How Did the COVID-19 Pandemic Disrupt and Reshape Student Learning Trajectories?

This study asks how the COVID-19 pandemic affected student learning trajectories in Michigan, particularly in math and reading. Researchers analyzed statewide M-STEP test data from 2016–2023 and benchmark assessments from 2020–2023 for 5th and 6th graders. They found that math achievement growth fell by about 0.20 standard deviations between 2019 and 2022, while declines in reading were small and statistically insignificant. Losses were larger for Black, Latino, and economically disadvantaged students, and for those who spent more time in remote learning. Although math scores partially rebounded after 2021, recovery stalled by 2023, leaving achievement below pre-pandemic levels.

The Policy Scientist’s Perspective

This article addresses one of the most consequential educational policy issues of the past half-century—the persistence of learning loss following the COVID-19 pandemic. The topic matters far beyond Michigan, as it highlights the structural fragility of instructional systems under large-scale disruption. Its findings—particularly the 0.20 standard deviation decline in math achievement and the stalled recovery—underscore how prolonged interruptions can entrench inequality across racial and economic lines. While the study relies on observational rather than causal methods, its large, longitudinal data set and careful statistical controls provide credible, policy-relevant estimates. The use of both statewide summative assessments and frequent benchmark tests enhances measurement reliability and temporal precision. Although results generalize most directly to states with similar demographics and assessment systems, the pattern of stalled recovery likely applies more broadly. Despite the absence of experimental design, this paper stands out among recent publications for its scope, data quality, and clarity in tracing long-term achievement trajectories.

Full Citation and Link to Article

Strunk, K. O., Kilbride, T., Imberman, S., Walker, S., Yu, D., & Hopkins, B. (2025). The path of student learning delay during the COVID-19 pandemic: Evidence from Michigan. Education Finance and Policy. Advance online publication. https://doi.org/10.1162/edfp.a.26

Extended Summary

Central Research Question

The central question addressed in this study is how the COVID-19 pandemic altered student learning trajectories in Michigan, particularly in mathematics and English language arts (ELA), and whether recovery efforts since 2021 have been sufficient to return achievement to pre-pandemic levels. The authors—Strunk, Kilbride, Imberman, Walker, Yu, and Hopkins—seek to quantify both the magnitude and duration of learning losses across different student subgroups and instructional modalities. Specifically, they ask how academic progress, as measured by standardized tests, evolved from the pre-pandemic period through 2023, whether disparities widened across racial and socioeconomic lines, and to what extent the mode of instruction (remote, hybrid, or in-person) influenced these patterns. The study’s primary policy relevance lies in identifying whether the learning losses observed early in the pandemic were temporary or enduring.

The research is motivated by the observation that national assessments have shown both severe learning disruptions and uneven recoveries. The authors argue that understanding these patterns within one state, where both summative and benchmark data can be linked, offers unusually fine-grained insight into how achievement changed across time and populations. This focus allows them to model not just the overall decline but also the trajectory of recovery—or stagnation—over several school years.

Previous Literature

The study builds upon a large and rapidly growing body of research on pandemic-related learning loss. Early national reports (Goldhaber et al., 2022; Kuhfeld & Lewis, 2022) established that student test scores fell sharply in 2020–2021 and remained below pre-pandemic baselines through 2022. These declines were particularly severe in mathematics, where the National Assessment of Educational Progress (NAEP) reported the steepest drops in over 50 years. The literature consistently documents disproportionate effects on Black, Latino, and economically disadvantaged students, as well as those attending high-poverty schools or receiving remote instruction. Studies such as Sass and Ali (2022) and Kogan and Lavertu (2021) found that remote learning environments were associated with significantly lower academic growth, especially in math.

Recent analyses have shown that the academic recovery observed in 2021–2022 slowed or even stalled by 2023 (Lewis & Kuhfeld, 2023). Moreover, national evidence suggests that pre-existing achievement gaps widened during the pandemic and have remained elevated. While much of this research uses national or multi-state data, relatively few studies combine state-level summative exams with high-frequency benchmark assessments. The authors argue that this approach allows them to measure both total achievement change and intra-year growth rates—something unavailable in most prior work.

By integrating these two data sources, the study extends beyond the single-year snapshots typical of national analyses. It also contributes to the literature by examining how instructional modalities shaped achievement over multiple years, adding a longitudinal perspective to what has been largely cross-sectional evidence. The paper situates itself within this literature as one of the first to use data through spring 2023, thereby addressing the question of whether recovery has continued or plateaued.

Data

The analysis draws on two primary sources of student achievement data from Michigan: (1) the Michigan Student Test of Educational Progress (M-STEP), the state’s annual summative assessment in math and ELA for grades 3–7, and (2) district-level benchmark assessments—NWEA MAP Growth and Curriculum Associates’ i-Ready—administered each fall and spring between 2020–2023. Together, these data sets allow the authors to compare both long-term growth and short-term learning dynamics during and after the pandemic.

The M-STEP data include student-level records from 2015–2016 through 2022–2023. Because the exam was canceled in spring 2020 and participation was optional in 2021, the analysis focuses on comparing a pre-pandemic cohort (tested in 2016 and 2019) with a pandemic cohort (tested in 2019 and 2022). This design permits a three-year growth comparison between otherwise similar groups of students, providing an estimate of pandemic-related changes in learning trajectories.

