Investigating customers’ continuous trust towards mobile banking apps

The study addresses how continuous trust—trust beyond initial adoption—forms among mobile banking users and to what extent various antecedents influence it. With rapid growth in mobile technology and banking apps, retaining long-term user trust is critical due to perceived risks (privacy leaks, hacking, lack of control) and absence of face-to-face interaction. Prior research emphasizes initial adoption; fewer studies target continuous-use trust. The authors pose two research questions: (RQ1) What factors contribute to mobile banking customers’ continuous trust? (RQ2) To what extent do these factors affect trust at the continuous-use stage? They propose and test a model integrating app-related (perceived ease of use, privacy assurance and security features, information quality), organization-related (reputation, customer support), and customer-related (propensity to trust, previous experience) antecedents using SEM on data from frequent users.

The conceptual framework builds on TAM, UTAUT, and the S-O-R framework, and on literature identifying clusters of trust antecedents: app-related, organization-related, and customer-related. Hypotheses: H1 PEU→continuous trust; H2 PASF→continuous trust; H3 Information quality→continuous trust; H4 Organization reputation→continuous trust; H5 Customer support→continuous trust. Indirect organizational effects on app perceptions are posited: H4a OR→PEU, H4b OR→PASF, H4c OR→IQ; H5a CS→PEU, H5b CS→PASF, H5c CS→IQ. Customer-related: H6 Propensity to trust→continuous trust; H7 Previous experience→continuous trust, plus indirect effects H7a PE→PEU, H7b PE→PASF, H7c PE→IQ, H7d PE→OR, H7e PE→CS. The integrated model positions seven latent constructs as direct and indirect antecedents to continuous trust at the continuous-use stage.

Design: Cross-sectional online survey using a 7-point Likert scale (1=Strongly disagree to 7=Strongly agree). Measurement items adapted from prior validated scales (Appendix 1) for constructs: perceived ease of use (PEU), privacy assurance and security features (PASF), information quality (IQ), organization reputation (OR), customer support (CS), propensity to trust (PTT), previous experience (PE), and continuous trust (CT). Demographics (age, gender) included. Sampling and administration: Approved by NTU IRB (IRB-2021-305). Conducted online in China among existing frequent mobile banking users (screened via a filter question). Recruitment from higher education communities and public residents in 10+ major cities. Questionnaire developed in English, translated into Chinese, and refined via focus group of six Chinese-proficient authors. Items were randomized by construct to mitigate bias. Data: 557 responses collected; duplicates (by IP), incomplete forms, and flat-lined responses removed, resulting in N=450. Sample: 53.8% male; age 19–50 with 37.6% aged 19–25. Frequently used banks included CCB (20.4%), ICBC (17.8%), BOC (16.9%), CMB (14.9%), ABOC (10.4%), BOCM (6.2%), PSBC (4.4%), others (8.9%). Common method bias: Harman’s single-factor test variance=43.65% (<50%), suggesting minimal bias. Analysis: Conducted in IBM SPSS Amos 26. Two-step approach: (1) Confirmatory Factor Analysis (CFA) to assess measurement reliability and validity; (2) Structural equation modeling (SEM) to test hypothesized paths in a full model, followed by a reduced model excluding non-significant paths/constructs. Measurement model fit: χ²=666.77, df=322, χ²/df=2.07, IFI=0.96, TLI=0.96, CFI=0.96, RMSEA=0.05. Reliability/validity: Composite Reliability (CR)>0.7; AVE>0.5; discriminant validity supported (AVE > squared inter-construct correlations). Model comparison used fit indices, R² for CT, and AIC for parsimony.

