Skip to main content
PLOS One logoLink to PLOS One
. 2026 Nov 16;21(11):e0362486. doi: 10.1371/journal.pone.0362486

Differentiated pathways from internationalization at home to global competence among Chinese university students: The roles of cross-cultural self-efficacy and perceived internationalization climate

Ji Zhang 1,*
Editor: Wang Zheng2
PMCID: PMC13773313  PMID: 43038239

Abstract

Internationalization at home has become an important route for expanding access to international learning opportunities beyond cross-border mobility. However, less is known about whether different at-home internationalization pathways are associated with students’ global competence in the same way. Based on 768 valid responses from Chinese university students, this study distinguished three pathways of internationalization at home: internationalized curricular and learning resource exposure, cross-cultural interaction engagement, and digital international learning participation. Structural equation modeling was used to examine their associations with self-reported global competence, the indirect role of cross-cultural self-efficacy, the moderating role of perceived internationalization climate, and group heterogeneity by overseas experience. The results showed that all three pathways were positively associated with global competence, but their association profiles differed. Digital international learning participation had the strongest direct association with global competence, suggesting its particular relevance for expanding accessible international learning opportunities. Cross-cultural interaction engagement showed the strongest association with cross-cultural self-efficacy and the largest indirect effect, indicating that interactional experiences were more closely linked to the psychological translation of internationalized learning into perceived competence. Internationalized curricular and learning resource exposure was also positively associated with global competence, functioning mainly as a foundational route of cognitive input. Perceived internationalization climate strengthened the associations of curricular/resource-based and digital pathways with cross-cultural self-efficacy, but did not significantly moderate the interaction pathway. Multi-group analysis indicated an overall structural difference between students with and without overseas experience. Follow-up path-specific tests showed that the association between digital international learning participation and cross-cultural self-efficacy was significantly stronger among students with overseas experience, whereas the remaining structural paths did not differ significantly. These findings suggest that internationalization at home should be understood not as a single aggregate input, but as a differentiated configuration of cognitive, interactional, digital, psychological, and institutional conditions related to students’ self-reported global competence.

1. Introduction

For a long time, higher education internationalization was organized primarily around cross-border mobility, international student recruitment, institutional partnerships, and global branding. Its dominant imagery was built around “going out” and “bringing in” [1,2]. As international learning opportunities remain unevenly distributed across student groups, however, the problem of the “non-mobile majority” has become increasingly visible. Research attention has therefore moved from the overseas experience of a relatively small group of students to the international and intercultural learning opportunities available to most students on their home campuses [3,4]. In this shift, internationalization at home is no longer merely a compensatory arrangement when cross-border mobility is limited. It is increasingly conceptualized as an institutional practice that embeds international and intercultural dimensions into curricula, campus environments, and learning support systems [5,6].

This shift is closely aligned with the global competence agenda. The OECD defines global competence as the capacity to examine local, global, and intercultural issues, understand and appreciate the perspectives of others, interact effectively with people from different backgrounds, and act for collective well-being [7]. Related scholarship further emphasizes that global competence is not a simple accumulation of international knowledge, but a composite outcome involving cognitive understanding, perspective taking, intercultural communication and collaboration, and action orientation [8,9]. Accordingly, the value of internationalization at home lies not simply in bringing international elements into the campus, but in whether such elements are associated with students’ capacity to handle difference, negotiate across difference, and collaborate within difference in everyday learning contexts.

Existing research has provided relatively stable positive evidence for the relationship between internationalization at home and global competence. Studies conducted in Western higher education contexts and among Chinese university students have reported significant positive associations between internationalized curricula, campus-based intercultural interaction, digital international learning, and students’ global competence [10,11]. More recent comparative studies also suggest that different internationalization modes vary in their organizational form, participation threshold, and learning depth, and therefore may not be associated with student development in the same way [12,13]. With the expansion of virtual exchange, online collaboration, and collaborative online international learning, digital pathways have become one of the fastest-growing components of internationalization at home [14,15]. Systematic and scoping reviews likewise indicate that student outcomes are shaped less by the nominal quantity of internationalized activities than by the design quality, degree of curricular embeddedness, and participatory depth of those experiences [16,17].

Nevertheless, existing research has more fully answered whether internationalization at home is related to global competence than how different pathways perform different educational functions. Curricular, interactional, and digital experiences are often merged into a general input variable or treated as interchangeable forms of internationalized exposure [18,19]. This approach is useful for identifying an average positive association, but it may obscure the different positions that these pathways occupy in the competence formation process. Curricular and learning resource exposure is more likely to provide stable cognitive input. Cross-cultural interaction is more likely to help students form subjective judgments of competence through negotiation and feedback. Digital international learning may operate by expanding reach, lowering participation barriers, and increasing the frequency of international learning opportunities. Without distinguishing these pathways, it is difficult to explain how internationalization at home matters and why its association with global competence may vary across school contexts and student backgrounds.

The institutional meaning of internationalization at home should also not be understood only through average effects. Research on inclusive internationalization has shown that different social groups differ in their access to internationalization opportunities, participation modes, and benefit pathways [20]. Research on Chinese university students similarly suggests that whether students can translate internationalized experiences into global competence depends not only on the supply of experiences, but also on the institutional support environment and on students’ perceptions of their own capacity for action [21]. This study therefore does not treat internationalization at home as a homogeneous input. Instead, it conceptualizes internationalized curricular and learning resource exposure, cross-cultural interaction engagement, and digital international learning participation as three potentially differentiated pathways. It examines their associations with self-reported global competence, the indirect association through cross-cultural self-efficacy, the boundary role of perceived internationalization climate, and group heterogeneity by overseas experience. The aim is not to re-confirm that internationalization at home is generally beneficial, but to specify how different pathways occupy different positions in the competence-formation process and how these associations vary across psychological, institutional, and experiential conditions. Although the equity rationale for internationalization at home emphasizes the non-mobile majority, the present study does not define internationalization at home as relevant only to students without cross-border mobility. Instead, it examines campus-based internationalized experiences among students with and without prior overseas experience and treats overseas experience as an experiential boundary condition.

2. Theoretical background and hypotheses

2.1. Integrated theoretical framework

This study integrates three complementary theoretical lenses while distinguishing explanatory theories from construct-defining frameworks. First, the internationalization-at-home perspective defines the institutional learning opportunities available to students within their home universities and provides the conceptual basis for distinguishing curricular/resource-based, interactional, and digital pathways. The OECD global competence framework, in turn, defines the multidimensional outcome examined in this study rather than serving as a causal theory.

Second, social cognitive theory explains the psychological translation mechanism proposed in the model. Internationalized learning experiences provide students with mastery experiences, vicarious information, interpersonal feedback, and opportunities for self-evaluation. These experiences may strengthen cross-cultural self-efficacy, which is subsequently associated with students’ perceived global competence. This theoretical lens underlies the mediation hypotheses concerning cross-cultural self-efficacy.

Third, ecological systems theory and the school-climate perspective explain why the same learning experience may not operate identically under different institutional conditions. Perceived internationalization climate can make internationalized opportunities more visible, accessible, legitimate, and usable, thereby conditioning the extent to which specific learning experiences are associated with cross-cultural self-efficacy. This perspective underlies the moderation and moderated-mediation hypotheses.

Overseas experience is not treated as an additional causal theory. Instead, it is conceptualized as an experiential boundary condition that may shape students’ dependence on and interpretation of internationalization-at-home opportunities. Together, these theoretical lenses map onto the direct-association, mediation, moderation, moderated-mediation, and multi-group components of the proposed model.

2.2. Three pathways of internationalization at home and global competence

The core of internationalization at home is not a loose combination of isolated internationalized activities, but the provision of sustained and organized international and intercultural learning opportunities for students who do not engage in cross-border mobility [3,4]. Within this framework, curriculum internationalization occupies a foundational position because it determines whether students can systematically encounter global issues, diverse knowledge systems, and comparative perspectives [5]. At the same time, cross-cultural interaction on campus and digital international learning extend the implementation sites of internationalization from formal courses to project collaboration, peer exchange, and online collaborative learning [6].

Global competence has become a key outcome in higher education internationalization research. It includes understanding global issues and others’ perspectives as well as intercultural communication, collaboration, and action orientation [7,9]. Prior studies have generally found significant positive associations between internationalization at home and global competence [10,11]. Yet an overall association does not imply that different pathways are homogeneous in their operation. Curricular and learning resource exposure may be connected with cognitive foundations by providing stable input on global issues and multiple knowledge frameworks. Cross-cultural interaction may be connected with students’ capacity to handle difference through real negotiation and feedback. Digital international learning may be associated with global competence by providing frequent, low-threshold, and scalable international learning opportunities [12,13].

If these pathways are collapsed into a single aggregate variable, differences in input mode, participation quality, and educational function are likely to be hidden. Given evidence that global competence has a stable and measurable structure among Chinese students [22], this study includes the three pathways in one model to compare their differentiated associations with global competence.

H1a: Internationalized curricular and learning resource exposure is positively associated with global competence.

H1b: Cross-cultural interaction engagement is positively associated with global competence.

H1c: Digital international learning participation is positively associated with global competence.

2.3. Cross-cultural self-efficacy as a psychological translation pathway

Social cognitive theory argues that external experiences are translated into stable capacities not merely through the accumulation of experiences, but through how individuals understand task requirements and form judgments about their own ability to act [23,24]. In intercultural learning contexts, this judgment is reflected in cross-cultural self-efficacy, or students’ confidence that they can understand people from different cultural backgrounds, communicate across difference, and maintain collaboration. Without this subjective confidence, experiences may remain at the level of exposure and may not settle into more stable competence perceptions [25].

Existing studies show that international experiences can enhance general or intercultural self-efficacy, and that self-efficacy is further associated with intercultural competence and global competence [26,27]. Studies more directly related to internationalization at home provide similar evidence. Internationalized on-campus programs and virtual exchange have been found to enhance students’ confidence in intercultural tasks and support their development in communication, collaboration, and perspective taking [28,29]. This suggests that the association between internationalization at home and global competence may involve an experience-confidence-competence pathway.

When the three pathways are examined simultaneously, cross-cultural interaction is theoretically expected to be more strongly associated with self-efficacy because it involves real negotiation, immediate feedback, and contextual adjustment. The curricular pathway provides more stable knowledge input, whereas the digital pathway provides frequent and repeatable learning spaces. It is therefore plausible that the three pathways are not only directly associated with global competence, but also indirectly associated with it through cross-cultural self-efficacy.

H2a: Internationalized curricular and learning resource exposure is positively associated with cross-cultural self-efficacy.

H2b: Cross-cultural interaction engagement is positively associated with cross-cultural self-efficacy.

H2c: Digital international learning participation is positively associated with cross-cultural self-efficacy.

H3: Cross-cultural self-efficacy is positively associated with global competence.

H4a: Cross-cultural self-efficacy plays an indirect association role between internationalized curricular and learning resource exposure and global competence.

H4b: Cross-cultural self-efficacy plays an indirect association role between cross-cultural interaction engagement and global competence.

H4c: Cross-cultural self-efficacy plays an indirect association role between digital international learning participation and global competence.

2.4. Perceived internationalization climate as a contextual boundary

Student development does not occur in a vacuum. Ecological systems theory emphasizes that the development of individual capacities is embedded in institutional environments and interactional contexts [30]. School climate research similarly shows that how students understand and use learning experiences is strongly shaped by institutional provision, relational norms, and perceived support [31,32]. In internationalized higher education, perceived internationalization climate refers not only to the existence of international projects or policy statements, but also to whether students perceive accessible curricular arrangements, inclusive interactional norms, and sustained resource platforms [33,34].

The environmental dependency of different pathways may not be identical. Curricular/resource-based pathways and digital pathways often depend more on institutional provision, platform infrastructure, and teacher organization, and their associations with cross-cultural self-efficacy may therefore be stronger under a more supportive internationalization climate. The interaction pathway is also shaped by the environment, but whether interaction strengthens confidence may depend more directly on the equality, continuity, and task structure of the interaction itself. Perceived internationalization climate may thus not be a universal amplifier for all pathways, but may condition some pathways more strongly than others.

H5a: Perceived internationalization climate positively moderates the relationship between internationalized curricular and learning resource exposure and cross-cultural self-efficacy.

H5b: Perceived internationalization climate positively moderates the relationship between cross-cultural interaction engagement and cross-cultural self-efficacy.

H5c: Perceived internationalization climate positively moderates the relationship between digital international learning participation and cross-cultural self-efficacy.

H6a: Perceived internationalization climate positively moderates the indirect pathway linking internationalized curricular and learning resource exposure to global competence through cross-cultural self-efficacy.

H6b: Perceived internationalization climate positively moderates the indirect pathway linking cross-cultural interaction engagement to global competence through cross-cultural self-efficacy.

H6c: Perceived internationalization climate positively moderates the indirect pathway linking digital international learning participation to global competence through cross-cultural self-efficacy.

2.5. Overseas experience as a group boundary

Under unequal access to internationalization opportunities, another key question is whether internationalization at home has the same meaning for students with different experiential backgrounds. Inclusive internationalization research suggests that different groups of students may differ in their need structures and benefit pathways [20]. Studies of Chinese university students also indicate that institutional provision and resource conditions shape whether students view global competence as an attainable goal [21]. It is therefore reasonable to expect that the structural associations among internationalization at home, cross-cultural self-efficacy, and global competence may vary between students with and without overseas experience.

On the one hand, prior cross-border exposure may provide students with cultural capital and interpretive frameworks that make it easier to reactivate subsequent internationalized experiences on campus. On the other hand, students without overseas experience may rely more heavily on internationalization at home to obtain international contact opportunities. This study does not assume in advance which group benefits more strongly. Instead, overseas experience is treated as a boundary condition to be examined through multi-group structural equation modeling.

H7: The overall structural relationships among the core variables differ between students with and without overseas experience.

Based on the theoretical reasoning above, this study constructs an integrated model in which three pathways of internationalization at home are associated with global competence both directly and indirectly through cross-cultural self-efficacy, while perceived internationalization climate and overseas experience respectively serve as contextual and group boundary conditions. The theoretical model is shown in Fig 1.

Fig 1. Theoretical model.

Fig 1

3. Methods

3.1. Research design, sample, and procedure

This study employed a cross-sectional online survey design to examine the structural relationships between three pathways of internationalization at home and self-reported global competence among university students in China. Specifically, the study examined the associations between internationalization-at-home learning experiences and global competence, the mediating role of cross-cultural self-efficacy, the moderating role of perceived internationalization climate, and structural heterogeneity across groups with different overseas-experience backgrounds.

Data were collected through the Wenjuanxing online survey platform between April 4 and April 6, 2026. The questionnaire was distributed primarily through WeChat forwarding, supplemented by direct-link access and mobile submissions. A total of 800 questionnaires were submitted. To ensure that respondents matched the target population, an eligibility screening question was placed before the formal survey, asking whether respondents were currently enrolled in a higher-education program. Thirty-two respondents who answered “No” were excluded, and the remaining 768 valid responses were retained for analysis.

All substantive questionnaire items were mandatory; therefore, the final dataset contained no item-level missing values. No respondents were excluded based on completion time or post hoc response-pattern screening criteria. Among the retained respondents, completion time averaged 188.75 seconds (SD = 64.19), with a median of 185 seconds, an interquartile range of 131–247 seconds, and a range of 82–301 seconds. These response-time statistics are reported to provide transparency regarding data quality rather than as exclusion criteria. The final sample size was considered adequate for confirmatory factor analysis, structural equation modeling, mediation and moderation analyses, and multi-group comparisons [35].

Regarding recruitment channels, 701 respondents (91.3%) accessed the questionnaire through WeChat forwarding, 38 (4.9%) through direct-link access, and 29 (3.8%) through mobile submissions. According to the IP-derived location metadata automatically recorded by the survey platform, responses were received from 29 provincial-level regions in China. These locations represent the geographic origins of online submissions rather than verified locations of participants’ universities or current residence. Because university or institutional affiliation information was not collected, the institutional distribution of the sample cannot be directly reported.

The study adopted a convenience sampling strategy based on practical accessibility through online dissemination. Although the sample covered respondents from multiple regions and included students from different levels of higher education, it should not be interpreted as a nationally representative sample of Chinese university students. Potential sample selectivity, including differences in international exposure and overseas-experience composition, should therefore be considered when interpreting the findings and is further discussed in the limitations section.

