The underrepresentation of female students, students of color, first-generation students, and low-income students in science, technology, engineering, and mathematics (STEM) fields is not a new challenge. STEM educators have worked for decades to address the “leaky” STEM pipeline, which begins in late elementary and middle school, continues through secondary school, and builds to a crescendo in the college years as students explore their careers and prospects for continued study at more advanced levels (Kitchen et al., 2018). Several research-based approaches to supporting underrepresented students’ persistence in STEM have been identified, including early research experiences, STEM mentoring, and the development of supportive learning groups, all of which seek to foster students’ motivation and efficacy in STEM (Chang et al., 2014; Hurtado et al., 2010).

Sadly, these worthy approaches are often undermined by persistent societal inequities that threaten students’ development as future STEM professionals. Lack of robust K–12 STEM preparation, unequal work and childcare responsibilities, lack of access to formal and informal learning spaces, and lack of technology can all conspire to create formidable barriers to STEM success in college (Hurtado et al., 2010). While institutions of higher education implement positive academic and social supports to address these challenges under typical conditions, the COVID-19 pandemic, which began to sweep through the United States in March 2020, created uncharted territories for colleges that needed to rapidly shift to online teaching and learning. This shift was especially daunting for disciplines such as science and engineering that rely heavily on hands-on and experiential learning through laboratories and fieldwork. The rapid shift to virtual college education also threatened academic and psychosocial supports typically available to students in person, such as office hours, study groups, libraries, computer labs, and informal interactions with peers, all of which could be particularly problematic for underrepresented students whose STEM identities—and specifically, their self-efficacy and motivation in STEM—could be most fragile. We wondered whether the move away from in-person to virtual learning exacerbated socioeconomic and other cultural inequalities in STEM education. To that end, we undertook a study to examine the following two research questions:

1. What demographic and sociocultural factors were associated with students’ self-efficacy and motivation in STEM during the rapid shift to online teaching and learning?
2. How did the loss of in-person laboratory experiences impact students across a range of STEM courses on indicators of self-efficacy and motivation in STEM?

In this article, we present the results of a quantitative study and address implications for both virtual and in-person STEM teaching and learning.

Theoretical context

Students from underserved groups—such as women, those from racial and ethnic minority backgrounds, first-generation students, and low-income students—face considerable barriers to STEM education (Bullock et al., 2017; Ong et al., 2018). Disparities in secondary school preparation, societal and cultural biases, lack of role models, and differential access to human and technological supports all threaten students’ persistence in the STEM pipeline (Chang et al., 2014; Ensmenger, 2010; Erete et al., 2021; Traweek, 1988; Xie & Shauman, 2003). The exclusion of participants from diverse backgrounds in STEM represents a loss to the field and to society at large, as STEM jobs are increasingly in demand and the success of STEM depends on the contributions of diverse perspectives and experiences (Powell, 2018).

Universities have allocated considerable attention and resources to supporting underrepresented students in STEM, including precollege and summer bridge programs, mentoring, and academic clubs, among other efforts (Chang et al., 2014). Many of these programs attempt to increase not only students’ understanding of STEM content but also their perceptions of their own potential for success in STEM, also known as their self-efficacy. Self-efficacy refers to one’s judgment about his or her own ability to accomplish specific courses of action (Bandura, 1997) and has been shown to be positively correlated with students’ academic achievement, goals, motivation, and persistence (Pajares, 2002; Schunk, 1991). A strong research base suggests that high self-efficacy in STEM impacts achievement, and vice versa—that is, as students gain confidence in themselves as scientists and engineers, they are more motivated to perform better and consequently do perform better (Zeldin et al., 2008). Self-efficacy is believed to be derived from four sources: mastery experiences (i.e., students’ own performance accomplishments), vicarious experiences (i.e., observations of peers and role models), social persuasions (i.e., verbal judgments and messages from others), and physiological states (i.e., physiological sensations and feedback, such as anxiety or joy; Bandura, 1986). Mastery experiences are widely viewed as the most influential factors in promoting self-efficacy (Bandura, 1997; National Research Council, 2001). Consequently, ensuring that all students have the tools, access, experiences, and support to succeed in STEM coursework is key to promoting self-efficacy and, therefore, motivation to persist in STEM.

