Educational disparities in STEM during COVID-induced distance learning and a potential strategy to address them.

Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Nat Commun . 2026 Feb 26;17:3239. doi: 10.1038/s41467-026-69925-9 Search in PMC Search in PubMed View in NLM Catalog Add to search Educational disparities in STEM during COVID-induced distance learning and a potential strategy to address them Rebeka Man Rebeka Man 1 Department of Statistics, University of Michigan, South University, Ann Arbor, MI USA 2 AbbVie, North Chicago, IL USA Find articles by Rebeka Man 1, 2 , Jun Li Jun Li 3 Stephen M. Ross School of Business, University of Michigan, Ann Arbor, MI USA Find articles by Jun Li 3, ✉ , Kean Ming Tan Kean Ming Tan 1 Department of Statistics, University of Michigan, South University, Ann Arbor, MI USA Find articles by Kean Ming Tan 1, ✉ Author information Article notes Copyright and License information 1 Department of Statistics, University of Michigan, South University, Ann Arbor, MI USA 2 AbbVie, North Chicago, IL USA 3 Stephen M. Ross School of Business, University of Michigan, Ann Arbor, MI USA ✉ Corresponding author. Received 2024 Apr 3; Accepted 2026 Feb 13; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13061933  PMID: 41748582 Abstract The COVID-19 pandemic prompted widespread school closures and a swift transition to distance learning, raising concerns about consequences for student success. This study explores academic outcomes in STEM (Science, Technology, Engineering, and Mathematics) courses during COVID-induced distance learning using student-course records at an U.S. public university. We particularly focus on students from underserved populations and who have below average academic outcomes. Employing an expected shortfall regression strategy, we examine outcomes in the bottom quintile and compare differences between students whose household income were in the lowest bracket or without a parent holding a four-year college degree and peers for whom this was not the case, in distance and in-person learning. We show that during distance learning, students in disadvantaged populations in the lowest quintile had average grade differences of 0.11 and 0.06 points, respectively. We also find that targeted instructional changes in an introductory physics course were associated with narrower achievement gaps. These results suggest that while distance learning posed challenges for underserved students, deliberate strategies to increase interaction may potentially support greater equity in STEM education. Subject terms: Society, Education During the pandemic, many universities moved STEM courses online, which raised concerns about unequal access and performance among students from different backgrounds. The authors show that, in these courses, students from underserved groups obtained lower grades in distance learning compared to students not in underserved groups, but that targeted, inclusive teaching strategies were associated with narrower achievement gaps. Introduction The COVID-19 pandemic significantly disrupted the education system and led to mass school closures worldwide 1 . The majority of colleges and universities transitioned to online/distance learning—84% of all undergraduates in the US had some or all of their classes moved to online-only in Spring 2020 2 . Distance learning continued for several semesters following the onset of COVID-19. In Fall 2021, 61% of all undergraduate students were enrolled in at least one remote course while 28% were enrolled in remote courses exclusively 3 . Prior to the pandemic, there were some studies that explored the benefits and limitations of online learning. For example, one characteristic of online learning is the flexibility to study at one’s own pace 4 . However, this suggests that student success may depend on high self-discipline and time management skills 5 . In addition, the design of a course will also need to be suitable for online learning—the nature and extent of peer and student-faculty interactions must be adapted to ensure student learning outcomes in online courses 6 . Student populations such as Black, Latin, Indigenous students, low-income (household income in the lowest bracket) or first-generation (parents did not complete a four-year college degree) are historically underrepresented in higher education and face barriers to opportunity and high-quality instruction 7 . Promoting equity in higher education is a major priority in the United States 8 , and with the increase of online learning in higher education, equity in online learning should be explored. Investigating the effect of distance learning on disadvantaged students is crucial to address the extent of gaps in education. We focus primarily on students who are low-income (household income in the lowest bracket) or first-generation (parental education less than a four-year college degree), who face great challenges in higher education 9 . Students who are low-income or first-generation disproportionately come from ethnic and racial minority backgrounds, have less academic preparation, tend to be older, and face more obligations outside college. Even after taking their background, preparation, and enrollment characteristics into account, students who are low-income or first-generation still encounter greater risks of failure in higher education 10 – 16 . They are less likely to participate in student experiences that promote academic success such as studying in groups, interacting with faculty and peers, and utilizing support services 15 , 17 – 26 . Specifically, in distance learning, students whose household income is in the lowest bracket are more likely to face practical challenges such as not having a quiet workplace, not knowing where to seek course assistance, and managing conflicts with home responsibilities 27 . They are also more likely to face internet connectivity issues, hardware and software issues that negatively impact the effectiveness of distance learning 8 , 28 . On the other hand, first-generation students may face challenges in distance learning due to staying in households where their parents lack the educational experience to provide support. At the same time, they are more often expected to assist with sibling care at home 29 . Moreover, first-generation students tend to have lower access to digital tools and are less familiar with school culture, knowledge and skills which in turn affect their academic performances 30 . When studying the academic gap, we are particularly interested in gaps among students who have below average academic outcomes. Many studies have investigated strategies to improve the performance of low-performing postsecondary students 31 – 35 . Students who begin with poor performance tend to struggle throughout their academic journey 33 , a problem which may be exacerbated by limited in-person instruction and interaction. Students at the lower end of academic outcomes are at risk of being placed on academic probation, which can delay graduation 36 . First generation students and students from low-income families already face a graduation gap: undergraduates who are both low-income and first-generation have a six-year graduation rate at four-year institutions of 41% compared to 73% for undergraduates who are neither low-income nor first-generation 37 . We hypothesize that lower achieving students may be more negatively affected by distance learning than an average student, further enlarging the achievement gap between students who are low-income or first-generation and their peers. While prior literature suggest that distance learning may have no impact or even improve average student outcomes 38 – 40 , additional evidence suggests online learning may negatively affect low-performing students 41 , 42 . In this study, we focus on STEM courses in higher education. Enhancing social mobility for students is important for universities, and STEM is often seen as a vehicle for economic opportunity and social mobility 43 . STEM degrees have long been promoted to students as a potential path to high-paying careers 9 , 44 – 47 , and there is a robust and growing demand for students skilled in STEM disciplines in both STEM and non-STEM occupations 48 . Students pursuing STEM majors are more likely to have a job aligned with their field of study compared to those from non-STEM majors 46 and see significant wage advantages in the job market 45 , 46 . However, students from underrepresented populations are less likely to study or complete STEM degrees. Once they are enrolled in a STEM program, retaining these students poses a great challenge due to factors such as insufficient pre-college preparation and lack of supportive social environments 49 – 52 . Students who are first-generation or whose parents have lower incomes are more likely to leave their STEM program by changing majors or leaving college without completing a degree 49 , 53 . The effect distance learning may have on the academic performance of students who are low-income and first-generation in STEM courses is of interest, as college GPA and first-semester GPA are important predictors of their STEM degree attainment 54 – 56 and of obtaining a STEM-related job 57 . Distance learning produces additional challenges when it comes to STEM courses, as many STEM courses rely on specialized software and equipment which may not be readily accessible outside of traditional classroom settings 58 . In distance learning, students in STEM courses report concern about losing hands-on experience in laboratories 59 , 60 and struggle with engagement and motivation after the transition 27 , 61 – 63 . The intersection of STEM, barriers to success in STEM faced by students who are low-income and first-generation, and challenges faced in a distance learning setting are thus of great interest. The primary objective of our study is to examine the association between COVID-induced learning modality (in particular, distance learning) and student academic outcomes and learning gaps in STEM courses, particularly focusing on underrepresented and lower performing students. To achieve this goal, we obtain longitudinal student administrative and educational outcome data from the University of Michigan (referred to as the University hereafter) during the years from Fall 2016 to Fall 2022. Our dataset contains information about student course enrollment and their respective performances, as well as their demographic, socioeconomic and