The benchmark assessments complement this by capturing more frequent data on the same cohorts. Benchmark results are available at the district-grade-subgroup level (e.g., by race and economic status) for students continuously enrolled from fall 2020 through spring 2023. The sample includes roughly 97,700 students—about two-thirds of Michigan’s 5th and 6th graders. Benchmark scores are standardized using national norms to allow comparability across testing vendors and grades.

Additional data include student demographic characteristics, district-level instructional modality data (measured by months of in-person instruction in 2020–21), and county-level COVID-19 death rates to approximate community-level pandemic intensity. This combination enables analyses of heterogeneity by race, socioeconomic status, and mode of instruction, while accounting for contextual pandemic severity. The data set is unusually rich for state-level education research, covering both pre-pandemic and multi-year post-pandemic periods. The main limitations are the absence of student-level benchmark data and potential measurement noise from differing test conditions, particularly for remote administrations in 2020–21.

Methods

The study employs two complementary statistical models. The first estimates differences in three-year growth between the pre-pandemic and pandemic cohorts using individual-level M-STEP data. The dependent variable is standardized score growth over three years, while key independent variables include a cohort indicator, base-year achievement, and a vector of student characteristics (gender, race, special education status, English learner status, economic disadvantage, etc.), with grade and district fixed effects. The estimated coefficient on the pandemic cohort captures the average learning loss relative to the pre-pandemic period. Interaction terms between the pandemic indicator and subgroup variables allow the authors to assess heterogeneous effects by race, economic status, and instructional modality.

The second model examines within-pandemic trajectories using the benchmark assessment data. Here, the dependent variable is the average standardized test score at the district-grade-semester level. Time indicators represent each testing period from fall 2020 through spring 2023, with coefficients interpreted as deviations from the pre-pandemic baseline (spring 2019). The model includes grade and district fixed effects, as well as controls for changing demographic composition and local COVID-19 incidence. This approach captures the shape of learning trajectories—sharp drops, partial rebounds, and later stagnation—over time.

Although both models rely on regression-based comparisons rather than causal identification, they are well-specified for descriptive inference. The inclusion of fixed effects reduces bias from unobserved, time-invariant district characteristics. However, the lack of random assignment or quasi-experimental variation means the findings should not be interpreted as causal. Nonetheless, the authors’ methodological rigor and the use of multiple data sources yield robust and internally consistent estimates of learning dynamics.

Findings/Size Effects

The results demonstrate clear and substantial learning losses during the pandemic, concentrated in mathematics. Students in the pandemic cohort exhibited 0.20 standard deviations lower growth in math compared to their pre-pandemic counterparts. By contrast, ELA growth declined by only 0.03 standard deviations, a difference that was not statistically significant. These effects are large by educational standards, roughly equivalent to half a year of missed learning in math.

The learning losses were not evenly distributed. Black, Latino, and economically disadvantaged students experienced the steepest declines, with math growth 0.04 to 0.07 standard deviations lower than comparable White or higher-income peers. Students in districts that spent part or all of 2020–21 in remote instruction performed 0.05 standard deviations lower in math than those with continuous in-person learning. However, differences across instructional modalities were smaller for ELA.

The benchmark assessment results show a more nuanced trajectory. Michigan students’ math and reading scores fell sharply during the 2020–21 school year—roughly 0.10 standard deviations below national norms by spring 2021—then partially rebounded in 2021–22. However, by spring 2023, recovery had stalled, with math and reading still about 0.07–0.08 standard deviations below pre-pandemic levels. The pattern was consistent across both M-STEP and benchmark data, lending confidence to the findings.

Subgroup analyses reveal that while initial losses were larger for minority and low-income students, these groups showed somewhat stronger recovery rates, narrowing the relative gap by 2023. Nonetheless, absolute achievement remained below pre-pandemic baselines for all groups. Students in districts with full in-person instruction recovered fastest, particularly in math, whereas those in districts with extended remote learning never fully closed the gap.

Overall, the size effects indicate that the pandemic caused substantial, durable, and demographically uneven reductions in learning. The persistence of these losses even after two full years of recovery efforts suggests that academic “catch-up” has largely plateaued.

Conclusion

The study concludes that Michigan students, particularly in mathematics, have not regained the ground lost during the pandemic. Despite partial improvement between 2021 and 2022, achievement in both math and ELA remains below pre-pandemic levels. The authors characterize the recovery as “stalled,” a term consistent with national evidence showing limited progress since 2022. They warn that unless student achievement growth accelerates beyond historical rates, pandemic-induced learning delays are likely to persist into later grades.

The policy implications are significant. The findings imply that returning to pre-pandemic instructional practices will be insufficient to restore achievement levels. The authors call for sustained monitoring of student performance and targeted interventions—particularly in mathematics—to prevent long-term scarring effects on cohorts affected by the pandemic. They also note that Michigan’s combination of benchmark and summative data provides a model for other states seeking to track learning recovery with greater precision.

In evaluating the research, the study’s strengths lie in its unusually comprehensive data set, multi-year perspective, and consistent patterns across distinct testing systems. While it does not employ causal inference methods such as randomized interventions or natural experiments, its internal consistency and analytic transparency make it one of the strongest descriptive accounts available to date. The findings generalize most directly to states with similar demographic profiles and pandemic-related schooling disruptions, though the broad trends likely apply nationwide.

In sum, this study represents one of the most rigorous and data-rich examinations of pandemic-era learning trajectories to date. It provides compelling evidence that the learning losses caused by COVID-19 were deep, inequitable, and enduring—and that recovery has, at least for now, reached a troubling plateau.