pursuit of science career paths (Alhadabi, 2021; Honicke & Broadbent, 2016; Perez et al., 2014; Stets et al., 2017).

A large body of empirical evidence exists in the science domain. For instance, using large-scale data (HSLS:09), researchers found that science self-efficacy was significantly associated with science interest (Alhadabi, 2021). In addition, competence beliefs in science, similarly measured to science academic self-efficacy, positively predicted GPA in science-related courses and negatively predicted the intent to leave a STEM major (Perez et al., 2014). Together, science academic self-efficacy is expected to positively relate to science-related outcomes, including interest, achievement, and persistence.

2.4. Role of Perceived Ability in Dweck’s Social-Cognitive Model

In prior empirical work on the moderating role of perceived ability, perceived ability was operationalized as perceived competence, which refers to one’s beliefs about ability to learn and do academic work in a given domain (Schunk & Pajares, 2005). Surprisingly, the hypothesized moderating role of perceived ability in the relations between different types of achievement goals and outcomes has not been strongly supported (Cho et al., 2011; Cho et al., 2018; Cury et al., 2006; Kaplan & Midgley, 1997; Ommundsen, 2001). For instance, in Cury and colleagues’ (2006) research with students aged 12-15, perceived competence in mathematics did not moderate the relation of any type of achievement goals to intrinsic motivation or performance in math-specific domain or general intelligence. Similar results were found among students 15 to 16 years old in physical education (PE) classes, such that perceived competence in PE lessons did not moderate the relation between achievement goals and anxiety and satisfaction in physical education (Ommundsen, 2001). Even with college students in a statistics course, no significant interaction effect was found between perceived competence and three types of achievement goals (i.e., mastery goals, performance-approach, and performance-avoidance goals) on interest, effort, and academic achievement in the course (Cho et al., 2011).

One study did find evidence for the moderating role of perceived competence in mathematics, but it moderated the relation between mastery goals and adaptive and maladaptive learning strategies (Kaplan & Midgley, 1997), contrary to the theoretical expectation for the moderating role of performance goals. Specifically, the positive relation between mastery goals and adaptive strategy use was stronger among students with high perceived competence than those with low perceived competence. Likewise, an increase in mastery goals was associated with a steeper decrease in maladaptive strategy use for students with lower perceived competence than those with higher perceived competence. We are aware of only one other study (Cho et al., 2018) that found evidence for the moderating role of perceived competence in the relations between performance goals and achievement outcomes: among fourth and fifth grade struggling readers, perceived competence had a positive moderating role against the potential negative effect of performance-avoidance goals on reading. That is, performance-avoidance goals more negatively predicted reading comprehension when reading self-efficacy was lower.

2.5. Summary

Prior research supports the idea that achievement goals and perceived ability (conceptualized as perceived competence or self-efficacy) are important predictors of academic outcomes including domain interest, achievement, and persistence. However, there is little support in the prior literature for Dweck’s (1986; Dweck & Leggett, 1988) hypothesis that perceived ability moderates the effects of performance goals on subsequent patterns of behavior. As we describe below, we sought to test the potential moderating role of perceived ability on the relation of achievement goals to academic outcomes more rigorously by examining this moderating effect as a latent interaction. In doing so, we assessed a domain-specific indicator of perceived ability (i.e., science academic self-efficacy), which is comparable to a domain-specific level of perceived competence (Marsh et al., 2019). This approach allows us to easily interpret the findings across the prior and current studies.

3. The Present Study

Our goal in this study was to return to the original theorizing of Dweck’s social-cognitive model (Figure 1). We speculate that a few factors might have hindered the identification of interaction in prior work. First, little research has focused on the relations among all three key constructs (mindsets, achievement goals, perceived ability) in a single model. For instance, testing only the interaction between perceived ability and achievement goals fails to take into consideration the potential influences of mindsets, thereby testing only a part of Dweck’s social-cognitive model (e.g., Cho et al., 2018). Second, most prior research has not accounted for measurement error while estimating the relations among these constructs, which may result in a magnification of measurement error when testing interactions because each main variable is multiplied to create interaction terms (Nagengast et al., 2011). To address these concerns, we treated crucial motivational variables as latent variables to estimate their relations utilizing structural equation modeling with latent interaction terms, an advanced statistical technique better suited for testing interactions (Marsh et al., 2004).