Measurement model was reliable and valid (CR>0.7; AVE>0.5; good CFA fit). Full SEM model fit well (χ²=847.666; df=332; χ²/df=2.55; IFI=0.943; TLI=0.934; CFI=0.942; RMSEA=0.059) and explained R²CT=0.823. Hypothesis testing in full model:

• Supported direct effects on CT: H1 PEU→CT (β=0.203, p=0.042); H2 PASF→CT (β=0.173, p=0.017); H4 OR→CT (β=0.578, p<0.001); H5 CS→CT (β=0.595, p<0.001).
• Not supported: H3 IQ→CT (β=−0.006, p=0.951); H6 PTT→CT (β=−0.070, p=0.099); H7 PE→CT (β=−0.085, p=0.121).
• Indirect/mediating relationships supported: OR→PASF (H4b β=0.579, p<0.001); OR→IQ (H4c β=0.307, p<0.001); CS→PASF (H5b β=0.259, p=0.001); CS→IQ (H5c β=0.283, p<0.001); PE→PEU (H7a β=0.918, p<0.001); PE→IQ (H7c β=0.380, p<0.001); PE→OR (H7d β=0.824, p<0.001); PE→CS (H7e β=0.828, p<0.001). Non-significant: OR→PEU (H4a), CS→PEU (H5a), PE→PASF (H7b). Reduced model: Fit acceptable and more parsimonious (χ²=493.254; df=179; χ²/df=2.756; IFI=0.955; TLI=0.947; CFI=0.954; RMSEA=0.063; AIC=597.254 vs 995.666). Explanatory power remained high (R²CT=0.818, only 0.5% less than full model). Significant paths included: OR→CT (β≈0.585, p<0.001); CS→CT (β≈0.584, p<0.001); PEU→CT (β≈0.219, p=0.027); PASF→CT (β≈0.195, p=0.003); OR→PASF (β≈0.584, p<0.001); PE→PEU (β≈0.893, p<0.001); PE→OR (β≈0.825, p<0.001); PE→CS (β≈0.826, p<0.001); PE→IQ (β≈0.596, p<0.001). Age and gender had no significant effects on CT. Overall, organization reputation and customer support were the strongest predictors of continuous trust, with PEU and PASF also contributing; information quality and propensity to trust did not directly affect CT at the continuous-use stage.

The findings address RQ1 by identifying key antecedents of continuous trust in mobile banking: organization reputation, customer support, perceived ease of use, and privacy assurance/security features as direct predictors; previous experience as an indirect antecedent shaping perceptions of the app and organization. RQ2 is addressed by quantifying these effects: organization reputation and customer support exhibit the largest standardized effects on continuous trust (≈0.58 each), followed by perceived ease of use (≈0.22) and privacy/security (≈0.20). Previous experience strongly shapes PEU, OR, CS, and IQ but does not directly predict trust at this stage. Information quality and propensity to trust do not significantly influence continuous trust, highlighting stage-specific dynamics in trust formation: dispositions and information presentation matter less once users are experienced, whereas organizational reliability, support, usability, and security assurances matter more. The models fit the data well, and the reduced model achieves comparable explanatory power with substantially better parsimony. Demographics (age, gender) did not significantly affect trust. Practical implications include prioritizing robust customer support, enhancing and communicating security/privacy safeguards, investing in corporate reputation management, and ensuring high usability; banks should leverage customer experience feedback loops to improve perceptions that ultimately bolster continuous trust.

The study advances theory by validating a parsimonious SEM model of continuous trust in mobile banking that integrates app-, organization-, and customer-related constructs. It shows that organization reputation and customer support are the strongest direct drivers of continuous trust, with additional contributions from perceived ease of use and privacy/security. Previous experience shapes these perceptions indirectly, while propensity to trust and information quality do not directly impact continuous trust at the continuous-use stage. The work contributes to distinguishing antecedents of initial versus continuous trust and provides an evidence-based framework for designing trust-enhancing strategies. Future research should examine cross-cultural generalizability, include broader demographics (including older users and rural populations), incorporate objective behavioral measures of continuance use, and explore additional contextual factors (e.g., temporal or environmental changes) and constructs that may affect trust over time.

Generalizability is limited by sampling frequent users in major Chinese cities; findings may not transfer to other cultures or rural contexts. Self-reported, cross-sectional survey data may not reflect actual long-term usage behavior; objective usage measures are needed. The study did not include users over 50, constraining age-related insights. The construct set, while theory-driven, is not exhaustive; omitted variables may influence continuous trust. Translation and online administration, although carefully managed, may introduce residual measurement bias.