The United States faces several significant STEM workforce and educational challenges. First, the demand for well-trained STEM workers continues to outpace the supply [Bureau of Labor Statistics, 2014; Office of Innovation and Improvement (OII), 2016]. Second, the majority of U.S. students continue to perform at or below proficiency levels in math and science across elementary, middle, and high school (National Academies of Sciences, Engineering, and Medicine, 2015; Organization for Economic Co-operation and Development, 2018). For example, only 36% of 4th graders reach proficiency levels in science and this rate drops to 22% in 12th grade [U.S. Department of Education, Institute of Education Sciences, National Center for Education Statistics, National Assessment of Educational Progress (NAEP), 2019]. From the 2015 to 2019 NAEP, data from 8th grade and 12th grade remained constant, but in fourth grade, a higher percentage scored below NAEP Basic compared to 2015, indicating scores were moving in the wrong direction even before the COVID-19 pandemic. Third, data from the National Student Clearinghouse Research Center on the impacts of the COVID-19 pandemic suggests that total postsecondary enrollment in 4-year schools declined by 3.05 percent in fall 2021, with STEM majors being most adversely impacted. Enrollment was down 7.2% in Physical Science, 5.6% in Mathematics and Statistics, and 4.2% in Biological and Biomedical sciences (National Student Clearing House Research Center, 2021) and there is evidence that enrollment rates in 2022 are worsening (National Student Clearing House Research Center, 2022).

Fourth, achievement disparities spanning racial/ethnic, socioeconomic, and gender lines emerge in elementary and widen over time (Reardon, 2011; U.S. Department of Education, Institute of Education Sciences, National Center for Education Statistics, National Assessment of Educational Progress (NAEP), 2019). For example, in the 2019 NAEP results, 4th grade White students scored an average of 163 on the science test while Black students scored 130. Similar disparities were observed between students identified has having a disability and those not (average score of 124 and 156 respectively), and students identified as English Language Learners those not identified as English Language Learners (average score of 122 and 155). The same trends are magnified in 8th and 12th grade; 12th-grade English Language learners scored an average of 99. These trends directly contribute to a continued lack of diversity in STEM occupations, as highlighted in a recent report produced by the Pew Charitable Trusts (Fry et al., 2021). Given the current trends with educational disparities worsening during the COVID-19 pandemic [U.S. Department of Education, Institute of Education Sciences, National Center for Education Statistics, National Assessment of Educational Progress (NAEP), 2022], it is probable that these disparities will worsen as well.

A significant body of work has shown that the beliefs and perceptions we hold about ourselves and our competence in a variety of domains (e.g., school, sports, physical appearance) influence our actions and behaviors. Self-concept and self-efficacy are two of the most well-researched self-perceptions, both formed and influenced by one’s experiences and interactions with the world. Self-concept refers to one’s awareness and understanding of their own self and attributes (Shavelson et al., 1976), whereas self-efficacy refers to an individual’s belief that he or she possesses the capabilities to accomplish a goal and successfully carry out a particular course of action (Bandura, 1977). While these two self-perceptions have historically been studied as separate constructs, recent work by Marsh and colleagues demonstrated the “murky” empirical distinction that exists between these two constructs, reporting that scores on measures of self-efficacy and self-concept were highly correlated in a sample of German 11 to 12-year-olds (r > 0.90; Marsh et al., 2019a, p. 331). Yet while observing that these two self-perceptions are closely related, Marsh et al. (2019a) also notes an important conceptual distinction between them: self-efficacy tends to be future-oriented (probing what a person can do in the future) whereas self-concept is based on past achievements and accomplishments.

Both self-concept and self-efficacy are conceptualized to have multifaceted hierarchical structures, with a global, general sense of self at the top and academic and nonacademic domains nested underneath (Marsh and Martin, 2011; Marsh and Scalas, 2011) that are further differentiated into subdomains (e.g., academic self-concept has subdomains in English, history, math, and science self-concepts). Academic self-perceptions have been a core domain of investigation for decades (e.g., Marsh et al., 2019b). Academic self-concept refers to a person’s knowledge and perceptions about their own academic competencies (Wigfield and Karpathian, 1991; Ferla et al., 2009). Self-concept beliefs are more general self-perceptions that can include evaluations of self or affective evaluations of one’s feelings. For example, “I am good at science” or “I look forward to science” (Beier and Rittmayer, 2008). Academic self-efficacy, on the other hand, refers to a student’s beliefs that s/he can master the knowledge and skills necessary to succeed in an academic discipline (Schunk, 1991; Ferla et al., 2009). As such, academic self-efficacy is related to both task-specific and future expectations. Academic self-efficacy judgments are often made without consideration of others’ capabilities but rather are focused on oneself (Gist and Mitchell, 1992; Bong and Skaalvik, 2003).

