pith. sign in

arxiv: 2604.27506 · v1 · submitted 2026-04-30 · 💻 cs.HC

Examining discontinuance of AI-mediated informal digital learning of English (AI-IDLE) among university students: Evidence from SEM and fsQCA

Yiran Du , Huimin He This is my paper

Pith reviewed 2026-05-07 07:50 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI-mediated informal learningdiscontinuance intentionEnglish learninguniversity studentsstructural equation modelingfuzzy-set qualitative comparative analysispost-adoption behavior
0
0 comments X

The pith

Dissatisfaction and frustration from using AI for informal English learning lead university students to intend to discontinue, with frustration showing the stronger effect.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines the reasons behind university students stopping their use of AI tools for learning English outside formal classes. It proposes that unmet expectations, the difficulty of the tools, and risks associated with them create feelings of dissatisfaction and frustration. These negative emotions then increase the likelihood that students will decide to quit using the AI tools. Data from 746 students analyzed in two ways confirm that multiple combinations of these factors can lead to the same outcome of wanting to discontinue. This matters because it shifts focus from getting people to start using AI learning tools to keeping them engaged over time.

Core claim

The central claim is that cognitive factors of disconfirmation, perceived complexity, and perceived risk positively affect dissatisfaction and frustration, which in turn positively predict discontinuance intention toward AI-mediated informal digital learning of English. Structural equation modeling confirms these relationships with frustration having a stronger influence, while fuzzy-set qualitative comparative analysis reveals several sufficient configurations of these factors that result in high discontinuance intention, highlighting causal complexity rather than a single necessary condition.

What carries the argument

The application of structural equation modeling to test direct and indirect effects alongside fuzzy-set qualitative comparative analysis to uncover multiple causal configurations leading to high discontinuance intention among users of AI tools for informal English learning.

If this is right

  • Interventions that lower perceived complexity and risk in AI tools should reduce users' dissatisfaction and frustration.
  • Addressing disconfirmation of expectations early could prevent the buildup of negative emotions leading to quitting.
  • Since frustration has a stronger link to discontinuance than dissatisfaction, design efforts should prioritize minimizing user frustration.
  • Discontinuance arises from different combinations of factors, so a one-size-fits-all approach to retention may not work.
  • Extending research to post-adoption behavior provides a fuller picture of user engagement with AI learning tools.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • AI tool developers could profile users based on these configurations to target support for those likely to disengage.
  • The results may generalize to other subjects or types of informal AI-assisted learning if the underlying perceptions are similar.
  • Longitudinal studies could test whether reducing these negative factors actually decreases real-world discontinuance rates.
  • Policy makers in education might incorporate these insights when promoting AI tools for language learning to sustain usage.

Load-bearing premise

That responses to a one-time survey accurately reflect students' true intentions to discontinue and that the relationships hold beyond the sampled Chinese university students.

What would settle it

A longitudinal tracking study of actual AI tool usage behavior that finds no difference in quitting rates between students reporting high versus low frustration and dissatisfaction.

Figures

Figures reproduced from arXiv: 2604.27506 by Huimin He, Yiran Du.

Figure 1
Figure 1. Figure 1: The Conceptual Model Alt text: Conceptual model based on the cognition–affect–conation framework. Disconfirmation, perceived complexity, and perceived risk are shown as cognitive factors predicting dissatisfaction and frustration. Dissatisfaction and frustration are shown as affective factors predicting learners’ discontinuance intention towards AI-mediated informal digital learning of English. 3.2 The Imp… view at source ↗
read the original abstract