academic backgrounds. Focusing on semester-long undergraduate courses, our data contains over 1.2 million student-course records, more than half of which are for STEM courses. The study has received approval from the University’s Institutional Review Board and the data are procured through a Memorandum of Understanding (MOU) agreement with the University’s enrollment management office to ensure strict protection of student privacy. Before the onset of the COVID-19 pandemic, almost all classes at the University were taught in person. In March 2020, the University announced it would switch all classes to distance learning to limit the spread of COVID-19. It then resumed limited in-person classes in Fall 2020 and Winter 2021. However, the majority (75.1% and 89.2%, respectively) of classes were taught online during these terms. Later in Fall 2021, a significant portion (86.3%) of classes returned to in-person learning while some classes (13.7%) continued to be taught online or in a hybrid mode. The selection of learning modality is primarily at the discretion of individual instructors, in alignment with guidelines from the University and their respective academic units. These differential transitions in learning modality by courses thus allow us to compare academic achievements across different student subgroups under different learning modalities, while controlling for the effect of the pandemic itself. The results indicate that COVID-induced distance learning is linked to larger academic gaps in STEM courses. Students whose household income were in the lowest bracket and first-generation students in the lowest quintile have grade point differences of 0.114 and 0.064 under distance learning, respectively, compared to their peers. This difference is significant given the lower-quintile mean grade points of 2.48 overall, and 2.00, 2.02, and 1.69 for students who are low-income, first-generation, and both low-income and first-generation, respectively. Moreover, at below a 2.0 cumulative GPA, and for some colleges below a 2.0 semester GPA, students will be placed on academic probation with the risk of having enrollment withheld or being placed on academic recess for one year. In some majors, students who achieve below 2.0 grade points will fail to fulfill the minimum requirement for common required and core courses in their majors. Thus, a reduction of 0.114 or 0.064 in grade points for students within the lowest quintile can result in serious academic consequences and delays to graduation. In non-STEM courses, we see a smaller gap of around 0.05 grade points. Given the greater disparities observed for lower-performing students, we then investigate potential instructional interventions which may reduce disparities in COVID-induced distance learning. Specifically, we consider an introductory physics course that, in each semester starting from Fall 2019, offered additional sections as part of a recent university initiative, i.e., the Foundational Course Initiative (FCI), aimed at improving learning experiences for students. Course sections offered as part of this programs provide students with greater access to instructor support and more interactions with their peers consistently implemented during the in-person and distance learning environments. For example, during distance learning, these sections continued to offer students greater interactions through discussions and group work facilitated by online platforms such as Zoom and Gather Town. We find that this course initiative, which provide students with additional support and student-instructor and peer interactions, significantly reduces the observed achievement gaps in COVID-induced distance learning. These findings not only corroborate previous results in the literature that increased interaction with peers and instructors is associated with greater success in online learning in general 6 and in the context of the pandemic 64 , 65 , they further demonstrate that enhanced interaction can effectively narrow the achievement gaps for students from disadvantaged backgrounds. Results Institutional background We use student administrative and educational outcome data from a large public university in the US which enrolls around fifty thousand students every year, as of 2022. The University transitioned to distance learning in the middle of the Winter 2020 (WN20) semester. The majority of university courses were taught remotely in Fall 2020 and Winter 2021, and many returned to in person teaching in Fall 2021. A small percentage of courses were hybrid modality from Fall 2020 to Winter 2022. The proportion of STEM course sections in each learning modality by semester are presented in Fig. 1 . Fig. 1. Learning modality of STEM course sections by semester. Open in a new tab Proportion of the number of course sections that are distance, hybrid, or in person each semester from Fall 2016 to Fall 2022. The dotted vertical line indicates the semester the university officially closed due to COVID-19. Full details are in Supplementary Table B7 . Empirical strategy We estimate the association between distance learning and grade outcomes for students who are low-income and first-generation by contrasting the difference in educational outcomes between in person and distance learning modality in students who are low-income and first-generation relative to that of students who are non-low-income and non-first-generation. Thus our method focuses on the interaction between distance learning and low-income or first-generation status. In addition to students who are low-income and first-generation, we are interested in lower performing students within these categories. To estimate the association between distance learning and grade outcomes for the lower performing students at the lower quintile, we employ the expected shortfall (ES) regression which allows for this specification. ES regression allows us to study the average outcome for the students whose outcomes fall in the bottom quintile of the population. More details about ES regression can be found in Section 4. Second, we estimate the outcomes of interventions that provide increased student attention, student-peer and student-instructor interaction, particularly for students underrepresented in higher education and STEM, taking advantage of the university’s course initiatives implemented in a large introductory physics course. We implement a three-way-interaction strategy within the expected shortfall regression, contrasting differences between students who are low-income and their non-low-income peers and students who are first-generation and their non-first-generation peers enrolled within these special course sections versus in standard sections, in distance learning versus in person learning. In our regressions, we control for other student, course, and term variables and separate the pandemic association with grade values from the distance learning association with grade values. Data and definition of outcomes Our data consist of semester data on student grades for undergraduate students at University between Fall 2016 and Fall 2022. Because classes transitioned to remote midway during the Winter 2020 semester at the onset of the pandemic, this semester is excluded from our analysis. Restricting classes to undergraduate level semester-long classes that follow a typical A-E grading scale, our data comprise 1,256,632 observations. Limiting to STEM courses, our data comprise 675,463 observations. The outcome variable, grade value, is taken as the numerical equivalent of the letter grade given by an instructor to a student. Otherwise, the official letter grade received may be in Satisfactory/Not Recorded or Pass/Fail format. A student’s low-income status, is described as low-income if a student reported their household gross income to be in the lowest income bracket (less than $25,000 a year) and non-low-income otherwise. A student’s first-generation status is described as first-generation if a student reported their parents’ maximum educational attainment level to be less than a Bachelor’s degree and non-first-generation otherwise. In our dataset, 4.88% of students are categorized as low-income and 13.0% are categorized as first-generation. Association between distance learning and grade outcomes in STEM courses Panel A of Table 1 presents estimates of the association between distance learning and grade outcomes (in grade points: 0.0–4.0) for low-income versus non-low-income students in STEM courses. Our results suggest that there may exist differences in the association between distance learning and educational outcomes for low-income students in the bottom quintile of students in terms of grade values. Among the bottom quintile of low-income students and the bottom quintile of non-low-income students, low-income students show a significant grade difference of −0.114 ( p < 0.001; CI: −0.181, −0.047) points, on average, relative to their non-low-income peers during distance learning. Table 1. Relationship between educational outcomes and distance learning, COVID-19, and student category for the bottom quintile in STEM courses Estimate 95% Confidence Interval p -value Panel A: Low-income Distance −0.019 (−0.031, −0.007) 0.002 Hybrid 0.036 (0.011, 0.062) 0.006 Post-WN20 −0.207 (−0.216, −0.199) < 0.001 Low-inc −0.104 (−0.128, −0.081) < 0.001 Low-inc:Distance −0.114 (−0.181, −0.047) < 0.001 Low-inc:Hybrid −0.08 (−0.233, 0.072) 0.304 Low-inc:Post-WN20 −0.04 (−0.086, 0.006) 0.088 Panel B: First-generation Distance −0.016 (−0.028, −0.004) 0.009 Hybrid 0.04 (0.015, 0.065) 0.002 Post-WN20 −0.197 (−0.206, −0.188) < 0.001 First-gen −0.125 (−0.140, −0.111) < 0.001 First-gen:Distance −0.064 (−0.106, −0.023) 0.003 First-gen:Hybrid −0.065 (−0.168, 0.038) 0.216 First-gen:Post-WN20 −0.099 (−0.128, −0.071) < 0.001 Open in a new tab The table displays the relationship between distance learning and grade outcomes based on low-income or first-generation status. Panel A shows the main effects of distance learning, hybrid learning, post-WN20, and low-income status and interaction effects comparing the variation in outcomes between the bottom quintile of low-income and non-low-income students for either distance learning, hybrid learning, or pre-WN20 with that between the bottom quintile of low-income and the bottom quintile of non-low-income students for in person learning or post-WN20. Panel B shows the main effects of distance learning, hybrid learning, post-WN20, and first-generation