In spring 2020, the COVID-19 pandemic prompted a rapid shift to online teaching and learning at many universities, including the site for our study. This shift threatened the development of students’ self-efficacy in STEM due to differential access to online educational spaces and the loss of traditional STEM laboratories. Students who cannot readily access educational spaces, including virtual spaces, run the risk of missing out on mastery experiences that can promote self-efficacy in STEM. Studies demonstrate the persistence of a “digital divide,” even in countries such as the United States that have a high gross domestic product (GDP) and high levels of general education. According to a 2021 survey from the Pew Research Center, only 57% of U.S. adults who make less than $30,000 a year have access to broadband internet, and only 59% of these adults have access to a desktop or laptop computer, compared to 92% of American adults who make more than $100,000 yearly (Vogels, 2021). In addition to the digital divide, early research on the effects of the COVID-19 pandemic cited social inequality as a key factor influencing whether individuals could remain socially distanced, work from home, or rely on health insurance (Bavel et al., 2020). Our investigation, then, could not simply focus on how many students’ self-efficacy and motivation may have been impacted by the shift to online teaching, but which students were affected. In this article, we draw on an intersectional approach (Collins & Bilge, 2016) to examine whether and if access to online educational spaces differentially impacts students according to characteristics such as race, social class, and gender. As we developed our hypotheses, we recognized that the move to online education had the potential to either level an uneven playing field or exacerbate inequalities. For instance, informal mentorship extended to students of privilege (Xie & Shauman, 2003) might be eliminated in an online environment, yet students living away from campus may have unequal access to the tacit tools necessary to succeed (Czerniewicz, 2018), such as reliable internet connections, parental expertise, and peers for study groups.

The second area of concern regarding the move to online teaching and learning is the loss of authentic laboratory experiences that provide students with the opportunity to succeed and see themselves as scientists. Moreover, authentic laboratory experiences, which typically group students into laboratory teams or partnerships, allow for enhanced vicarious sources of self-efficacy, such as seeing how others learn and navigate the activities, sharing challenges and concerns, and providing opportunities for interactions with professors or teaching assistants, all of which have been shown to support the self-efficacy of underrepresented students in STEM (Hutchison et al., 2006). The importance of authentic laboratory experiences in STEM courses is based on constructivism and has long been seen as the epistemic foundation for science learning, with both social (Vygotsky, 1978) and cognitive (Piaget, 1957; von Glasersfeld, 1982) aspects contributing to pedagogic and curricular best practices. The idea that learners must be active rather than passive receivers of knowledge through interaction with others and their environment in order to make sense of the world is key to contemporary understandings of learning in STEM (Hodson & Hodson, 1998; Liu & Chen, 2010). Hands-on, authentic laboratory experiences have been found to be particularly robust not only for science content and process understanding but also for psychological factors, including motivation and self-efficacy (Dunlap, 2005; Luzzo et al., 1999). Therefore, providing such opportunities for all students, but particularly for those from underrepresented groups, is critical for supporting retention in STEM.

What happens when the “trappings” of constructivist, lab-based courses are removed during a time such as the COVID-19 pandemic? We hypothesized that students who “missed out” on the anticipated laboratory experiences, such as those in engineering and physics courses, might feel less confident about themselves as learners compared to students in courses in which the move to online learning did not lead to such stark changes, such as in computer science. We also hypothesized that we would see associations between demographic and socioeconomic factors related to the loss of laboratories in online teaching because the students whose self-efficacy in STEM might be most fragile could be most affected by these changes. We believed that examining these relationships during the rapid shift to online teaching and learning had the potential to inform the field going forward for both in-person and virtual STEM learning.

Methods

We present the results of a quantitative survey on the student population at a residential university located in the northeastern United States, targeting students in three departments in STEM fields: civil and environmental engineering, where laboratory work is common and expected; computer science, where dependence on materials such as servers and computational resources is necessary but can be accessed remotely; and physics, which has a combination of theory and lab-based work. We contacted 962 students who participated in courses in these departments during spring 2020 semester and received 186 complete surveys, for a response rate of 19.33%.