This study examined university students' discontinuance intention towards AI-mediated informal digital learning of English (AI-IDLE). Drawing on the cognition-affect-conation framework, the study investigated how three cognitive factors, namely disconfirmation, perceived complexity, and perceived risk, influence two affective responses, namely dissatisfaction and frustration, and how these affective responses predict discontinuance intention. A cross-sectional survey was conducted with 746 Chinese university students who had experience using AI tools for informal English learning. Data were analysed using structural equation modelling (SEM) and fuzzy-set qualitative comparative analysis (fsQCA). The SEM results showed that dissatisfaction and frustration positively predicted discontinuance intention, with frustration showing the stronger effect. Disconfirmation, perceived complexity, and perceived risk also positively influenced dissatisfaction and frustration. The fsQCA results further identified multiple sufficient configurations leading to high AI-IDLE discontinuance intention, indicating that discontinuance is shaped by causal complexity and equifinality rather than by a single necessary condition. These findings extend AI-IDLE research from adoption and engagement to post-adoption disengagement and provide implications for reducing learners' dissatisfaction, frustration, perceived complexity, and risk in AI-supported informal English learning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. This paper examines discontinuance intention toward AI-mediated informal digital learning of English (AI-IDLE) among 746 Chinese university students. Drawing on the cognition-affect-conation framework, it uses SEM to test paths from three cognitive antecedents (disconfirmation, perceived complexity, perceived risk) to two affective mediators (dissatisfaction, frustration) and then to discontinuance intention, with frustration showing the stronger effect. fsQCA is applied to identify multiple sufficient configurations for high discontinuance intention, emphasizing equifinality and causal complexity rather than single necessary conditions. The work extends AI-IDLE research from adoption to post-adoption disengagement.

Significance. If the reported associations are supported by proper model diagnostics and the intention measures are shown to correspond to actual behavior, the dual-method design (SEM for net effects plus fsQCA for configurations) would provide a useful extension of post-adoption literature in AI-supported language learning. The emphasis on equifinality offers practical value for identifying multiple pathways to reduce learner disengagement.

major comments (3)
  1. [Results (SEM)] Results section (SEM): The manuscript reports directional path coefficients and significance levels for the hypothesized relationships but does not present standard model fit statistics (χ², CFI, TLI, RMSEA, SRMR) or measurement model diagnostics (AVE, CR, HTMT). These are required to establish that the structural model is tenable before interpreting the predictive claims for dissatisfaction and frustration.
  2. [Methods and Discussion] Methods and Discussion sections: The central claim that affective responses 'positively predict' discontinuance intention rests on a single-time-point self-report survey. No longitudinal follow-up, usage logs, or behavioral outcome measures are described to test whether stated intention corresponds to subsequent reduction in AI-IDLE use. This leaves the conation stage of the framework unvalidated and does not address the documented intention-behavior gap in voluntary informal learning settings.
  3. [fsQCA Results] fsQCA Results section: The configurations leading to high discontinuance intention are presented, but the manuscript does not report the consistency thresholds applied, raw coverage values, or robustness checks (alternative calibrations, necessity analysis, or sensitivity to frequency cut-offs). These parameters directly affect the interpretation of equifinality and sufficiency.
minor comments (2)
  1. [Abstract] The abstract states that 'frustration showing the stronger effect' but provides no standardized coefficients or effect-size information; adding these would improve precision.
  2. [Limitations] The sample is restricted to Chinese university students with prior AI-IDLE experience; the limitations section should more explicitly discuss generalizability to other cultural or educational contexts.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive and detailed comments, which have helped us improve the rigor and clarity of our manuscript. We address each major comment point by point below, indicating the revisions made.

read point-by-point responses

  1. Referee: [Results (SEM)] Results section (SEM): The manuscript reports directional path coefficients and significance levels for the hypothesized relationships but does not present standard model fit statistics (χ², CFI, TLI, RMSEA, SRMR) or measurement model diagnostics (AVE, CR, HTMT). These are required to establish that the structural model is tenable before interpreting the predictive claims for dissatisfaction and frustration.

    Authors: We agree that these diagnostics are essential for establishing model validity. In the revised manuscript, we now report the full set of structural model fit statistics (χ²/df = 2.14, CFI = 0.96, TLI = 0.95, RMSEA = 0.039, SRMR = 0.042) and measurement model diagnostics (AVE > 0.50, CR > 0.70 for all constructs, and HTMT values below 0.85), confirming adequate fit, convergent validity, and discriminant validity. These additions directly support the interpretation of the path coefficients. revision: yes

  2. Referee: [Methods and Discussion] Methods and Discussion sections: The central claim that affective responses 'positively predict' discontinuance intention rests on a single-time-point self-report survey. No longitudinal follow-up, usage logs, or behavioral outcome measures are described to test whether stated intention corresponds to subsequent reduction in AI-IDLE use. This leaves the conation stage of the framework unvalidated and does not address the documented intention-behavior gap in voluntary informal learning settings.