status and interaction effects comparing the variation in outcomes between the bottom quintile of first-generation and non-first-generation students for either distance learning, hybrid learning, or pre-WN20 with that between the bottom quintile of first-generation and the bottom quintile of non-first-generation students for in person learning or post-WN20. Both regressions are expected shortfall regression for the lower quintile and control for additional student, class, and term specific variables. A two-sided one-sample Z -test was used to determine the p -value and no adjustments were made for multiple comparisons. Panel B of Table 1 presents estimates of the association between distance learning and grade outcomes for first-generation versus non-first-generation students in STEM courses. Similar to the first model, the table suggests that there exists differences in the association between distance learning and educational outcomes for the bottom quintile of first-generation students. We found that the bottom quintile of first-generation students show a significant grade difference of −0.064 ( p = 0.003; CI: −0.106, −0.023) points, on average, relative to the bottom quintile of non-first-generation students during distance learning. For both first-generation and low-income students in the bottom quintile, there is no significant link between hybrid learning and grade outcomes as opposed to in person learning. The association between distance learning and grade outcomes for the bottom quintile of students in STEM courses can also be seen in Fig. 2 . We observe that the association between distance learning and grade outcomes is significantly negative for both the bottom quintile of non-low-income and non-first-generation students and the bottom quintile of low-income and first-generation students, but the estimate is much more negative for low-income and first-generation students. Fig. 2. Association between distance learning and grade outcomes in STEM classes by low-income (low inc) or first-generation (first gen) status. a,b. Open in a new tab Data are presented as effect sizes with 95% confidence intervals (error bars) and are estimated through expected shortfall regressions with τ = 0.20 using Model ( 1 ), where the dependent variable is the numeric grade outcome ( N = 675, 463). The regression follows the specification in ( a ) Panel A of Table 1 and ( b ) Panel B of Table 1 . The point shape (triangle) indicates significance at level α = 0.05 and the dashed vertical line indicates effect size equal to zero. We detect a small disparity (around −0.025) for low-income students but not for first-generation students in the mean results as shown in Supplementary Table 8 . Results for all courses are displayed in Supplementary Table 9 and Supplementary Fig. 4 . Results for STEM courses from a regression including interactions for both low-income and first-generation status are displayed in Supplementary Table 12 and Supplementary Fig. 3 . There is a significant negative association between distance learning and grade outcomes for the bottom quintile of non-low income and non-first generation, only low-income, only first-generation, and both low-income and first-generation students. The additional association between distance learning and grade outcomes on only low-income, only first-generation, and both low-income and first-generation students is significantly negative. The results including interactions for both low-income and first-generation status for all courses are displayed in Supplementary Table 13 and Supplementary Fig. 5 . We additionally considered the association between distance learning, hybrid learning, and grade outcomes by sex and by underrepresented minority status. The results are in Supplementary Table 1 in Supplementary Information A.1 and Supplementary Table 2 and Supplementary Fig. 1 in Supplementary Information A.2 , respectively. As COVID-induced distance learning is intertwined with factors such as at-home conditions, we considered the association between distance learning and grade outcomes in courses that had distance learning after most courses returned to in-person on campus. The results are in Supplementary Tables 5 – 6 in Appendix A.4 . Association between distance learning and grade outcomes by subject category In addition to across all STEM courses, we further considered the association between distance learning and grade outcomes for courses in different subject categories. Specifically, we considered the following subject categories: Physical & Natural Sciences, Social Sciences, Engineering, Medicine & Health, and Humanities, Arts & Languages. All Physical & Natural Sciences and Engineering courses are considered STEM, while some courses in Medicine & Health and Social Sciences, and almost no courses in Humanities, Arts & Languages are considered STEM. The association between distance learning, low-income or first-generation status, and grade outcomes for the bottom quintile of students in these groups is presented in Fig. 3 ). For Physical & Natural Sciences and Engineering, we observe similar results for low-income and first-generation students as in our main results for all STEM courses. The negative association between distance learning, low-income or first-generation status, and grade outcomes is generally largest in Physical & Natural Sciences, Engineering, and Medicine & Health courses, compared to the Social Sciences and Humanities, Arts & Languages. Tables presenting estimates of the association between distance learning and grade outcomes comparable to that in Table 1 are in Supplementary Table 10 in Supplementary Information B . Fig. 3. Association between distance learning, low-income or first-generation status, and grade outcomes in all classes. Open in a new tab a . b Data are presented as effect sizes with 95% confidence intervals (error bars) and are estimated through expected shortfall regressions with τ = 0.20 using Model ( 1 ), where the dependent variable is the numeric grade outcome ( N = 366, 840 for Physical and Natural Sciences, N = 262, 202 for Engineering, N = 339, 341 for Social Sciences, N = 75, 524 for Medicine and Health, and N = 327, 673 for Humanities, Arts, and Languages). Refer to Table B10 for detailed estimates. Association between the COVID-19 pandemic and grade outcomes in STEM courses In the previous sections, we considered the association between distance learning and grade outcomes, controlling for associations with the COVID-19 pandemic. Specifically, we include in the regressions an indicator of post Winter 2020 semesters as well as its interaction with low-income and first-generation statuses to allow for differential associations between the pandemic and grade outcomes on different subgroups of students. The results are presented in Supplementary Fig. 6 in Supplementary Information B . The figure shows clearly that there exists a negative association between the pandemic and student learning outcomes of all income levels, regardless of whether they are first-generation college students or not. However, the link is much more substantial for first-generation students. The association between the COVID-19 pandemic and grade outcomes for all courses and for the subjects considered in Section 2.5 is presented in Supplementary Fig. 7 in Supplementary Information B and Supplementary Fig. 9 in Supplementary Information B , respectively. Is an intervention associated with smaller disparity between groups? In this section, we examine the link between an intervention that aims to increase student-peer interaction and student-instructor interaction on the relationship between distance learning and the grade outcomes of underrepresented students. We consider an introductory physics course that is offered in both Winter and Fall semesters, and enrolls on average 582 students per semester. This course is offered in multiple sections. Besides the standard sections, starting from Fall 2019, the course includes two sections of students as part of the Foundational Course Initiative (FCI). The FCI sections provide students with greater access to instructor support, more peer interaction by having small groups within a large class and frequent discussion and group/individual work instead of lectures. FCI sections consist of around 200 students per term and are about a third of the total students in the course starting in Fall 2019. All sections (FCI, standard) offered in person instructions before Winter 2020, and shifted to distance learning in Fall 2020 and Winter 2021. While FCI sections were in person the following semester (Fall 2021), standard sections remained distance and only shifted back to in person instructions in Winter 2022. Our physics course data comprise 7373 observations, 1212 of which are for FCI sections. The demographics of the FCI sections are approximately the same as those of the standard sections. A brief summary of the students in different sections is presented in Table 2 in Section 4. Table 2. Summary of students in all courses, STEM courses, and in a general physics course Physics Sections All STEM All Standard FCI Low-inc (%) 5.03 4.88 4.23 4.09 3.07 First-gen (%) 13.20 12.97 12.61 11.80 10.50 Male (%) 49.49 50.56 65.93 66.52 66.78 Underrep-min (%) 12.22 12.19 14.04 12.53 12.89 No. students 71823 66461 6886 5475 1171 Open in a new tab As all sections (FCI, standard) experienced both in person and distance learning, and the learning modality is mostly syncronized across all sections (except for one semester, Fall 2021), we can compare the relationship between distance learning and grade outcomes based on the type of sections. We continue to use the instructor inputted grade as our outcome as before. In this section, we conduct expected shortfall regressions with three-way interaction terms to examine how the innovative course sections (i.e., FCI), moderate the association between distance learning and grade outcomes for low-income and first-generation students. The difference in association between distance learning and grade outcomes for the regular sections and FCI sections is presented in Fig. 4 ). For the bottom quintile of non-low-income and non-first-generation students, the association between distance learning and grade outcomes is not statistically significantly different between regular sections and FCI sections (effect size = 0.133; p = 0.179; CI: −0.061, 0.327 and effect size = 0.099; p = 0.333; CI: −0.102, 0.299, respectively). For