Decades of research have shown that self-perceptions significantly influence a person’s functioning and well-being both in school and beyond. In academic settings, these self-perceptions can predict and explain students’ motivation, behaviors, and achievement (Bong and Clark, 1999). For example, individuals with high self -efficacy in a particular academic subject, like math, tend to engage in activities and tasks they feel capable of accomplishing while avoiding those in which they feel less efficacious. Self-efficacy also influences how much people persist and persevere in challenging situations (Lent et al., 1984; Schunk, 1995). Individuals with high self-efficacy are also more likely to persist in the pursuit of desired outcomes and tend to be more resilient to setbacks. As such, academic self-efficacy influences a student’s engagement and persistence in academic activities and the resulting academic outcomes (Schunk and Pajares, 2002).

The same is broadly true of research on self-concept; studies in educational settings have shown that having a positive academic self-concept (e.g., feeling you are good at school) facilitates educational behaviors (persistence, course-work selections; Eccles, 2005), and academic achievement and attainment (Guay et al., 2004; Marsh and Craven, 2006; Pinxten et al., 2010; Chen et al., 2013; Trautwein and Möller, 2016). Meta-analyzes (Valentine et al., 2004; Huang, 2011) and longitudinal studies (Marsh and O'Mara, 2008) have also shown that academic achievement and self-concept are positively and reciprocally related to each other.

Research has shown the subdomains of academic self-perceptions are distinct (Marsh and Martin, 2011; Marsh and Scalas, 2011), suggesting that one can perceive oneself as more capable in in one subject than another. A handful of studies have explored students’ self-perceptions in STEM subject areas and have shown that individuals with high STEM self-perceptions perform better and persist longer in STEM disciplines relative to those with lower self-perceptions in STEM. For example, science self-perceptions and beliefs are related to STEM persistence and achievement in middle school (Britner and Pajares, 2006) and beyond. Science self-efficacy predicts a variety of academic behaviors and outcomes, including science grades in middle school (Britner and Pajares, 2001), science capacities in high school (Jansen et al., 2015a), science performance and persistence in college students (Andrew, 1998; Byars-Winston et al., 2010; Hanauer et al., 2016). Heilbronner (2011) found that self-efficacy beliefs in one’s ability to achieve in STEM predicted the choice of STEM majors in college. Students who believed that they could excel in STEM were more likely to graduate with a STEM degree compared with those of lacked these beliefs. Similarly, high-school students who express positive science self-efficacy beliefs tend to persist in STEM majors compared to those who do not express these beliefs (Shaw and Barbuti, 2010). Recently, a handful of studies has shown that self-concept in the sciences is distinguishable from self-concept in math and language (Jansen et al., 2014; Marsh et al., 2015; Jansen et al., 2015a, 2019) and that students hold differentiated beliefs across subjects such as chemistry, biology, and physics.

Evaluation of the extant literature shows several prominent gaps in the literature that limit our understanding of how science self-concept and self-efficacy develop over time. First, the majority of research has primarily focused on high school and college student populations. We were unable to find any research that examined science self-perceptions in elementary student populations, which represents a key period when students begin formal science education and influence burgeoning self-perceptions in this academic domain. Second, this age-related gap in research is likely due to the lack of developmentally-appropriate science self-perception measures. Third, research often probes general STEM self-perception beliefs rather than specific beliefs about science, and evidence suggests that students’ self-perceptions may, indeed, vary by subject. Understanding and predicting how motivational factors influence educational outcomes in the sciences is seen as critical in both research and educational practice and policy (Osborne et al., 2003; Organization for Economic Co-operation and Development, 2007; Bybee and McCrae, 2011; Taskinen et al., 2013; Jansen et al., 2015b). Despite the value of such knowledge, we currently lack age-appropriate, comprehensive measures to probe science self-perceptions in elementary school students.

