The E-cigarette KAP Questionnaire for young students was developed following established psychometric principles for instrument construction. Based on the KAP theoretical framework, an initial pool of items was generated through a systematic literature review and expert brainstorming sessions. Six experts specializing in public health, health promotion, sociology, and cross-cultural psychology participated in a two-round Delphi consultation. All experts held at least an associate professor rank and had more than 5 years of professional experience in their respective domains.

The item-level content validity index (I-CVI) ranged from 0.86 to 1.00, and the scale-level content validity index (S-CVI) reached 0.94, well above the recommended threshold of 0.90, indicating excellent content relevance and semantic clarity.

After developing the Chinese version, professional linguists translated the questionnaire into English. Subsequently, Indonesian linguistic experts produced a culturally adapted Bahasa Indonesia version.

To ensure cross-cultural equivalence and linguistic clarity, a pilot test was conducted in April 2023 with 20 young students (10 from China and 10 from Indonesia). Cognitive interviews assessed item clarity, cultural appropriateness, and logical coherence. Based on participant feedback, minor modifications were made to ambiguous or culturally sensitive expressions.

A preliminary reliability assessment of the knowledge domain showed a Cronbach’s α of 0.99 and an inter-item Kappa coefficient of 0.69, demonstrating strong internal consistency and initial measurement stability, providing a foundation for large-scale validation.

A cross-sectional online survey was conducted between May and June 2023. Sampling sites were selected based on comparable smoking prevalence and urban demographics. Kunming, China (adult smoking rate: 33.7%) (32), and Jakarta, Indonesia (adult smoking rate: 32.2%) (33), were chosen as representative regions.

Sample sizes were calculated using standard formulas for cross-sectional surveys, yielding 567 and 553 participants from Kunming and Jakarta, respectively.

A multistage stratified cluster sampling approach was employed. In each city, three universities were selected (one medical and two non-medical institutions). Within each institution, classes were randomly selected by grade level, and all students in the selected classes were invited to participate. Eligible participants were undergraduate students aged 18–25 years, currently enrolled at their universities, and willing to provide informed consent.

Ethical approval was obtained from the Ethics Committee of Kunming Medical University (Approval No. KMMU2022MEC027), and the study adhered to the principles of the Declaration of Helsinki (2013 revision).

Data collection was conducted through an encrypted online platform. All questionnaire items were set as mandatory-response fields to minimize missing data. In pilot testing, the average completion time in pilot testing was 10 min (range: 8–12 min); responses completed in less than 3 min or more than 30 min were excluded as invalid. Each class participated under supervision coordinated by class advisors, following a standardized protocol. Instructions were read aloud to ensure procedural consistency, respondent independence, and a uniform testing environment. Data integrity was verified through double data entry and cross-checking by two independent researchers.

Item selection was guided by both CTT and SEM. Reliability testing, exploratory factor analysis (EFA), CFA, and discrimination analysis was combined to identify and remove items with weak psychometric performance or ambiguous factor loadings. The final instrument was refined to ensure clear theoretical dimensionality and measurement precision.

Data were entered into Microsoft Excel 2019 and analyzed using SPSS 27.0 and AMOS 29.0 software. Categorical variables were reported as frequencies and percentages. Continuous variables conforming to a normal distribution were described as mean ± standard deviation (SD); non-normally distributed variables were reported as median (P25, P75). A two-tailed p-value < 0.05 was considered statistically significant.

A total of 1,400 questionnaires were distributed, and 1,333 valid responses were obtained after excluding incomplete and invalid submissions (631 from Kunming and 702 from Jakarta), yielding an effective response rate of 95.2%. Among the respondents, 594 (44.6%) were male and 739 (55.4%) were female; 47.3% were Chinese and 52.7% Indonesian. By grade level, 549 (41.8%) were freshmen, 582 (44.3%) sophomores, and 182 (13.7%) juniors or above. Regarding academic major, 412 (30.9%) were medical students, and 921 (69.1%) were non-medical.