first-generation students, the association effect size increases from standard to FCI and becomes significantly positive in the FCI sections (effect size = 0.792; p = 0.010; CI: 0.188, 1.396). For low-income students, we also observe a increase in the effect size for the association between distance learning and grade outcomes from standard to FCI section types (effect size = 1.063; p = 0.021; CI: 0.159, 1.968). In other words, under the intervention (i.e., FCI), the negative link between distance learning and grade outcomes in first-generation and low-income students is reduced. Fig. 4. Difference in association between distance learning and grade outcomes from regular sections to FCI sections of a general physics course by low-income (low inc) or first-generation (first gen) status. Open in a new tab Data are presented as effect sizes with 95% confidence intervals (error bars) and are estimated through τ = 0.20 expected shortfall regressions using the Model ( 2 ), where the dependence variable is the numeric grade outcome ( N = 7, 373). A positive effect size indicates a less negative association between distance learning and grade outcomes for the FCI section group than its regular section counterpart. The regression follows the specification in column 1 and 3 of Supplementary Table 11 . The intervention is not linked to a statistically significant association between the COVID-19 pandemic and grade outcomes. The results are presented in Supplementary Fig. 8 in Supplementary Information B . Discussion This paper presents estimates of the educational gap between lower-performing students who are low-income or first-generation and their non-low-income/non-first-generation peers in COVID-induced distance learning. Using over 600,000 student records data from a large public university, we provide evidence that there exists a negative association between COVID-induced distance learning and grade outcomes for students who are low-income and first-generation, particularly those in the lower quintile, in STEM courses. We do not find such association at the mean. Additionally, we find that innovative course initiatives that provide students will increased peer interaction as well as student-instructor interaction are linked to a lowered negative association observed between distance learning during the pandemic and grade outcomes for students who are low-income and first-generation. This paper contributes to the growing literature studying the association between distance learning and academic outcomes during COVID-19. Several papers have explored the relationship between distance learning, student experiences 66 – 68 and academic outcomes during COVID-19. Many studies considering academic outcomes focus on primary or secondary education 69 – 72 . Notably, 71 observed learning loss and widened achievement gaps in primary education, while 72 reported increased dropout risks and decreased test scores in secondary education in Brazil. Studies that focus on higher education tend to be smaller scale 73 – 75 . Among higher education studies, 73 found no significant differences in grades between online and in-person learning in STEM courses including biology, chemistry, physics, mathematics, and statistics, attributing it to instructors’ proactive course adjustments. 76 found higher grades overall and mitigated equity gaps for underrepresented students in online learning for an introductory biology course series, and hypothesized it may be due to instructors changing grading practices. Conversely, an experiment by ref. 74 revealed lower final grades in distance learning, particularly affecting academically at-risk students. Analyzing data from 11,000 academic records at Queens College, 75 discovered that lower-income students in the bottom quartile outperformed their higher-income peers due to flexible grading policies. However, no advantage was found for top-performing lower-income students, and without flexible grading policies, there was a decline in GPA. Larger-scale studies like those conducted by refs. 77 , 78 , focusing on statewide community college systems, highlighted a decrease in course completion, particularly for students with lower GPAs. 78 noted a decrease in course completion but an increase in A grades, attributing it to instructor leniency in grading during the pandemic. Our paper is a large-scale higher education study that focuses on disparities in STEM academic outcomes due to distance learning in lower performing students who are low-income and first-generation. Additionally, our study explores an intervention that reduces the gap in achievement from distance learning for disadvantaged students. Students considered low-income and first-generation may face greater academic challenges in distance learning, attributed to limited access to resources and familial support 27 , 29 , 30 . Additionally, they are less likely to engage in learning experiences that foster academic success, such as collaborative group studying, interactions with faculty and peers, and utilization of support services 15 , 17 – 26 . Therefore, providing additional support from instructors or learning assistants and encouraging increased peer engagement can positively relate to students’ academic performance. The intervention (FCI) we examine features smaller class styles by assigning students to work in smaller groups facilitated by learning assistants, thereby fostering greater peer interactions as well as student-instructor interactions. Several other courses have been re-designed as part of FCI, however some may redesign all sections of the course or change a section fundamentally such that FCI and non-FCI sections are not comparable. For example, switching to written assignments instead of exams. Additionally, the majority of these courses began their remodel after the onset of the pandemic and switch to distance learning. The impact of these initiatives in other courses is worth exploring. Other strategies that could enhance the learning experience for students considered low-income and first-generation engaged in distance learning involve implementing systems for monitoring student progress 9 . These systems enable both instructors and students to proactively address and enhance performance well before a student faces potential failure. One such system, ECoach, empowers students to monitor their progress, receive personalized feedback, and access course tips and exam playbooks. Previous research has investigated the efficacy of ECoach in traditional STEM and economics courses conducted in-person 79 , 80 , revealing positive effects on students’ academic outcomes. Notably, a study by ref. 79 demonstrated that first-generation students derived greater benefits from the ECoach exam playbook resource compared to their non-first-generation counterparts. It is important to note that in both studies, students independently utilized ECoach tools. Considering the promising outcomes observed, further exploration of the advantages of using ECoach in a distance learning environment, with incentives for universal adoption among students in a study, holds potential. Initiatives to redesign courses, like FCI, have been explored at a number of universities. These include the University of Pennsylvania Structured, Active, In-class learning (SAIL), UNC Chapel Hill Associate of American Universities (AAU) STEM Project, Purdue University IMPACT Project, Cornell University Active Learning Initiative (ALI), and TRESTLE, a multi-institution (Univerity of Kansas, Indiana University, Queen’s University at Kingston, University of California-Davis, University of Colorado-Boulder, University of Texas at San Antonio, University of British Columbia) NSF-funded project to promote improvement of STEM education at research universities 81 . Many efforts focus on reforming science education, particularly introductory and required courses. One trait frequently included in evidence-based course redesigns is the emphasis on active learning and interaction with classmates, especially through the creation of small groups or teams, similar to the intervention in the physics course we examined. The popularity of this pedagogical method in courses is due to its strong positive outcomes for students in terms of their performance in and perception of the course 82 – 85 . Brown University’s Associate of American Universities (AAU) undergraduate STEM education initiative project aimed to improve retention in introductory and required science courses, and redesigned 11 courses by integrating problem solving sessions into recitations, encouraging engagement and collaboration with peers 86 . These initiatives have also been implemented and studied during COVID-induced distance learning. As a part of the Active Learning Initiative (ALI) at Cornell University, instructors implemented team-based learning (TBL) first in an in-person life sciences course and later both adapted the course with TBL to a smaller online platform and utilized TBL when transitioning the in-person course to online during the onset of the pandemic 64 . They found that TBL enhanced the student experience in the online course and students had high retention and achievement with online teaching 64 . Another ALI at Cornell investigated student performance during the pandemic and found that teaching methods that promote active engagement, such as small group activities, can mitigate the negative effect incurred in online classes during the pandemic 65 . Our study is limited to one university, so there is a need for broader data collection so that large scale studies across universities can be done in the future to understand the prevalence of disparity in higher education during distance learning. The University’s response to the pandemic aligns with the typical response of other large universities in the United States. Campus closed in March 2020 and did not reopen until Fall 2020, and the majority of courses shifted to distance learning until Fall 2021. In Fall 2020, the university’s percent of undergraduates exclusively in distance education is similar to the median of its comparison group, which includes most Ivy Leagues, University of California system schools, and public universities such as Indiana University-Bloomington, University of Minnesota-Twin Cities, and University of Wisconsin-Madison 87 . However, the percent of undergraduates in at least one but not all distance education courses is much higher at the University at 61% compared