The study protocol was reviewed by the Research Office of Zhejiang Wanli University and determined to qualify for exemption from formal ethics review (Document No. WL20250701; determination date: July 6, 2025). In an institutional confirmation letter dated October 19, 2026, Zhejiang Wanli University confirmed, after examining its existing institutional responsibility and delegation records and the original study documentation, that the Research Office was already institutionally authorized to assess the applicability of ethics review and determine eligibility for exemption for this category of non-interventional educational and social science research when the original determination was issued. The University further verified that the original application, exemption determination, and submitted manuscript concerned the same study, and that the questionnaire, participant information and consent page, survey procedures, and data collection conducted in April 2026 fell within the scope of the original determination. The confirmation letter documents the historical authority of the Research Office and the validity of the original exemption within that scope; it does not constitute a new ethics approval.

The exemption covered a non-interventional, minimal-risk online questionnaire survey of university students. Before accessing the questionnaire, participants were provided with an electronic study information sheet and informed consent statement. Participation was voluntary, respondents could proceed only after providing electronic informed consent, and they could discontinue participation before submission. The ethics-review exemption did not constitute a waiver of informed consent. The questionnaire did not solicit respondents’ names, student numbers, university affiliations, telephone numbers, or other directly identifying information. The survey platform automatically recorded limited technical metadata, including IP-derived location information. These metadata were used only for duplicate-response checking and aggregated geographic description and were not used to identify individual respondents. Analysis and data-sharing files excluded direct identifiers and detailed IP information. The collected data were used solely for academic research and were analyzed and reported only in aggregate form.

3.2. Instrument development and measurement

The questionnaire was developed directly in Chinese based on established construct definitions and measurement dimensions rather than translated item by item from a single English-language scale. Because the instrument was developed specifically for the Chinese higher-education context instead of being adapted through translation from an existing English-language questionnaire, a conventional back-translation procedure was not applicable. This approach differs from cross-language scale transfer studies, where translation and back-translation are typically required to ensure semantic equivalence [36].

The development process followed a construct-driven approach. First, the conceptual definitions and measurement dimensions of internationalization at home, cross-cultural self-efficacy, perceived internationalization climate, and global competence were identified from relevant theoretical and empirical studies. Second, an initial Chinese-language item pool was developed and adapted to reflect the learning experiences and institutional contexts of Chinese university students. The items were organized according to the prespecified construct definitions and were measured using a uniform 7-point Likert scale to maintain response consistency and reduce respondent burden [37,38].

Before formal data collection, the initial item pool was reviewed by a panel of five experts, including two scholars specializing in higher-education internationalization, two scholars with experience in intercultural education, and one scholar specializing in educational measurement and questionnaire design. The expert review focused on whether the items adequately represented their intended constructs, whether the wording was clear and appropriate for Chinese university students, and whether conceptually adjacent constructs—particularly internationalized curricular and learning-resource exposure (X1) and digital international learning participation (X3)—were sufficiently distinguishable. Based on the experts’ comments, ambiguous or overlapping expressions were revised, and the wording and construct alignment of the items were refined. The resulting item set and nine-factor measurement structure were finalized before the formal survey was administered. The expert review was qualitative and content-oriented; no formal Delphi procedure or quantitative content-validity index was used.

No separate student cognitive interviews, small-sample internal pretest, independent pilot survey, or pre-survey EFA was conducted before the formal data collection. This limitation is reported explicitly because expert content review cannot substitute for respondent-based pretesting or independent psychometric validation. To strengthen the available internal validation evidence, the revised study therefore incorporated several complementary post hoc assessments, including HTMT analysis, comparisons with theoretically plausible alternative measurement models, and a split-sample procedure combining EFA and CFA. Because the two subsamples were drawn from the same formal dataset, the split-sample procedure should be interpreted as evidence of internal replicability rather than independent external validation.

All focal constructs were modeled as latent variables. The three internationalization-at-home pathways were measured as separate dimensions representing internationalized curricular and learning resource exposure, cross-cultural interaction engagement, and digital international learning participation. Cross-cultural self-efficacy was modeled as a mediator, perceived internationalization climate as a contextual moderator, and global competence as a multidimensional higher-order construct consisting of global issue understanding, perspective taking, intercultural communication and collaboration, and global responsibility and action.

3.3. Measurement and operationalization

Except for background variables, all focal variables were measured on a 7-point Likert scale ranging from 1 = strongly disagree to 7 = strongly agree. Definitions, dimensions, and item numbers of the focal constructs are shown in Table 1. The complete items, variable names, and scoring rules are provided in S1 Table and S2 Table in the Supporting Information.

Table 1. Definitions and measurement dimensions of the focal constructs.

Construct Code Definition Indicators/ dimensions Items
Internationalized curricular and learning resource exposure X1 Students’ exposure to institutionally embedded internationalized curriculum contents and structured learning resources within formal teaching and learning arrangements International cases and comparative perspectives; global issue integration in coursework; institutionally provided internationalized learning resources 4
Cross-cultural interaction engagement X2 Students’ participation in cross-cultural interaction in classes, projects, and campus activities Classroom interaction; project collaboration; campus-based intercultural exchange 4
Digital international learning participation X3 Students’ participation in digitally mediated international learning activities through online platforms, virtual interaction, and cross-border collaborative learning Virtual exchange and online international courses; digitally mediated cross-border interaction and collaboration; active use of international digital learning resources in task completion 4
Cross-cultural self-efficacy M Students’ confidence in understanding, communicating, and collaborating across cultures Understanding confidence; communication confidence; collaboration confidence 5
Perceived internationalization climate W Students’ perceived campus climate of internationalization Institutional support; inclusive norms; accessible international opportunities 5
Global competence (second-order) Y Higher-order latent construct Y1 global issue understanding; Y2 perspective taking; Y3 intercultural communication and collaboration; Y4 global responsibility and action 12

Internationalized curricular and learning resource exposure (X1) measured students’ exposure to internationalized content and resources within formal teaching arrangements. The construct covered the inclusion of international cases, global issues, and cross-national comparisons in courses; access to course materials, lectures, and academic resources in English or other foreign languages; teacher guidance from different national or regional perspectives; and the university’s provision of internationalized courses, lectures, and learning resources. The construct therefore represents a primarily curricular and cognitive-input pathway in the development of global competence. Its design drew mainly on literature concerning curriculum internationalization, internationalization at home, and global competence among Chinese university students [5,10].

Cross-cultural interaction engagement (X2) measured students’ perceived opportunities for and self-reported participation in cross-cultural communication in courses, projects, and campus activities. It covered opportunities to communicate with people from different national or cultural backgrounds, participation in classroom discussions, student activities, and project practices involving intercultural communication, willingness to express one’s own views while listening to different perspectives, and institutional provision of platforms for interaction with international students, foreign teachers, or other cross-cultural groups. The item construction was informed by studies of internationalization at home and research on intercultural interaction and cross-cultural relationships on campus [11,13,39,40].

Digital international learning participation (X3) measured students’ experiences of and opportunities for international learning through digital media. In contrast to X1, which emphasizes internationalized content and resources embedded in formal curricular provision, X3 focuses on digitally mediated participation, including online international courses, virtual exchange, cross-border collaborative learning, access to digital resources concerning international universities, organizations, and global issues, task-related use of international databases and foreign-language materials, and opportunities for online communication or collaboration with overseas teachers, students, or international teams. The construct was informed by research on virtual exchange, collaborative online international learning, and digitally mediated international learning [14,41].

The distinction between X1 and X3 was based on the organizing locus and mode of engagement rather than on the presence of digital media alone. X1 captures whether international content, comparative perspectives, and international learning resources are embedded in curricular and institutional provision. An international resource may therefore remain part of X1 even when it is accessed digitally, provided that the resource is embedded in a course or formal teaching arrangement. By contrast, X3 captures participation, resource use, and interaction organized through digital media, such as online international courses, virtual exchange, cross-border collaborative tasks, and the use of international digital resources. Curricular provision and digitally mediated participation were therefore treated as conceptually related but distinct attributes.

Cross-cultural self-efficacy (M) measured students’ subjective confidence in handling intercultural communication and learning tasks. It did not measure objectively demonstrated intercultural competence. Instead, it assessed whether students believed that they could communicate effectively with people from different cultural backgrounds, remain open and respectful when viewpoints differed, meet communication, collaboration, and understanding demands in intercultural learning tasks, identify key differences between cultural perspectives and understand why they arise, and integrate different viewpoints into a balanced conclusion. The construct was grounded in the self-efficacy framework of social cognitive theory [23,24] and informed by empirical studies of intercultural competence and internationalized learning [25,26,28,29].

Perceived internationalization climate (W) measured students’ overall evaluation of the institutional environment for internationalization. It captured whether the university valued students’ international outlook and intercultural competence, fostered an inclusive and diversity-respecting campus climate, provided support from teachers and administrators, offered internationalized learning opportunities that were continuous, accessible, and relatively equitable, and enabled students to obtain international learning opportunities without necessarily studying abroad. The measurement was informed by school-climate research concerning institutional support and inclusiveness [31,32] and by studies of learning environments in international higher education [33,34].

Global competence (Y) was modeled as a second-order latent construct consisting of four first-order dimensions: global issue understanding, perspective taking, intercultural communication and collaboration, and global responsibility and action orientation. This structure reflects the multidimensional nature of global competence while preserving its overall higher-order structure. The construct was based primarily on the OECD global competence framework and related conceptual work [7,9] and was also informed by scale-development research involving Chinese samples [22]. Importantly, the study measured students’ self-reported global competence rather than externally assessed performance. The conclusions concerning global competence therefore refer to perceived understanding, competence, and action orientation rather than objectively verified ability.

Several background variables were included as controls: gender, education level, major field, self-rated foreign-language proficiency, overseas experience, and parents’ highest level of education. Major field comprised seven nominal categories: humanities and social sciences; economics, business, and law; education; science and engineering; medicine and life sciences; arts and sports; and other fields. In the controlled structural, mediation, moderation, moderated-mediation, and control-robustness models, major field was represented by six dummy variables, with humanities and social sciences as the reference category.

Overseas experience was measured using four mutually exclusive categories: no overseas experience; tourism or family visits; short-term exchange lasting three months or less; and long-term study or exchange lasting more than three months. The original categories were retained in descriptive reporting. In the controlled single-group models, overseas experience was represented by three dummy variables, with no overseas experience as the reference category. For the multi-group SEM, overseas experience was recoded into a binary grouping variable distinguishing respondents with no overseas experience from those reporting any form of overseas experience. This binary grouping preserved adequate group sizes and allowed overseas exposure to be examined as a broad experiential boundary. Because tourism or family visits differ from educational mobility in purpose, duration, and learning intensity, the binary comparison should not be interpreted as a fine-grained causal comparison among different forms of mobility. Gender, education level, foreign-language proficiency, and parental education were entered using their original questionnaire coding.

3.4. Data quality control and common method bias

Because the study employed a cross-sectional, self-reported, single-survey design, data quality and potential common method bias were important methodological concerns. Data screening was conducted before hypothesis testing and was limited to eligibility and completeness checks. The opening screening item was used to determine whether respondents were currently enrolled in higher education. Of the 800 submitted questionnaires, 32 failed this criterion and were excluded, leaving 768 eligible cases. Because all substantive questionnaire items were mandatory, no retained case contained item-level missing values.

Completion times were inspected descriptively, but no fixed minimum or maximum response-time threshold had been specified or applied. No additional case was excluded solely because of response duration, straightlining, or another post hoc response-pattern criterion. Among retained cases, the median completion time was 185 seconds, the interquartile range was 131–247 seconds, and the full range was 82–301 seconds. These statistics are reported transparently rather than used retrospectively as exclusion thresholds. No duplicate IP address was observed among the retained submissions.

To reduce misunderstanding and careless responding, the questionnaire included an eligibility screen, clear instructions, modular organization, concise item wording, and a consistent response format. To address common method bias arising from same-time and same-source measurement, the study combined procedural precautions with supplementary statistical diagnostics. Participants were assured that participation was voluntary, that the questionnaire did not request names or student identification information, and that findings would be reported only in aggregate form. Non-evaluative wording was used to reduce evaluation anxiety and social-desirability concerns [38].

Statistically, Harman’s single-factor test was first used as an auxiliary diagnostic of whether one factor dominated the item variance. This test cannot rule out common method bias and cannot substitute for multi-source or multi-wave research designs. It was therefore supplemented by a CFA-based sensitivity model in which all 34 focal indicators loaded on their theoretically specified trait factors and on an additional orthogonal common method factor. The method-factor loadings were constrained to equality to reduce the risk of instability or non-identification associated with a fully unrestricted method factor. The method factor was specified as orthogonal to the nine trait factors. These analyses were intended to assess whether a single common method component materially changed the measurement-model results, rather than to establish the absence of all forms of common method bias [38]. Accordingly, all structural paths are interpreted as statistical associations based on cross-sectional self-reported data.

3.5. Data analysis strategy

The analysis followed the two-step approach of first assessing the measurement model and then estimating the structural models [37]. Data preparation, descriptive statistics, distribution diagnostics, and reliability analyses were conducted using IBM SPSS Statistics. CFA, SEM, latent interaction analysis, and multi-group analysis were conducted using Mplus Version 9. HTMT coefficients were calculated from item-level correlations.

All questionnaire items were mandatory. After the eligibility screen, the retained analytical sample contained no item-level missing values; therefore, no imputation, full-information maximum likelihood procedure, or other missing-data treatment was required. Because the focal indicators used seven response categories and showed no serious departures from univariate normality, they were treated as approximately continuous variables. Measurement models, the baseline structural model, moderation models, the common-method-factor model, and measurement-invariance models were estimated using robust maximum likelihood estimation (MLR). Models requiring nonparametric bootstrap confidence intervals, including the mediation and moderated-mediation models, were estimated using maximum likelihood estimation (ML) with 5,000 bootstrap replications.

First, descriptive statistics and distribution diagnostics were conducted for the background variables and focal constructs, and the suitability of the data for latent-variable modeling was assessed. Because global competence was theoretically defined as a multidimensional construct, both first-order and second-order CFA models were estimated. The theoretical nine-factor measurement model was specified without cross-loadings or correlated indicator residuals. No indicator was deleted and no modification-index-driven respecification was performed to improve model fit.

At the measurement-model stage, internal consistency was assessed using Cronbach’s alpha and composite reliability. Convergent validity was assessed using average variance extracted. Discriminant validity was evaluated using both the Fornell–Larcker criterion and HTMT [42,43]. Particular attention was given to the HTMT value between X1 and X3 because these two constructs both involve access to international learning resources but differ in their organizing locus and mode of participation.

Several supplementary measurement analyses were conducted. First, the theoretical nine-factor model was compared with four plausible alternatives: a model combining X1 and X3; a model combining X1, X2, and X3; a model treating the four dimensions of global competence as a single first-order factor; and a one-factor model combining all focal indicators. Second, the equal-loading orthogonal common-method-factor model was compared with the baseline nine-factor CFA. Third, a post hoc split-sample validation was conducted. Using random seed 20260808, the full sample was randomly divided into a discovery subsample and a validation subsample of 384 respondents each. A nine-factor EFA was conducted in the discovery subsample, after which the theoretical nine-factor CFA and the second-order global-competence model were estimated in the validation subsample. The random group assignments were saved to permit exact replication. Because both subsamples came from the same original dataset, the procedure provides evidence of internal replicability rather than external validation.

At the structural-model stage, SEM was used to estimate the relationships among the three internationalization-at-home pathways, cross-cultural self-efficacy, perceived internationalization climate, and global competence. The primary structural model estimated the paths from X1, X2, and X3 to cross-cultural self-efficacy and global competence and the path from cross-cultural self-efficacy to global competence.

Latent moderation was estimated using the latent moderated structural equations approach implemented through the XWITH command in Mplus, with TYPE = RANDOM and numerical integration. No manifest product indicators were constructed, and the observed indicators were not mean-centered. Three interaction models were estimated separately for X1 × W, X2 × W, and X3 × W. Each model included the main effects of the focal internationalization-at-home pathway and perceived internationalization climate on cross-cultural self-efficacy, the latent interaction term, the other two internationalization-at-home pathways, and the background controls. The global-competence equation included cross-cultural self-efficacy, the three internationalization-at-home pathways, perceived internationalization climate, and the same controls. Thus, the direct effect of W was estimated in both the M and Y equations. This latent interaction procedure followed established methodological recommendations [44].