3.1. Proposed Model

Our proposed model (see Figure 2) was designed to test how mindsets (2nd year, T1) related to achievement goals (3rd year, T2), which in turn related to science outcomes (4th year, T3), as well as how achievement goals related to science outcomes depending on the level of science academic self-efficacy (3rd year, T2). Since Dweck’s social-cognitive model suggests only two types of achievement goals—mastery goals and performance goals—we also included these two types of goals, which was also supported by the statistical decisions (see Measures). The outcome variables included students’ science interest value as well as GPA in science-related courses and the number of science-related courses. Interest value, GPA, and course completion reflect students’ affective, achievement, and choice behaviors, respectively. Including these different kinds of persistence indicators provides a richer portrait into the persistence of science students explained by social-cognitive processes. Furthermore, including self-reported measure of interest value as well as behavioral indicators of GPA and course completion provides a robust measurement of persistence in science. Finally, we controlled for students’ gender and math SAT scores by specifying them as predictors of these three outcomes.

Figure 2. Conceptual Model Depicting the Hypothesized Relations among the Key Variables.

Note. This hypothesized model was specified by building upon Dweck’s social-cognitive model as depicted in Figure 1, where straight directional arrows represent regression paths. Science academic self-efficacy was set as a moderating variable in the relation between each type of achievement goals and the three science-related outcomes.

3.2. Hypotheses

First, we hypothesized that fixed mindset would positively relate to performance goals, while growth mindset would positively relate to mastery goals. Growth mindset was also hypothesized to positively relate to science academic self-efficacy and science outcomes, including interest value, GPA in science-related courses, and the number of science-related courses completed.

In terms of the outcomes of achievement goals, we hypothesized that mastery goals would be positively related to all three science outcomes. Performance goals were also expected to relate positively to all outcomes, but we expected smaller effect sizes than mastery goals. Science academic self-efficacy was hypothesized to positively relate to all three science outcomes. In addition, we hypothesized that there would be significant indirect effects of mindsets on outcomes via achievement goals and science academic self-efficacy. Specifically, growth mindset would be positively associated with mastery goals and science academic self-efficacy, which in turn would relate positively to science outcomes. Likewise, fixed mindset would relate positively to performance goals, which in turn would relate positively to science outcomes.

Finally, regarding the interaction effects, we hypothesized that science academic self-efficacy would moderate the relation of achievement goals to science outcomes, particularly for performance goals, as suggested by Dweck’s model. Thus, we expected that higher science academic self-efficacy would strengthen the positive relations between performance goals and outcomes and lower science academic self-efficacy would reduce or reverse these relations. Conversely, science academic self-efficacy would not interact with mastery goals because the outcome patterns of these goals are expected to be consistent regardless of science academic self-efficacy level, as theoretically expected.

4. Method

4.1. Participants and Procedure

The sample for this study was drawn from an ongoing, multi-cohort intervention study at a highly selective university in the Southeastern United States. This larger study examines motivation and persistence in science among undergraduate students and received Institutional Review Board approval from the last author’s former and current institutions (IRB numbers are hidden for review). All procedures in this study were conducted in accordance with ethical principles and legal regulations. Students in the intervention (n = 193) were excluded from this study because the central motivation constructs in this study were targeted by the intervention.

Using a longitudinal, cohort-sequential study design¹, participants were invited to complete a baseline survey during the fall semester of their first year of college and follow-up surveys during the spring semester of their subsequent undergraduate years. Additionally, participants were invited to complete three post-graduation surveys one, two, and seven years after graduation. For the current study, we utilized data from students’ follow-up surveys collected in the second to fourth years.

