Cross-Subject Predictive Validity for Learning Outcomes of Delayed Start Behavior
Jordan Gutterman, Ashish Gurung, Lee Branstetter, Kenneth Koedinger, Vincent Aleven · 2026
Read Cross-Subject Predictive Validity for Learning Outcomes of Delayed Start Behavior by Jordan Gutterman, Ashish Gurung, Lee Branstetter, Kenneth Koedinger, Vincent Aleven (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-25308.txt.
Introduction
Software logs can reveal how students regulate their own learning, down to how long they sit idle before starting a session. Jordan Gutterman and colleagues study that pause, called delayed start, in 2026 using practice data from seven hundred eleven seventh graders. They ask whether delay habits during maths practice predict standardised test results in other subjects too. Mixture modelling also splits students into behavioural subgroups so the measure can be applied in real classrooms.
Delays during maths predicted later scores in both maths and English, a cross subject signal. Students who typically began within five minutes, about a fifth of the group, grew more than average in both subjects. Another fifth, averaging over thirteen minutes of delay, trended in the opposite direction. The authors argue session level behaviour adds a content free engagement measure beside usage and progress data.
Catalog record
| CREATOR | Jordan Gutterman, Ashish Gurung, Lee Branstetter, Kenneth Koedinger, Vincent Aleven |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.25308
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Cross-Subject Predictive Validity for Learning Outcomes of Delayed Start Behavior Jordan Gutterman, Ashish Gurung, Lee Branstetter, Kenneth Koedinger, Vincent Aleven Abstract Behavioral detectors provide valuable insights into learner motivation and self-regulation. Among these, delayed start, a new session-level detector, has shown great promise as a valid behavioral measure that generalizes well across systems. In this paper, we examine cross-subject predictive validity of delayed start behavior. Using iReady data from 711 grade 7 students, we find delayed starts during Math practice are predictive of standardized test performance in both Math ( β \beta =.07 SD, p=.02) and English ( β \beta =.10 SD, p=<.001). Additionally, using mixture modeling and sensitivity analyses, we use a data-driven strategy to operationalize the identification of delayed starters in practice. We identify two underlying sub-groups of interest: “early starters” (<5 minute average delay, 20% of students) and “chronic delayers” (>13 minutes average delay, 20% of students). Relative to students in neither sub-group, early starters experienced greater growth (Math β \beta =.11 SD, p=.07; ELA β \beta =.15 SD, p=.02), while chronic delayers had the opposite trends (Math β \beta = − - .13 SD, p=0.05; ELA β \beta = − - .11 SD, p=0.11). Session-level measures provide a new opportunity for content-independent detectors, adding a behavioral component to the traditional usage and progress based on student engagement with content. This work aims to bridge education research with classroom practice by developing interpretable measures that align with behavioral cues teachers already use during classwork sessions to monitor and support students. 1 Introduction Self-regulated learning helps conceptualize and interpret learner behaviors, using frameworks that describe how students plan, execute, and control their learning. AIED research has informed the development of detectors of learner behavior in digital learning platforms (DLPs), which identify both productive and unproductive patterns of engagement. These detectors capture a range of learner behaviors, including effort [ 32 , 12 ] , gaming the system [ 3 ] , off-task behavior [ 34 ] , and procrastination [ 1 , 7 ] . Beyond insights into productive and unproductive engagement, these behaviors are predictive of short- and long-term academic outcomes [ 27 , 31 ] . Building on this foundation, behavioral detectors have been integrated into teaching augmentation tools (e.g., teacher-facing dashboards and augmented reality glasses) to support real-time pedagogical decision-making. Recent trends suggest the emerging need for identifying and intervening on behaviors related to learner self-regulation. Survey research has shown a worrying trend of a rising proportion of students with lower levels of conscientiousness and diligence than prior cohorts [ 5 , 36 ] . There is an opportunity to use such a detector of student behavior that is relevant to educators’ experience in the classroom: teachers understand the process of telling their students, “It’s time to get started”. Advances in session-level behavioral analytics have expanded our understanding of student behavior during individual academic work within classwork sessions. Gurung et