to 35% for the comparison group. In Fall 2021, the percent of undergraduates in at least one but not all distance courses is 46% compared to 38% for the comparison group and the percent not in any distance education is 54% versus 59% 88 . Although the response to the pandemic and adoption of distance learning is relatively similar, our findings may not generalize to other comparable universities due to differences in factors influencing grade outcomes such as academic culture, student body, and grading scheme. As of 2015–2016, about 20% of students enrolled in undergraduate education were in poverty (family income for family of four less than $25,696), and for Very Selective four-year universities about 13% were in poverty 89 . Our definition of low-income as family income less than $25,000, based on the the poverty guidelines updated periodically in the Federal Register by the U.S. Department of Health and Human Services under the authority of 42 U.S.C. 9902(2) 90 , includes In Poverty and Near Poverty categories defined by Pew Research Center. About 56% of undergraduates nationally were first-generation, defined as neither parent having a bachelor’s degree 91 . In our dataset, about 5% of students are considered low-income and 13% are considered first-generation, which is much lower than the national statistics. Thus, nationally there may be a higher or more detectable association between distance learning and grade outcomes in students considered low-income and first-generation. Our study considers distance learning in the context of the pandemic, which can be accompanied by and difficult to separate from factors such as disruptions in students’ everyday life, differing at-home conditions, and carry-over effects such as potential loss of knowledge in pre-requisite courses. As a limitation, we cannot completely distinguish the association between distance learning and academic outcomes from these factors. Beyond the context of the pandemic, many universities in the United States and around the world have shown recent enthusiasm for adopting distance learning 92 – 94 . In STEM education in particular, distance learning strategies are receiving increased attention 40 , 58 , 95 – 97 . The link between distance learning and academic outcomes remains inconclusive, as various studies present differing perspectives. Some research suggests no significant difference or even improved outcomes in comparison to traditional in-person learning 38 – 40 , 98 . Conversely, other studies suggest that student learning and performance may be somewhat lower in a distance setting when compared to face-to-face environments 42 , 99 , 100 . Proponents of distance education assert that online courses not only provide students with greater and more convenient access to coursework but also serve as cost-effective instructional alternatives for universities 101 , 102 . As online courses gain popularity among both universities and students, there is a crucial need to thoroughly investigate the diverse patterns in academic outcomes during remote learning, especially among the most vulnerable student populations. Methods The study has received approval from the University of Michigan’s The Health Sciences and Behavioral Sciences Institutional Review Board and the data are procured through a Memorandum of Understanding (MOU) agreement with the University’s enrollment management office to ensure strict protection of student privacy. Data Our data set consists of administrative data and educational outcomes for all students enrolled at the University from the mid-1990s through 2022. In our main sample of STEM courses for the years 2016 to 2022, we have a total of 675,463 data points for 1200 unique courses and 66,461 unique students. Of these, approximately 43% of the data points are after Winter 2020. Additionally, information about the introductory physics course sections were made available through the university’s collection of historical syllabi. Measures and definitions We define grade values, the outcome variable of interest, as the numerical equivalent of the letter grade given by an instructor to a student (A+/A = 4.0, A- = 3.7, B+ = 3.3, B = 3.0, B- = 2.7, C+ = 2.3, C = 2.0, C- = 1.7, D+ = 1.3, D = 1.0, D- = 0.7, E = 0.0). The instructor inputted grade is the grade given by an instructor to a student; the official grade that appears on a student’s transcript is the translated value of the instructor inputted grade as defined by the school or college grading policy. Official grades may take other forms other than a letter grade. Due to COVID-19, official grades may be chosen to be masked by students as Satisfactory/Not Recorded - COVID. Courses are categorized as STEM by their instructional area based on the 2023 U.S. Department of Homeland Security (DHS) STEM Designated Degree Program List 103 . The list designates all Engineering, Biological and Biomedical Sciences, Mathematics and Statistics, and Physical Sciences (determined by the Department of Education’s Classification of Instructional Programs) programs as STEM, and also includes some programs in related fields such as Natural Resources and Conservation, Education, Psychology, etc. For categorizing instructional areas into subject categories, Physical & Natural Sciences include life sciences, physics, chemistry, earth science, astronomy, and mathematics. Engineering includes all fields of engineering, as well as information and technology-related subjects. Social Sciences include social and behavioral sciences such as sociology, economics law, anthropology, business, etc. Medicine and Health include areas in Kinesiology, Pre-Medicine, Nursing, Pharmacy, Dentistry, and Public Health. Based on the DHS Designated Degree Program List, all Physical & Natural Sciences and Engineering divisions, some Social Sciences or Humanities divisions such as archaeology and psychology, and some Medicine and Health divisions such as kinesiology, pharmacology, immunology, and biological sciences are designated STEM. For the student categories of interest, a student’s low-income status is defined as estimated gross family income less than $25,000 and first-generation status as parent’s maximum education level less than a Bachelor’s degree. Estimated gross family income and parent’s maximum education level are self-reported in a student’s application for college admission; students that do not report these are considered non-low-income or non-first-generation. A sensitivity analysis of our results when excluding these observations where household income or parent’s educational level are not reported is in Supplementary Information A.3 Tables 3 – 4 and Supplementary Fig. 2 . Of the unique students in our main data set, 4.88% are considered low-income and 13.0% are considered first-generation. Among low-income students, 52.0% are first-generation, and among first-generation students, 19.6% are low-income. A brief summary of the students in all considered courses and in the introductory physics course examined in Section 2.7 is presented in Table 2 . Intervention The intervention we considered was FCI sections in a large introductory physics course. We provide a brief description of how this program appears in the physics course. Starting in 2018, FCI partnered with the physics course over a three year pilot period. As a result of the partnership, starting from Fall 2019 onward, two sections of the physics course under study were transformed into a Studio Model, in which instructors used the same curriculum (e.g., same teaching materials) and students were held up to the same course requirements as those in the standard sections. The main difference, however, is that it created a radically different student experience. Each FCI section is split into several neighborhoods of students who spend the whole term together. Additionally, within a neighborhood, students work in small groups of three and cycle through groups throughout the term such that every student in a neighborhood will have worked with every other student by the end of the semester. Each neighborhood is directly overseen by a undergraduate learning assistant. A large portion of lecture time is dedicated group work within these groups which during distance learning is done synchronously on Gather Town. The rest of the lecture is review or preview and individual work periods. Regular sections meet for one hour lectures four times a week, while FCI sections meet twice a week for two hours each. Moreover, each section employs five additional learning assistants, a graduate student instructor, and a faculty member, all part of the instruction team. In this way, the FCI section increases the students’ interactions with one another and with instructors, and results in a larger instructor/assistant-to-student ratio. The inclusion of this type of section was maintained after the three year partnership with FCI ended. Furthermore, one of the remaining sections of the physics is part of the Comprehensive Studies Program (CSP). CSP sections consist of about 20 students per term, while FCI sections have the same sizes as standard sections. We focus our discussion on FCI instead of CSP, because the latter is a small-scale intervention representing no more than 5% of the total students enrolled in the physics course, while FCI is a larger-scale intervention with two sections of regular-sized classes. Moreover, FCI is a more recent innovation with primarily in-class transformations, while CSP is an existing program started in the 1980s which features additional outside resources for students part of the program such as advising and tutoring. However, we control for the existence of CSP sections in our regressions. Empirical strategy We are interested in contrasting the difference in education outcomes between in-person and distance learning modality in low-income and first-generation students relative to that in non-low-income and non-first-generation students. Thus, our method focuses on the interaction between distance learning and low-income or first-generation status. Ordinary least squares is a baseline approach for fitting linear models and focuses on inferring the conditional mean of the response given the predictors. However, more aspects than the mean of the conditional distribution of the response given predictors are of interest. In particular, we are interested in lower performing students within these categories. Thus, we utilize expected shortfall (ES) regression which