With respect to religion, 731 (54.8%) reported religious affiliation and 602 (45.2%) did not. Urban residents accounted for 62.6% of the sample, while 37.4% were from rural areas. Monthly living expenses were <1,000 yuan for 31.4%, 1,000–1,999 yuan for 41.4%, and ≥2,000 yuan for 27.2%. Additionally, 8.7% had family members who used e-cigarettes, 29.3% had close friends who used them, and 74.3% had been exposed to secondhand smoke within the past week.

Based on literature review, expert consultation, and pilot testing, the initial scale consisted of 32 items (9 knowledge, 15 attitude, 8 behavior items), with the total score of 40 points (knowledge: 9; attitude: 15; behavior: 16). The mean total score was 18.76 ± 2.23, with subscale means of 5.74 ± 1.38 (knowledge), 12.30 ± 3.24 (attitude), and 6.89 ± 2.22 (behavior).

Cronbach’s α for the total scale was 0.605, while the α values for the knowledge, attitude, and behavior subscales were 0.702, 0.856, and 0.775, respectively, indicating acceptable to good internal consistency. The total scale reliability (α = 0.605) falls within the range deemed acceptable for initial research applications of a multidimensional tool.

The SEM framework was applied to evaluate the model’s fit and theoretical consistency. Three candidate models (M1–M3) were compared: Model M1, a basic three-factor model without cross-loading; Model M2, a three-factor correlated model with adjusted measurement errors; Model M3, a modified SEM model after iterative parameter optimization.

The final model (M3) exhibited the best fit: χ²/df = 1.350, RMSEA = 0.050 (90% CI: 0.022–0.072), SRMR = 0.0643, CFI = 0.968, TLI = 0.960 (see Table 2).

All fit indices met or exceeded the recommended thresholds (CFI, TLI > 0.90; RMSEA < 0.08), indicating excellent model fit and parsimony. Standardized factor loadings ranged from 0.52 to 0.84, confirming strong relationships between latent variables and observed indicators. The initial and final model structure is presented in Figures 1, 2, respectively.

Comprehensive evaluation of the model’s internal structure demonstrated that the knowledge-attitude correlation was r = 0.682 (p < 0.001); the attitude-practice correlation was r = 0.715 (p < 0.001); the knowledge-practice correlation was r = 0.639 (p < 0.001). These findings indicate moderate associations among the three dimensions, aligning with the theoretical logic of the KAP model, namely, that knowledge influences attitudes, which in turn drive behavioral practice.

The overall Cronbach’s α coefficient of the questionnaire was 0.643, with the subscale coefficients as follows: knowledge = 0.819, attitude = 0.847, and practice = 0.775. All subscale coefficients exceeded the recommended threshold of 0.70. Although the overall Cronbach’s α of the questionnaire was slightly below 0.7, it remains acceptable for early-stage psychometric instrument research. When combining the knowledge and attitude subscales, the Cronbach’s α coefficient was 0.889. In contrast, when knowledge, attitude, and practice were analyzed together, the overall Cronbach’s α of the questionnaire was 0.643. Split-half reliability further confirmed the internal consistency, indicating that each subscale reliably measures its intended construct.

Convergent validity was verified through factor loadings and composite reliability (CR). All standardized loadings exceeded 0.50, and Average Variance Extracted (AVE) values for each factor were above 0.50, satisfying Fornell–Larcker criteria for convergent validity. Discriminant validity analysis showed that the square root of AVE for each dimension was greater than its inter-factor correlations, indicating good construct separation and theoretical distinctness.

Following reliability and validity optimization, the final E-cigarette Knowledge, Attitude, and Practice Questionnaire consisted of 15 items distributed across three dimensions: (1) Knowledge—4 items, (2) Attitude—8 items, (3) Practice—3 items.

The model structure was both statistically stable and theoretically consistent with the KAP framework, providing a scientifically sound foundation for subsequent cross-cultural and behavioral studies.