Because conventional chi-square-based fit indices are not produced for latent interaction models estimated using XWITH and numerical integration, a corresponding no-interaction reference model was estimated. The reference model contained the same measurement model, the same main structural paths, perceived internationalization climate, and the same background controls but contained no latent interaction term. The no-interaction model was evaluated using conventional fit indices. The interaction models were evaluated using convergence diagnostics, log-likelihood, AIC, BIC, robust likelihood-ratio comparisons with the no-interaction model, and the interaction coefficients.

For significant interactions, simple slopes were calculated at the mean and at one latent standard deviation below and above the mean of perceived internationalization climate. The latent standard deviation of W was estimated from its model variance rather than fixed arbitrarily at one. Conditional indirect effects were calculated at the same three levels of W. For each pathway, the index of moderated mediation was calculated as the product of the latent interaction coefficient and the path from cross-cultural self-efficacy to global competence.

Indirect associations in the ordinary mediation model were estimated using ML with 5,000 nonparametric bootstrap draws and 95% percentile bootstrap confidence intervals, implemented through CINTERVAL(BOOTSTRAP) in Mplus [45]. Conditional indirect associations and indices of moderated mediation were estimated using ML with 5,000 nonparametric bootstrap draws and 95% bias-corrected bootstrap confidence intervals, implemented through CINTERVAL(BCBOOTSTRAP). For the three moderated-mediation models (analyses 08–10), resampling was performed using the built-in Mplus bootstrap procedure with TYPE = RANDOM and ALGORITHM = INTEGRATION. Each of these three archived outputs reports 5,000 requested and 5,000 completed bootstrap computations and normal termination; no incomplete bootstrap computations were reported. The moderated-mediation hypotheses were evaluated from the 95% bias-corrected bootstrap confidence intervals of the indices, rather than from the statistical significance of the individual conditional indirect effects.

To examine group heterogeneity by overseas experience, multi-group SEM was conducted. Before structural paths were compared, measurement invariance was evaluated sequentially using configural, first-order metric, higher-order metric, and scalar invariance models. In the configural model, all non-marker first-order loadings, non-marker second-order loadings, and observed-item intercepts were allowed to vary across groups. The first-order metric model constrained the first-order loadings to equality. The higher-order metric model additionally constrained the freely estimated second-order loadings to equality. The scalar model further constrained observed-item intercepts to equality, fixed latent means to zero in the reference group, and allowed latent means to be estimated in the focal group. Changes in CFI, RMSEA, and SRMR were used to evaluate invariance across successive models [46,47]. A partial scalar model was specified as a potential fallback if full scalar invariance was not supported; however, it was not required because the full scalar model satisfied the adopted invariance criteria.

After measurement invariance had been established, the seven core structural paths were freely estimated across the two overseas-experience groups. The multi-group structural models were estimated without background covariates. Overseas experience was used solely to define the two groups, comprising 354 respondents without overseas experience and 414 with overseas experience. Group differences were evaluated first using an omnibus Wald test of the joint equality of all seven unstandardized coefficients and then using seven one-degree-of-freedom Wald tests for the individual paths. These tests therefore concern unadjusted group differences in the seven structural coefficients. Unstandardized coefficients were used for formal group-difference testing, whereas standardized coefficients were reported to facilitate interpretation within each group.

Two sets of supplementary structural robustness analyses were conducted. First, the partial-mediation, no-mediation, and full-mediation models in Table 2 were estimated using MLR on the same 768 respondents and the same 29 indicators of X1, X2, X3, M, and the four first-order dimensions of global competence. Global competence was specified as a second-order factor. These models excluded the background covariates and W and its indicators and differed only in their direct and mediating path restrictions. The partial-mediation model included both the indirect paths through cross-cultural self-efficacy and the direct paths from X1, X2, and X3 to global competence. The full-mediation model omitted the three direct X-to-Y paths. The no-mediation model omitted the M-to-Y path and fixed the residual covariance between M and Y to zero. Model fit and information criteria were compared only among these three unadjusted models.

Table 2. Comparison of unadjusted alternative structural models.

Model χ² df CFI TLI RMSEA SRMR AIC BIC
Partial mediation 386.588 363 0.998 0.998 0.009 0.023 68041.032 68510.055
No mediation 413.483 364 0.995 0.995 0.013 0.035 68067.617 68531.996
Full mediation 464.762 366 0.991 0.990 0.019 0.058 68118.356 68573.447

Note: All models used the same 768 respondents, 29 substantive indicators, second-order specification of global competence, and MLR estimator. No background covariates or W indicators were included. Information criteria are compared only among the three models in this table.

Second, the standardized core path estimates were compared with and without background covariates to assess their sensitivity to covariate adjustment. The primary SEM in Table 3 included 13 observed covariates in both the M and Y equations, whereas the alternative models in Table 2 did not include these covariates. The different degrees of freedom therefore reflect the covariate specification rather than differences in respondents or substantive indicators. Table 4 compares the unadjusted partial-mediation model with the covariate-adjusted primary SEM reported in Table 3.

Table 3. Structural model results.

Hypothesis Path Unstandardized B SE Standardized β Result
H2a X1 → M 0.156 0.038 0.151 Supported
H2b X2 → M 0.384 0.04 0.38 Supported
H2c X3 → M 0.294 0.038 0.287 Supported
H1a X1 → Y 0.122 0.035 0.145 Supported
H1b X2 → Y 0.146 0.036 0.176 Supported
H1c X3 → Y 0.224 0.034 0.267 Supported
H3 M → Y 0.209 0.041 0.255 Supported

Note: Structural model fit: χ²(714) = 758.158, p = 0.122, CFI = 0.996, TLI = 0.996, RMSEA = 0.009, 90% CI [0.000, 0.014], SRMR = 0.029. N = 768. B denotes the unstandardized coefficient. The model controlled for gender, education level, foreign-language proficiency, parental education, six major-field dummy variables, and three overseas-experience dummy variables. Control coefficients are omitted because they are not focal to the hypotheses; the complete Mplus output is provided in S4 File.

Table 4. Control-robustness comparison of core paths.

Path Without controls β With controls β Change
X1 → M 0.151 0.151 0
X2 → M 0.384 0.38 −0.004
X3 → M 0.288 0.287 −0.001
X1 → Y 0.143 0.145 0.002
X2 → Y 0.179 0.176 −0.003
X3 → Y 0.266 0.267 0.001
M → Y 0.252 0.255 0.003

Note: ** p < 0.01; *** p < 0.001.

Because the study was based on cross-sectional survey data, all direct, indirect, moderation, moderated-mediation, and multi-group results were interpreted as theoretically consistent statistical associations. The terms mediation, moderation, and pathway describe the specified statistical model and should not be interpreted as establishing strict temporal ordering or causal identification.

4. Results

4.1. Descriptive statistics and common method bias

Based on 768 valid responses, this study systematically tested the structural relationships among internationalization-at-home learning experiences, cross-cultural self-efficacy, and global competence. Of the 768 participants, 354 (46.1%) reported no overseas experience, 222 (28.9%) reported overseas experience for tourism or family visits, 115 (15%) reported a short-term exchange lasting three months or less, and 77 (10%) reported long-term study or exchange lasting more than three months. The three overseas-experience categories together comprised 414 participants (53.9%). Descriptive statistics for background variables are reported in S3 Table. Descriptive statistics for the focal constructs are shown in Table 5. The means of the three internationalization-at-home pathways, cross-cultural self-efficacy, perceived internationalization climate, and the four first-order dimensions of global competence were all above the midpoint of the 7-point scale, indicating that the sample generally reported medium-to-high levels of internationalized learning experience and global competence. Skewness ranged from −0.552 to −0.185, and kurtosis ranged from −0.809 to −0.434, suggesting no serious non-normality.

Table 5. Descriptive statistics of key constructs.

Construct Mean SD Skewness Kurtosis
X1 4.636 1.202 −0.247 −0.569
X2 4.689 1.205 −0.316 −0.434
X3 4.744 1.236 −0.185 −0.637
M 4.957 1.226 −0.337 −0.809
W 5.086 1.276 −0.427 −0.781
Y1 4.764 1.251 −0.324 −0.607
Y2 4.804 1.269 −0.302 −0.708
Y3 5.007 1.317 −0.552 −0.477
Y4 5.046 1.331 −0.503 −0.634

Note: N = 768. X1 = internationalized curricular and learning resource exposure; X2 = cross-cultural interaction engagement; X3 = digital international learning participation; M = cross-cultural self-efficacy; W = perceived internationalization climate; Y1-Y4 = the four first-order dimensions of global competence.

Bivariate correlations among composite constructs are reported in S4 Table. Overall, the three internationalization-at-home pathways were significantly and positively correlated with cross-cultural self-efficacy, perceived internationalization climate, and the four dimensions of global competence. The correlation coefficients were generally in the low-to-moderate or moderate range. These results provide preliminary support for the expected directions of association and indicate that the constructs are related but not so highly correlated as to suggest serious overlap.

Before CFA, data suitability was assessed. The KMO value was 0.922, and Bartlett’s test of sphericity was significant (χ² = 14735.609, df = 561, p < 0.001), indicating that the variables had a suitable common factor structure for latent variable modeling. Harman’s single-factor test results are reported in S5 Table. The unrotated principal component analysis showed that the first factor explained 28.898% of the variance, below the commonly used 50% heuristic threshold. This result cannot fully rule out common method bias, but it provides preliminary auxiliary evidence that a single common factor did not dominate item variance. Given that all focal variables were collected from the same self-reported questionnaire at one time point, subsequent results should still be interpreted as structural statistical associations rather than as strict causal evidence.

Harman’s single-factor test was supplemented by a CFA-based equal-loading common-method-factor sensitivity analysis. The baseline nine-factor model showed χ²(491) = 515.774, CFI = 0.998, TLI = 0.998, RMSEA = 0.008, and SRMR = 0.021. After an orthogonal common method factor with equal loadings across all 34 indicators was added, model fit remained virtually unchanged, χ²(490) = 514.724, CFI = 0.998, TLI = 0.998, RMSEA = 0.008, and SRMR = 0.021. The common method loading was 0.004 (SE = 0.014, p = 0.805), and the corresponding Wald test was not significant, Wald χ²(1) = 0.061, p = 0.805. The results indicate that a single equal-loading common method component did not materially account for the observed measurement structure. Nevertheless, this statistical sensitivity analysis cannot eliminate all forms of common method bias, and the results remain subject to the limitations of same-source, cross-sectional self-report data [38]. Complete results are reported in Panel B of S5 Table.

4.2. Measurement model

At the measurement model stage, a first-order CFA model including nine first-order latent variables was first estimated. As shown in Table 6, the first-order CFA model fitted the data well: χ² = 515.774, df = 491, χ²/df = 1.050, CFI = 0.998, TLI = 0.998, RMSEA = 0.008, and SRMR = 0.021. According to conventional SEM criteria, these fit indices indicate good correspondence between the measurement model and the sample data [42,43].

Table 6. Measurement model fit indices.

Model χ² df CFI TLI RMSEA SRMR
First-order CFA 515.774 491 0.998 0.998 0.008 0.021
Second-order CFA (Y) 71.988 50 0.994 0.992 0.024 0.025

Note: The first-order model included X1, X2, X3, M, W, and the four first-order dimensions of global competence. The second-order model specified global competence as a higher-order latent factor measured by Y1-Y4.

Standardized item loadings from the first-order CFA are reported in S7 Table. All loadings were significant at p < 0.001 and ranged from 0.629 to 0.889. M3, which assessed confidence in meeting communication, collaboration, and understanding demands in internationalized or cross-cultural learning tasks, had the lowest loading of 0.629. It was retained to preserve this aspect of the prespecified self-efficacy construct. Its loading remained statistically significant, and the self-efficacy construct showed acceptable composite reliability (CR = 0.875) and average variance extracted (AVE = 0.585) [42,43].

Reliability and convergent validity results are shown in Table 7 and S6 Table. Cronbach’s alpha values ranged from 0.873 to 0.914, indicating good internal consistency. Composite reliability values ranged from 0.840 to 0.892, and AVE values ranged from 0.585 to 0.667, meeting commonly used standards for latent variable measurement [42]. Latent variable correlations were significant but remained within acceptable ranges (S8 Table). The Fornell–Larcker results are reported in Panel A of S9 Table. The square root of AVE for each construct exceeded its correlations with the other constructs, supporting discriminant validity [42].

Table 7. Reliability, convergent validity, and higher-order loadings.

Construct/ Path Items/ Loading CR/ R² AVE
X1 4 0.876 0.641
X2 4 0.875 0.638
X3 4 0.877 0.640
M 5 0.875 0.585
W 5 0.892 0.623
Y1 3 0.840 0.636
Y2 3 0.846 0.647
Y3 3 0.857 0.667
Y4 3 0.847 0.649
Y → Y1 0.849 0.720
Y → Y2 0.836 0.699
Y → Y3 0.808 0.653
Y → Y4 0.771 0.594

Note: All standardized factor loadings were significant at p < 0.001. Detailed Fornell-Larcker results are provided in the Supporting Information.

Discriminant validity was additionally evaluated using HTMT. The first-order CFA yielded an X1–X3 latent correlation of 0.254 (SE = 0.037, p < 0.001; S8 Table), while the corresponding HTMT coefficient was 0.249 (S9 Table). These results provide complementary evidence that the two constructs were related but empirically distinguishable. The largest HTMT value among the six focal constructs was 0.484, and the largest value in the complete nine-factor measurement model was 0.749, occurring between Y1 and Y2. All HTMT coefficients were below the commonly used threshold of 0.85, providing additional support for discriminant validity [42,43]. The complete HTMT matrices are reported in Panels B and C of S9 Table.

Global competence was then modeled as a second-order latent construct. The second-order CFA model also showed good fit (Table 6): χ² = 71.988, df = 50, χ²/df = 1.440, CFI = 0.994, TLI = 0.992, RMSEA = 0.024, and SRMR = 0.025. Detailed second-order loadings are reported in Table 7 and S10 Table. The standardized loadings of the four first-order dimensions on second-order global competence were 0.849, 0.836, 0.808, and 0.771, with R² values of 0.720, 0.699, 0.653, and 0.594, respectively. These results indicate that global issue understanding, perspective taking, intercultural communication and collaboration, and global responsibility and action orientation were valid reflective dimensions of global competence.

A post hoc split-sample procedure was conducted to examine the internal replicability of the measurement structure. Using random seed 20260808, the full sample was randomly divided into a discovery subsample and a validation subsample of 384 respondents each. A nine-factor EFA conducted in the discovery subsample showed acceptable fit, χ²(291) = 364.667, CFI = 0.988, TLI = 0.978, RMSEA = 0.026, and SRMR = 0.014. The theoretical nine-factor CFA was subsequently estimated in the validation subsample and also showed good fit, χ²(491) = 511.931, CFI = 0.997, TLI = 0.996, RMSEA = 0.011, and SRMR = 0.029. The second-order global-competence model was also supported in the validation subsample, χ²(508) = 530.717, CFI = 0.997, TLI = 0.996, RMSEA = 0.011, and SRMR = 0.031. Because both subsamples were drawn from the same original dataset, these findings provide evidence of internal replicability rather than independent external validation. Full results and the random-assignment rule are reported in Panel A of S13 Table.

The theoretical measurement structure was further compared with several plausible alternative models. A model combining X1 and X3 showed substantially poorer fit, χ²(499) = 1985.874, CFI = 0.882, TLI = 0.868, RMSEA = 0.062, and SRMR = 0.087. The models combining X1, X2, and X3, treating global competence as a unidimensional first-order construct, and combining all indicators into a single factor also fitted the data substantially worse than the theoretical nine-factor model. Across these simplified alternatives, CFI ranged from 0.427 to 0.921 and RMSEA ranged from 0.050 to 0.134. The complete alternative-model results are reported in Panel B of S13 Table. These comparisons support the empirical distinction among X1, X2, and X3 and the multidimensional higher-order specification of global competence.

