💻 computer science

Coasting Through Class: Learning Opportunity Loss from Practice Avoidance During Individual Seatwork

This study analyzes ASSISTments data to reveal that students "coast" through 60% of available math practice time due to delayed starts, idling, and early stopping, a consistent behavioral pattern that significantly reduces learning opportunities and negatively impacts standardized test performance.

Original authors: Ashish Gurung, Jordan Gutterman, Danielle R. Thomas, Mingyu Feng, Vincent Aleven, Kenneth R. Koedinger

Published 2026-04-29

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Ashish Gurung, Jordan Gutterman, Danielle R. Thomas, Mingyu Feng, Vincent Aleven, Kenneth R. Koedinger

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Empty Bus" Problem

Imagine a school bus that leaves the garage every morning with a full tank of gas and a strict schedule to get to the destination. The bus has enough fuel to drive for 60 minutes. However, the driver only actually drives the bus for 24 minutes. For the remaining 36 minutes, the engine is running, the bus is sitting at the curb, or the driver is just waiting to start, but the bus isn't moving toward the destination.

This paper studies exactly that scenario, but instead of a bus, it's a digital math class. The researchers wanted to know: How much time are students actually spending learning, and how much time are they just "sitting there" without doing the work?

They call this lost time "Coasting." Just like a car coasting downhill without the engine working, students are "coasting through class" when they aren't actively practicing math, even though they are sitting at their computers.

What is "Coasting"?

The researchers broke "coasting" down into three specific ways students waste time during a class session:

The "Delayed Start" (Waiting at the Red Light): The class period begins, but the student doesn't start working immediately. They might be chatting, looking out the window, or just staring at the screen. This is time lost at the very beginning.
The "Mid-Practice Idle" (The Engine Idling): The student starts working, but then stops for a long time (more than 2 minutes) without doing anything. They aren't answering questions; they are just sitting there.
The "Early Stop" (Parking the Car Early): The class period isn't over, and there is still more math to do, but the student shuts down their computer or stops working because they finished their assigned homework. They stop "coasting" before the bell rings.

The Formula:

Total Coasted Time = Delayed Start + Idle Time + Early Stop

What Did They Find?

The researchers looked at data from over 1,400 students using a math platform called ASSISTments. Here is what they discovered:

The 40/60 Split: Students only used 40% of their available class time to actually practice math. The other 60% was "coasted time."
Where the time went:
- 62% of the lost time was because students stopped early.
- 36% was because they waited too long to start.
- Only 2% was because they were just sitting in the middle of a problem doing nothing.
- Analogy: It's not that students are stuck in traffic (idling) in the middle of the lesson; it's that they are late to the party (delayed start) or they leave the party as soon as they get a snack (early stop).
It's a Habit: The researchers checked if this behavior was random or a pattern. They found that if a student tends to start late or stop early one day, they are likely to do it again the next day. It's a consistent habit, not just a one-time mistake.
Even "Good" Students Coast: Even when the researchers removed the time students stopped early because they had finished all their work, students still wasted about 32% of their time. This means even when there is more work available, many students just stop.

Does It Matter Who the Student Is?

The researchers wondered if certain groups of students coasted more than others (e.g., boys vs. girls, different races, or students with special needs).

Gender: At first, it looked like boys coasted more than girls. However, when they adjusted for the fact that boys often finished their assigned work faster, the difference disappeared. It wasn't that boys were avoiding work; they just finished the required tasks quicker.
Special Needs (IEP): Surprisingly, students with Individualized Education Programs (IEPs) coasted less than others. The researchers guess this might be because these students get more teacher help or have modified workloads that keep them engaged longer.
The Big Takeaway: Coasting happens to everyone. It doesn't matter if you are rich or poor, live in a city or a farm, or speak English as a second language. The "coasting" behavior is universal across all these groups.

The Secret to Better Grades: "Extra Effort"

This is the most important finding of the paper.

The researchers asked: Does spending more time actually help?

The Answer: Yes, but only if you choose to keep going.
The Analogy: Imagine two runners. Runner A runs the required 5 miles and stops. Runner B runs the required 5 miles, but then decides to run 2 more miles because they want to.
The Result: The students who did "Extra Effort" (stayed after finishing their first assignment to do more math) scored significantly higher on standardized tests.
The Catch: Just sitting at the computer for a long time (Time on Task) didn't predict better grades if the student wasn't actually working. But voluntarily choosing to keep working after finishing the minimum requirement was a huge predictor of success.