Recruitment took place in first-year undergraduate chemistry courses that were mandatory for science majors, with permission from course instructors. Students aged 18 and above in their first college year were eligible; those under 18 (n = 56) were invited to participate upon turning 18. Out of 2,581 students in the recruitment courses, n = 1,934 (75%) agreed to participate in the study. All participants received $10 as compensation for each survey completed. From the larger pool of students who completed the first-year baseline survey, a comparison group was randomly selected from those not involved in the intervention, with stratification to oversample women and students from underrepresented ethnic and racial groups. This subsample of non-intervention participants (n = 1,036) received annual invitations for follow-up surveys via email during their second (T1), third (T2), and fourth (T3) years, resulting in a one-year interval between the two time points. Non-participation in a follow-up survey did not preclude invitations to future surveys.

Therefore, the final sample for this study included the entire comparison group of 1,036 undergraduate students (58% female²; 43% Asian, 25% White, 13% African American, 9% Hispanic, and 10% multiracial). Of the sample of 1,036 non-intervention participants, 49% (n = 509) completed the second-year survey (T1), 44% (n = 458) completed the third-year survey (T2), and 46% (n = 476) completed the fourth-year survey (T3). Of the 1,036 students included in this study, 34% (n = 353) completed both T1 and T2 surveys, 33% (n = 344) completed both T1 and T3 surveys, 33% (n = 345) completed both T2 and T3 surveys, and 28% (n = 286) completed all three surveys. Results of missing data analyses are detailed in Data Analyses and Supplemental Materials.

4.2. Measures

Data included survey responses and institutional records. All survey items except mindsets (see below) were assessed using a 5-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree). All items are presented in the supplemental materials.

4.2.1. Mindsets (T1)

Students responded to four items each for fixed mindset (α = .914) and growth mindset (α = .913) of general intelligence (Dweck, 1999). A six-point Likert scale was used, ranging from 1 (strongly disagree) to 6 (strongly agree). A confirmatory factor analysis (CFA) demonstrated that the model fit with these two mindsets modeled as separate latent factors was adequate³, χ²(19) = 53.85, p < .001, CFI = .980, RMSEA [90% CI] = .061 [.042, .081], SRMR = .033 (standardized factor loadings > .748). We tested an alternative CFA model with one factor (i.e., a single factor including the reversed scores on growth mindset and the raw scores on fixed mindset); the fit was substantially worse than the two-factor model, χ²(20) = 1115.10, p < .001, CFI = .365, RMSEA [90% CI] = .335 [.319, .352], SRMR = .102 (standardized factor loadings > .688), supporting the separability of fixed mindset and growth mindset.

4.2.2. Science Academic Self-Efficacy (T2)

Students responded to five items for science academic self-efficacy from the Patterns of Adaptive Learning Survey (PALS; Midgley et al., 2000), which assesses students’ confidence in doing science-related coursework (α = .891). This measure was used as an indicator of perceived ability in science in the current study. A CFA indicated the one-factor model fit was adequate, χ²(5) = 7.28, p = .200, CFI = .996, RMSEA = .032 [.000, .077], SRMR = .013 (standardized factor loadings > .752).

4.2.3. Achievement Goals (T2)

Students responded to the items about three different types of achievement goals in science including mastery(-approach) (5 items; α = .838), performance-approach (5 items; α = .912), and performance-avoidance (4 items; α = .846) goals, measured using PALS (Midgley et al., 2000). We conducted two CFAs, one with three types of goals (i.e., 3-factor model) and the other with mastery goals and combined performance goals (i.e., 2-factor model). Both models demonstrated adequate fit, including the 3-factor model, χ²(74) = 154.57, p < .001, CFI = .973, RMSEA = .049 [.038, .060], SRMR = .043 (standardized factor loadings > .634), and the 2-factor model, χ²(76) = 172.88, p < .001, CFI = .967, RMSEA = .053 [.042, .063], SRMR = .044 (standardized factor loadings > .634). For the model conciseness and theoretical consistency, we selected the 2-factor model.