Although the fit indices of the proposed measurement model were unusually strong, the model was not obtained through post hoc optimization. No indicator was deleted, no cross-loading was added, no residual covariance was freed, and no modification-index-driven respecification was made. The nine-factor measurement structure and its corresponding item assignments were specified before formal data collection on the basis of the theoretical construct definitions and the qualitative expert review described in Section 3.2. They were not developed or modified in response to the CFA results. The strong fit should therefore be interpreted together with the split-sample validation and the comparisons with theoretically plausible alternative measurement models, rather than as evidence that the model reproduces the population covariance structure without uncertainty [42,43].

Taken together, the first-order and second-order CFA results, reliability and convergent-validity estimates, Fornell–Larcker and HTMT assessments, split-sample validation, and alternative-model comparisons provided consistent support for the proposed measurement structure.

4.3. Structural model and direct associations

After the measurement model was supported, the main structural model was estimated. The results are reported in Table 3. The model showed good overall fit, χ²(714) = 758.158, p = 0.122, CFI = 0.996, TLI = 0.996, RMSEA = 0.009, 90% CI [0.000, 0.014], and SRMR = 0.029. These results indicate that the structural model remained consistent with the observed data after the three internationalization-at-home pathways, cross-cultural self-efficacy, global competence, and the demographic control variables were included.

The standardized structural model with standardized path coefficients is shown in Fig 2.

Fig 2. Standardized structural model with standardized path coefficients.

Fig 2

All three internationalization-at-home pathways were significantly and positively associated with cross-cultural self-efficacy. Among them, cross-cultural interaction engagement showed the strongest association with self-efficacy, followed by digital international learning participation and internationalized curricular and learning resource exposure. This suggests that, compared with curricular and resource input, real interaction is more strongly associated with students’ subjective judgment that they can handle intercultural tasks. A cautious interpretation is that interactive experiences are more closely connected with confidence formation through immediate feedback, ongoing negotiation, and contextual adjustment, rather than merely increasing internationalized exposure.

All three internationalization-at-home pathways were also significantly and positively associated with global competence. Digital international learning participation showed the strongest direct association, followed by cross-cultural interaction engagement and internationalized curricular and learning resource exposure. Taken together with the previous results, the digital pathway was more prominent in direct association, whereas the interaction pathway was more prominent in psychological translation through self-efficacy. This does not mean that the digital pathway is educationally superior to the other pathways. Rather, the three pathways should be understood as complementary mechanisms with different educational functions. Curricular exposure provides stable cognitive input, interaction supports experiential internalization, and digital learning expands opportunity access and participation frequency.

4.4. Indirect associations through cross-cultural self-efficacy

To examine whether cross-cultural self-efficacy played an indirect association role between the three internationalization-at-home pathways and global competence, bootstrapping with 5,000 resamples was conducted. The results are reported in Table 8, and standardized direct, indirect, and total effects are provided in S11 Table.

Table 8. Bootstrap indirect association results.

Hypothesis Path Direct effect Indirect effect Total effect 95% CI of indirect effect
H4a X1 → M → Y 0.122** 0.033** 0.155*** [0.014, 0.056]
H4b X2 → M → Y 0.146*** 0.080*** 0.226*** [0.046, 0.119]
H4c X3 → M → Y 0.224*** 0.061*** 0.285*** [0.035, 0.092]

Note: The confidence intervals are 95% percentile bootstrap confidence intervals based on 5,000 bootstrap samples.

The 95% percentile bootstrap confidence intervals for all three indirect associations excluded zero. The indirect association of internationalized curricular and learning-resource exposure with global competence through cross-cultural self-efficacy was 0.033, 95% percentile bootstrap CI [0.014, 0.056]. The corresponding estimates were 0.080, 95% percentile bootstrap CI [0.046, 0.119], for cross-cultural interaction engagement and 0.061, 95% percentile bootstrap CI [0.035, 0.092], for digital international learning participation. Because the corresponding direct associations remained significant, these results were consistent with partial mediation in the specified cross-sectional model. They should not be interpreted as evidence of causal transmission.

Because the direct associations between the three pathways and global competence remained significant after self-efficacy was included, all three indirect pathways should be interpreted as partial mediation rather than full mediation. The standardized effect decomposition in S11 Table shows that the indirect effects followed the pattern X2 > X3 > X1, whereas the total effects followed the pattern X3 > X2 > X1. In other words, cross-cultural interaction engagement was most strongly associated with global competence through self-efficacy, whereas digital international learning participation showed the strongest total association. These findings are consistent with the idea that self-efficacy bridges intercultural learning experiences and competence development [25,27], while also showing that different internationalization-at-home pathways may be linked to global competence through different psychological and opportunity structures.

4.5. Moderation and moderated indirect associations

After examining the direct and indirect associations, latent interaction models were estimated to determine whether the associations of the three internationalization-at-home pathways with cross-cultural self-efficacy varied across levels of perceived internationalization climate. The analysis further examined whether these conditional first-stage associations were reflected in the corresponding indirect pathways to global competence. The latent interaction terms were estimated using the LMS procedure implemented through the XWITH command in Mplus. The results are reported in Table 9, with model-comparison statistics, the direct associations of W, simple-slope estimates, and plotting data provided in S12 Table.

Table 9. Moderation and moderated indirect association results.

Panel A. Latent interaction effects
Hypothesis Path Coefficient/ conditional effect SE p Interpretation
H5a X1 × W → M 0.116 0.029 <0.001 Supported
H5b X2 × W → M 0.034 0.032 0.285 Not supported
H5c X3 × W → M 0.162 0.028 <0.001 Supported
Panel B. Conditional indirect effects and indices of moderated mediation
Hypothesis Path Low W effect

[BC 95% CI]
Mean W effect

[BC 95% CI]
High W effect

[BC 95% CI]
Index

[BC 95% CI]
Result
H6a X1 → M → Y −0.006

[−0.031, 0.021]
0.026

[0.007, 0.051]
0.057 [0.030, 0.094] 0.025

[0.012, 0.045]
Supported
H6b X2 → M → Y 0.075

[0.042, 0.120]
0.084

[0.050, 0.124]
0.093

[0.054, 0.140]
0.007

[−0.006, 0.024]
Not supported
H6c X3 → M → Y 0.009

[−0.016, 0.038]
0.054

[0.030, 0.086]
0.098

[0.060, 0.146]
0.036

[0.021, 0.058]
Supported

Note: B denotes an unstandardized coefficient. Low, mean, and high W correspond to one latent standard deviation below the mean, the latent mean, and one latent standard deviation above the mean, respectively. Panel A estimates were obtained using MLR. The conditional indirect associations and indices of moderated mediation in Panel B were estimated using ML with 5,000 bootstrap replications and 95% bias-corrected bootstrap confidence intervals. H6a–H6c were evaluated from the bias-corrected confidence interval of the index of moderated mediation rather than from the significance of individual conditional indirect associations. All models controlled for gender, education level, foreign-language proficiency, parental education, six major-field dummy variables, three overseas-experience dummy variables, and the other two internationalization-at-home pathways. W was included as a predictor of both M and Y.

Because conventional global fit indices such as CFI, TLI, RMSEA, and SRMR are not available for LMS models containing latent interactions, the interaction models were evaluated relative to a common no-interaction reference model using log-likelihood values, AIC, BIC, and robust likelihood-ratio tests. The reference model contained the same measurement model, structural paths, and control variables as the interaction models but excluded the latent interaction term. The no-interaction reference model had a log-likelihood of −39852.833, an AIC of 79999.666, and a BIC of 80682.303. Relative to this model, adding X1 × W significantly improved model fit, robust LR χ²(1) = 16.775, p < 0.001, as did adding X3 × W, robust LR χ²(1) = 36.705, p < 0.001. In contrast, adding X2 × W did not significantly improve model fit, robust LR χ²(1) = 1.150, p = 0.284. The complete relative-fit comparisons are reported in Panel B of S12 Table.

As shown in Panel A of Table 9, the X1 × W interaction was positive and statistically significant, B = 0.116, SE = 0.029, p < 0.001, supporting H5a. The X3 × W interaction was also positive and significant, B = 0.162, SE = 0.028, p < 0.001, supporting H5c. By contrast, the X2 × W interaction was not significant, B = 0.034, SE = 0.032, p = 0.285; therefore, H5b was not supported. These findings indicate that higher perceived internationalization climate was associated with stronger relationships of internationalized curricular and learning-resource exposure and digital international learning participation with cross-cultural self-efficacy. The relationship between cross-cultural interaction engagement and cross-cultural self-efficacy did not vary significantly across levels of perceived internationalization climate.

W was entered as a predictor of both M and Y in each interaction model. Its association with M was positive and statistically significant in the X1 interaction model, B = 0.133, p = 0.003; the X2 interaction model, B = 0.116, p = 0.011; and the X3 interaction model, B = 0.144, p = 0.001. By contrast, its direct association with Y was not statistically significant in the X1 model, B = −0.067, p = 0.062; the X2 model, B = −0.066, p = 0.065; or the X3 model, B = −0.066, p = 0.067. These estimates are reported in Panel C of S12 Table.

The two significant interactions were probed at one latent standard deviation below the mean, at the latent mean, and at one latent standard deviation above the mean of W. For X1, the simple slope was small and statistically non-significant at low W, B = −0.025, p = 0.662. It was positive and significant at mean W, B = 0.117, p = 0.005, and became stronger at high W, B = 0.259, p < 0.001. The small negative estimate at low W should not be interpreted as evidence that curricular and learning-resource exposure reduced cross-cultural self-efficacy because the slope was not statistically distinguishable from zero (Fig 3).

Fig 3. Simple-slope plots for the significant latent interactions.

Fig 3

Panel (a) presents X1 × W → M at one latent standard deviation below and above the mean of W. Panel (b) presents X3 × W → M using the same probing values. Predicted values are centered partial predictions, with the other predictors and background controls held constant.

A similar pattern was observed for X3. The simple slope was positive but non-significant at low W, B = 0.043, p = 0.458. It was positive and significant at mean W, B = 0.242, p < 0.001, and was stronger at high W, B = 0.441, p < 0.001. Thus, the associations of X1 and X3 with cross-cultural self-efficacy became progressively stronger as perceived internationalization climate increased. The complete simple-slope estimates and plotting coordinates are reported in Panels D–F of S12 Table.

The moderated indirect association hypotheses were evaluated using the indices of moderated mediation. For X1, the conditional indirect association was not significant at low W, B = −0.006, BC 95% CI [−0.031, 0.021], but was significant at mean W, B = 0.026, BC 95% CI [0.007, 0.051], and high W, B = 0.057, BC 95% CI [0.030, 0.094]. More importantly, the index of moderated mediation was significant, index = 0.025, BC 95% CI [0.012, 0.045], supporting H6a.

For X2, the conditional indirect associations were significant at low, mean, and high levels of W. However, the index of moderated mediation was not significant, index = 0.007, BC 95% CI [−0.006, 0.024]. The significance of the three conditional indirect associations does not demonstrate moderated mediation because the relevant test concerns whether the indirect association changes significantly as a function of W. Accordingly, H6b was not supported.

For X3, the conditional indirect association was not significant at low W, B = 0.009, BC 95% CI [−0.016, 0.038], but was significant at mean W, B = 0.054, BC 95% CI [0.030, 0.086], and high W, B = 0.098, BC 95% CI [0.060, 0.146]. The index of moderated mediation was positive and significant, index = 0.036, BC 95% CI [0.021, 0.058], supporting H6c.

Taken together, perceived internationalization climate was associated with stronger first-stage and indirect associations for the curricular/resource-based and digital pathways, but not for the cross-cultural interaction pathway. It should therefore not be interpreted as a universal moderator of all internationalization-at-home experiences. One possible explanation is that curricular resources and digital opportunities depend more directly on whether students perceive an institutionally supportive environment, whereas the relevance of cross-cultural interaction may depend more on the quality, equality, continuity, and task structure of the interaction itself. This interpretation is consistent with research on school climate and learning environments [31–34], although these specific mechanisms were not directly measured and should therefore be treated as tentative. Because the data are cross-sectional and self-reported, the conditional associations should not be interpreted as evidence of causal effects.

4.6. Multi-group SEM by overseas experience

To examine whether prior cross-border exposure constituted a boundary condition, multi-group SEM was conducted for 354 respondents without overseas experience and 414 respondents with overseas experience. Overseas experience was used solely as the grouping variable. The multi-group structural models did not include background covariates; therefore, the omnibus and path-specific Wald tests evaluated unadjusted group differences in the seven structural coefficients. The grouping was intended to examine structural differences by broad experiential background rather than differences among specific types of overseas experience.

Measurement invariance was examined sequentially, and the results are reported in Panel C of S13 Table. The configural model showed acceptable fit, χ²(726) = 846.497, CFI = 0.989, TLI = 0.987, RMSEA = 0.021, and SRMR = 0.035. Constraining the first-order factor loadings to equality produced minimal changes in fit, χ²(747) = 870.922, CFI = 0.988, TLI = 0.987, RMSEA = 0.021, and SRMR = 0.038. Equality constraints were then imposed on the second-order factor loadings, and the higher-order metric model also showed acceptable fit, χ²(750) = 874.744, CFI = 0.988, TLI = 0.987, RMSEA = 0.021, and SRMR = 0.039. Finally, the scalar invariance model fitted the data well, χ²(774) = 894.109, CFI = 0.989, TLI = 0.988, RMSEA = 0.020, and SRMR = 0.040. Changes in CFI, RMSEA, and SRMR were within the recommended thresholds, supporting full scalar invariance. A partial scalar model was therefore not required [46,47].

After measurement invariance had been established, the seven core structural paths were freely estimated across the two overseas-experience groups. The multi-group structural models were estimated without background covariates. Overseas experience was used solely to define the two groups, comprising 354 respondents without overseas experience and 414 with overseas experience. Group differences were evaluated first using an omnibus Wald test of the joint equality of all seven unstandardized coefficients and then using seven one-degree-of-freedom Wald tests for the individual paths. These tests therefore concern unadjusted group differences in the seven structural coefficients. Unstandardized coefficients were used for formal group-difference testing, whereas standardized coefficients were reported to facilitate interpretation within each group. As shown in Table 10, the omnibus Wald test was significant, Wald χ²(7) = 14.488, p = 0.043, indicating that the seven unstandardized structural coefficients differed jointly across the two groups.

Table 10. Overall group difference test results for multi-group SEM by overseas experience.

Test Wald χ² df p Conclusion
Joint equality of the seven structural paths 14.488 7 0.043 Significant overall group difference

Note: The multi-group models did not include background covariates. Overseas experience was used only as the grouping variable. The null hypothesis was that the seven unstandardized structural coefficients were jointly equal between students with and without overseas experience. The seven paths were X1 → M, X2 → M, X3 → M, M → Y, X1 → Y, X2 → Y, and X3 → Y. The significant omnibus Wald test indicates that the coefficients differed jointly across groups. Path-specific one-degree-of-freedom Wald tests are reported in Table 11.

The freely estimated coefficients and path-specific Wald tests are reported in Table 11. Only the X3 → M path differed significantly between groups. The association between digital international learning participation and cross-cultural self-efficacy was stronger among students with overseas experience (B = 0.389, β = 0.379) than among those without overseas experience (B = 0.170, β = 0.165), Wald χ²(1) = 8.387, p = 0.004. None of the other six path-specific differences reached statistical significance. Accordingly, the significant omnibus test indicates an overall difference in the structural system, but the follow-up results locate this difference primarily in the X3 → M pathway.

Table 11. Multi-group SEM results by overseas experience.

Path No overseas B No overseas β Overseas B Overseas β Wald χ²(1) p
X1 → M 0.15 0.148 0.168 0.158 0.053 0.817
X2 → M 0.387 0.396 0.382 0.368 0.004 0.951
X3 → M 0.17 0.165 0.389 0.379 8.387 0.004
M → Y 0.273 0.325 0.135 0.17 3.067 0.08
X1 → Y 0.131 0.154 0.126 0.149 0.007 0.931
X2 → Y 0.101 0.123 0.19 0.23 1.544 0.214
X3 → Y 0.225 0.26 0.241 0.295 0.061 0.805

Note: B denotes unstandardized coefficients and β denotes within-group standardized coefficients. The multi-group models did not include background covariates. Overseas experience was used only as the grouping variable. Path-specific Wald tests evaluated equality of the unstandardized coefficients between groups.