The survey began with demographic characteristics, including gender, race, and immigration status; parental education and financial aid status served as proxies for socioeconomic indicators. We asked about students’ context during the semester, including whether or not they were employed and if that status changed and whether they had access to resources such as labs, libraries, equipment, instructors, or peers. We also asked them to rate their interactions with professors, teaching assistants, and peers both before and after the transition online and inquired about the characteristics of their new environment for studying, including whether they had dedicated study space, good internet access, or caregiving tasks in the home. Finally, we asked about their academic outcomes in a series of self-reported questions: if they made the decision to pass or fail the class or change majors; their level of concern regarding the impact of the transition on their grades and on their trajectories; their sense of motivation and the performance of their best academic work; and if they felt prepared for the next course in the sequence. These questions gauged students’ self-efficacy and motivation in science, as those who experienced lower self-efficacy would be more likely to change majors, elect to take courses pass or fail, and feel least prepared to do well in subsequent semesters and least motivated to do their best work. Our survey questions were influenced by valid and reliable instruments, including those developed by Chen et al. (2001) and Gungor et al. (2007). We developed an abridged survey using questions we deemed both relevant to the specific circumstances of our study and sufficient to derive a snapshot of indicators related to student self-efficacy and motivation, while remaining mindful of the importance of survey brevity during an already challenging time for students. We shared the questions drawn from the validated instruments with colleagues who employ surveys in their work to assess face validity (Nevo, 1985)—that is, whether the items appeared to be suitable for their intended use in this study. Once we had a high level of interrater agreement, we distributed the survey via email and collected data using Qualtrics software. We identified several correlations between intersectional student identities and reported outcomes using a Pearson’s chi-squared test for categorical variables and a Kruskal-Wallis test for Likert or rated variables.

Results

As shown in Table 1, our data set included slightly more female respondents than male respondents, with three quarters of respondents born in the United States and nearly two thirds having at least one foreign-born parent. Twenty percent of respondents received Pell Grants, a level of financial aid reserved for students in low socioeconomic circumstances; the remainder of those who reported financial aid may have been recipients of a variety of aid packages aimed at enabling middle-class students to attend the university. Black students were slightly more represented in this study compared to their representation in the sciences overall.

When the university closed midway through the semester, 66% of respondents reported that they shared a workspace with others at home sometimes or always, and 32% rated the reliability of their home internet connection between 1 and 3 out of 5 (1 = poor and 5 = excellent). Students reported a decrease in the quality and frequency of their interactions with professors and instructional aids, with stark decreases in the quality of interactions with peers. Interestingly, as shown in Figure 1, access to specialized equipment or laboratories was largely reported as “never needed” for success in the course (labs = 45%, equipment = 34%) or as “instructor changed,” which meant that the resources were needed but the instructor adapted the resources for online (labs = 34%, equipment = 12%). As shown in Figure 2, the majority of students (69%) disagreed or strongly disagreed with the statement “I was motivated to get my work done during the second half of the semester.” Among those who responded, 53.4% disagreed or strongly disagreed with the statement “I was able to get my work done effectively and efficiently in the second half of the semester.” Meanwhile, 42% disagreed or strongly disagreed with the statement “My living arrangements … allowed me to perform my best academic work,” compared to 35% who agreed or strongly agreed with the statement.

In terms of which students were affected, we observed some important correlations in our data. Pessimistic outlooks that reflected low self-efficacy were not significantly correlated with change in instructional contact or loss of laboratories or physical tools. Low levels of self-efficacy and motivation were associated with situational factors related to socioeconomic status (SES) and cultural positioning, such as when students had poor internet access, had family members who had lost employment, were in caregiving situations, did not have a study space, had lost access to the library, and were Pell Grant recipients. These characteristics were statistically significantly correlated with intersectional categories of race, gender, and socioeconomic class.

For example, as shown in Figure 3, students who were Black, had low socioeconomic status, or were first generation (neither they nor their parents were born in the United States) reported poorer internet access than students from other demographics. First-generation women reported having the most caregiving responsibilities at home, while second-generation women (those born in the United States to parents who were born elsewhere) were more likely to share a workspace and unlikely to have their own place to work. These results were strongly correlated with students’ negative self-reports on their ability to perform their best academic work. Students who experienced someone in their family having COVID-19 were concerned about its effects on their preparedness for the next class; however, students with family members who had preexisting conditions reported more performance and motivation impacts than those students whose families actually experienced an infection.

We examined which students were more likely to select the option to pass or fail the class instead of receiving a grade and found that students who were in more precarious positions socially were more likely to choose the nongraded option. Female students (p = 0.41), Black students (p = 0.08), and students receiving Pell grants (p = 0.07) were far more likely to select an option in which they did not receive a grade for the class. We also noted some students in precarious situations who chose to receive grades: first-generation students and students receiving financial aid other than the Pell Grant. We suggest that cultural factors may dissuade these students from choosing a nongraded option, such as the expectations of the families of recent immigrants or that forms of financial aid other than the Pell Grant may limit their ability to select the pass-or-fail option.

