Factors influencing continuance intention to use mobile banking: an extended expectation-confirmation model with moderating role of trust

The study addresses why and how users continue using mobile banking after initial adoption, focusing on Vietnam where mobile banking adoption is growing but continuance remains underexplored. Prior literature emphasizes adoption rather than continuance, with mixed findings across frameworks (TAM, ECM, UTAUT). Key gaps include limited empirical research on user behavioral adaptation as a driver of continuance, scarce studies on the moderating role of trust, and few Vietnam-specific investigations. The study integrates the expectation-confirmation model (ECM), decomposed theory of planned behavior (DTPB), and adaptive structuration theory for individuals (ASTI), plus trust, to explain continuance intention. Research questions: (1) What determinants affect user adaptation and continuance intention to use mobile banking? (2) How do these determinants influence adaptation and lead to continuance intention? (3) How does trust moderate the effects of adaptation and satisfaction on continuance intention? The work aims to extend IT continuance theory and offer actionable insights for banks to retain mobile banking users.

The literature review defines mobile banking within mobile commerce and outlines its benefits and growth. Continuance intention (CI) is framed as post-adoption intention to keep using IS, with ECM (Bhattacherjee, 2001) positing that confirmation affects satisfaction and perceived usefulness (PU), which drive CI. The study inherits ECM’s core paths (confirmation→PU; confirmation→satisfaction; PU→satisfaction; PU→CI; satisfaction→CI). DTPB decomposes beliefs into subjective norm (SN) and control beliefs (self-efficacy, SE) affecting intention and behavior; prior studies link SN and SE to post-adoption usage. ASTI conceptualizes individual-level IT adaptation (ADP) as adjustments to technology and work practices that can enhance satisfaction and continuance. Trust is widely recognized in online contexts for mitigating risk and influencing continuance; however, its moderating role is underexamined in mobile banking. Hypotheses: H1 satisfaction→CI (+); H2 PU→CI (+); H3 PU→satisfaction (+); H4 confirmation→satisfaction (+); H5 confirmation→PU (+); H6 adaptation→CI (+); H7 SE→CI (+); H8 SN→CI (+); H9 PU→adaptation (+); H10 SN→adaptation (+); H11 SE→adaptation (+); H12 adaptation→satisfaction (+); H13 trust moderates adaptation→CI (+ stronger with higher trust); H14 trust moderates satisfaction→CI.

Design: Cross-sectional survey of mobile banking users in Vietnam. Sampling and data collection: Convenience and snowball sampling targeting customers of three large banks (Vietcombank, Vietinbank, Techcombank) across Hanoi, Ho Chi Minh City, and Danang. Data collected Jan–Jun 2022 via face-to-face (69%) and online (31%) responses. 700 questionnaires distributed; 523 valid responses (74.7% response rate). Eligibility: mobile banking account holder, app downloaded, prior use of at least one service (e.g., bill payment, transfer). Pre-test: Group discussions with 20 bank officials/experts/users to refine items. Measures: Seven-point Likert scales (1=strongly disagree to 7=strongly agree). Constructs and sources: CI (3 items; Bhattacherjee, 2001; Baabdullah et al., 2019), Confirmation (3; Susanto et al., 2016), PU (originally 4, PU3 dropped; Davis et al., 1989; Baabdullah et al., 2019), SE (4; Taylor & Todd, 1995; Susanto et al., 2016), SN (4; Bhattacherjee & Lin, 2015; Park et al., 2019), Adaptation (ADP, 4; Nguyen & Ha, 2022; Barki et al., 2007), Trust (TR, 4; Gefen et al., 2003; Gao et al., 2015). Analysis: PLS-SEM using SmartPLS 4 with 5000 bootstrap subsamples. Assessed CMV via Harman’s single-factor test (22.067% variance <50%) and full collinearity VIFs (<3). Measurement model: Reliability acceptable (CR 0.703–0.889), AVE 0.527–0.693, HTMT <0.85 for discriminant validity. Structural model: Checked collinearity (inner VIF <3), reported path coefficients, t-stats, p-values, effect sizes (f²), R², and predictive relevance (Q² >0 for PU, SA, ADP, CI). R²: CI=0.243; ADP=0.096; SA=0.222. Moderation tested for TR×ADP→CI and TR×SA→CI interaction terms. Demographics (n=523): 51% female; age 18–23 (17%), 24–55 (47%), 56–65 (36%); occupation student (17%), officer (55%), other (28%); monthly income below $335 (17%), $335–$835 (38%), $835–$1250 (26%), above $1250 (19%).