H7 was therefore supported at the overall structural level, with the most clearly identifiable source of heterogeneity occurring in the X3 → M pathway.

From an interpretive perspective, this overall heterogeneity may mean that prior cross-border exposure is related to how students understand and use on-campus internationalized experiences. It does not mean that internationalization at home serves only already advantaged students. Rather, the result suggests that opportunity access, experiential participation, and outcome translation do not automatically coincide, and that prior experience may shape the way on-campus experiences are connected with global competence. This interpretation is consistent with research on inclusive internationalization and Chinese students’ global competence, which indicates that equal access to opportunities does not necessarily imply identical pathways of benefit [20,21].

4.7. Robustness checks

The first robustness analysis compared three unadjusted alternative structural models. These models used the same sample and measurement specification but differed in the restrictions imposed on the direct and mediating paths. The results are reported in Table 2. The results are reported in Table 2. The partial mediation model showed the best overall fit and information criteria: χ² = 386.588, df = 363, CFI = 0.998, RMSEA = 0.009, AIC = 68041.032, and BIC = 68510.055. The no mediation and full mediation models had slightly weaker fit and higher information criteria. Although the difference between the no mediation and partial mediation models was not extremely large, the overall fit and information criteria support the partial mediation model as the more plausible specification. This indicates that cross-cultural self-efficacy is an important but not exclusive pathway connecting the three internationalization-at-home experiences with global competence.

Second, the stability of core path estimates was compared with and without control variables. The results are reported in Table 4. For all core paths, including X1 → M, X2 → M, X3 → M, X1 → Y, X2 → Y, X3 → Y, and M → Y, the direction and magnitude of the coefficients remained highly stable after controls were included. The absolute changes were close to zero. This suggests that the main structural associations were robust to adjustment for the specified background covariates.

4.8. Summary of findings

Overall, the results show that the three internationalization-at-home pathways are all positively associated with global competence, but the associations are not homogeneous. Digital international learning participation shows the strongest direct association, cross-cultural self-efficacy partially explains the relationships between all three pathways and global competence, and perceived internationalization climate mainly amplifies the curricular/resource-based and digital pathways. Students with and without overseas experience differed at the overall structural level. The clearest path-specific difference occurred in the association between digital international learning participation and cross-cultural self-efficacy, which was stronger among students with overseas experience; the remaining six structural paths did not differ significantly. These results provide further evidence for the relationship between internationalization at home and global competence, while specifying the psychological translation and contextual conditions under which different pathways are associated with students’ self-reported global competence.

5. Discussion and conclusions

Rather than merely confirming a positive relationship between internationalization at home and global competence, the more explanatory finding of this study is that different pathways exhibit different association profiles. Previous research has already reported relatively stable positive associations between internationalization at home and global competence [10,11], and more recent comparative studies have begun to note differences across internationalization modes [12,13]. By incorporating curricular, interactional, and digital pathways into one model, this study shows that even when all three are significantly associated with global competence, they differ in direct association, psychological translation, and contextual dependency. This suggests that treating internationalization at home as a single aggregate input may conceal its internal functional structure.

The relatively strong direct association of digital international learning participation is consistent with recent research on virtual exchange and COIL, which highlights the role of digital pathways in expanding access to internationalized learning [14,15]. Reviews also suggest that the advantages of digital pathways often lie in scalability, lower participation thresholds, and organizational flexibility across time and space [48,49]. However, the present results do not support the strong claim that digital international learning is generally superior to other pathways. A more defensible interpretation is that the digital pathway shows a stronger direct association with global competence perceptions, while it does not replace the distinctive role of interaction in psychological internalization.

Cross-cultural interaction engagement was especially prominent in its association with cross-cultural self-efficacy and in its indirect effect. This clarifies the educational meaning of the interaction pathway. Self-efficacy theory emphasizes that stable competence judgments are formed through authentic tasks, feedback, and continuous action [23,24]. Research on intercultural relationships and cross-cultural group work in universities also repeatedly shows that the quality of interaction often matters more than the frequency of contact [39,40]. Accordingly, the value of the interaction pathway lies not only in providing more contact opportunities, but also in making intercultural experiences more likely to be translated into the belief that one can handle such tasks.

The role of perceived internationalization climate was asymmetric. It significantly amplified the curricular/resource-based pathway and the digital pathway, but did not significantly amplify the interaction pathway. This indicates that institutional context matters, but it may function more as an institutional amplifier for curricular provision and digital opportunities than as a universal enhancer of all pathways. School climate and learning environment research has consistently shown that perceived support, institutional provision, and resource platforms affect how students use learning experiences [31,32], and the same logic applies in internationalized education [33,34]. Yet the present findings also suggest that whether interaction translates into self-efficacy still depends on the relational quality, continuity, and task structure of the interaction itself. Internationalization at home should therefore avoid superficial accumulation of activities and should not assume that a generally positive climate can automatically substitute for high-quality interaction design [41,50].

The group heterogeneity results refine the interpretation of overseas experience as an experiential boundary. The significant omnibus Wald test indicated an overall difference in the structural system, and the follow-up tests identified the X3 → M pathway as the principal source of this heterogeneity. Digital international learning participation was more strongly associated with cross-cultural self-efficacy among students with overseas experience. One possible interpretation is that prior cross-border exposure provides experiential reference points that help students translate digitally mediated international learning into confidence in handling intercultural tasks. This interpretation remains associational rather than causal. Moreover, the other six path comparisons were not significant, so the finding should not be generalized into a claim that all internationalization-at-home pathways operate differently across the two groups. The result is consistent with research suggesting that opportunity access and the translation of experience into competence may vary with students’ prior experiential and cultural resources [20,21].

5.1. Theoretical implications

The main theoretical significance of this study is not simply that it confirms a positive relationship between internationalization at home and global competence, but that it shows why internationalization at home should not be conceptualized as a homogeneous input captured by a total exposure indicator. Existing studies often demonstrate the educational value of internationalization at home through overall associations [10,13]. Such treatment is useful, but it tends to understate the differentiated positions of different pathways in the competence formation chain. By placing curricular and learning resource exposure, cross-cultural interaction engagement, and digital international learning participation in one model, this study demonstrates that the three pathways differ in direct association, psychological translation, and contextual dependency. Internationalization at home is therefore better understood as a functionally differentiated configuration of pathways rather than as a single undifferentiated input.

A second theoretical contribution is the positioning of cross-cultural self-efficacy within the association between internationalization-at-home experiences and self-reported global competence. Existing research has noted relationships among international experience, self-efficacy, and intercultural competence [25,27], but has less often compared different internationalization-at-home pathways through the same psychological translation mechanism. The present results show that the interaction pathway is more prominent in self-efficacy formation and indirect association, while the digital pathway is more prominent in direct association. This indicates that the relevance of internationalization at home is not only derived from the accumulation of international elements, but also from whether experiences are internalized as confidence in handling intercultural tasks.

A third theoretical contribution is the simultaneous incorporation of perceived internationalization climate and overseas experience into the analytical framework. This extends the analysis from average associations to pathway-specific contextual conditions and experiential boundaries. Perceived internationalization climate was not uniformly associated with stronger relationships across all three pathways. Instead, it moderated the curricular/resource-based and digital pathways but did not significantly moderate the cross-cultural interaction pathway. The multi-group results further indicated a significant overall difference in the structural system between students with and without overseas experience. Follow-up path-specific tests localized this difference primarily to the X3 → M pathway: digital international learning participation was more strongly associated with cross-cultural self-efficacy among students with overseas experience. The other six structural paths did not differ significantly between the two groups. Overseas experience should therefore not be conceptualized as a universal boundary condition affecting all internationalization-at-home pathways in the same way. Rather, its relevance appears to be pathway-specific and may be particularly important for how students connect digitally mediated international learning with confidence in handling intercultural tasks. These findings suggest that research on internationalization at home should move beyond asking whether it is generally effective and should instead examine which pathways are associated with which outcomes, through which psychological processes, and under what contextual and experiential conditions.

5.2. Policy and practical implications

From a policy and practice perspective, the findings suggest that higher education institutions should not evaluate internationalization at home mainly by counting activities, project scale, or platform coverage. If internationalization at home is understood only as an issue of supply volume, curricular, interactional, and digital pathways may be treated as interchangeable arrangements. This would ignore their different functions in student development. The present study supports a functionally differentiated approach: the curricular/resource-based pathway primarily provides global issue input and comparative cognitive expansion; the interaction pathway is more closely connected with the translation of difference-based experience into confidence for action; and the digital pathway is particularly relevant for expanding access, reach, and frequency of internationalized learning opportunities.

Curriculum development should therefore go beyond adding international cases, foreign-language materials, or internationalized labels. Comparative problem awareness, multi-perspective analysis, and global issue discussion need to be embedded into teaching processes so that the curricular pathway can perform its cognitive input function. Similarly, interaction platforms should not focus only on activity frequency or participation numbers. They should emphasize shared tasks, sustained collaboration, and feedback mechanisms, because the translation of interactional experience into stronger cross-cultural self-efficacy depends heavily on the quality of interaction. Digital international learning should also be treated as more than platform provision or resource accumulation. Its value lies in how digital technologies reorganize access to international learning opportunities, participation in collaboration, and task completion. This means that course embedding, teacher organization, and assessment design are central to converting digital opportunities into student development outcomes.

The findings on perceived internationalization climate further suggest that institutional environments should not be regarded as a universal enhancer of all pathways. A supportive climate appears more relevant for strengthening curricular provision and digital opportunities than for replacing the quality of interaction itself. Institutions should therefore improve curricular systems, platform infrastructure, and resource accessibility while also designing more concrete interaction tasks, collaboration rules, and feedback structures. A generally positive internationalization climate is not enough if cross-cultural interaction remains superficial, short-term, or poorly structured.

The path-specific group difference also has implications for differentiated student support, particularly in digitally mediated international learning. Students without overseas experience may benefit from more explicit task orientation, intercultural scaffolding, guided reflection, and feedback when participating in virtual exchange or other forms of digital international learning. Students with prior overseas experience may be more able to connect digitally mediated tasks with existing experiential reference points, although this interpretation requires further investigation. Because the other six structural paths did not differ significantly between groups, the results do not justify designing entirely separate curricular or interactional systems for students with and without overseas experience. A more appropriate approach would combine broadly accessible internationalization-at-home opportunities with targeted support for students who have fewer prior opportunities to interpret and use digitally mediated intercultural experiences.

Overall, the practical significance of this study is not that higher education institutions should simply create more types of internationalized activities. Instead, the findings call for a rethinking of the organizational logic of internationalization at home. Rather than stacking curricular, interactional, and digital resources, institutions should configure the three pathways more clearly around cognitive input, psychological translation, and opportunity support. Only in this way can internationalization at home move from the surface-level introduction of international elements into the campus toward the sustained development of student capacities in everyday learning contexts.

6. Limitations and directions for future research

This study used cross-sectional survey data. It is therefore more suitable for identifying whether variables show theoretically consistent structural associations than for establishing strict causal identification. The temporal order among different pathways and their dynamic translation processes should be examined in future research using longitudinal tracking, multi-wave designs, or appropriate intervention studies. The terms pathway, mediation, and moderation in this study serve the interpretation of the theoretical and statistical model and should not be read as confirming strict temporal or causal transmission.

The focal variables were mainly collected through a single self-reported questionnaire at one time point. Although Section 3.4 combined procedural controls with supplementary statistical diagnostics to address potential common method bias, the present quantitative design identifies patterned associations among the focal constructs but cannot fully reveal how students interpret the quality, meaning, or turning points of curricular, interactional, and digital international learning experiences. Future research should therefore consider an explanatory sequential mixed-method design. Structural equation modeling could first identify pathway-specific, contextual, and subgroup patterns, followed by purposive interviews or focus groups with students representing contrasting pathway profiles and overseas-experience categories. The qualitative phase could then explain why apparently similar exposures generate different levels of cross-cultural self-efficacy and how students experience the institutional internationalization climate in practice. In particular, global competence in this study refers to students’ self-reported competence perceptions rather than externally assessed objective performance. Future studies that combine subjective perceptions with behavioral performance or third-party evaluation would provide stronger evidence on how different pathways relate to students’ actual competence development.

The sample was recruited through convenience-based WeChat forwarding and online dissemination. This procedure may have disproportionately reached students who were already interested in international learning, were connected to internationally oriented peer networks, or studied in institutions with stronger internationalization provision. The relatively high proportion of participants reporting some overseas experience (53.9%) and the medium-to-high construct means may therefore reflect upward sample selection. Such selectivity may influence not only estimated mean levels but also structural coefficients if the range or meaning of internationalization-at-home experiences differs between this sample and the wider population of Chinese higher education students. The findings should consequently be interpreted as within-sample structural associations rather than nationally representative population estimates. Future research could include more institutional types, disciplinary structures, regional contexts, and school internationalization ecologies to examine whether the differentiated pathway patterns vary across organizational environments.

At the construct level, this study supported the distinction among the three internationalization-at-home pathways through the measurement model and discriminant validity tests. Conceptually, however, internationalized curricular and learning resource exposure and digital international learning participation remain adjacent. The former emphasizes institutionalized curricular input, whereas the latter emphasizes participation modes organized through digital media. In practice, the two are not completely separate. Future research can use more refined item design, contextual experiments, or multi-level learning behavior data to distinguish more clearly between the internationalization of curricular input and the digitalization of internationalized participation.

The group heterogeneity findings should be interpreted cautiously. Because the multi-group models did not adjust for background covariates, differences in group composition may partly account for the observed structural differences. Although the omnibus Wald test indicated a significant overall difference in the structural system, only one of the seven follow-up path-specific tests was statistically significant. Specifically, the X3 → M association was stronger among students with overseas experience, whereas the other six structural paths did not differ significantly between groups. This result provides pathway-specific evidence of heterogeneity rather than evidence that the entire internationalization-at-home process operates differently across the two groups. In addition, overseas experience was represented by a broad binary grouping that combined family visits, short-term exchanges, and longer-term study or exchange experiences. These forms of cross-border exposure may differ substantially in purpose, intensity, duration, and opportunities for intercultural learning. The observed X3 → M difference should therefore be regarded as a finding requiring replication rather than as a definitive group distinction. Future research should distinguish different types, durations, and intensities of overseas experience and examine them together with family background, language capital, institutional context, and patterns of digital international learning. Such work could clarify when and why prior cross-border experience changes the association between digitally mediated international learning and cross-cultural self-efficacy.

In sum, the findings show that the research agenda on internationalization at home should not stop at identifying average positive associations. Future research should further examine which pathways are related to which outcomes, under what conditions, and through what psychological processes. Such work can help transform internationalization at home from a broad policy agenda into a more internally structured and analytically powerful research object.

Supporting information

S1 Table. Questionnaire constructs, final items, response scale, and dataset variable names.

(DOCX)

pone.0362486.s001.docx (19.2KB, docx)
S2 Table. Source literature, adaptation basis, and development history of the questionnaire.

(DOCX)

pone.0362486.s002.docx (14.4KB, docx)
S3 Table. Data cleaning and sample distribution.

(DOCX)

pone.0362486.s003.docx (26.9KB, docx)
S4 Table. Bivariate correlations among composite constructs.

(DOCX)

pone.0362486.s004.docx (13.4KB, docx)
S5 Table. Statistical diagnostics of potential common method bias.

(DOCX)

pone.0362486.s005.docx (18.6KB, docx)
S6 Table. Reliability statistics.

(DOCX)

pone.0362486.s006.docx (11.4KB, docx)
S7 Table. First-order CFA standardized factor loadings.

(DOCX)

pone.0362486.s007.docx (14KB, docx)
S8 Table. Correlation coefficients among latent variables (StdYX).

(DOCX)

pone.0362486.s008.docx (12.5KB, docx)
S9 Table. Discriminant-validity assessment using the Fornell–Larcker criterion and HTMT.

(DOCX)

pone.0362486.s009.docx (18.9KB, docx)
S10 Table. Detailed second-order CFA item loadings.

(DOCX)

pone.0362486.s010.docx (14KB, docx)
S11 Table. Standardized direct, indirect, and total effects across mediation paths.

(DOCX)

pone.0362486.s011.docx (11.5KB, docx)
S12 Table. Supplementary results of moderation analyses and simple-slope plotting data.