4.2.4. Science Outcomes (T3)

We assessed students’ science interest value, science GPA, and the number of science courses taken. For interest value, students responded to five items assessing the degree to which they like and enjoy science (Conley, 2012; α = .924). A CFA indicated the one-factor model fit was adequate, χ²(5) = 30.01, p < .001, CFI = .986, RMSEA = .102 [.069, .139], SRMR = .017.

As an indicator of academic achievement in science, we used students’ average GPA across all graded science courses completed from their spring semester of third year to the spring semester of fourth year. Science courses included the required courses for science majors at the university, and elective courses in the domain of science such as chemistry, biology, or physics. We also summed the total number of science courses that students had completed during the same period as their science GPA was assessed; this was used as the last outcome variable. Students’ GPA and the list of completed courses were obtained from institutional records.

4.2.5. Covariates

To control for students’ prior achievement, we used students’ SAT math scores, as math SAT scores have commonly been used as a prior achievement indicator in prior research on college students’ science persistence and SAT scores in general have good predictive validity for college outcomes (e.g., GPA, college retention). We also used gender as a covariate given the gender differences in science-related outcomes (Miller et al., 2015). Students’ math SAT scores and gender (female = 1; male = 0) were collected from institutional records. These covariates were specified as predictors for three science-related outcomes, including interest value, science GPA, and science course completion.

4.3. Data Analyses

We tested a structural equation model (SEM) using Mplus v.8. Within the SEM framework, latent interactions between science academic self-efficacy and each type of achievement goal (i.e., mastery goals and performance goals) were tested as predictors of three outcomes. Testing latent interaction allows researchers to estimate measurement error and provides unbiased estimates of interaction effects, distinct from conventional multiple moderated regression analyses. Before the analysis, we standardized all indicators, and created product indicators for the latent interactions based on the preliminary CFAs, using the unconstrained approach and matched pair-strategy (Marsh et al., 2004; see Supplemental Materials). We also tested indirect effects of mindsets on different outcomes via each type of achievement goal and science academic self-efficacy. We used a robust maximum likelihood estimator (MLR), which has commonly been used in testing latent interaction because it provides standard errors that are corrected for the potential non-normality of the product-indicators.

Finally, our analysis of the missing data suggests that the mechanism of missing data is considered missing at random (MAR), as the missingness is largely associated with other observed variables (Graham, 2009; see Missing Data Analyses in the supplemental materials). To reduce potential bias in the estimates, we accounted for these patterns of missingness in main analyses by including all relevant variables including gender, math SAT scores, science GPA, and the number of science courses completed within the models. This approach aids with full information maximum likelihood (FIML) estimation when the data falls under MAR (Enders, 2010).

5. Results

5.1. Correlations

The correlational patterns were generally consistent with our expectations (Table 1). Fixed and growth mindsets were negatively correlated with each other (r = −.62). Fixed mindset was negatively correlated with science academic self-efficacy (r = −.14) whereas growth mindset was positively correlated with it (r = .20). Fixed mindset was positively correlated with performance goals (r = .16), whereas growth mindset was positively correlated with mastery goals (r = .13). Fixed mindset was not significantly correlated with any of the science outcomes, but unexpectedly, growth mindset was negatively and weakly correlated with science GPA (r = −.15).

Table 1.