We noted, too, that changes in the frequency and quality of interactions with instructional staff and peers also had an intersectional aspect. Pell Grant recipients reported less-frequent interactions with their teaching assistants, while those students who reported being born outside the United States had lower associations with professor interaction. We note that this could be because these students returned to their home countries, so this change might not represent their social status but rather their geographical location and time zone. Among classmates, female students were more likely to report a greater decline in the quality of interactions with their peers. While female students and students from low socioeconomic backgrounds were the least likely to feel prepared for the next class in the sequence, students from higher socioeconomic backgrounds were much more likely to report feeling like they got their most effective and best work done when they returned home. Finally, students lamented the loss of the library as essential to their coursework, noting that they missed the opportunity for communal, peer-supported, quiet study.

Discussion

To determine the effects of a rapid shift to online STEM teaching and learning during the COVID-19 pandemic, we examined correlations between various demographic and sociocultural factors and students’ self-assessments on measures of self-efficacy and motivation. Our findings suggest that students from underrepresented groups were differentially negatively impacted by the shift to online STEM teaching and learning: Specifically, Black students who had low socioeconomic status had the poorest internet access; second-generation female students experienced a loss of study space and heightened caregiving responsibilities that impinged on their schoolwork; female students across the board experienced confidence loss; and students from low socioeconomic backgrounds were more likely to lose contact with their instructional assistants. Each of the impacts can be interpreted as a potential contributor to reduced self-efficacy in science. For example, the lack of stable internet and study space and increased caregiving responsibilities make successful completion of assignments more difficult, potentially reducing opportunities for mastery experiences. Similarly, loss of contact with instructional assistants and peers reduces vicarious experiences, a factor that is particularly critical for women’s self-efficacy (National Research Council, 2001). Each of these “losses” can increase anxiety, a physiological state that can negatively impact self-efficacy (Bandura, 1986). While situational factors such as loss of study space or increased caregiving responsibilities are not specific to STEM fields, the differential impact on underrepresented students in STEM courses has serious implications for shoring up the “leaks” in the postsecondary STEM pipeline. Our findings demonstrate the need for access to quiet study spaces, support for students who have caretaking responsibilities, concerted efforts by faculty and instructional assistants to form relationships with students, and comprehensive technology support (including WiFi hotspots) for virtual courses for all students, including those underrepresented in STEM. Our findings also confirm those of other researchers who found that marginalized populations can be differentially impacted by rapid shifts to virtual STEM teaching (Abou-Khalil et al., 2021). Ironically, although technology is often viewed as an equalizer, our results revealed that the environment of the university—rather than the virtual environment—is a “great leveler” for underrepresented students. The access to libraries, study spaces, peers, and instructors guaranteed by the residential university provided better and more equal support to students from all backgrounds, as opposed to them relying on what could be provided at home during a period of online access. The results lend credence to the rationale for STEM living-learning communities (Soldner et al., 2012), community-building in online STEM courses (Kim et al., 2021), and efforts to reduce barriers to STEM instructors (e.g., welcoming and flexible office hours and demystifying course syllabi; Jack, 2016; Whittaker & Montgomery, 2012).

Perhaps most interestingly, the loss of traditional laboratories did not seem to differentially impact students in courses with traditional labs (i.e., engineering and physics) and those with labs that did not require a specific laboratory space or equipment (i.e., computer science). Moreover, there did not appear to be any difference in concerns over loss of laboratory experiences among students based on demographics. In other words, the loss of authentic laboratory spaces, equipment, and investigations was not impactful in the way that, for example, students’ experiences with COVID-19 in their families or even worries about family members with preexisting conditions impacted their motivation or feelings of preparedness for the next class.