(DOCX)

pone.0362486.s012.docx (15.6KB, docx)
S13 Table. Supplementary measurement validation and measurement-invariance analyses.

(DOCX)

pone.0362486.s013.docx (15.8KB, docx)
S1 File. Questionnaire.

(DOCX)

pone.0362486.s014.docx (25.7KB, docx)
S2 File. Raw Data.

(XLSX)

pone.0362486.s015.xlsx (309.1KB, xlsx)
S3 File. Dataset.

(SAV)

pone.0362486.s016.sav (100.8KB, sav)
S4 File. Analysis code and Mplus input/output files.

(ZIP)

pone.0362486.s017.zip (1.5MB, zip)

Data Availability

The minimal data set is available at Zenodo via https://doi.org/10.5281/zenodo.19738164.

Funding Statement

This paper is a phased achievement of the 2025 Zhejiang Provincial Education Association for International Exchange Special Project on “Sino-Foreign Cooperative Education”, titled “Research on the Linkage Mechanism and Practical Path of Sino-Foreign Cooperative Education and Overseas Education from the Perspective of Two-way Drive” (Project No. ZRCCFC202504), and the 2026 Zhejiang Provincial Philosophy and Social Sciences Planning Special Project on “Research on Ideological and Political Work in Colleges and Universities”,titled “Research on the Mechanism and Path of ‘Comprehensive Ideological and Political Courses’ Construction in Sino-Foreign Cooperative Education Institutions” (Project No. 26GXSZ030YB). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Knight J. Internationalization remodeled: Definition, approaches, and rationales. Journal of Studies in International Education. 2004;8(1):5–31. doi: 10.1177/1028315303260832 [DOI] [Google Scholar]
  • 2.Hudzik JK. Comprehensive internationalization: Institutional pathways to success. London/New York: Routledge; 2015. doi: 10.4324/9781315771885 [DOI] [Google Scholar]
  • 3.Nilsson B. Internationalisation at home from a swedish perspective: The case of malmö. Journal of Studies in International Education. 2003;7(1):27–40. doi: 10.1177/1028315302250178 [DOI] [Google Scholar]
  • 4.Beelen J, Jones E. Redefining Internationalization at Home. The European Higher Education Area. Springer International Publishing; 2015. p. 59–72. doi: 10.1007/978-3-319-20877-0_5 [DOI] [Google Scholar]
  • 5.Leask B. Internationalizing the Curriculum. London: Routledge; 2015. doi: 10.4324/9781315716954 [DOI] [Google Scholar]
  • 6.Almeida J, Robson S, Morosini M, Baranzeli C. Understanding internationalization at home: Perspectives from the Global North and South. European Educational Research Journal. 2018;18(2):200–17. doi: 10.1177/1474904118807537 [DOI] [Google Scholar]
  • 7.OECD. Preparing our youth for an inclusive and sustainable world: The OECD PISA global competence framework. Paris: OECD Publishing; 2018. [Google Scholar]
  • 8.Boix Mansilla V, Jackson A. Educating for Global Competence: Preparing Our Youth to Engage the World. New York: Asia Society/CCSSO; 2011. doi: 10.13140/2.1.3845.1529 [DOI] [Google Scholar]
  • 9.Mansilla VB, Wilson D. What is global competence, and what might it look like in Chinese schools?. Jnl of Research in Internatl Education. 2020;19(1):3–22. doi: 10.1177/1475240920914089 [DOI] [Google Scholar]
  • 10.Soria KM, Troisi J. Internationalization at home alternatives to study abroad: Implications for students’ development of global, international, and intercultural competencies. Journal of Studies in International Education. 2014;18(3):261–80. doi: 10.1177/1028315313496572 [DOI] [Google Scholar]
  • 11.Meng Q, Zhu C, Cao C. An exploratory study of Chinese University undergraduates’ global competence: Effects of internationalisation at home and motivation. Higher Education Quarterly. 2017;71(2):159–81. doi: 10.1111/hequ.12119 [DOI] [Google Scholar]
  • 12.Sercu L. Internationalization at home as a factor affecting intercultural competence. A study among Belgian university students. European Journal of Higher Education. 2022;13(4):536–57. doi: 10.1080/21568235.2022.2094815 [DOI] [Google Scholar]
  • 13.Yang X, Li T. Impacts of different internationalization modes on the global competence of Chinese university students. ECNU Review of Education. 2024;7(4):945–70. doi: 10.1177/20965311241256363 [DOI] [Google Scholar]
  • 14.O’Dowd R. Virtual exchange: Moving forward into the next decade. Computer Assisted Language Learning. 2021;34(3):209–24. doi: 10.1080/09588221.2021.1902201 [DOI] [Google Scholar]
  • 15.Shen Z, de la Garza A. How to develop global competence in Chinese learners? Collaborative online international learning versus research-based self-study. Asia-Pacific Edu Res. 2025;35(2):425–34. doi: 10.1007/s40299-025-01038-1 [DOI] [Google Scholar]
  • 16.Soulé MV, Parmaxi A, Nicolaou A. Internationalization at home in higher education: A systematic review of teaching and learning practices. Journal of Applied Research in Higher Education. 2024;17(7):29–60. doi: 10.1108/jarhe-10-2023-0484 [DOI] [Google Scholar]
  • 17.Mittelmeier J, Lomer S, Al-Furqani S, Huang D. Developing meaningful internationalisation that impacts students’ outcomes in higher education: A scoping review of the literature 2011–2022. Journal of Studies in International Education. 2024;28(4):503–25. doi: 10.1177/10283153231222278 [DOI] [Google Scholar]
  • 18.Agnew M, Kahn HE. Internationalization-at-home: Grounded practices to promote intercultural, international, and global learning. Metropolitan Universities. 2014;25(3):31–46. [Google Scholar]
  • 19.Robson S, Almeida J, Schartner A. Internationalization at home: Time for review and development?. European Journal of Higher Education. 2017;8(1):19–35. doi: 10.1080/21568235.2017.1376697 [DOI] [Google Scholar]
  • 20.Van Mol C, Perez-Encinas A. Inclusive internationalisation: Do different (social) groups of students need different internationalisation activities?. Studies in Higher Education. 2022;47(12):2523–38. doi: 10.1080/03075079.2022.2083102 [DOI] [Google Scholar]
  • 21.Li X, Xiao J, Gao H, Zheng W, Zhao X. “It is not a top priority”: Exploring how Chinese university students enact their agency in developing global competence. International Journal of Educational Research. 2025;133:102667. doi: 10.1016/j.ijer.2025.102667 [DOI] [Google Scholar]
  • 22.Liu Y, Yin Y, Wu R. Measuring graduate students’ global competence: Instrument development and an empirical study with a Chinese sample. Studies in Educational Evaluation. 2020;67:100915. doi: 10.1016/j.stueduc.2020.100915 [DOI] [Google Scholar]
  • 23.Bandura A. Social Foundations of Thought and Action: A Social Cognitive Theory. Englewood Cliffs: Prentice-Hall; 1986. [Google Scholar]
  • 24.Bandura A. Self-efficacy: The exercise of control. New York: W H Freeman; 1997. doi: 10.1037/10522-094 [DOI] [Google Scholar]
  • 25.Deardorff DK. Identification and assessment of intercultural competence as a student outcome of internationalization. Journal of Studies in International Education. 2006;10(3):241–66. doi: 10.1177/1028315306287002 [DOI] [Google Scholar]
  • 26.Petersdotter L, Niehoff E, Freund PA. International experience makes a difference: Effects of studying abroad on students’ self-efficacy. Personality and Individual Differences. 2017;107:174–8. doi: 10.1016/j.paid.2016.11.040 [DOI] [Google Scholar]
  • 27.Huang Q, Cheung ACK, Xu J, Jiang Y. Intercultural learning and adolescents’ global competence: The mediating effect of self-efficacy. International Journal of Intercultural Relations. 2025;105:102143. doi: 10.1016/j.ijintrel.2025.102143 [DOI] [Google Scholar]
  • 28.Galan-Lominchar M, Roque IM-S, Del Campo Cazallas C, Mcalpin R, Fernández-Ayuso D, Zerolo BE. Internationalization at home program significantly increases the self-efficacy of nursing students: A pre-post study. Nurse Educ Today. 2024;143:106361. doi: 10.1016/j.nedt.2024.106361 [DOI] [PubMed] [Google Scholar]
  • 29.Heymans Y, Strosnider C, Pool J, Jansen van Vuuren M. Fostering intercultural competence through virtual exchange: Perspectives of undergraduate health students. Open Praxis. 2024;16(2):119–29. doi: 10.55982/openpraxis.16.2.607 [DOI] [Google Scholar]
  • 30.Bronfenbrenner U. The Ecology of Human Development: Experiments by Nature and Design. Cambridge, MA: Harvard University Press; 1979. doi: 10.4159/9780674028845 [DOI] [Google Scholar]
  • 31.Thapa A, Cohen J, Guffey S, Higgins-D’Alessandro A. A review of school climate research. Review of Educational Research. 2013;83(3):357–85. doi: 10.3102/0034654313483907 [DOI] [Google Scholar]
  • 32.Wang M-T, Degol JL. School climate: A review of the construct, measurement, and impact on student outcomes. Educ Psychol Rev. 2015;28(2):315–52. doi: 10.1007/s10648-015-9319-1 [DOI] [Google Scholar]
  • 33.Xu X, Schönrock-Adema J, Jaarsma ADC, Duvivier RJ, Bos NA. A conducive learning environment in international higher education: A systematic review of research on students’ perspectives. Educational Research Review. 2022;37:100474. doi: 10.1016/j.edurev.2022.100474 [DOI] [Google Scholar]
  • 34.Dauber D, Spencer-Oatey H. Global communication skills: Contextual factors fostering their development at internationalised higher education institutions. Studies in Higher Education. 2023;48(7):1082–96. doi: 10.1080/03075079.2023.2182874 [DOI] [Google Scholar]
  • 35.Wolf EJ, Harrington KM, Clark SL, Miller MW. Sample size requirements for structural equation models: An evaluation of power, bias, and solution propriety. Educ Psychol Meas. 2013;76(6):913–34. doi: 10.1177/0013164413495237 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Brislin RW. Back-Translation for Cross-Cultural Research. Journal of Cross-Cultural Psychology. 1970;1(3):185–216. doi: 10.1177/135910457000100301 [DOI] [Google Scholar]
  • 37.Anderson JC, Gerbing DW. Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin. 1988;103(3):411–23. doi: 10.1037/0033-2909.103.3.411 [DOI] [Google Scholar]
  • 38.Podsakoff PM, MacKenzie SB, Lee J-Y, Podsakoff NP. Common method biases in behavioral research: A critical review of the literature and recommended remedies. J Appl Psychol. 2003;88(5):879–903. doi: 10.1037/0021-9010.88.5.879 [DOI] [PubMed] [Google Scholar]
  • 39.Kudo K, Volet S, Whitsed C. Intercultural relationship development at university: A systematic literature review from an ecological and person-in-context perspective. Educational Research Review. 2017;20:99–116. doi: 10.1016/j.edurev.2017.01.001 [DOI] [Google Scholar]
  • 40.Montgomery C. A decade of internationalisation: Has it influenced students’ views of cross-cultural group work at university?. Journal of Studies in International Education. 2009;13(2):256–70. doi: 10.1177/1028315308329790 [DOI] [Google Scholar]
  • 41.Simões AV, Sangiamchit C. Internationalization at home: Enhancing global competencies in the EFL classroom through international online collaboration. Education Sciences. 2023;13(3):264. doi: 10.3390/educsci13030264 [DOI] [Google Scholar]
  • 42.Fornell C, Larcker DF. Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research. 1981;18(1):39. doi: 10.2307/3151312 [DOI] [Google Scholar]
  • 43.Henseler J, Ringle CM, Sarstedt M. A new criterion for assessing discriminant validity in variance-based structural equation modeling. J of the Acad Mark Sci. 2014;43(1):115–35. doi: 10.1007/s11747-014-0403-8 [DOI] [Google Scholar]
  • 44.Marsh HW, Wen Z, Hau K-T. Structural equation models of latent interactions: Evaluation of alternative estimation strategies and indicator construction. Psychol Methods. 2004;9(3):275–300. doi: 10.1037/1082-989X.9.3.275 [DOI] [PubMed] [Google Scholar]
  • 45.Cheung GW, Lau RS. Testing mediation and suppression effects of latent variables: Bootstrapping with structural equation models. Organizational Research Methods. 2008;11(2):296–325. doi: 10.1177/1094428107300343 [DOI] [Google Scholar]
  • 46.Vandenberg RJ, Lance CE. A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods. 2000;3(1):4–70. doi: 10.1177/109442810031002 [DOI] [Google Scholar]
  • 47.Chen FF. Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal. 2007;14(3):464–504. doi: 10.1080/10705510701301834 [DOI] [Google Scholar]
  • 48.Chan CKY. A scoping review of empirical research on Collaborative Online International Learning (COIL) implementation. Educational Research Review. 2025;49:100749. doi: 10.1016/j.edurev.2025.100749 [DOI] [Google Scholar]
  • 49.Sierra-Huedo ML, Bruton L, Fernández C. Becoming global at home: An analysis of existing cases and a proposal for the future of internationalization at home. Journal of Education. 2022;204(1):241–54. doi: 10.1177/00220574221108053 [DOI] [Google Scholar]
  • 50.De Wit H. Internationalization in higher education, a critical review. sfuer. 2019;12(3):9–17. doi: 10.21810/sfuer.v12i3.1036 [DOI] [Google Scholar]

Decision Letter 0

Wang Zheng

31 Jul 2026

Dear Dr. Ji,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Sep 14 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

  • A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only  the individual author can complete the verification step; PLOS staff cannot  verify ORCID iDs on behalf of authors.

We look forward to receiving your revised manuscript.

Kind regards,

Wang Zheng

Guest Editor

PLOS One

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for stating the following financial disclosure:

[This paper is a phased achievement of the 2025 Zhejiang Provincial Education Association for International Exchange Special Project on "Sino-Foreign Cooperative Education", titled "Research on the Linkage Mechanism and Practical Path of Sino-Foreign Cooperative Education and Overseas Education from the Perspective of Two-way Drive" (Project No. ZRCCFC202504), and the 2026 Zhejiang Provincial Philosophy and Social Sciences Planning Special Project on "Research on Ideological and Political Work in Colleges and Universities",titled "Research on the Mechanism and Path of 'Comprehensive Ideological and Political Courses' Construction in Sino-Foreign Cooperative Education Institutions" (Project No. 26GXSZ030YB).].

Please state what role the funders took in the study.  If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."

If this statement is not correct you must amend it as needed.

Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf.

3. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager.

4. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Partly

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously? -->?>

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available??>

The PLOS Data policy

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #1: Yes

Reviewer #2: Yes

**********

Reviewer #1: 1. Analysis software, estimator, and missing-data handling are never reported

Nowhere in the Methods or Results does the manuscript state which software was used, which estimator was applied (ML, MLR, WLSMV?), or how missing data were handled.

2. Implausibly good model fit requires explanation and additional diagnostics

The first-order CFA fit (chi-square = 515.774, df = 491, chi-square/df = 1.050, CFI = 0.998, RMSEA = 0.008) and the structural model fit (chi-square/df = 1.075, RMSEA = 0.010) are extraordinarily close to perfect for a 34-item, nine-factor model with N = 768. Fit this clean is rare in applied survey research and, combined with the very uniform descriptive profile of the constructs (all means 4.6–5.1, all SDs 1.20–1.33, skewness and kurtosis in narrow bands), it invites scrutiny. I do not allege any impropriety, but the authors should (a) confirm whether any post-hoc modifications (correlated residuals, item deletion after seeing the data) were made and disclose them; (b) report the full item-development history, including any pilot testing or exploratory factor analysis on an independent or split sample; and (c) provide response-time distributions and the exact data-cleaning thresholds (what counted as "obviously abnormal response time"?) so readers can assess data quality.

3. Measurement instrument: validation evidence is insufficient

All six focal constructs were measured with items "developed through a literature-based borrowing and contextual adaptation approach" without back-translation and, apparently, without any independent validation study. The distinction between X1 (internationalized curricular/resource exposure) and X3 (digital international learning participation) is conceptually adjacent, as the authors themselves acknowledge in the Limitations. Fornell–Larcker results are relegated to supporting information and no HTMT ratios are reported, even though Henseler et al. (2015) is cited in the reference list.