Means, Standard Deviations, and Correlations of Observed Variables

	1	2	3	4	5	6	7	8	9	10
1. Gender (0 = male; 1 = female)	-
2. Math SAT	−.12**	-
3. T1 Fixed mindset	−.07	.10*	-
4. T1 Growth mindset	−.04	−.15**	−.62***	-
5. T2 Science academic self-efficacy	−.18***	.10	−.14**	−.20***	-
6. T2 Mastery goals	−.13**	.05	−.09	.13*	.47***	-
7. T2 Performance goals	−.14**	.07	.16**	−.04	.14**	.32***	-
8. T3 Interest value	−.12*	.19***	−.04	.07	.34***	.50***	.20***	-
9. Science GPA	.06	.36***	.06	−.15**	.10	−.01	.003	.12*	-
10. Science course completion	−.13***	.02	−.08	.02	.11*	.27***	.06	.29***	.26***	-
M	0.58	735.89	2.88	3.84	3.87	4.03	2.94	4.17	3.46	3.81
SD	0.49	63.89	1.12	1.01	0.73	0.63	0.88	0.70	0.58	3.61

Note. The possible maximum value of math SAT scores was 800. Mindsets were measured based on a 6-point scale, and science academic self-efficacy, achievement goals, and interest value were measured based on a 5-point scale. Science GPA was based on the 4.0 scale.

^*p < .05

^**p < .01.

Science academic self-efficacy was more strongly correlated with mastery goals (r = .47) than performance goals (r = .14). Science academic self-efficacy was also positively correlated with interest value (r = .34) and science course completion (r = .11) but not significantly correlated with science GPA. Mastery goals were more strongly correlated with interest value (r = .50) than science course completion (r = .47), but these goals were not significantly correlated with science GPA. Performance goals were only positively correlated with interest value (r = .20), not with science GPA and science course completion. All three science outcomes were positively correlated with one another (r’s = .12 to .29).

5.2. Testing Dweck’s Model in the SEM Framework

Results from the final model are presented in Figure 3. The model fit was adequate: χ²(947) = 1575.23, p < .001, CFI = .928, TLI = .922, RMSEA = .031 [90% CIs = .028, .033], SRMR = .058. In line with our hypothesis, fixed mindset was positively related to performance goals (β = .25) whereas growth mindset was positively related to mastery goals (β = .17). Neither mindset was significantly related to science academic self-efficacy. Mastery goals, in turn, positively related to interest value (β = .60) and science course completion (β = .28) but did not significantly relate to science GPA. Neither performance goals nor science academic self-efficacy were significantly related to any of the science-related outcomes. There were also no significant interaction effects between mastery goals and science academic self-efficacy. In contrast to our hypotheses, the relation of performance goals to the science outcomes was not significantly moderated by science academic self-efficacy.

Figure 3. Results from the Final Structural Equation Model.

Note. Statistically significant standardized path coefficients are presented. Only one indirect effect, the effect of growth mindset on interest via mastery goals, was statistically significant, which is depicted as the curved directional arrow. Dotted lines represent statistically nonsignificant paths. SAS = science academic self-efficacy; Mastery = mastery goals; Perf = performance goals.

*p < .05; **p < .01; ***p < .001.

One significant indirect effect was found. Growth mindset had a positive indirect effect on interest value (β = .10) via mastery goals. For covariates, SAT math scores were positively related to interest value (β = .002) and science GPA (β = .01), and female students were more likely to complete more science courses than their male peers (β = .29). Figure 3 presents standardized coefficients for all significant paths, and Tables S2 and S3 in the supplemental materials present direct effects and indirect effects, respectively.

The effect sizes (f²) calculated based on R² values. According to Cohen’s (1988) guidelines (i.e., f² ≥ 0.02, f² ≥0.15, and f² ≥ 0.35 represent small, medium, and large effect sizes, respectively), the effect sizes for interest value (f² = .79) and science GPA (f² = .18) explained by the model were interpreted as large and medium, respectively. Specifically, the variance in interest value explained (R² = .44, p < .001) was notably larger than that in science GPA (R² = .15, p < .001), although these effect sizes were also derived from the predictions of covariates. The effect size for science course completion (f² = .14) was close to medium based on the explained variance in the variable (R² = .12, p < .001). For mastery goals (R² = .02, p = .270), performance goals (R² = .04, p = .096), and science self-efficacy (R² = .03, p = .234), the effect sizes were f² = .02, f² = .04, f² = .03, respectively, which all indicated the small effect sizes.