al. [ 13 ] found that behavioral measures of self-regulation (e.g., when to start, how long to work, and when to stop) are highly predictive of learning outcomes, including end-of-year teacher grades in math and performance on standardized tests in math. These session-level measures also exhibit strong reliability and cross-system generalizability across DLPs. When combined with more established predictors of learning outcomes such as gaming the system [ 3 ] and prior performance, they found that delayed start (–0.25 standard deviation [SD]) was twice as predictive as gaming-the-system (–0.12 SD) and half as predictive as prior performance (0.47 SD) in MATHia. Beyond their predictive strength and reliability, these measures are also easily interpretable and closely aligned with teaching practice, making them well-suited for integration into educator-facing tools. In this paper, we extend the exploration of delayed start behavior by examining its generalizability across subjects, and explore practical ways to operationalize delayed-start behavior and define an interpretable, practitioner-aligned threshold for real-time use during classwork sessions. This paper aims to answer these research questions: RQ 1. Is students’ delayed start behavior during math practice predictive of learning outcomes in English? RQ 2. Are there sub-groups of learners that are revealed in their patterns of delayed start behavior and associated learning outcomes? 2 Background 2.1 Self-regulated learning Self-regulated learning (SRL) provides a framework for understanding how learners monitor, manage, and regulate their learning. Prominent frameworks by Zimmerman [ 39 ] , Pintrich [ 29 ] , and Winne and Hadwin [ 38 ] differ in scope and emphasis, particularly with respect to the relative emphasis on individual versus contextual factors. Pintrich’s model, for instance, emphasizes individual control while acknowledging group dynamics, whereas Winne and Hadwin conceptualize SRL as a recursive cycle in which students continuously define tasks, set goals, and adjust strategies in response to internal states and external conditions. Pintrich further specifies four regulatory phases (planning, monitoring, control, and reflection) that operate across cognition, motivation/affect, behavior, and context. Effective self-regulation is fundamental to learner success, with habits forming early in development [ 25 ] . Education research has drawn on self-regulated learning to characterize skills and behaviors underlying students’ decision-making [ 26 ] , while research in psychology has emphasized executive functions, including working memory, inhibitory control, and cognitive flexibility [ 8 ] . During learning, these regulatory capacities shape how students organize and direct their efforts across tasks and contexts. Although many such processes are internal and difficult to observe directly, their effects often manifest in how students allocate and use available time. For instance, a motivated student with well-developed self-regulation skills will likely begin tasks promptly, maintain focus, and reflect on how to optimize effort when encountering difficult content. 2.2 Diligence and procrastination Beyond SRL, productive time use in learning, particularly for task initiation, has been studied through the related constructs of diligence and procrastination. Diligence is characterized by sustained effort and perseverance over time [ 39 ] . Much behavioral research has explored diligence as a component of behavioral regulation within SRL, often conceptualized as effort regulation [ 39 ] or linked to conscientiousness [ 19 ] . Galla et al. [ 11 ] , however, explored diligence as a stable pattern of effective time management and timeliness, rather than as a situational response; for example, compared to their peers, a diligent student is consistently on time. The Academic Diligence Task (ADT) [ 11 ] operationalizes diligence as a latent construct inferred from DLP log data. Their diligence measure, which quantifies how students allocate time between academic and leisure activities, predicted academic outcomes and was correlated with math interest, self-efficacy, and effort regulation. Similarly, effective time management and conscientiousness have also been studied as procrastination. Procrastination is defined as “voluntarily delay[s] an intended course of action despite expecting to be worse off for the delay,” [ 35 ] . Procrastination has a long history – warnings of the pitfalls of delays date back to Greek [ 15 ] and Roman [ 6 ] times. It has been conceptualized both as a stable individual trait [ 35 , 10 ] , and as contextual task avoidance in response to situational and motivational factors [ 23 ] . More recent approaches propose trait-state models that integrate self-regulation and emotional regulation [ 30 , 37 ] .