There are a few explanations for these findings. First, it is possible that instructors in engineering and physics adjusted course expectations in such a way as to assuage student concerns about the loss of this foundational element of the course. STEM assessment during the rapid shift to online teaching has been cited as among the most vexing pedagogical challenges, as these assessments are nearly impossible to retrofit to revised content (García-Alberti et al., 2021). Consequently, the faculty at our research site may have adjusted assessments and course expectations in such a way that students perceived themselves as being more or less “in the same boat,” particularly insofar as preparation for future courses in the STEM sequence. Second, faculty may have made substantial changes to their courses in ways that provided students with robust experiences, leading to students’ feelings of self-efficacy. We are aware that some faculty developed “found materials” labs (e.g., materials found at home), online simulations, videos of ongoing lab experiments on campus, and materials kits that were sent home, which may have contributed to students’ lack of concern over the loss of labs and the unremarkable difference between the three courses we examined on this variable. That said, it should not be lost on science educators that students in STEM courses in a competitive college environment were more concerned about the potential of family members getting sick than the actual loss of their planned laboratory experiences. This finding aligns with Maslow’s hierarchy of needs (Maslow, 1943), which positions physiological, safety, and relationship needs as precursors and foundations for esteem and self-fulfillment and actualization needs. When students’ basic needs for space and time for work were not met, or when they were concerned about family members, the loss of STEM labs took a back seat. We believe this finding supports the need for increased emphasis on sociocultural and humanistic frameworks for teaching STEM—for example, teaching engineering concepts by making direct connections to improving the lives of humans or using socioscientific issues (Zeidler, 2014), which are complex societal issues related to science in order to teach STEM concepts, to tap into students’ empathy and desire to engage with and learn from others. Other such frameworks, including place-based education (Johnson et al., 2020) and culturally relevant pedagogies in STEM (Johnson & Elliott, 2020), connect to students’ environments and interests, unlike more traditional science teaching, which often emphasizes decontextualized, impersonal STEM content.

In sum, students’ motivation and self-efficacy in STEM were more clearly associated with their social and structural conditions than with the lack of access to the material conditions of the laboratory. Researchers suggest that as individuals experience institutional inequalities in their lives, they transform these experiences into personal choices: “Constraints” become translated into “preferences” that impact students’ career choices (Correll, 2004). Unless such problems are rectified through concerted policies at the university level to measure and close these gaps, personal experiences of structural inequalities associated with the transition to online education during the COVID-19 pandemic may translate into preferences against STEM for the very students we hope to maintain in the pipeline.

Limitations, future research, and conclusion

We note that our online survey may have undersampled at-risk populations who might have limited time for or access to the survey; however, this situation would likely have understated, not overstated, our findings. It is also possible that we may have oversampled students who found the semester to be more challenging and wished to express their concerns. Additionally, self-reported lower motivation and performance concerns may not directly translate to poorer performance, and students from less-privileged backgrounds may be more likely to underevaluate their own performance, in line with issues such as imposter syndrome. We relied on self-reported indicators because we were unable to use students’ grades as reliable performance indicators due to a university-wide policy, introduced at the end of the semester, to institute pass-fail grades for all courses. We therefore aim to conduct a difference-in-difference cohort analysis following our students’ return to in-person instruction to determine quantitatively which students are best prepared for the next course in the sequence. Finally, we note that our use of an abridged survey instrument, while adequate for developing a snapshot of student motivation and self-efficacy across courses and demographic groups within our study, may have limited generalizability of the findings across studies.

While the COVID-19 pandemic created innumerable challenges in higher education, it also provided a novel opportunity for STEM education researchers to better understand and remedy the overt and tacit inequities that lead to underrepresentation in the field. Moreover, the finding that college students’ primary areas of concern during COVID-19 crystalized around their interactions with peers and concerns over loved ones, as opposed to academic concerns, emphasizes the need for STEM educators to ensure that curriculum and pedagogy maintain human connection and personal and societal relevance for all students.

Sami Kahn (samik@princeton.edu) is the executive director of the Council on Science and Technology, Janet Vertesi (Vertesi@princeton.edu) is an associate professor in the Department of Sociology, Sigrid Adriaenssens (sadriaen@princeton.edu) is an associate professor in the Department of Civil and Environmental Engineering, Julia Byeon (ybyeon@princeton.edu) is a doctoral student in the Department of Sociology, Mona Fixdal (mfixdal@princeton.edu) is the senior associate director for online education in the McGraw Center for Teaching and Learning, Kelly Godfrey (kellygb@princeton.edu) is the assistant director for educational and program assessment in the McGraw Center for Teaching and Learning, Jérémie Lumbroso (lumbroso@cs.princeton.edu) is a lecturer in the Department of Computer Science, and Kasey Wagoner (kwagoner@princeton.edu) is a lecturer in the Department of Physics, all at Princeton University in Princeton, New Jersey.