4. Latent interaction and moderated mediation: method and key statistics are missing

Section 3.4 states that "moderation was estimated through latent interaction terms" following Marsh, Wen, and Hau (2004), but the specific approach (LMS/XWITH in Mplus, unconstrained product indicators, mean-centering strategy) is not identified, and no fit or pseudo-fit information is given for the interaction models. More importantly, for the moderated mediation hypotheses (H6a–H6c) the manuscript reports only conditional indirect effects at three levels of W; the index of moderated mediation with its bootstrap confidence interval should be reported for each pathway, since the difference between conditional indirect effects, not their individual significance, is the test of H6. As currently presented, the conclusion for H6b ("not supported") and H6a/H6c ("supported") cannot be verified. Please also clarify how W itself was entered in the Y equation (was the direct effect of W on M and on Y estimated?).

5. Figure 3 is internally inconsistent

In both panels of Fig 3, the legend is titled "Internationalization climate level" but the legend entries read "Low X1 (-1 SD) / High X1 (+1 SD)" (and "Low X3 / High X3" in panel b), while the x-axis is also labeled with X1 (or X3). As drawn, the figure appears to plot X-by-X rather than X-by-W. Presumably the legend lines should be labeled Low W / High W. Please correct the figure and verify the plotted values against S12 Table. Also note that in panel (a) the low-climate slope is negative, implying that under a weak climate, more curricular exposure is associated with slightly lower self-efficacy; if the corrected figure retains this pattern, it deserves explicit discussion rather than being described simply as an amplification effect.

6. Measurement invariance testing stops at metric invariance

For the multi-group analysis, only configural and metric invariance are reported (S13 Table). Although comparisons of unstandardized structural paths technically require metric invariance, reporting scalar invariance is standard practice and would strengthen the comparison; if scalar invariance fails, this should be disclosed and partial invariance explored. Additionally, the overall Wald test (p = 0.043) is marginal, and the authors appropriately refrain from single-path claims; however, the manuscript should state clearly whether any single-path Wald or chi-square-difference tests were conducted, and if so report them (even as null results), rather than presenting only descriptive coefficient comparisons in Table 9.

7. Common method bias controls are minimal

Harman's single-factor test is widely regarded as insensitive, as the authors acknowledge. Given that every focal variable comes from the same instrument at the same time point, at least one stronger statistical diagnostic should be added, such as a CFA-based unmeasured latent method factor model or a marker-variable analysis, with results reported alongside the Harman test. Otherwise, the possibility that a common evaluative response tendency inflates all structural paths, particularly the M-to-Y path linking two self-perception constructs, remains a serious threat to the substantive interpretation.

8. Sample representativeness and the overseas-experience variable

A convenience sample recruited via WeChat snowballing in which 53.9% of respondents report overseas experience is highly atypical of the Chinese undergraduate population and directly relevant to the paper's framing around the "non-mobile majority." Please (a) report the composition of the overseas-experience categories (how many are travel/family visits versus study exchanges) in the main text, since pooling brief tourism with long-term study into one "overseas experience" group is a coarse and potentially misleading dichotomy; (b) discuss explicitly how sample selectivity may bias both mean levels and structural paths; and (c) report the distribution of institutions/regions rather than only stating that 29 provincial-level regions are covered.

Terminology: "consistency motifs" (Section 3.3) should read "consistency motives." "S17 File. Mpuls files" should read "Mplus files."

Reviewer #2: Manuscript Review Report – Pathways of internationalization at home-cross-cultural Self efficacy-Global competence

1. Specific theory/theories underlying the manuscript versus the conceptual framework and hypothesis constructed by the authors were not clearly delineated or the specific theories were missed

2. More sound if qualitative analysis is integrated with SEM in the form of explanatory sequential mixed methods, since the constructs, content-wise, need further deep qualitative meaning.

3. Submission declarations were not included in the manuscript as in the system. So, consider it.

4. For the remaining manuscript issues, it is well done and suitable for publication in an indexed journal

**********

what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review?  For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

Reviewer #2: Yes:  Ayetenew Abie (PhD)

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

PLoS One. 2026 Nov 16;21(11):e0362486. doi: 10.1371/journal.pone.0362486.r002

Author response to Decision Letter 1


20 Aug 2026

RESPONSE TO REVIEWERS

Manuscript ID: PONE-D-26-20413

Differentiated pathways from internationalization at home to global competence among Chinese university students: The roles of cross-cultural self-efficacy and perceived internationalization climate

Dear Academic Editor and Reviewers,

I sincerely thank the Academic Editor and both reviewers for their careful and constructive evaluation. I have revised the manuscript and Supporting Information extensively, and a clean copy, a marked-up copy, and the requested analysis materials accompany this response.

I am especially grateful to Reviewer 1 for the detailed methodological review. It prompted substantial improvements in reporting transparency, measurement validation, latent-interaction analysis, invariance testing, sample description, and interpretive caution. Reviewer 2’s comments also helped clarify the theoretical architecture and strengthen the future-research agenda. Reviewer comments are reproduced below, followed by concise responses and the principal revision locations.

Response to the Academic Editor

Comment E1

Please ensure that your manuscript meets PLOS ONE’s style requirements, including those for file naming.

Response:

Thank you for this reminder. I reviewed all revised files against the PLOS ONE style and file-naming requirements. The clean and marked-up manuscripts, figures, Supporting Information, questionnaire, data files, and analysis archive have been organized consistently, and the references, captions, Supporting Information citations, and section formatting were checked.

Revised: Submission files and revised manuscript.

Comment E2

Please state what role the funders took in the study. If the funders had no role, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” Please include this amended Role of Funder statement in your cover letter.

Response:

Thank you for pointing this out. The requested statement has been added verbatim to the revised cover letter: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” The related funding information was also checked in the submission system.

Revised: Revised cover letter and online financial disclosure.

Comment E3

Please ensure that you have an ORCID iD and that it is validated in Editorial Manager.

Response:

The corresponding author’s ORCID iD has been validated in Editorial Manager.

Revised: Editorial Manager author record.

Comment E4

Please include your full ethics statement in the Methods section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

Response:

The Methods now identifies the Research Office of Zhejiang Wanli University as the reviewing body, gives the exemption determination and document number, explains the anonymous minimal-risk basis for the exemption, and states that electronic informed consent was required before questionnaire access. It also explains the limited technical metadata recorded and the removal of direct identifiers and detailed IP information from analysis and shared files.

Revised: Section 3.1.

Response to Reviewer 1

I am especially grateful for the reviewer’s rigorous and constructive assessment. The specificity of these comments enabled me to identify where the original manuscript lacked methodological transparency and where additional analyses were necessary. We believe the revised manuscript is substantially stronger as a direct result of this review.

Comment 1

Analysis software, estimator, and missing-data handling are never reported. Nowhere in the Methods or Results does the manuscript state which software was used, which estimator was applied (ML, MLR, WLSMV?), or how missing data were handled.

Response:

Section 3.5 now reports the complete analysis workflow. Data preparation, descriptive statistics, and reliability analyses were conducted in IBM SPSS Statistics; CFA, SEM, latent-interaction, and multi-group analyses were conducted in Mplus Version 9. MLR was used for the measurement, baseline SEM, moderation, common-method-factor, and invariance models, whereas ML with 5,000 bootstrap replications was used when nonparametric bootstrap confidence intervals were required. Because all substantive items were mandatory, the 768 retained cases had no item-level missing values; no imputation or FIML treatment was therefore needed.

Revised: Sections 3.1 and 3.5.

Comment 2

Implausibly good model fit requires explanation and additional diagnostics. The first-order CFA fit (chi-square = 515.774, df = 491, chi-square/df = 1.050, CFI = 0.998, RMSEA = 0.008) and the structural model fit (chi-square/df = 1.075, RMSEA = 0.010) are extraordinarily close to perfect for a 34-item, nine-factor model with N = 768. Fit this clean is rare in applied survey research and, combined with the very uniform descriptive profile of the constructs, it invites scrutiny. I do not allege any impropriety, but the authors should (a) confirm whether any post-hoc modifications (correlated residuals, item deletion after seeing the data) were made and disclose them; (b) report the full item-development history, including any pilot testing or exploratory factor analysis on an independent or split sample; and (c) provide response-time distributions and the exact data-cleaning thresholds so readers can assess data quality.

Response:

I am particularly grateful for this careful and appropriately cautious observation. It prompted the most extensive set of revisions to the manuscript. I agree that an unusually strong fit should not be presented without a transparent account of model specification, item development, screening decisions, and additional diagnostics. I now state that the nine-factor model was specified before confirmatory analysis: no items were deleted, no cross-loadings or residual covariances were added, and no modification-index-driven respecification was used.

The instrument history now explains that the questionnaire was developed directly in Chinese from prespecified constructs and qualitatively reviewed by five experts. It also explicitly discloses that no student cognitive interviews, small-sample pretest, independent pilot survey, formal Delphi process, quantitative content-validity index, or pre-survey EFA was conducted.

Following the reviewer’s suggestion, I conducted a post hoc split-sample EFA/CFA using seed 20260808 (n = 384 in each half). Both analyses supported the nine-factor structure; four alternative measurement models fitted substantially worse. Full fit statistics and the saved group assignment are reported in S13 Table and the analysis archive. I describe this as internal split-sample replication, not independent external validation.

The screening flow and timing distribution are also reported. Of 800 submissions, 32 failed the higher-education eligibility screen, leaving 768 cases. No response-time, straightlining, or other post hoc pattern threshold was applied. Completion time had a mean of 188.75 seconds (SD = 64.19), a median of 185, an interquartile range of 131–247, and a range of 82–301 seconds; these statistics are reported transparently rather than used retroactively for exclusion.

Revised: Sections 3.1, 3.2, and 4.2; S3 and S13 Tables.

Comment 3

Measurement instrument: validation evidence is insufficient. All six focal constructs were measured with items developed through a literature-based borrowing and contextual adaptation approach without back-translation and, apparently, without any independent validation study. The distinction between X1 and X3 is conceptually adjacent. Fornell–Larcker results are relegated to supporting information and no HTMT ratios are reported, even though Henseler et al. (2015) is cited.

Response:

The revised manuscript clarifies that the questionnaire was developed directly in Chinese for the Chinese higher-education context rather than translated item by item from a single English-language instrument; conventional back-translation was therefore not applicable. The construct-development process, five-expert review, and specific attention to X1–X3 overlap are now documented.

Complete HTMT matrices have been added. The X1–X3 HTMT was 0.249; the maximum was 0.484 among the six focal constructs and 0.749 in the full nine-factor model, all below 0.85. The X1–X3 merged model also fitted substantially worse than the theoretical model. Together with the split-sample results, these analyses strengthen internal discriminant-validity evidence while preserving the limitation that no independent external validation sample was available.

Revised: Sections 3.2 and 4.2; S9 and S13 Tables.

Comment 4

Latent interaction and moderated mediation: method and key statistics are missing. Section 3.4 states that moderation was estimated through latent interaction terms following Marsh, Wen, and Hau (2004), but the specific approach (LMS/XWITH in Mplus, unconstrained product indicators, mean-centering strategy) is not identified, and no fit or pseudo-fit information is given for the interaction models. More importantly, for H6a–H6c the manuscript reports only conditional indirect effects at three levels of W; the index of moderated mediation with its bootstrap confidence interval should be reported for each pathway. Please also clarify how W itself was entered in the Y equation.

Response:

Sections 3.5 and 4.5 now specify the LMS approach implemented with XWITH in Mplus, TYPE = RANDOM, and numerical integration. No manifest product indicators were created, the observed indicators were not mean-centered, and the three interaction models were estimated separately.

A common no-interaction model with the same measurement structure, main paths, W, and controls was added as the reference. Robust likelihood-ratio comparisons showed improvement for X1 × W and X3 × W, but not X2 × W; detailed likelihood, AIC, and BIC values are provided in S12 Table.

Table 7 now reports the 5,000-replication bias-corrected bootstrap index of moderated mediation for each pathway. The index was significant for X1 (0.025, 95% CI [0.012, 0.045]) and X3 (0.036, 95% CI [0.021, 0.058]) but not X2 (0.007, 95% CI [−0.006, 0.024]); H6a and H6c are therefore supported and H6b is not. W was included as a predictor in both the M and Y equations, and its direct associations are reported in S12 Table.

Revised: Sections 3.5 and 4.5; Table 7; S12 Table.

Comment 5

Figure 3 is internally inconsistent. The legend entries appear to identify Low X1/High X1 or Low X3/High X3 rather than Low W/High W. Please correct the figure and verify the plotted values against S12 Table. Also, the negative low-climate slope in panel (a) deserves explicit discussion.

Response:

Fig 3 now labels the lines as Low W (−1 SD) and High W (+1 SD), with X1 and X3 on the respective horizontal axes. The plotting coordinates were recalculated and checked against S12 Table.

The revised discussion reports that the low-W X1 slope was small, negative, and non-significant (B = −0.025, p = 0.662). It is not interpreted as evidence of a harmful effect; the defensible conclusion is that the X1 association was indistinguishable from zero at low W and became positive and stronger at mean and high W. The X3 simple slopes are also reported.

Revised: Section 4.5; Fig 3; S12 Table.

Comment 6

Measurement invariance testing stops at metric invariance. Reporting scalar invariance is standard practice and would strengthen the comparison; if scalar invariance fails, this should be disclosed and partial invariance explored. Additionally, the overall Wald test (p = 0.043) is marginal, and the manuscript should state whether any single-path Wald or chi-square-difference tests were conducted and report them.

Response:

The invariance sequence now includes configural, first-order metric, higher-order metric, and scalar models. Full scalar invariance was supported (χ²(774) = 894.109, CFI = 0.989, TLI = 0.988, RMSEA = 0.020, SRMR = 0.040), so partial scalar invariance was unnecessary.

In addition to the omnibus test, Wald χ²(7) = 14.488, p = 0.043, seven one-degree-of-freedom tests are now reported. Only X3 → M differed significantly: it was stronger among students with overseas experience (B = 0.389, β = 0.379) than among those without it (B = 0.170, β = 0.165), Wald χ²(1) = 8.387, p = 0.004. The discussion now describes this single pathway-specific difference rather than uniform structural heterogeneity.

Revised: Sections 4.6 and 6; Tables 8 and 9; S13 Table.

Comment 7

Common method bias controls are minimal. Harman’s single-factor test is widely regarded as insensitive. Given that every focal variable comes from the same instrument at the same time point, at least one stronger statistical diagnostic should be added, such as a CFA-based unmeasured latent method factor model or a marker-variable analysis.

Response:

I added a CFA-based equal-loading orthogonal common-method-factor sensitivity model. The common method loading was 0.004 (SE = 0.014, p = 0.805), and its Wald test was non-significant, χ²(1) = 0.061, p = 0.805; adding the factor did not materially change model fit. The manuscript also states that this diagnostic cannot eliminate all common method bias or overcome the limitations of same-source, cross-sectional self-report data.

Revised: Sections 3.4 and 4.1; S5 Table.

Comment 8

Sample representativeness and the overseas-experience variable. A convenience sample recruited via WeChat snowballing in which 53.9% of respondents report overseas experience is highly atypical of the Chinese undergraduate population. Please report the composition of the overseas-experience categories in the main text, discuss how sample selectivity may bias mean levels and structural paths, and report the distribution of institutions/regions rather than only stating that 29 provincial-level regions are covered.

Response:

The revised Results reports all four categories: no overseas experience, 354 (46.1%); tourism or family visits, 222 (28.9%); short-term exchange of three months or less, 115 (15.0%); and longer study or exchange, 77 (10.0%). Three dummy variables were used in controlled single-group models, with no overseas experience as the reference. The binary grouping was retained only for multi-group SEM to preserve group sizes and is not presented as a fine-grained comparison of mobility types.

Recruitment is now reported as 701 WeChat-forwarding entries, 38 direct-link entries, and 29 mobile submissions, with IP-derived submissions from 29 provincial-level regions. University affiliation was not collected, so an institutional distribution cannot be validly reported; this limitation is stated explicitly.

The limitations section now explains how convenience recruitment and the high overseas-experience proportion may produce upward selection in both mean levels and structural coefficients. The results are therefore framed as within-sample associations rather than nationally representative population estimates.