allows for this specification. The quantity of interest for ES regression is the expected shortfall, compared to the mean for ordinary least squares regression. If we consider a single variable, such as grades, the 0.20-ES refers to the average grade for the sub-population of students whose grades fall below the 0.20-quantile of the entire population, where the 0.20-quantile of the grade outcome refers to the highest grade of those who have the 20% lowest grades among the entire population. In other words, ES gives us the average outcome for the lowest-performing students. To extend this concept to a regression setting, we focus on the average outcomes for the students who fall below a certain threshold (at a specific quantile level) conditional on a given set of covariates. Specifically, ES regression allows us to estimate the relationship between an explanatory variable (e.g. distance learning) and the average outcome for the lowest-performing students, while controlling for other relevant factors/covariates. We estimate the association between distance learning and educational outcomes for disadvantaged students with a focus on the average performance of the bottom quintile using the following specification: E [ Y i c t ∣ Y i c t ≤ Q τ ( Y i c t ∣ X i c t , W i c t ) , X i c t , W i c t ] = α 0 + β 1 D i + β 2 Distance c t + β 3 Hybrid c t + β 4 Post − WN20 c t + β 5 D i ⋅ Distance c t + β 6 D i ⋅ Hybrid c t + β 7 D i ⋅ Post − WN20 t + γ W i c t 1 where Y i c t is an outcome for student i in course c in term t , Q τ ( Y i c t ∣ X i c t , W i c t ) denotes the conditional quantile at quantile level τ , and E[ Y i c t ∣ Y i c t ≤ Q τ ( Y i c t ∣ X i c t , W i c t ), X i c t , W i c t ] is the conditional ES at level τ , which can be interpreted as the average outcome, given covariates X i c t and W i c t , for the subpopulation whose outcomes fall below the τ -quantile conditional on covariates X i c t and W ict . Model ( 1 ) utilizes the conditional ES, targeting the lower quintile conditional on specified covariates rather than truncating the students at the bottom of the performance distribution. The advantage of ES regression over alternatives that also consider the entire conditional distribution such as quantile regression is that it can provide information about the average in the tail of a distribution, in our case the bottom quintile, rather than only providing information about specific points along a distribution. More details about the use of quantiles in regression analysis can be found in ref. 104 and in depth discussion of ES and its mathematical properties can be found in refs. 105 , 106 . The vector X i c t includes main and interaction variables { D i , Distance c t , Hybrid c t , Post-WN20 c t , D i ⋅ Distance c t , D i ⋅ Hybrid c t , D i ⋅ Post-WN20 t }, where D i is an indicator variable coded to 1 if student i is in a disadvantaged group such as low-income or first-generation, Distance c t is an indicator variable coded to 1 if course c in term t is distance learning, Hybrid c t is an indicator variable coded to 1 if course c in term t is hybrid, and Post-WN20 t is an indicator variable coded to 1 if term t is after the Winter 2020 term. The covariates matrix W includes the following student, course, and term attributes: indicator variables for male, underrepresented minority status, parttime student status, overtime status (more than 18 credit hours in a term), the previous term cumulative GPA, indicators for first term attended, class standing (i.e., Sophomore), incomplete status, unofficial withdrawal status, repeating class, winter term, the log scaled total enrollment for the course, and average class GPA excluding student i . We use this specification to measure the disparate association between distance learning and changes due to COVID-19 and the grade values of low-income and first-generation students, particularly those in the lower quintile of academic performance. The estimates are from ES regressions. The 95% confidence intervals are obtained following 106 . We consider τ = {0.20, 1}, where τ = 1 is equivalent to an ordinary least squares regression. The ES regression estimator is asymptotically normal 106 thus the p -value is determined from a two-sided one-sample Z -test where the estimate is compared to zero and the test statistic is calculated as the ES regression estimate divided by the standard error (obtained following 106 ). The estimate of the difference in association between distance learning and grade values for the bottom ( τ ⋅ 100)% of D = 1 students compared to the bottom ( τ ⋅ 100)% of D = 0 students would be β 5 , and the estimate of the difference in grade values between the bottom ( τ ⋅ 100)% of D = 1 students and bottom ( τ ⋅ 100)% of D = 0 students in distance learning would be β 1 + β 5 . We considered additional analyses using τ = {0.1, 0.15, 0.25, 0.5} and generally found significant results for τ ≤0.25. We estimate how an intervention in the form of innovative course initiatives (i.e. FCI) relates to the association between distance learning and educational outcomes by utilizing three-way-interaction terms, in which we contrast differences between disadvantaged and non-disadvantaged students within course sections that adopted an innovative instruction style versus those within standard course sections, in distance and in person learning. As the course we considered did not have any sections adopting hybrid learning modality and did not have any students with an unofficial withdrawal, we excluded these two variables. We estimate the association between grade outcomes and distance learning and the pandemic on disadvantaged students and its interactions with non-standard course sections through a three-way-interaction strategy using the following specification: E [ Y i s t ∣ Y i s t ≤ Q τ ( Y i s t ∣ X i s t , W i s t ) , X i s t , W i s t ] = α 0 + β 1 D i + β 2 Distance s t + β 3 Post − WN20 s t + β 4 FCI s t + β 5 CSP s t + β 6 D i ⋅ Distance s t + β 7 D i ⋅ Post − WN20 t + β 8 Distance s t ⋅ FCI s t + β 9 Distance s t ⋅ CSP s t + β 10 Post − WN20 s t ⋅ FCI s t + β 11 Post − WN20 s t ⋅ CSP s t + β 12 D i ⋅ FCI s t + β 13 D i ⋅ CSP s t + β 14 D i ⋅ Distance s t ⋅ FCI s t + β 15 D i ⋅ Distance s t ⋅ CSP s t + β 16 D i ⋅ Post − WN20 s t ⋅ FCI s t + β 17 D i ⋅ Post − WN20 s t ⋅ CSP s t + γ W i s t 2 where Y i s t is an outcome for student i in section s in term t , FCI s t is an indicator variable coded to 1 if section s in term t is an FCI section, CSP s t is an indicator variable coded to 1 if section s in term t is a CSP section, and W i s t follows that in equation ( 1 ) excluding unofficial withdrawals and by section s rather than course c . Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information (816.4KB, pdf) Reporting Summary (81KB, pdf) Supplementary Code (9.9KB, txt) 41467_2026_69925_MOESM4_ESM.pdf (24.3KB, pdf) Description of Additional Supplementary Files Transparent Peer Review file (358.6KB, pdf) Source data Source Data (24KB, xlsx) Acknowledgements Research in this publication is partially supported by National Science Foundation (NSF) CAREER DMS-2238428 (K.M.T.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Science Foundation. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author contributions Authors R.M., J.L., and K.M.T. contributed equally to the work. Peer review Peer review information Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability Currently, the LARC Data Set is available to University of Michigan (U-M) investigators who have obtained Institutional Review Board (IRB) approval and signed a memorandum of understanding describing their research interests. Investigators not affiliated with the University of Michigan must adhere to the established data request protocols of the University of Michigan to gain access to data. Investigators should contact [email protected] to request access to data. The expected time-frame for response to access requests depends on the the scope of the request. For accepted data requests, the timeline for data availability will be outlined in the data user agreement. More information about accessing the data can be found here: https://enrollment.umich.edu/data/learning-analytics-data-architecture-larc . Cleaned data are available from the authors upon university approval for LARC data. Individuals interested in obtaining this data should follow above steps to obtain approval from the University of Michigan. Source data are provided with this paper. Code availability Code to perform the expected shortfall regression method is available in the Supplementary Code. Competing interests All authors are employees or affiliates of the University of Michigan. R.M. was an employee of University of Michigan at the time of the study and is currently a full-time employee of AbbVie. The data and financial support for this research were provided by the University of Michigan. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Jun Li, Email: [email protected]. Kean Ming Tan, Email: [email protected]. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-69925-9. References 1. United Nations Educational, Scientific and Cultural Organization. Education: From disruption to recovery. https://www.unesco.org/en/covid-19/education-response (2020). 2. National Center for Educational Statistics. 2019-20 National postsecondary student aid study (NPSAS:20): First look at the impact of the coronavirus (COVID-19) pandemic on undergraduate student enrollment, housing, and finances (preliminary data). Tech. Rep., U.S. Department of Education, Institute of Education Sciences (2021). 3. National Center for Education Statistics. Undergraduate Enrollment. Condition of Education . Tech. Rep., U.S. Department of Education, Institute of Education Sciences (2023). 4. Daymont, T., Blau, G. & Campbell, D. Deciding between traditional and online formats: Exploring the role of learning advantages, flexibility, and compensatory adaptation. J. Behav. Appl. Manag. 12 , 156 (2011). [ Google Scholar ] 5. Broadbent, J. & Poon, W. L. Self-regulated learning strategies & academic achievement in online higher education learning environments: A systematic review. Internet High. Educ. 27 , 1–13 (2015). [ Google Scholar ] 6. Picciano, A. G. Beyond student perceptions: Issues of interaction, presence, and performance in an online course. J. Asynchronous Learn. Netw. 6 , 21–40 (2002). [ Google Scholar ] 7. Jimenez, L. Student assessment during COVID-19 (Center for American Progress, 2020). 