Students’ motivation to achieve in a particular subject is influenced not only by their self-perceptions but also by the value that they assign to a particular area (Wigfield and Eccles, 2000). The most common type of value belief is individual interest in a given topic or area. With respect to STEM, students’ interests in STEM fields and STEM jobs influence their pursuit of math and science courses, their performance in these courses, and their entry into STEM majors in college (Lichtenberger and George-Jackson, 2013). For example, interest in math and science in middle school influences students’ selection of high-school math and science courses (Farenga and Joyce, 1998; Köller et al., 2001). Students who express interest in STEM in eighth grade are up to three times more likely to ultimately pursue STEM degrees later in life than students who do not express such an interest (The President’s Council of Advisors on Science and Technology (PCAST), 2010). This relationship extends to high school and college. Harackiewicz et al. (2002) reported that interest predicts college students’ choice of academic major, and that interest combines with external factors to predict academic outcomes. Similarly, Maltese and Tai (2011) found that high-school students’ interest in mathematics and science is more predictive of entry into a STEM major than achievement test scores. Like science self-perceptions, less is known about elementary students’ interests in science-related activities in and out of school and how these interests are associated with future careers. Understanding interest in early elementary science is critical, as these interests have the potential to influence engagement in future STEM courses and informal learning opportunities throughout elementary, middle, and high school.

There is agreement amongst researchers, educators, and policymakers that expanding access to high-quality STEM education has the potential to reduce inequities in education and beyond. Efforts to improve STEM education in the U.S. were well underway prior to the pandemic (National Science and Technology Committee (NSTC), 2018), and included (1) innovative standards for science literacy (e.g., the Next Generation Science Standards), (2) novel recommendations to maximize science teachers’ instructional effectiveness (National Academies of Sciences, Engineering, and Medicine, 2015), (3) investments in STEM education and programs for underrepresented students, and (4) a research agenda focused on whether these programs can foster the development of students’ self-efficacy, self-concept, interest, and ultimately achievement in STEM fields. The continued effects of the pandemic on education renew the importance of programs aimed to foster high-quality STEM experiences for all students. Yet, understanding what programs lead to changes in students’ STEM attitudes, competencies, and achievement hinges upon having suitable measures to examine change in diverse populations.

In particular, understanding how science self-perceptions and interest in students from historically underrepresented backgrounds in STEM develop and change over time requires measures to be developed and validated with diverse populations. However, many of the measures that are most widely used in psychological research were developed with homogenous samples of White, relatively affluent students (e.g., Stewart et al., 2012). There is evidence that measures display different psychometric properties when administered to samples of students drawn from different sociodemographic and socioeconomic backgrounds, and that these differences have consequences for how we interpret the scores on those measures as well as estimating program intervention outcomes (Nagengast and Marsh, 2014; Dong and Dumas, 2020; Soland, 2021). For this reason, a priority for STEM research is to examine the psychometric properties of measures developed with diverse samples of students, beginning with these measures’ reliability and validity.

In sum, there is a lack of age-appropriate, reliable, and valid measures to assess science-related self-perceptions and interest in elementary student populations from diverse backgrounds. We took the first step to address this gap by developing new age-appropriate measures of students’ science self-efficacy, science self-concept, interest in science activities, and interest in future STEM careers. In this paper, we describe the development and preliminary pilot test of these measures with a sample of third grade students (ages 8–9) representative of a population that has historically been underrepresented in STEM careers (94% of our sample was African American and Hispanic). In this preliminary study, we focused on the measures’ reliability and validity, and addressed three specific research questions: (1) Do the measures demonstrate reliability and construct validity? (i.e., are they internally consistent and measuring what we think they should?), (2) How are scores on each of these measures distributed? and (3) Do the measures demonstrate convergent validity? (i.e., do the scores on these measures relate to one another?).