6. Discussion

6.1. Testing Dweck’s Social-Cognitive Model in Science

Few studies found the significant moderating role of perceived ability in the association between performance goals and behavior patterns, as theoretically hypothesized by Dweck’s social-cognitive model (Dweck, 1986; Dweck & Leggett, 1988). In the current study, we aimed to test Dweck’s social-cognitive model using a latent moderated structural approach, while focusing on college students’ motivational processes in science based on their mindsets, science academic self-efficacy, science achievement goals, and science-related outcomes. Although we did not identify statistically significant latent interactions between science academic self-efficacy and performance goals, we observed theoretically and practically meaningful paths that provide implications for supporting students in their pursuit of science careers through motivation.

First, we expected fixed mindset to be positively related to performance goals, and growth mindset to be positively related to mastery goals, which was supported by our data. The association of fixed mindset with performance goals and that of growth mindset with mastery goals were consistent with Dweck’s theorizing (Figure 1) as well as empirical evidence (Burnette et al., 2013; Dinger & Dickhäuser, 2013). These findings confirm the role of mindsets as antecedents of achievement goals, and the current study, in particular, suggests that students’ mindset regarding general intelligence relates to science-specific achievement goals.

We also hypothesized that growth mindset would be positively related to science academic self-efficacy, as well as other science-related outcomes. In our findings, however, growth mindset was not directly associated with subsequent science academic self-efficacy, science interest value, science GPA, and the number of science courses completed. These non-significant relations of growth mindset to the outcomes were different from prior findings that reported the significant associations between growth mindset and outcomes (Burnette et al., 2020; Dai & Cromley, 2014; Lee et al., 2022; Yu & McCllan, 2020). However, there have also been claims that the associations expected by the theory are not always found (Li & Bates, 2020), or if found, have very small effect sizes (Lou & Li, 2023; Sisk et al., 2018; Macnamara & Burgoyne, 2023). Consequently, there is a call for a greater understanding of potentially heterogeneous effects of mindsets on outcomes across samples and contexts (Hecht et al., 2023; Tipton et al., 2023; Walton & Yeager, 2020; Yeager & Dweck, 2020). In fact, mindset theory primarily concerns responses to challenges or setbacks rather than academic achievement in general. Therefore, attempting to explain the variance in students’ science motivation and achievement in general situations, rather than focusing on the variance in behavior in specific challenging situations, may be one of the reasons for the non-significant findings in the current study. Another reason may be related to the significant indirect effect of growth mindset on the outcomes through mastery goals. In other words, growth mindset may likely contribute to student outcomes through more specific forms of motivation in a certain domain, rather than predicting them directly. The significant direct effects of mastery goals are detailed below.

Regarding achievement goals, our hypotheses were only partially supported, as only mastery goals—not performance goals—positively related to science interest value and the number of science-related courses completed. Prior research has generally shown that mastery goals are positively associated with academic-related outcomes (Hulleman et al., 2010; Van Yperen et al., 2014; Wirthwein et al., 2013). These patterns from prior findings may be reflected in our results, as mastery goals were significantly related to the outcomes, except for cumulative science GPA in the current study. The positive relation between mastery goals and interest value has been well-documented by empirical evidence, including findings with meta-analysis (Baranik et al., 2010; Huang, 2011; Hulleman et al., 2010; Scherrer et al., 2020). Similarly, the positive association between mastery goals and science course completion is expected, supported by empirical findings demonstrating generally positive educational outcomes of mastery goals, including engagement and perseverance (Cho et al., 2019; Millet et al., 2021; Yu & McLellan, 2020). Additionally, the definition of mastery goals, focusing on mastering learning materials, is closely linked to completing coursework or not giving up on coursework, contributing to science course completion. Together, the current findings regarding the outcomes of achievement goals suggest that encouraging mastery goals, rather than performance goals, may be more beneficial for science-related outcomes. Nonetheless, it is important to note that one reason for the more beneficial effects of mastery goals compared to performance goals is that performance goals in this study comprised both performance-approach and performance-avoidance goals.