Why This Matters for Teachers and Schools

The paper suggests that digital learning platforms are great tools, but they can't fix the problem if students aren't using the time.

The Problem: Schools give students a set amount of time to learn. If students only use 40% of it, the system is failing, even if the software is perfect.
The Solution: Teachers need to set clear expectations. Students need to know that "finishing the assignment" doesn't mean "stop working." They need to know there is more to do.
The Tech Fix: The software itself could help by showing prompts like, "You're done with the first part! Here are 5 more problems to try," or by alerting teachers when a student stops working early.

Summary

This paper introduces a new way to look at student behavior called "Coasting." It reveals that in digital math classes, students waste a massive amount of time (60%) by starting late, stopping early, or just waiting. This happens to almost everyone, regardless of their background. However, the students who break the habit of coasting and choose to do "Extra Effort"—voluntarily practicing more than required—are the ones who end up with the best grades. The lesson is simple: Time spent isn't the same as time learned; you have to actually use the time.

1. Problem Statement

While digital learning platforms (DLPs) are designed to optimize learning through adaptive support, their efficacy is contingent on students actively utilizing allocated practice time. Existing research heavily focuses on unproductive engagement (e.g., "gaming the system," "mind-wandering," or "wheel-spinning") where students are present but not learning effectively. However, absolute disengagement—where students avoid the task entirely—is understudied because it often leaves little to no log trace.

The authors identify a gap in quantifying practice avoidance during structured classwork sessions (individual seatwork). Specifically, they seek to measure the loss of learning opportunities caused by:

Delayed Starts: Students waiting to begin work after the session begins.
Mid-practice Idling: Periods of inactivity while the session is ongoing.
Early Stops: Students ceasing work before the session ends, even if more practice time is available.

The paper posits that this "lost time" represents a significant barrier to learning that current metrics fail to capture comprehensively.

2. Methodology

Dataset

Source: Publicly available logs from ASSISTments, a mathematics DLP providing immediate feedback and on-demand help.
Scope: Two school years (2022–2023, 2023–2024).
Participants: $N = 1,425$ students across 71 classes, 34 teachers, and 22 schools in 17 U.S. states.
Data Types: Transaction logs (~3 million), Measure of Academic Progress (MAP) pre-test (Fall) and post-test (Spring) scores, and demographic data (gender, race/ethnicity, SES, IEP, ELL, urbanicity).

Operational Definitions & Measures

The authors define "Coasted Time" as the sum of three session-level behavioral measures:
$\text{Coasted Time} = \text{Delayed Start} + \text{Idle Time} + \text{Early Stop}$

Session Identification: Used a temporal inference algorithm (Gurung et al., 2026) to classify logs into Classwork, At-school Independent Work, and Homework. Classwork was defined as sessions during school hours (7 AM–3 PM) with >5 concurrent active students.
Delayed Start: Time elapsed between the session start (inferred from peer activity) and the student's first action.
Idle Time: Inactivity periods exceeding the 99th percentile of response times (rounded to 2 minutes).
Early Stop: Time remaining in the session after the student stops working.
- Adjustment: To distinguish avoidance from efficiency, an "adjusted" metric was calculated where early stop time was set to 0 if the student completed at least one assignment.
Extra Effort: Defined as continued practice after the completion of the first required assignment within a classwork session.

Analytical Approach

Temporal Stability: Used Generalizability Theory (G-Theory) to calculate G-coefficients and $\phi$ -coefficients, assessing the reliability of coasting behaviors across time.
Demographic Analysis: Employed multilevel linear regression (students nested within classes) to examine variance in coasting based on gender, race/ethnicity, SES, IEP, ELL, and school locale, controlling for prior performance.
Predictive Validity: Used multilevel regression to test if Extra Effort predicts Spring MAP scores, controlling for prior performance, total time on task, and class-level variance. Model fit was evaluated using Bayesian Information Criterion (BIC) and $R^2$ .