- Model fit and prediction: R²(CI)=0.243; R²(ADP)=0.096; R²(SA)=0.222; Q² >0 for PU, SA, ADP, CI, indicating predictive relevance. - Supported direct effects (bootstrapping results, Table 4): • Satisfaction→CI: β=0.109, t=2.345, p=0.019 (H1 supported). • Perceived usefulness→CI: β=0.211, t=4.688, p=0.000, f²=0.119 (H2 supported). • PU→Satisfaction: β=0.185, t=4.445, p=0.000 (H3 supported). • Confirmation→Satisfaction: β=0.257, t=6.239, p=0.000 (H4 supported). • Confirmation→PU: β=0.327, t=7.541, p=0.000 (H5 supported). • Adaptation→CI: β=0.240, t=5.523, p=0.000 (H6 supported). • Self-efficacy→CI: β=0.125, t=2.727, p=0.000 (H7 supported). • PU→Adaptation: β=0.111, t=2.564, p=0.010 (H9 supported). • SN→Adaptation: β=0.162, t=3.433, p=0.001 (H10 supported). • SE→Adaptation: β=0.152, t=3.089, p=0.002 (H11 supported). • Adaptation→Satisfaction: β=0.218, t=5.446, p=0.000 (H12 supported). - Non-significant direct effect: • Subjective norm→CI: β=0.035, t=0.817, p=0.414 (H8 not supported). - Moderation effects: • Trust×Adaptation→CI: β=0.133, t=3.047, p=0.002 (H13 supported). • Trust×Satisfaction→CI: β=−0.077, t=1.626, p=0.104 (H14 not supported). - Measurement validity: All constructs showed acceptable reliability (CR 0.703–0.889), convergent validity (AVE 0.527–0.693), and discriminant validity (HTMT <0.85). - Practical ranking of CI drivers (by β): Adaptation (0.240), PU (0.211), SE (0.125), Satisfaction (0.109); SN has no direct effect on CI but influences CI indirectly via Adaptation and Satisfaction.

Findings confirm ECM’s core mechanisms: confirmation enhances PU and satisfaction, which both raise continuance intention. Extending ECM with DTPB and ASTI demonstrates that adaptation is a central post-adoption behavior shaping both satisfaction and continuance. PU, SE, and SN significantly foster adaptation; adaptation, in turn, increases satisfaction and CI. SN does not directly affect CI, suggesting that social pressures matter for shaping adaptive behaviors rather than the ultimate continuance decision in this context. Trust strengthens the positive link between adaptation and CI, implying that when users trust mobile banking, their adaptive efforts translate more strongly into continued use. However, trust does not moderate the satisfaction→CI relationship, indicating that once users are satisfied, trust adds little incremental influence on continued intention. Overall, the integrated model elucidates how perceptions (confirmation, PU), social/control beliefs (SN, SE), and behavioral adaptation jointly form CI in mobile banking, addressing the research questions and highlighting adaptation and trust’s contingent role.

The study proposes and validates an extended ECM integrating DTPB and ASTI with trust to explain continuance intention for mobile banking in Vietnam. Empirical results from 523 users show that adaptation, perceived usefulness, self-efficacy, and satisfaction are significant predictors of continuance, while subjective norm affects adaptation but not CI directly. Trust moderates the adaptation→continuance link but not the satisfaction→continuance link. Contributions include: (1) a novel integrated continuance model combining ECM, DTPB, and ASTI with adaptation and trust; (2) evidence of trust’s moderating role on adaptation’s impact on CI; and (3) unpacking DTPB components (SN, SE) to show their roles in adaptation and CI. Managerially, banks should facilitate user adaptation through customizable, easy-to-use, and useful features, build user self-efficacy, and enhance trust to amplify adaptation’s effect on retention. Future research could test the model in developed countries and incorporate individual innovation characteristics to broaden generalizability.

- Context specificity: Data from Vietnam and three major banks may limit generalizability; replication in other countries and market types is needed. - Omitted variables: Individual innovation/technology traits and other potential determinants were not included. - Cross-sectional design: Limits causal inference over time; longitudinal designs could capture temporal dynamics, especially regarding trust. - Self-reported measures and nonprobability sampling (convenience, snowball) may introduce bias despite CMV and collinearity checks.