Participants were 204 third-grade students (54.9% female, 45.1% male; M_age = 9.09 years, SD = 0.39) attending four elementary schools in a single school district. The school district is a large (one of the 25^th largest in the country), urban district that prioritizes science education, even in elementary school. While most students were African American (73.5%), participants included students who were Hispanic/Latino (20.6%), White (3.9%), Asian (1.0%), American Indian (0.5%), or Multiracial (0.5%). Five students (2.5%) were classified by the district as having a disability, and 30 (14.7%) were classified as having limited English proficiency (LEP). Over 40 % of the sample (84 students, or 41.2%) qualified for the district’s free or reduced-price meals program. Eligibility for this program is based on receipt of other benefits (Food Supplement Program or Temporary Cash Assistance) or falling within the Federal Income Eligibility Guidelines, which required that a family’s gross income be no more than 1.85 times the federal poverty level (which was $25,750 for a family of four in the year that data were collected).

Below we describe how we developed a new measure and adapted established measures for middle childhood students for use in this study.

Students completed the battery of surveys during a 30-min session that was administered in the following order: Science Self-efficacy Scale, Science Self-concept Scale, Student Interest in Science Activities (Middle Childhood) Scale, and Student Attitudes Toward STEM Careers Scale (S-STEM Upper Elementary Your Future Survey). The measures were group administered in the classroom by a team of two researchers. Before beginning the surveys, students were encouraged to think about how they (not others) feel about each statement and were assured that there were no right or wrong answers and to answer what is true for themselves. Students were given a single, paper-and-pencil packet of all measures and one of the research team members led students through each measure, reading each statement aloud (while students followed along silently) and pausing for students to circle their response before moving on to the next statement for each measure. The team member read each statement twice pausing between each to ensure students had a chance to follow along and understand the statement. While one member of the research team read each statement according to an experimental script, the other team member walked around the classroom ensuring that students were on the correct statement and circling an answer choice for each statement. Students were told that the team would like to learn more about their feelings about and interests in school and science and that all students will have different answers because everyone has different feelings. See the Supplementary Appendix for specific instructions provided for each measure.

To establish initial evidence for the reliability and validity of these measures, data analysis steps were designed to answer our research questions:

Scores for each scale or sub-scale were calculated as the mean of the completed items. Means, standard deviations, and statistics reporting skew and kurtosis are presented in Table 10, while histograms of the distributions for each scale and sub-scale are presented in Figures 1–4. As can be seen in Figure 1, the distribution of scores for the Science Self-efficacy scale is slightly negatively skewed, indicating more positive feelings of self-efficacy in science. Figure 2 shows the two factors for Science Self-concept: (1) Competence and (2) Feelings about Science (Affect). As can be seen in the figure, the Competence subscale approximates a normal distribution while the Affect subscale exhibited a moderate degree of negative skew, indicating overall positive feelings about science. Figure 3 shows the distribution of scores for the 2 factors of the Student Interest in Science Activities (Middle Childhood) scale. As can be seen in the figures, factor 1: informal/less structured activities approximated a normal distribution while factor 2: formal/structured activities exhibited a slight degree of negative skew. The negative skew for this factor again indicates students displayed more interest in engaging in formal activities around science. Finally, Figure 4 shows the distributions of the three subscales for Student Attitudes Toward STEM Careers. As can be seen in Figure 4, factors 1 Basic/Foundational, and 2, Applied/Service-oriented approximated a normal distribution. For Factor 3, Computational Sciences, the distribution is more spread out, indicating that some students are expressing strong disinterest in these types of careers, while others are expressing interest in them.

When one is developing a measure, it is important to understand how performance on the measure relates to performance on other measures that assess the same or similar construct. This idea, known as convergent validity, explores whether two measures that should be related to each other, are related. For example, we hypothesize that measures of height and weight should be related to each other because as one grows taller one also gains more mass. This is also true of psychological constructs. As self-efficacy and self-concept both “measure” one’s perceptions of self, we hypothesize that performance on one will be closely related to performance on another.

While a larger sample is needed to examine this question for these measures, we can examine this question preliminarily. Thus, we explored the correlations (i.e., associations) among the scales and sub-scales for each measure, using scale and sub-scale scores calculated based on the factor loadings for each confirmatory model. For example, our final confirmatory factor model indicated that our measure of science Self-concept displayed a two-factor structure (rather than a single-factor structure): (1) Perceptions of Competence and (2) Feelings about Science (Affect). For each student we calculated a Perceptions of Competence sub-scale score and a Feelings about Science (Affect) sub-scale score as the mean of the items that loaded on each factor. These scores are reported in Table 11, while the correlations among scales and sub-scales are reported in Table 11.