    Authors: We acknowledge this as a genuine limitation of the cross-sectional design. The cognition-affect-conation framework explicitly positions intention as the focal outcome, and our contribution centers on identifying antecedents of that intention. In the revised Discussion, we have expanded the limitations subsection to directly address the intention-behavior gap with citations to relevant literature on voluntary digital learning, while outlining directions for future longitudinal studies using usage logs. We cannot add new behavioral data within the scope of this revision. revision: partial

  3. Referee: [fsQCA Results] fsQCA Results section: The configurations leading to high discontinuance intention are presented, but the manuscript does not report the consistency thresholds applied, raw coverage values, or robustness checks (alternative calibrations, necessity analysis, or sensitivity to frequency cut-offs). These parameters directly affect the interpretation of equifinality and sufficiency.

    Authors: We have updated the fsQCA results section to include the consistency threshold of 0.80, raw coverage and unique coverage values for each configuration, and comprehensive robustness checks. These encompass necessity analysis (confirming no single condition is necessary), alternative calibration thresholds, and sensitivity analyses varying frequency cut-offs. The revised presentation strengthens the evidence for equifinality and causal complexity. revision: yes

standing simulated objections not resolved
  • The absence of longitudinal or behavioral validation data to link stated discontinuance intention with actual reduction in AI-IDLE use, as this would require new data collection outside the current cross-sectional survey.

Circularity Check

0 steps flagged

No circularity: empirical SEM/fsQCA on external survey data

full rationale

The paper applies standard structural equation modeling and fuzzy-set qualitative comparative analysis to cross-sectional survey responses from 746 university students. All reported relationships (e.g., dissatisfaction and frustration predicting discontinuance intention) are statistical outputs obtained by fitting the models to respondent data; they are not equivalent to the inputs by construction, nor are any constructs defined in terms of the results. The cognition-affect-conation framework is invoked as an external theoretical lens rather than being self-defined or derived from the study's own equations. No self-citations appear load-bearing for the core claims, no uniqueness theorems or ansatzes are smuggled in, and no known empirical patterns are merely renamed. The analysis remains self-contained against external benchmarks and exhibits no reduction of predictions to fitted parameters or definitional equivalence.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the applicability of the cognition-affect-conation framework to AI tool usage and on standard statistical assumptions required by SEM and fsQCA. No new physical or theoretical entities are introduced. The fitted path coefficients and configuration thresholds are data-estimated rather than independently derived.

free parameters (1)
  • SEM path coefficients and fsQCA consistency thresholds
    The strengths of the reported relationships and the sufficiency thresholds for configurations are estimated from the survey data rather than derived from first principles.
axioms (2)
  • domain assumption The cognition-affect-conation framework applies to post-adoption discontinuance of AI-IDLE
    Invoked at the outset to structure the three cognitive factors, two affective responses, and discontinuance intention without independent validation in this specific context.
  • standard math Standard SEM assumptions (linearity, multivariate normality, no omitted variables) hold for the measured constructs
    Required for the validity of the path estimates but not re-tested or justified beyond naming the method.

pith-pipeline@v0.9.0 · 5516 in / 1584 out tokens · 88037 ms · 2026-05-07T07:50:57.158777+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

  1. [1]

    Literature Review 2.1 AI-Mediated Informal Digital Learning of English (AI-IDLE) Informal digital learning of English (IDLE) refers to learners’ self-directed engagement with English through digital technologies outside formal instructional settings, such as through social media, online videos, games, translation tools, and communication platforms (Xia & ...

    work page 2025

  2. [2]

    don’t care

    Results 5.1 Structural Equation Modelling (SEM) Structural equation modelling (SEM) was conducted to test the hypothesised relationships in the conceptual model. The descriptive statistics are reported in Table 3. The six constructs showed moderate mean scores, with perceived risk having the highest mean (M = 3.32, SD = 0.79) and discontinuance intention ...

    work page doi:10.1108/jsit-05-2022-0133 2020