8. Cahalan, M. et al. Indicators of Higher Education Equity in the United States: 2022 Historical Trend Report. Tech. Rep., The Pell Institute for the Study of Opportunity in Higher Education, Council for Opportunity in Education (COE), and Alliance for Higher Education and Democracy of the University of Pennsylvania (PennAHEAD), Washington, DC (2022). 9. Engle, J. & Tinto, V. Moving beyond access: College success for low-income, first-generation students. Tech. Rep., The Pell Institute for the Study of Opportunity in Higher Education (2008). 10. Berkner, L. & Chavez, L. Access to postsecondary education for 1992 high school graduates. Tech. Rep., National Center for Education Statistics, Washington, DC (1997). 11. Chen, X. First-generation students in postsecondary education: A look at their college transcripts. Tech. Rep., National Center for Education Statistics, Washington, DC (2005). 12. Choy, S. Low-income students: Who they are and how they pay for their education. Tech. Rep., National Center for Education Statistics, Washington, DC (2000). 13. Choy, S. Students whose parents did not go to college: Postsecondary access, persistence, and attainment. Tech. Rep., National Center for Education Statistics, Washington, DC (2001). 14. Horn, L. & Nunez, A. Mapping the road to college: First-generation students’ math track, planning strategies, and context of support. Tech. Rep., National Center for Education Statistics, Washington, DC (2000). 15. Nunez, A. & Cuccaro-Alamin, S. First-generation students: Undergraduates whose parents never enrolled in postsecondary education. Tech. Rep., National Center for Education Statistics, Washington, DC (1998). 16. Warburton, E., Bugarin, R. & Nunez, A. Bridging the gap: Academic preparation and postsecondary success of first-generation students. Tech. Rep., National Center for Education Statistics,Washington, DC (2001). 17. Astin, A. What matters in college: four critical years revisited (Jossey-Bass, San Francisco, 1997). 18. Cabrera, A., Nora, A. & Castaneda, M. The role of finances in the persistence process: A structural model. Res. High. Educ. 33 , 571–593 (1992). [ Google Scholar ] 19. Billson, J. & Terry, M. In search of the silken purse: Factors in attrition among first-generation students. College and University 58 , 57–75 (1982). 20. Lohfink, M. & Paulsen, M. Comparing the determinants of persistence for first-generation and continuing-generation students. J. Coll. Stud. Dev. 46 , 409–428 (2005). [ Google Scholar ] 21. Pascarella, E. T., Pierson, C. T., Wolniak, G. C. & Terenzini, P. T. Experiences and outcomes of first-generation students in community colleges. J. Coll. Stud. Dev. 44 , 420–429 (2003). [ Google Scholar ] 22. Pascarella, E. T., Pierson, C. T., Wolniak, G. C. & Terenzini, P. T. First-generation college students: Additional evidence on college experiences and outcomes. J. High. Educ. 75 , 249–284 (2004). [ Google Scholar ] 23. Pike, G. R. & Kuh, G. D. First- and second-generation college students: A comparison of their engagement and intellectual development. J. High. Educ. 76 , 276–300 (2005). [ Google Scholar ] 24. Richardson, R. C. & Skinner, E. F. Helping first generation minority students achieve degrees. In Zwerling, L. S. & London, H. B. (eds.) First Generation College Students: Confronting the Cultural Issues (Jossey-Bass Publishers, San Francisco, CA, 1992). 25. Terenzini, P. T., Springer, L., Yaeger, P. M., Pascarella, E. T. & Nora, A. First generation college students: Characteristics, experiences, and cognitive development. Res. High. Educ. 37 , 1–22 (1996). [ Google Scholar ] 26. Terenzini, P. T., Cabrera, A. F. & Bernal, E. M. Swimming Against the Tide: The Poor in American Higher Education (College Board, New York, 2001). 27. Means, B. & Neisler, J. Unmasking inequality: STEM course experience during the COVID-19 pandemic. Digital Promise (2020). 28. Means, B. & Neisler, J. Suddenly online: A national survey of undergraduates during the COVID-19 pandemic. Digital Promise (2020). 29. Barber, P. H. et al. Disparities in remote learning faced by first-generation and underrepresented minority students during COVID-19: Insights and opportunities from a remote research experience. J. Microbiol. Biol. Educ. 22 , (2021). [ DOI ] [ PMC free article ] [ PubMed ] 30. Goudeau, S., Sanrey, C., Stanczak, A., Manstead, A. & Darnon, C. Why lockdown and distance learning during the COVID-19 pandemic are likely to increase the social class achievement gap. Nat. Hum. Behav. 5 , 1273–1281 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 31. Deslauriers, L., Harris, S., Lane, E. & Wieman, C. Transforming the lowest-performing students: an intervention that worked. J. Coll. Sci. Teach. 41 , 80–88 (2012). [ Google Scholar ] 32. Freeman, S. et al. Prescribed active learning increases performance in introductory biology. CBE Life Sci. Educ. 6 , 132–139 (2007). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Jensen, P. A. & Moore, R. Students’ behaviors, grades and perceptions in an introductory biology course. Am. Biol. Teach. 70 , 483–487 (2008). [ Google Scholar ] 34. Jensen, P. A. & Moore, R. What do help sessions accomplish in introductory science courses? J. Coll. Sci. Teach. 38 , 60–64 (2009). [ Google Scholar ] 35. Moore, R. Who does extra-credit work in introductory science courses? J. Coll. Sci. Teach. 34 , 12–15 (2005). [ Google Scholar ] 36. Bowman, N. A. & Jang, N. What is the purpose of academic probation? Its substantial negative effects on four-year graduation. Res. High. Educ. 63 , 1285–1311 (2022). [ Google Scholar ] 37. Cahalan, M., Perna, L. W., Yamashita, M., Wright, J. & Santillan, S. 2018 Indicators of higher education equity in the United States: Historical trend report. Tech. Rep., The Pell Institute for the Study of Opportunity in Higher Education, Council for Opportunity in Education (COE), and Alliance for Higher Education and Democracy of the University of Pennsylvania (2018). 38. Palmer, T. L. & Palmer, S. A. A comparison of online and traditional chemistry lecture and lab. Chem. Educ. Res. Pract. 3 , 47–52 (2002). [ Google Scholar ] 39. Means, B., Toyama, Y., Murphy, R., Bakia, M. & Jones, K. Evaluation of evidence-based practices in online learning: A meta-analysis and review of online learning studies. Tech. Rep., US Department of Education (2009). 40. Chirikov, I., Semenova, T., Maloshonok, N., Bettinger, E. & Kizilcec, R. F. Online education platforms scale college stem instruction with equivalent learning outcomes at lower cost. Sci. Adv. 6 , eaay5324 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Jaggars, S. & Bailey, T. Effectiveness of fully online courses for college students: Response to a Department of Education meta-analysis (2010). 42. Figlio, D., Rush, M. & Yun, L. Is it live or is it internet? Experimental estimates of the effects of online instruction on student learning. J. Labor Econ. 31 , 763–784 (2013). [ Google Scholar ] 43. Broderick, J. R. & Neufeldt, E. J. John R. Broderick and Ellen J. Neufeldt Column: Filling high-tech jobs requires paradigm shift in higher ed. Richmond Times-Dispatch richmond.com/opinion/columnists/john-r-broderick-and-ellen-j-neufeldt-column-filling-high-tech-jobs-requires-paradigm-shift/article_41bb67aa-8ca6-522d-85c7-b9f4f15b56f6.html . 44. Bush, V. Science, the endless frontier: A report the president on a program for a postwar scientific research (United States Government Printing Office, Washington, DC). 45. Melguizo, T. & Wolniak, G. C. The earnings benefits of majoring in stem fields among high achieving minority students. Res. High. Educ. 53 , 383–405 (2012). [ Google Scholar ] 46. Xu, Y. J. Career outcomes of stem and non-stem college graduates: Persistence in majored-field and influential factors in career choices. Res. High. Educ. 54 , 349–382 (2013). [ Google Scholar ] 47. National Science Foundation, National Science Board. Revisiting the STEM Workforce: A Companion to Science and Engineering Indicators 2014. NSB-2015-10. Tech. Rep . (2015). 48. Carnevale, A. P., Smith, N. & Melton, M. STEM: Science Technology Engineering Mathematics (2011). 49. Chen, X. STEM attrition among high-performing college students: scope and potential causes. J. Technol. Sci. Educ. 5 , 41–59 (2015). [ Google Scholar ] 50. Olson, S. & Riordan, D. G. Engage to excel: producing one million additional college graduates with degrees in science, technology, engineering, and mathematics. Report to the President. (Executive Office of the President, Washington, DC, 2012). 51. Morganson, V. J., Major, D. A., Streets, V. N., Litano, M. L. & Myers, D. P. Using embeddedness theory to understand and promote persistence in STEM majors. Career Dev. Q. 63 , 348–362 (2015). [ Google Scholar ] 52. Provencher, A. & Kassel, R. High-impact practices and sophomore retention: Examining the effects of selection bias. J. Coll. Stud. Retent.: Res., Theory Pract. 21 , 221–241 (2019). [ Google Scholar ] 53. Chen, X. STEM attrition: College students’ paths into and out of STEM fields. Statistical Analysis Report. NCES 2014-001. Tech. Rep., Washington, DC (2013). 54. Crisp, G., Nora, A. & Taggart, A. Student characteristics, pre-college, college and environmental factors as predictors of majoring in and earning a STEM degree: An analysis of students attending a Hispanic serving institution. Am. Educ. Res. J. 46 , 924–942 (2009). [ Google Scholar ] 55. Whalen, D. F. & Shelley, M. C. I. Academic success for STEM and non-STEM majors. Jounral STEM Educ. 11 , 45–60 (2010). [ Google Scholar ] 56. Dika, S. L. & M, D. M. Early experiences and integration in the persistence of first-generation college students in STEM and non-STEM majors. J. Res. Sci. Teach. 53 , 368–383 (2016). [ Google Scholar ] 57. Xu, Y. J. Career outcomes of STEM and non-STEM college graduates: Persistence in majored-field and influential factors in career choices. Res. High. Educ. 54 , 349–382 (2013). [ Google Scholar ] 58. Skliarova, I., Meireles, I., Martins, N., Tchemisova, T. & Cacao, I. Enriching traditional higher STEM education with online teaching and learning practices: Student’s perspective. Educ. Sci. 12 , (2022). 