The five correlations shaded in gray in Table 11 correspond to the correlations between subscales taken from the same measure. As they come from the same measure, we predict that these correlations would be strong and positive. In all cases, the correlations are positive, but they are not all strong. For example, the correlation between the Science Self-Concept Competence subscale (e.g., I get good grades in science) and the Affect subscale (e.g., I look forward to science) is 0.52 (see the gray box from row “2. Science Self-Concept: Competence” and column 3). The relation between these two subscales is not very strong, suggesting that while they both measure a form of science self-concept, the subscales comprising this measure appear to assess related but distinct aspects of self-concept, which also confirm our factor analysis. An inspection of Table 11 reveals this is true for all cases – indicating that sub-scales taken from the same measure appear to assess related but also distinct constructs in all cases.

The remaining correlation coefficients reported in Table 11 correspond to the associations between scales and subscales from different measures, and in each case, the values for these coefficients are also positive. This is perhaps to be expected, given that all the measures administered in this study were designed to assess the same over-arching domain: students’ attitudes about STEM. However, it is also apparent that the strength of these positive associations varies widely, ranging from r = 0.08 to 0.62, reflecting differing degrees of relatedness among measure scales and subscales. For example, the correlation between Science Self-Concept: Competence and Interest towards Basic/Foundational Science Careers (Row 2 and Column 6) is 0.14, suggesting that students’ perceptions of their science competence is not strongly related to their attitudes toward basic/foundational sciences careers.

Our results revealed that the Science Self-Efficacy Scale displayed a single-factor structure, which suggests that a single score can be calculated from this measure to indicate elementary students’ science self-efficacy. This single-factor structure is consistent with measurement work on the structure of self-efficacy (Bandura, 2006), as well as the suggestion that self-efficacy can be divided broad domains (academic and non-academic) and that academic self-efficacy can be further divided into sub-domains (e.g., science self-efficacy, math self-efficacy) that function as single subject-specific constructs. In contrast, the Science Self-Concept Scale displayed a 2-factor structure that included Perceptions of Science Competence and Feelings about Science (Affect). This division replicates and extends previous work that established the multidimensional nature of academic self-concept in elementary school children. For example, Arens et al. (2011) administered Marsh’s academic self-concept questionnaire to 3rd – 6th graders in Germany and found that the self-concept domain was separated into competence and affect components. Here, we parallel these findings and extend them to the domain of science. While previous work has established that academic self-concept can be divided into domain-specific concepts such as verbal and mathematical self-concept (Marsh, 1990; Skaalvik and Rankin, 1990) and that academic self-concept includes both affective and competency-related self-perceptions, to our knowledge, our results are the first to demonstrate that this distinction holds for science among elementary school students.

The Student Interest in Science Activities (Middle Childhood) Scale was designed to examine students’ current interest in science activities in middle childhood. The data collected using this measure yielded 2 underlying factors representing two broad types of science-related activities common in middle childhood. The first factor reflected informal, less structured, self-guided activities, with the exception of item 1 (“Doing science activities in school”). The fact that “Doing science activities in school” loaded on this first factor was surprising, but may be explained, in part, by the way in which students engage in science in the elementary grades in this school district, in which there is greater emphasis on hands-on, exploratory and investigative science activities. This instructional approach may be perceived by students as more self-directed in nature. The second factor reflected more formal, structured activities that are guided by an adult, such as science clubs after school, science camps, and science fairs. To our knowledge, this represents the first measure that explores students’ interest in age-appropriate, science activities commonly associated with middle childhood.

Our analyzes revealed that Student Attitudes Towards STEM Careers Scale displayed a 3-factor structure; Factor 1Basic/Foundational Sciences, Factor II, Applied/Service-Oriented Sciences, and Factor III Computational Sciences. This measure was originally developed by the Friday Institute for upper elementary students, ages 4-5th grade, and the measure codebook does not articulate the number of factors probed by these statements. Therefore, the present study is the first to examine the factor structure of this measure in a diverse sample of 3rd-grade students.

Our results have implications for research in measurement development and offer guidance for using these measures to evaluate program effectiveness.

This paper establishes preliminary evidence for the reliability and validity of our measures and additional work should attempt to replicate their factor structure. While our analyzes indicate these measures are promising, psychometrically, future research should address the limitations of this preliminary pilot study and expand on this work in several ways.