In terms of the outcomes of science academic self-efficacy, unexpectedly, it was not significantly related to any of the science outcomes in the current study. This non-significant finding differs from prior work that demonstrated students with higher science academic self-efficacy tend to have greater science interest, achieve higher scores in these courses, and enroll in more science courses (Alhadabi, 2021; Honicke & Broadbent, 2016; Perez et al., 2014; Stets et al., 2017). In the bivariate correlations of observed variables in the current study, science academic self-efficacy was also significantly correlated with interest value (r = .34) and the number of science course completed (r = .11). However, when mastery goals were included in the structural equation model, science academic self-efficacy no longer related to the outcomes. This indicates that mastery goals, which still significantly related to the outcomes, may be stronger predictors than science self-efficacy, at least for interest value (Harackiewicz et al., 1997), as well as course completion. This finding aligns with prior findings of task values being a stronger predictor of course completion than self-efficacy (Linnenbrink-Garcia et al., 2018). Like task values, achievement goals also address the “why do I want to do this?” question, which is more related to persistence and choice behavior, such as interest value and course completion in the case of this study.

Different from our hypothesis, no significant interaction effect emerged between performance goals and science academic self-efficacy on the outcomes, also found in most of the prior empirical evidence (Cho et al., 2011; Cury et al., 2006; Ommundsen, 2001). The prior and current findings taken together suggest that such a non-significant moderating effect for perceived ability seems to be quite robust, regardless of study design, including the age of participants, analytic method, or inclusion of covariates. Our findings help to extend this prior literature by confirming the non-significant interaction effects even though we designed the present study to probe for them by including all key variables in the same model, creating latent variables to reduce measurement error, considering temporal ordering of variables for potential causal relations, and ruling out the possibility of the confounding effects of gender and prior achievement. Also, given the grounding of our models in Dweck’s theoretical hypotheses, neither methodological nor theoretical characteristics of the study are likely reasons for the lack of interaction effects. As such, our findings bring into question one of the main hypotheses in Dweck’s social-cognitive model, namely that the relation of achievement goals to patterns of behaviors is moderated by perceived ability.

6.2. Practical Implications

Our findings suggest that establishing a growth-mindset culture in the classroom may be the most effective way to enhance mastery goal endorsement in college science, furthering the positive roles of mastery goals in science outcomes. Current interventions aimed at supporting growth mindset (reduced fixed mindset) have shown small but promising results, particularly for academically at-risk and economically-disadvantaged students (Sisk et al., 2018; Yeager & Dweck, 2012). However, it is also true that the effectiveness of mindset interventions has been questioned, and their implementation in all classrooms is not always feasible (Hecht et al., 2023; Li & Bates, 2020; Lou & Li, 2023; Walton & Yeager, 2020). Instead, it may be more realistic to focus on teachers’ comments and actions that could impact students’ mindsets.

In light of this, some researchers have investigated teaching practices that contribute to the establishment of growth mindset culture in the classroom (Kroeper et al., 2022a, 2022b; Seo & Lee, 2021). These practices include conveying messages about everyone’s capability for success, providing opportunities for additional practice, responding to struggling students with supportiveness, and emphasizing the values of learning processes and development. These behavioral cues provided by professors may be even more critical among science students, as science, and STEM fields more broadly, are often perceived to be difficult and requiring fixed, innate talent for high-achieving performance (Leslie et al., 2015). Such cues are expected to have downstream consequences for achievement goal endorsement and STEM persistence (e.g., Hulleman & Harackiewicz, 2009; Linnenbrink-Garcia et al., 2018).