3. Key Contributions

Conceptual Framework: Introduces "Coasting Behavior" as a unified framework for absolute disengagement, distinguishing it from unproductive engagement. It combines session-level measures (delayed start, early stop) with within-task measures (idling).
Methodological Extension: Extends session-level inference algorithms previously validated on Cognitive Tutor and iReady to the ASSISTments platform, demonstrating cross-system generalizability.
Quantification of Opportunity Loss: Provides empirical evidence that a majority of allocated classwork time is lost to avoidance behaviors, challenging the assumption that time allocated equals time utilized.
Differentiation of Effort: Distinguishes between "Time on Task" (working toward completion) and "Extra Effort" (working beyond completion), showing that the latter is a stronger predictor of learning outcomes.

4. Key Results

Extent and Composition of Coasting

Utilization Rate: Students dedicated only 40.15% of available classwork time to active math practice.
Lost Time: 59.85% of time was "coasted."
- Early Stop: 62.25% of coasted time (the largest component).
- Delayed Start: 35.79% of coasted time.
- Idle Time: 1.96% of coasted time (minimal).
Adjusted Findings: Even when excluding early stops attributable to assignment completion (setting early stop = 0 for those who finished), 32% of available time remained unutilized.

Reliability (Temporal Stability)

Delayed Start: Moderate reliability ( $G = 0.73$ ).
Early Stop: Moderate reliability ( $G = 0.71$ ).
Idle Time: Low reliability ( $G = 0.53$ ), suggesting idling is highly context-dependent (e.g., problem difficulty) rather than a stable trait.
Implication: Delayed starts and early stops are consistent behavioral patterns across sessions.

Demographic Variance

Gender: Unadjusted data showed males coasted more than females (approx. 32 seconds/session). However, this difference disappeared after adjusting for assignment completion, suggesting males simply finished assignments faster rather than avoiding work.
IEP Status: Students with IEPs coasted less than non-IEP peers in unadjusted models, but this difference also vanished after adjustment.
Other Factors: No significant differences were found based on race/ethnicity, SES, ELL status, or school locale (urban vs. rural). Coasting appears to be a universal phenomenon across demographics.

Impact on Learning (Extra Effort)

Time on Task: The proportion of time spent working toward initial completion was not a significant predictor of Spring MAP scores ( $p = .919$ ).
Extra Effort: Continued practice beyond the first assignment was a significant positive predictor of Spring MAP scores ( $\beta = 1.44, p < .001$ $β = 1.44, p < .001$ ).
- A 10% increase in extra effort proportion correlated with a 0.14 SD increase in test scores.
- Adding "Extra Effort" to the model resulted in a very strong improvement in model fit ( $\Delta BIC = 11$ ).

5. Significance and Implications

Theoretical Significance

Multilevel Disengagement: The study demonstrates that disengagement operates at two levels: the session level (avoiding the start or stopping early) and the task level (idling).
Opportunity Loss: It reframes "time on task" not just as a duration metric, but as a measure of opportunity utilization. The disconnect between allocated time and actual practice time is a critical, previously under-measured variable in learning science.
Self-Regulation: The predictive power of "Extra Effort" suggests that voluntary continuation of practice reflects deeper self-regulatory traits (e.g., mastery orientation, intrinsic motivation) that are more critical for learning than mere exposure time.

Practical Implications

For Educators: Teachers must set explicit expectations for sustained engagement throughout the entire class period, not just until the first assignment is done. Interventions should focus on "what to do next" for early finishers.
For Platform Designers: DLPs need affordances that support sustained engagement, such as:
- Progress indicators.
- Prompts for inactivity.
- Clear pathways for extension activities to prevent early stopping due to lack of direction.
For Policy: Evaluations of DLP efficacy must account for "coasting." A platform may be pedagogically sound, but if 60% of allocated time is lost to avoidance, the return on investment is severely diminished.

Limitations

Correlational Nature: Causality between coasting and achievement cannot be definitively established without experimental designs.
Inferred Boundaries: Session start/end times are inferred from log data, which may introduce noise if students deviate significantly from class norms.
Generalizability: The sample is limited to middle school math; patterns may differ in other subjects or grade levels.

Conclusion

The paper concludes that "coasting" is a pervasive, consistent behavioral pattern resulting in substantial learning opportunity loss (approx. 60% of class time). While demographic factors do not significantly drive this behavior, the distinction between minimal compliance (finishing the first task) and extra effort is crucial. Students who voluntarily extend their practice time achieve significantly better outcomes, highlighting the need for educational systems to foster and support sustained engagement beyond the minimum requirements.