First, our study was conducted with a sample of 204 3rd graders (8–9 years old) attending a public elementary school in a large, urban district. The size of our sample imposed certain limitations on our analyzes, which were most evident when attempting to fit higher-dimensional models to certain measures (e.g., the Student Interest in Science Activities (Middle Childhood) Scale). Therefore, our results should be replicated in future research employing larger samples. This research will also serve to inform the ongoing debate regarding the conceptual and empirical distinctions between self-efficacy and self-concept constructs at the general and domain-specific levels. For instance, Marsh et al. (2019b) explored the distinctions between self-efficacy and self-concept with a large sample of 11- to 12-year-old German students using 5 established and well-validated measures. They found generalized math self-efficacy and math outcome expectancies were statistically indistinguishable from math self-concept, suggesting that they may represent a single construct, but that these measures were empirically distinct from test-related and functional measures of self-efficacy. Given that understanding elementary school children’s science self-efficacy and self-concept is a burgeoning area of research with clear implications for workforce development, more research is clearly needed to further explore the conceptual and empirical distinctions between these constructs. Our newly devised measures could serve as a springboard for more extensive studies featuring larger samples that would allow fitting models with alternative factor structures, including those in which science self-efficacy and self-concept items loaded on a single factor. These studies could also seek to systematically vary items according to their temporal frame of reference (past/future) to investigate whether this difference in frame of reference accounts for much of the distinction between self-efficacy and self-concept, as proposed by Marsh et al. (2019b).

A second limitation to this work arises from a potentially unique educational experience within our sample. The pilot study was conducted in a school district that prioritizes science education, even in elementary school. This district hosts yearly science fairs and includes a variety of science opportunities for their students. This may partially explain the distributions for each measure’s scale and subscale (shown in Figures 1–4); distributions that suggest a population of students who are interested and moderately confident in their science skills. Additional work should replicate this in other districts to examine the variability in student self-perceptions and interests in this age range from districts that may not have a similar focus on science.

Future research should also expand upon this work with additional student populations (e.g., students from other racial/ethnic backgrounds, across rural and suburban populations, and across older and younger elementary populations). One point of particular interest would be the extent to which older students’ responses to the items that comprised certain sub-scales might be more internally consistent than the responses provided by their younger counterparts. The internal consistency values for the items that comprised each factor of the Student Interest in Science Activities (Middle Childhood) measure were marginal, as was the value for the items that corresponded to the third factor (Computational Sciences) of the Student Attitudes About STEM Careers measure. This may reflect, in part, 3rd grade students’ relative unfamiliarity with the outlined science activities or STEM-related careers (e.g., computer science). For instance, while the science activities described in the Student Interest in Science Activities measure were developmentally appropriate, not all children have access to these activities, which would likely lead to uncertainty about their interest level. Future research with students representing a wider range of ages and background experiences would allow this hypothesis to be tested.

Relatedly, future research should also explore whether students’ demographic characteristics (e.g., gender identity, race/ethnicity) are associated with different responses on the scales and subscales that comprise the measures featured in this study. However, this should only be done after establishing that these scales and sub-scales display properties of measurement invariance across these demographically-defined groups. This will allow us to assess whether our measures are assessing the same constructs and on the same scales for students with different gender identities and from different racial and ethnic backgrounds. Conducting measurement development work with diverse populations necessitates ensuring that the developed measures are assessing the same constructs for all populations.

A third limitation of our study is that our capacity to establish convergent and divergent validity is limited. While we can glean some evidence of these from the sub-scales of each measure, additional work should include additional measures that should and should not theoretically be related to these measures to provide additional evidence of convergent and divergent validity. Finally, a natural question that arises from this work relates to the predictive capacity of these measures to determine how they relate to and predict students’ performance in science and mathematics at school. Our work is limited in that it does not allow us to answer this question. However, the measurement development work presented here paves the way for others to answer such questions.

Lastly, another critical line of research should elucidate how science self-efficacy, self-concept, and interest develop over time, beginning in the elementary years when students start to engage in formal and informal science activities. This work should examine how different types of learning experiences influence the developmental trajectory, and whether students’ science self-perceptions and interest predict academic achievement, higher education fields of study, and entrance into STEM fields. Information from this line of research will be critical to the development and expansion of evidence-based programs and practices aimed at growing the STEM workforce.