Does Remote Learning Exposure Harm Student Attendance?

Singer (2026) examines whether the duration of remote learning in 2020–21 affected student attendance after the pandemic. He uses longitudinal administrative data on nearly one million Michigan students from 2017–18 through 2023–24, combined with district-level measures of remote learning duration. Using difference-in-differences and instrumental variables, he finds that each additional month of remote learning reduced post-pandemic attendance by about 0.46 percentage points. Effects are non-linear: students with four to six months missed about 3–4 more days, while those with seven to nine months missed roughly 6–9 additional days annually.

Why This Article Was Selected for The Policy Scientist

This article addresses a policy issue with broad implications: sustained declines in student attendance following the COVID-19 pandemic and the institutional factors that may have contributed to them. Attendance is a foundational input into learning, human capital formation, and long-term labor market outcomes, making this topic central to education policy and economic productivity. The study is timely given persistent national concerns about chronic absenteeism and incomplete recovery to pre-pandemic norms. It contributes to a limited but growing literature by isolating remote learning duration as a mechanism, extending prior descriptive and cross-sectional work. The longitudinal administrative dataset is strong, though limited to Michigan, which may constrain generalizability. The use of difference-in-differences and instrumental variables represents a credible causal inference strategy, though not equivalent to randomized designs.

Full Citation and Link to Article

Singer, J. (2026). Remote learning in 2020–21 and student attendance since the COVID-19 pandemic. Education Finance and Policy. https://doi.org/10.1162/EDFP.a.449

Central Research QuestionThis study examines whether, and to what extent, exposure to remote-only learning during the 2020–21 school year contributed to persistent declines in student attendance in the post-pandemic period. The central inquiry is explicitly causal in orientation: what is the effect of the duration of remote learning on subsequent student attendance outcomes? The analysis focuses not simply on whether remote learning mattered, but on how the intensity of exposure—measured in months of remote-only instruction—shaped later attendance behavior. By emphasizing duration, the study moves beyond binary distinctions between remote and in-person learning and instead evaluates whether marginal increases in exposure produced measurable differences in absenteeism. This framing allows the author to assess both linear and non-linear relationships, as well as the persistence of any observed effects across multiple post-pandemic school years.

Previous LiteratureThe article situates itself within several adjacent strands of research, including studies on pandemic-era learning modalities, student attendance, and broader educational disruptions. Prior work has documented substantial increases in chronic absenteeism during and after the pandemic, with national rates nearly doubling relative to pre-pandemic baselines. However, much of this literature remains descriptive, identifying trends without isolating specific mechanisms. A smaller set of studies has examined the relationship between remote learning and attendance, though these have largely relied on cross-sectional or district-level data, limiting causal interpretation.

Existing research suggests that longer exposure to virtual instruction is associated with higher rates of absenteeism, particularly in higher-poverty contexts, though findings are not entirely consistent across settings. Other work has emphasized related channels, such as deteriorations in student mental health, weakened school-family relationships, and shifts in parental attitudes toward attendance. The present study builds on this literature by introducing longitudinal, student-level data and applying quasi-experimental designs to better approximate causal inference. In doing so, it addresses a key gap: the lack of evidence linking pandemic-era instructional choices to persistent behavioral outcomes in the post-pandemic period.

DataThe empirical analysis draws on a comprehensive longitudinal dataset of Michigan public school students spanning the 2017–18 through 2023–24 academic years. The dataset includes over 5 million student-year observations and nearly one million unique students, providing substantial statistical power and the ability to track individual outcomes over time. Student-level administrative records include attendance, demographic characteristics, enrollment histories, and indicators of economic disadvantage.

These data are linked to district-level measures of instructional modality during the 2020–21 school year, constructed from monthly reports of whether districts offered in-person, hybrid, or remote-only learning. The key independent variable is the number of months a student’s district provided remote-only instruction, which serves as a proxy for exposure to remote learning. Additional contextual variables include district-level political partisanship, COVID-19 case rates, and demographic composition, which are used both as controls and as instruments in the causal analysis.