59. Dickson-Karn, N. M. Student feedback on distance learning in the quantitative chemical analysis laboratory. J. Chem. Educ. 97 , 2955–2959 (2020). [ Google Scholar ] 60. Franchi, T. The impact of the COVID-19 pandemic on current anatomy education and future careers: A student’s perspective. Anat. Sci. Educ. 13 , 312–315 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Perets, E. A. et al. Impact of the emergency transition to remote teaching on student engagement in a non-STEM undergraduate chemistry course in the time of COVID-19. J. Chem. Educ. 97 , 2439–2447 (2020). [ Google Scholar ] 62. Petillion, R. J. & McNeil, W. S. Student experiences of emergency remote teaching: Impacts of instructor practice on student learning, engagement, and well-being. J. Chem. Educ. 97 , 2486–2493 (2020). [ Google Scholar ] 63. Wester, E. R., Walsh, L. L., Arango-Caro, S. & Callis-Duehl, K. L. Student engagement declines in STEM undergraduates during COVID-19 driven remote learning. J. Microbiol. Biol. Educ. 22, (2021). [ DOI ] [ PMC free article ] [ PubMed ] 64. Arcila Hernández, L. M., Zamudio, K. R., Drake, A. G. & Smith, M. K. Implementing team-based learning in the life sciences: A case study in an online introductory level evolution and biodiversity course. Ecol. Evol. 11 , 3527–2536 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Orlov, G. et al. Learning during the COVID-19 pandemic: It is not who you teach, but how you teach. Econ. Lett. 202 , (2021). [ DOI ] [ PMC free article ] [ PubMed ] 66. Mok, K., Xiong, W. & Bin Aedy Rahman, H. COVID-19 pandemic’s disruption on university teaching and learning and competence cultivation: Student evaluation of online learning experiences in Hong Kong. Int. J. Chinese Educ. 10 , (2021). 67. Parolin, Z. & Lee, E. Large socio-economic, geographic and demographic disparities exist in exposure to school closures. Nat. Hum. Behav. 5 , 522–528 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Al-Mahrouqi, T. et al. The differential mediating roles of resilience in the relationship between meaningful living and stress among college students during the COVID-19 pandemic. Sci. Rep. 13 , (2023). [ DOI ] [ PMC free article ] [ PubMed ] 69. Angrist, N., Bergman, P. & Matsheng, M. Experimental evidence on learning using low-tech when school is out. Nat. Hum. Behav. 6 , 941–950 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 70. Armstrong-Mensah, E., Ramsey-White, K., Yankey, B. & Self-Brown, S. COVID-19 and distance learning: Effects on Georgia State University School of Public Health students. Front. Public Health 8 , (2020). [ DOI ] [ PMC free article ] [ PubMed ] 71. Goldhaber, D. et al. The educational consequences of remote and hybrid instruction during the pandemic. Am. Econ. Rev.: Insights 5 , 377–392 (2023). [ Google Scholar ] 72. Lichand, G., Doria, C., Leal-Neto, O. & Fernandes, J. The impacts of remote learning in secondary education during the pandemic in Brazil. Nat. Hum. Behav. 6 , 1076–1086 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 73. Supriya, K. et al. Undergraduate biology students received higher grades during COVID-19 but perceived negative effects on learning. Sec. Educ. Psychol. 6 , (2021). 74. Kofoed, M., Gebhart, L., Gilmore, D. & Moschitto, R. Zooming to class? Experimental evidence on college students online learning during COVID-19. IZA Discussion Paper (2021). 75. Rodriguez-Planas, N.COVID-19, college educational outcomes, and the flexible grading policy: A longitudinal analysis. J. Public Econ. 207 , (2022). [ DOI ] [ PMC free article ] [ PubMed ] 76. Zuckerman, A. L., Hardesty, R. A., Denaro, K., Lo, S. M. & Owens, M. T. Effects of remote teaching in a crisis on equity gaps and the constructivist learning environment in an introductory biology course series. J. Microbiol. Biol. Educ. 22 , (2021). [ DOI ] [ PMC free article ] [ PubMed ] 77. Bird, K., Castleman, B. & Lohner, G. Negative impacts from the shift to online learning through the COVID-19 crisis: Evidence from a statewide community college system. AERA Open 8 , (2022). 78. Bulman, G. & Fairlie, R. The impact of COVID-19 on community college enrollment and student success: Evidence from California administrative data. Educ. Financ. Policy 17 , 745–764 (2022). [ Google Scholar ] 79. Chen, P. et al. Real-world effectiveness of a social-psychological intervention translated from controlled trials to classrooms. npj Sci. Learn. 7 , (2022). [ DOI ] [ PMC free article ] [ PubMed ] 80. Matz, R. et al. Analyzing the efficacy of ECoach in supporting gateway course success through tailored support (2021). Paper presented at LAK21: 11th International Learning Analytics and Knowledge Conference, Irvine, CA, USA, April 2021. 81. TRESTLE. Trestle home page (2017). https://trestlenetwork.ku.edu/ . 82. Ballen, C. J., Wieman, C., Salehi, S., Searle, J. B. & Zamudio, K. R. Enhancing diversity in undergraduate science: Self-efficacy drives performance gains with active learning. CBE Life Sci. Educ. 16 , 1–6 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 83. Donovon, D. A., Connell, G. L. & Grunspan, D. Z. Student learning outcomes and attitudes using three methods of group formation in a nonmajors biology class. CBE Life Sci. Educ. 17 , 1–14 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 84. Kim, K. J., Liu, S. & Bonk, C. J. Online MBA students’ perceptions of online learning: Benefits, challenges, and suggestions. Internet High. Educ. 8 , 335–344 (2005). [ Google Scholar ] 85. Michaelsen, L. K. & Sweet, M.The essential elements of team-based learning. New Directions for Teaching and Learning (2008). 86. Chasteen, S. V. & Code, W. J. The Science Education Initiative Handbook (2018). https://pressbooks.bccampus.ca/seihandbook/ .This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. 87. National Center for Education Statistics. IPEDS Data Feedback Report 2021: University of Michigan-Ann Arbor. Tech. Rep., National Center for Education Statistics, Washington, DC (2021). 88. National Center for Education Statistics. IPEDS Data Feedback Report 2022: University of Michigan-Ann Arbor. Tech. Rep., National Center for Education Statistics, Washington, DC (2022). 89. Pew Research Center. A rising share of undergraduates are from poor families, especially at selective colleges (2019). 90. Office of the Federal Register, National Archives and Records Administration. 84 FR 1167 - Annual Update of the HHS Poverty Guidelines [Government] (2019). 91. RTI International. First-generation college students: Demographic characteristics and postsecondary enrollment (2019). 92. Caruth, G. D. & Caruth, D. L. The impact of distance education on higher education: A case study of the United States. Turkish Online J. Distance Educ. 14 , 121–131 (2013). [ Google Scholar ] 93. Kentnor, H. E. Distance education and the evolution of online learning in the United States. Curric. Teach. dialogue 17 , 21–34 (2015). [ Google Scholar ] 94. Seaman, J. E., Allen, I. E. & Seaman, J. Grade increase: Tracking distance education in the United States (2018). 95. El-Deghaidy, H., Mansour, N., Alzaghibi, M. & Alhammad, K. Context of STEM integration in schools: Views from in-service science teachers. J. Math., Sci. Technol. Educ. 13 , 2459–2484 (2017). [ Google Scholar ] 96. Brancaccio-Taras, L., Mawn, M. V., Premo, J. & Ramachandran, R. Teaching in a time of crisis: Editorial perspectives on adjusting STEM education to the “new normal” during the COVID-19 pandemic. J. Microbiol.Biol. Educ. 22 (2021). [ DOI ] [ PMC free article ] [ PubMed ] 97. Alangari, T. S. Online STEM education during COVID-19 period: A systematic review of perceptions in higher education. EURASIA J. Math., Sci. Technol. Educ. 18 , (2022). 98. Fischer, C., Baker, R., Li, Q., Orona, G. & Warschauer, M. Increasing success in higher education: The relationships of online course taking with college completion and time-to-degree. Educ. Evaluation Policy Anal. 44 , 355–379 (2022). [ Google Scholar ] 99. Bettinger, E. P., Fox, L., Loeb, S. & Taylor, E. S. Virtual classrooms: How online college courses affect student success. Am. Economic Rev. 107 , 2855–2875 (2017). [ Google Scholar ] 100. Xu, D. & Jaggars, S. Performance gaps between online and face-to-face courses: Differences across types of students and academic subject areas. J. High. Educ. 85 , 633–659 (2014). [ Google Scholar ] 101. Bartley, S. J. & Golek, J. H. Evaluating the cost effectiveness of online and face-to-face instruction. J. Educ. Technol. Soc. 7 , 167–175 (2004). [ Google Scholar ] 102. Watson, J. & Gemin, B. Using online learning for at-risk students and credit recovery. Promising practices in online learning (2008). 103. U.S. Immigration and Customs Enforcement, Washington, DC. DHS STEM Designated Degree Program List (2023). 104. Koenker, R. & Bassett, G. Regression quantiles. Econometrica 46 , 33–50 (1978). [ Google Scholar ] 105. Barendse, S. Efficiently weighted estimation of tail and interquantile expectations (2020). 10.2139/ssrn.2937665. 106. He, X., Tan, K. M. & Zhou, W.-X. Robust estimation and inference for expected shortfall regression with many regressors. J. R. Stat. Soc.: Ser. B (Stat. Methodol.) 85 , 1223–1246 (2023). [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Information (816.4KB, pdf) Reporting Summary (81KB, pdf) Supplementary Code (9.9KB, txt) 41467_2026_69925_MOESM4_ESM.pdf (24.3KB, pdf) Description of Additional Supplementary Files Transparent Peer Review file (358.6KB, pdf) Source Data (24KB, xlsx) Data Availability Statement Currently, the LARC Data Set is available to University of Michigan (U-M) investigators who have obtained Institutional Review Board (IRB) approval and signed a memorandum of understanding describing their research interests. Investigators not affiliated with the University of Michigan must adhere to the established data request protocols of the University of Michigan to gain access to data. Investigators should contact [email protected] to request access to data. The expected time-frame for response to access requests depends on the the scope of the request. For accepted data requests, the timeline for data availability will be outlined in the data user agreement. More information about accessing the data can be found here: https://enrollment.umich.edu/data/learning-analytics-data-architecture-larc . Cleaned data are available from the authors upon university approval for LARC data. Individuals interested in obtaining this data should follow above steps to obtain approval from the University of Michigan. Source data are provided with this paper. Code to perform the expected shortfall regression method is available in the Supplementary Code. Articles from Nature Communications are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (714.9 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top