It is also possible to provide a mastery-oriented context directly in college science. University professors could work to create mastery-structured undergraduate courses using the TARGET framework (i.e., task, authority, recognition, grouping, evaluation, and time; Ames, 1992). Indeed, college students’ mastery goals were more adopted when more engaged in lectures and discussion, perceiving autonomy in activities, and experiencing an environment without a focus on evaluation or harsh evaluations (Church et al., 2001; Karabenick, 2004; Lerdpornkulrat et al., 2018). In college science, often believed to be a highly demanding and competitive discipline (Canning et al., 2019; Muenks et al., 2020), it may be critical to guide students toward pursuing goals for mastering and understanding the learning materials themselves, rather than outperforming others.

6.3. Limitations

We note several limitations for future research. First, we focused on mindsets at a domain-general level, while other motivational variables were measured at a domain-specific level (i.e., science). The different measurement levels of these variables might hinder some significant findings. For example, science-specific outcomes might be expected to have stronger relations with mindsets about ability in science. Therefore, it is essential to interpret the current findings with caution, given the assessment of variables at different levels (domain-specific versus domain-general).

Related to this point, it is crucial for future research to validate our proposed models in other domains to assess generalizability of the findings. Although our focus was on the science domain, we view that students’ behavior patterns influenced by achievement goals, perceived ability, and mindsets can manifest in other subject domains, provided that students encounter challenges in academic tasks. Consequently, we suggest testing Dweck’s model in other academic disciplines with the same measurement levels for these constructs.

Second, our model did not account for the stability of each construct, as the repeated measures were not included. Utilizing cross-lagged modeling or controlling for the previous levels of each construct could enhance the validity of findings. Finally, we did not consider the potential cultural differences in the effects of mindset type (Huang et al., 2022), which is an important question for future research.

6.4. Conclusion

In this study, our aim was to evaluate the applicability of Dweck’s social-cognitive model in a real college science context, with a focus on testing the theoretical model using advanced statistical strategies. Explaining students’ science-related outcomes by their mindsets and achievement goals, which exhibit some stability reflecting individual differences (Kaplan & Maehr, 2007; Molden & Dweck, 2006), may provide a valuable contribution to the understanding of how students’ personal mindset and goal orientations impact their behavior patterns within the academic discipline of science. More importantly, however, as more recent research suggests the changeability of mindset beliefs (Dai & Cromley, 2014; Lee & Seo, 2019) and achievement goals (Lerdpornkulrat et al., 2018), the current study proposes the need for establishing a growth mindset atmosphere and mastery-structured educational contexts.

Overall, our findings partially supported Dweck’s original social-cognitive model. The current data did support aspects of Dweck’s model that highlight the importance of increasing growth mindset and mastery goals in promoting science persistence. This study is the first to test the moderating role of perceived ability in Dweck’s social-cognitive model using a latent moderated structural approach. However, we found no evidence of a significant interaction between science academic self-efficacy and achievement goals in predicting science outcomes, in line with previous empirical work. Our findings also informed practice by providing insights for how to intervene in support of students who are pursuing science-related career in college.

Highlights.

• Dweck’s social-cognitive model was tested in a college science context.

• Fixed mindset related to performance goals.

• Growth mindset related to mastery goals and the number of courses completed.

• There was an indirect effect of growth mindset on interest value via mastery goals.

• No latent interaction effect between self-efficacy and achievement goals appeared.

Educational Relevance and Implications Statement.

Understanding how students’ beliefs about the nature of ability, as innate and stable (fixed mindset) versus malleable and changeable (growth mindset), relate to interest, achievement, and persistence in science may help to better understand and address the long-lasting problem of high attrition out of undergraduate science degrees. Our findings highlight the importance of supporting increased endorsement of growth mindset and mastery goals (i.e., goals focusing on developing competence) among undergraduate students. It may be possible to implement teaching practices that contribute to the establishment of growth mindset culture in the classroom, such as conveying messages about everyone’s capability for success, providing opportunities for additional practice, responding to struggling students with supportiveness, and emphasizing the values of learning processes and development. University professors could also work to create mastery-structured undergraduate courses by providing students with opportunities to engage in lectures and environments without a focus on harsh evaluations.