The dataset is strong in several respects. It combines longitudinal depth with broad coverage, allowing for precise measurement of pre- and post-treatment outcomes. It also includes rich covariates that help mitigate confounding. However, the reliance on district-level exposure measures introduces some measurement error, as individual students may have experienced different modalities within the same district. The geographic concentration in a single state also raises questions about external validity, particularly given variation in policy responses across states.

MethodsThe study employs two complementary quasi-experimental strategies: a difference-in-differences (DiD) design and an instrumental variables (IV) approach. The DiD framework compares changes in attendance before and after the pandemic across students with varying levels of exposure to remote learning. By incorporating baseline attendance measures, student-level covariates, and fixed effects for grade and time, the model aims to isolate the contribution of remote learning duration from other confounding factors.

An event-study extension is used to assess pre-treatment trends and evaluate the parallel trends assumption, which is critical for causal interpretation. The absence of significant pre-trends strengthens the plausibility of the DiD estimates. The analysis also models remote learning duration both continuously and categorically, allowing for the detection of non-linear effects.

The IV strategy serves as a robustness check, addressing potential selection bias in district decisions about instructional modality. The instrument is constructed from predictors of reopening decisions, including local political partisanship, COVID-19 case rates, and district demographics. These variables are used to generate predicted values of remote learning duration, which are then used to estimate its effect on attendance. While the IV approach strengthens causal claims, it identifies a local average treatment effect for a specific subset of students, limiting generalizability.

Overall, the methodological approach reflects a strong commitment to causal inference using observational data. Although not equivalent to a randomized controlled trial, the combination of DiD and IV represents a credible strategy for estimating treatment effects in this context.

Findings/Size EffectsThe results indicate a statistically significant and substantively meaningful relationship between remote learning duration and post-pandemic attendance, with effects that are both persistent and non-linear. In the linear specification, each additional month of remote-only learning is associated with a 0.46 percentage point decline in attendance, equivalent to approximately one additional missed day in a 180-day school year.

However, the categorical estimates reveal important threshold effects. Students exposed to one to three months of remote-only learning show no statistically significant decline in attendance relative to those with no exposure. In contrast, students with four to six months of exposure experience attendance declines of roughly 2 percentage points, corresponding to 3–4 additional absences per year. For students with seven to nine months of remote learning, the effects are substantially larger, with declines ranging from approximately 3.5 to 5 percentage points, or 6–9 additional missed days annually.

These effects persist across multiple post-pandemic years, though there is some attenuation over time, particularly for students with the highest levels of exposure. Parallel analyses using chronic absenteeism as the outcome yield consistent results, with increases of 5–10 percentage points in the probability of being chronically absent for students with extended remote learning exposure.

The findings also indicate modest heterogeneity by socioeconomic status and race/ethnicity, with larger initial effects for disadvantaged students, though these differences diminish over time. Importantly, the study concludes that while remote learning contributed to attendance declines, it explains only a modest share of the overall statewide trend, given that most students experienced relatively short periods of remote-only instruction.

ConclusionThe study provides evidence that extended exposure to remote-only learning during the pandemic had lasting effects on student attendance, particularly for those with prolonged exposure. By leveraging longitudinal data and quasi-experimental methods, it advances the literature beyond descriptive accounts and offers a more precise estimate of the relationship between instructional modality and behavioral outcomes.

At the same time, the findings underscore the limits of remote learning as a singular explanatory factor. The modest aggregate impact suggests that broader structural and behavioral changes also contributed to post-pandemic attendance patterns. The study’s reliance on observational data, despite the use of DiD and IV techniques, leaves open the possibility of residual confounding. Future research using randomized or more tightly identified causal designs could further strengthen inference.

In sum, the article makes a meaningful contribution by clarifying the role of remote learning duration in shaping post-pandemic attendance, while highlighting the complexity of the mechanisms underlying persistent absenteeism.