Revised: Sections 3.1, 4.1, and 6; S3 Table.

Comment on terminology

“Consistency motifs” should read “consistency motives.” “S17 File. Mpuls files” should read “Mplus files.”

Response:

Both errors have been corrected, and the manuscript and Supporting Information labels were checked for the same issues.

Revised: Revised manuscript and Supporting Information list.

Response to Reviewer 2

I sincerely thank Reviewer 2 for the positive assessment of the statistical analysis and overall presentation, as well as for the constructive recommendations concerning theory, qualitative inquiry, and submission declarations.

Comment 1

Specific theory/theories underlying the manuscript versus the conceptual framework and hypothesis constructed by the authors were not clearly delineated or the specific theories were missed.

Response:

The theoretical background now distinguishes explanatory theories from construct-defining frameworks. Internationalization at home defines the opportunity structure and three learning pathways, and the OECD framework defines the multidimensional outcome. Social cognitive theory explains the self-efficacy-mediated translation process; ecological systems theory and the school-climate perspective explain the conditional role of perceived internationalization climate. Overseas experience is treated as an experiential boundary condition. The hypothesis sections have been reorganized so that each hypothesis is linked to its corresponding mechanism.

Revised: Sections 2.1–2.5.

Comment 2

More sound if qualitative analysis is integrated with SEM in the form of explanatory sequential mixed methods, since the constructs, content-wise, need further deep qualitative meaning.

Response:

I agree that qualitative evidence could deepen interpretation of these learning experiences. However, the present study was designed and ethically reviewed as a cross-sectional quantitative survey, and no qualitative data were collected; a post hoc qualitative component would therefore lack appropriate sampling, consent, and analytic procedures.

Section 6 now proposes a concrete explanatory sequential design for future research: SEM would first identify pathway-specific and subgroup patterns, followed by purposive interviews or focus groups with contrasting pathway profiles and overseas-experience categories. Claims throughout the manuscript have also been limited to cross-sectional associations rather than causal or temporal processes.

Revised: Section 6; interpretive qualifications in Sections 3.5, 4, and 5.

Comment 3

Submission declarations were not included in the manuscript as in the system. So, consider it.

Response:

The required declarations have been completed and checked in the PLOS submission system. The full ethics and informed-consent statement is included in Section 3.1; the funding disclosure and Role of Funder statement appear in the revised cover letter and submission fields; and the competing-interests, data-availability, and author-information fields have been reviewed for consistency.

Revised: Section 3.1; revised cover letter; submission-system declarations.

Comment 4

For the remaining manuscript issues, it is well done and suitable for publication in an indexed journal.

Response:

I appreciate this encouraging assessment. No additional manuscript change was required for this comment.

Closing statement

I again thank the Academic Editor and both reviewers for the time and care devoted to the manuscript. Their comments materially improved its theoretical clarity, methodological transparency, statistical verifiability, and interpretive precision. I hope that the revised manuscript and Supporting Information now address all concerns, and I respectfully submit the revision for further consideration by PLOS ONE.

Attachment

Submitted filename: Response to Reviewers.docx

pone.0362486.s019.docx (27KB, docx)

Decision Letter 1

Wang Zheng

18 Oct 2026

Dear Dr. Zhang,

Thank you for submitting your manuscript to PLOS One. After careful consideration, we feel that it has merit but does not fully meet PLOS One’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Dec 02 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

  • A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS One offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only  the individual author can complete the verification step; PLOS staff cannot  verify ORCID iDs on behalf of authors.

We look forward to receiving your revised manuscript.

Kind regards,

Wang Zheng

Guest Editor

PLOS One

Journal Requirements:

If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions??>

Reviewer #1: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously? -->?>

Reviewer #1: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available??>

The PLOS Data policy

Reviewer #1: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #1: Yes

**********

Reviewer #1: Thank you for the substantial revision and the detailed responses. The additional estimation details, HTMT analyses, split-sample validation, moderated-mediation indices, expanded measurement-invariance testing, and corrected Figure 3 address many of my previous concerns. However, several discrepancies remain that prevent me from recommending acceptance at this stage.

1. Section 3.5 and the note to Table 6 specify percentile bootstrap confidence intervals. However, Sections 4.4–4.5, Table 7, and the response letter describe bias-corrected intervals. These procedures are not interchangeable. Please identify the procedure actually used for each analysis and reconcile the Methods, Results, table headings and notes, and response letter with the archived analysis syntax and output. For the latent-interaction models, please also identify the resampling implementation, the number of requested and successfully completed bootstrap draws, and how any nonconvergent draws were handled.

2. The main structural model in Table 5 is reported as χ²(714) = 758.158, whereas the partial-mediation model in Table 10 is reported as χ²(363) = 386.588. Different specifications could explain these results, but their relationship is insufficiently explicit. Please identify the indicators, covariates, estimator, and sample used for the alternative-model comparisons, and state clearly whether these models include the controls used in Table 5. Please likewise clarify whether the multi-group structural comparisons include covariates. Model comparisons should use a consistent set of observed variables and respondents, with any differences in specification clearly documented.

3. Please explicitly confirm whether the Research Office of Zhejiang Wanli University is institutionally authorized to issue the reported human-participant research exemption. Please also report the X1–X3 latent correlation alongside the HTMT evidence and briefly identify the content of item M3 when explaining its retention. Finally, ensure that the revised supplementary files and repository materials correspond to the corrected study description and the analyses reported in this revision.

**********

what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review?  For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

PLoS One. 2026 Nov 16;21(11):e0362486. doi: 10.1371/journal.pone.0362486.r004

Author response to Decision Letter 2


25 Oct 2026

Response to Reviewers

Manuscript PONE-D-26-20413R1

Differentiated pathways from internationalization at home to global competence among Chinese university students: The roles of cross-cultural self-efficacy and perceived internationalization climate

Dear Dr. Wang and Reviewer 1,

I thank the Academic Editor and Reviewer 1 for the careful follow-up assessment. The remaining comments helped me resolve reporting inconsistencies and clarify the model relationships and the institutional basis of the ethics exemption. I have revised the manuscript and respond below.

Response to the Academic Editor

The decision letter requests reference checking and disclosure of reference changes.

Response. I have corrected the ordering of References 44 and 45 to match their first appearance: Marsh, Wen, and Hau is Reference 44 for latent interaction analysis, and Cheung and Lau is Reference 45 for mediation bootstrapping. No references were added or removed in this revision.

Revised: Section 3.5 and References 44–45.

Response to Reviewer 1

Comment 1 Bootstrap confidence intervals and implementation

Section 3.5 and the note to Table 6 specify percentile bootstrap confidence intervals. However, Sections 4.4–4.5, Table 7, and the response letter describe bias-corrected intervals. These procedures are not interchangeable. Please identify the procedure actually used for each analysis and reconcile the Methods, Results, table headings and notes, and response letter with the archived analysis syntax and output. For the latent-interaction models, please also identify the resampling implementation, the number of requested and successfully completed bootstrap draws, and how any nonconvergent draws were handled.

Response. The ordinary mediation model used ML with 5,000 nonparametric bootstrap replications and 95% percentile confidence intervals, specified by CINTERVAL(BOOTSTRAP). The moderated-mediation models used ML with 5,000 replications and 95% bias-corrected confidence intervals, specified by CINTERVAL(BCBOOTSTRAP). These distinctions are now consistent across the Methods, Results, and Tables 6–7. Table 7 also distinguishes MLR moderation estimates from ML/bootstrap conditional indirect estimates.

For the three moderated-mediation analyses (08–10), resampling used the built-in Mplus bootstrap procedure with TYPE = RANDOM and ALGORITHM = INTEGRATION. Each archived output reports 5,000 requested and 5,000 completed bootstrap computations and normal termination, with no incomplete computations reported. Resampling and estimation were handled internally by Mplus rather than by an external resampling script. Section 3.5 now reports this implementation and the completion figures explicitly.

Revised: Section 3.5; Sections 4.4–4.5 and Tables 6–7. The corresponding syntax and outputs are in S17 File.

Comment 2 Model specifications and covariate adjustment

The main structural model in Table 5 is reported as χ²(714) = 758.158, whereas the partial-mediation model in Table 10 is reported as χ²(363) = 386.588. Different specifications could explain these results, but their relationship is insufficiently explicit. Please identify the indicators, covariates, estimator, and sample used for the alternative-model comparisons, and state clearly whether these models include the controls used in Table 5. Please likewise clarify whether the multi-group structural comparisons include covariates. Model comparisons should use a consistent set of observed variables and respondents, with any differences in specification clearly documented.

Response. I have made the relationship among these models explicit. The primary SEM in Table 5 included 13 observed background covariates in both the M and Y equations: gender, education level, foreign-language proficiency, parental education, six major-field dummy variables, and three overseas-experience dummy variables.

The three alternative models in Table 10 were estimated using MLR on the same 768 respondents and the same 29 substantive indicators of X1, X2, X3, M, and the four first-order dimensions of global competence. Global competence was specified as a second-order factor. All three models excluded the background covariates and W and its indicators. They differed only in their direct and mediating path restrictions: the partial-mediation model included both sets of paths; the no-mediation model omitted M → Y and fixed the M–Y residual covariance to zero; and the full-mediation model omitted the three direct X-to-Y paths.

The difference between the degrees of freedom in Tables 5 and 10 therefore reflects the covariate specification, not different respondents or substantive indicators. Fit and information-criterion comparisons are restricted to the three consistently specified unadjusted models in Table 10. Table 11 separately compares the unadjusted partial-mediation model with the covariate-adjusted primary SEM reported in Table 5.

The multi-group structural models did not include background covariates. Overseas experience was used only to define the groups of 354 respondents without overseas experience and 414 with overseas experience. The omnibus and path-specific Wald tests therefore evaluate unadjusted group differences in the seven unstandardized structural coefficients. This is stated in the methods, results, and notes to Tables 8 and 9. Section 6 also acknowledges that differences in background composition may partly account for the observed structural differences.

Revised: Section 3.5; Sections 4.6–4.7 and Tables 5, 8–11; Section 6.

Comment 3 Institutional authority and measurement clarification

Please explicitly confirm whether the Research Office of Zhejiang Wanli University is institutionally authorized to issue the reported human-participant research exemption. Please also report the X1–X3 latent correlation alongside the HTMT evidence and briefly identify the content of item M3 when explaining its retention. Finally, ensure that the revised supplementary files and repository materials correspond to the corrected study description and the analyses reported in this revision.

Response. Zhejiang Wanli University has provided a signed institutional confirmation dated October 19, 2026. After examining its existing responsibility and delegation records and the original study documentation, the University confirmed that the Research Office was already authorized to determine eligibility for ethics-review exemption for this category of research when it issued determination WL20250701 on July 6, 2025. It also confirmed that the original application, exemption determination, and submitted manuscript concerned the same study, and that the questionnaire, participant information and consent page, survey procedures, and April 2026 data collection fell within the original determination’s scope.

Section 3.1 now states this authority explicitly and distinguishes the confirmation of the original determination from a new ethics approval. Electronic informed consent was required; the ethics-review exemption did not waive informed consent. The original exemption documentation and the signed Chinese and English institutional confirmations accompany this revision for editorial verification.

Section 4.2 now reports the X1–X3 latent correlation alongside HTMT: r = 0.254, SE = 0.037, p < 0.001, and HTMT = 0.249. It also identifies M3 as measuring confidence in meeting communication, collaboration, and understanding demands in internationalized or cross-cultural learning tasks. M3 was retained to preserve this aspect of the prespecified self-efficacy construct; its loading was 0.629, and the construct had CR = 0.875 and AVE = 0.585.

I have checked the relevant supplementary tables and archived Mplus outputs against these clarified descriptions. S8 and S9 Tables contain the latent-correlation and HTMT results, and S17 File contains the syntax and outputs supporting the reported model specifications and estimates. The questionnaire and data files are supplied as S14–S16 Files. These clarifications did not require changes to the underlying data or re-estimation of the models.

Revised: Sections 3.1 and 4.2; S8–S9 Tables and S14–S17 Files; institutional confirmation documents.

I appreciate the opportunity to clarify these points. The reviewer’s further scrutiny has improved the precision and transparency of the report, and I hope that the revised manuscript and supporting documentation address the remaining concerns.

Sincerely,

Ji Zhang

Attachment

Submitted filename: Response_to_Reviewers_Round_2.docx

pone.0362486.s020.docx (40KB, docx)

Decision Letter 2

Wang Zheng

26 Oct 2026

<p>Differentiated pathways from internationalization at home to global competence among Chinese university students: The roles of cross-cultural self-efficacy and perceived internationalization climate

PONE-D-26-20413R2

Dear Dr. Zhang,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Wang Zheng

Guest Editor

PLOS One

Additional Editor Comments (optional):

Reviewers' comments:

Acceptance letter

Wang Zheng

PONE-D-26-20413R2

PLOS One

Dear Dr. Zhang,

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS One. Congratulations! Your manuscript is now being handed over to our production team.

At this stage, our production department will prepare your paper for publication. This includes ensuring the following:

* All references, tables, and figures are properly cited

* All relevant supporting information is included in the manuscript submission,

* There are no issues that prevent the paper from being properly typeset

You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps.

Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing.

If we can help with anything else, please email us at customercare@plos.org.

Thank you for submitting your work to PLOS One and supporting open access.

Kind regards,

PLOS One Editorial Office Staff

on behalf of

Dr. Wang Zheng

Guest Editor

PLOS One

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    S1 Table. Questionnaire constructs, final items, response scale, and dataset variable names.

    (DOCX)

    pone.0362486.s001.docx (19.2KB, docx)
    S2 Table. Source literature, adaptation basis, and development history of the questionnaire.

    (DOCX)

    pone.0362486.s002.docx (14.4KB, docx)
    S3 Table. Data cleaning and sample distribution.

    (DOCX)

    pone.0362486.s003.docx (26.9KB, docx)
    S4 Table. Bivariate correlations among composite constructs.

    (DOCX)

    pone.0362486.s004.docx (13.4KB, docx)
    S5 Table. Statistical diagnostics of potential common method bias.

    (DOCX)

    pone.0362486.s005.docx (18.6KB, docx)
    S6 Table. Reliability statistics.

    (DOCX)

    pone.0362486.s006.docx (11.4KB, docx)
    S7 Table. First-order CFA standardized factor loadings.

    (DOCX)

    pone.0362486.s007.docx (14KB, docx)
    S8 Table. Correlation coefficients among latent variables (StdYX).

    (DOCX)

    pone.0362486.s008.docx (12.5KB, docx)
    S9 Table. Discriminant-validity assessment using the Fornell–Larcker criterion and HTMT.

    (DOCX)

    pone.0362486.s009.docx (18.9KB, docx)
    S10 Table. Detailed second-order CFA item loadings.

    (DOCX)

    pone.0362486.s010.docx (14KB, docx)
    S11 Table. Standardized direct, indirect, and total effects across mediation paths.

    (DOCX)

    pone.0362486.s011.docx (11.5KB, docx)
    S12 Table. Supplementary results of moderation analyses and simple-slope plotting data.

    (DOCX)

    pone.0362486.s012.docx (15.6KB, docx)
    S13 Table. Supplementary measurement validation and measurement-invariance analyses.

    (DOCX)

    pone.0362486.s013.docx (15.8KB, docx)
    S1 File. Questionnaire.

    (DOCX)

    pone.0362486.s014.docx (25.7KB, docx)
    S2 File. Raw Data.

    (XLSX)

    pone.0362486.s015.xlsx (309.1KB, xlsx)
    S3 File. Dataset.

    (SAV)

    pone.0362486.s016.sav (100.8KB, sav)
    S4 File. Analysis code and Mplus input/output files.

    (ZIP)

    pone.0362486.s017.zip (1.5MB, zip)
    Attachment

    Submitted filename: Response to Reviewers.docx

    pone.0362486.s019.docx (27KB, docx)
    Attachment

    Submitted filename: Response_to_Reviewers_Round_2.docx

    pone.0362486.s020.docx (40KB, docx)

    Data Availability Statement

    The minimal data set is available at Zenodo via https://doi.org/10.5281/zenodo.19738164.


    Articles from PLOS One are provided here courtesy of PLOS

    RESOURCES