This study constructed and psychometrically evaluated an e-cigarette KAP Questionnaire for young students in China and Indonesia using an SEM framework. The final 15-item scale demonstrates acceptable reliability and construct validity. The three-factor structure (knowledge, attitude, and practice) was empirically supported and the final model (M3) achieved excellent fit indices (χ²/df = 1.350, RMSEA = 0.050, CFI = 0.968, TLI = 0.960). Importantly, model refinement was limited to theoretically justified within-factor error covariances, with no cross-loadings or cross-factor error correlations introduced, thereby preserving the conceptual integrity of the KAP framework. These findings validate the theoretical proposition of the KAP model that cognitive understanding influences behavioral intentions and practices through attitudinal pathways (29). Moreover, the positive correlations observed among the three domains (r = 0.639–0.715) indicate that higher e-cigarette-related knowledge is associated with more rational attitudes and healthier behavioral tendencies, consistent with previous studies on smoking cognition and prevention among young students (40, 41).

The final Practice subscale was distilled into three core items through Classical Test Theory and Structural Equation Modeling: P1. Have you ever used e-cigarettes? P2. In the past 7 days, on how many days did you use e-cigarettes? P3. On average, how many times per day do you use e-cigarettes? These items capture the fundamental dimensions of use behavior: lifetime adoption (P1), recent engagement (P2), and use intensity (P3). This minimal set efficiently classifies respondents into public health-relevant categories, such as never-users, experimenters, occasional users, and regular users. However, this set cannot fully reflect all aspects of use behavior. Therefore, we retained item P8 (What is your primary reason for using e-cigarettes?) as an open-ended inquiry into motivations, which aids in the interpretability of the quantitative findings across different cultural contexts.

Psychometric analysis indicated that the combined knowledge and attitude subscales yielded a Cronbach’s α of 0.889. However, when the practice subscale was included, the overall reliability of the scale dropped to a moderate level (α = 0.643). The measurement of sensitive behaviors like e-cigarette use among youth necessitates a critical appraisal of potential biases and the sociocultural contexts that shape both behavior and reporting. First, social desirability bias likely influenced responses, particularly on the ‘Practice’ items. Despite our use of anonymous surveys to minimize this bias, participants may still have underreported use to conform to perceived social expectations. Second, cultural differences between China and Indonesia may underlie the divergence in e-cigarette use patterns among young students. Social desirability and policy environments may exert differential influences on self-reporting. In China, where tobacco control policies and school anti-smoking norms are widely implemented, respondents may underreport or deny e-cigarette use to align with societal expectations of being a “good student” or a “law-abiding citizen” (42, 43). In contrast, within the socioeconomic and cultural context of Indonesia, the tobacco industry wields substantial economic influence, adult smoking prevalence is high, and e-cigarettes lack clear regulation. Tobacco products are widely available through various retail channels, such as small stalls and street vendors, making them easily accessible to young students (44, 45). Therefore, future applications of this scale must account for the profound influence of local policy environments and socio-cultural norms on behavioral reporting. This insight is a principal finding of our validation study, highlighting that the instrument’s value lies not only in measuring constructs but also in revealing the contextual boundaries of those measurements.

The findings align with international evidence suggesting that knowledge and attitude are key determinants of youth e-cigarette behaviors (46). Prior research has shown that young students with insufficient awareness of nicotine dependence or e-cigarette toxicity are more likely to engage in experimental or routine vaping (47) whereas high knowledge and unfavorable attitudes toward e-cigarettes as protective factors against initiation (47, 48).

This study extends these findings by confirming that these relationships remain stable across culturally distinct populations, namely China and Indonesia. Notably, while earlier instruments have primarily emphasized behavioral dependence (e.g., Penn State Electronic Cigarette Dependence Index) (26, 49), our scale integrates cognitive, affective, and behavioral dimensions, enabling a more comprehensive assessment of youth perceptions. Furthermore, the SEM-based validation approach enhances methodological rigor beyond traditional CTT tools, strengthening interpretability and facilitating cross-cultural comparison.

A key strength of this study is the methodological integration of CTT and SEM. The complementary nature of the two psychometric paradigms ensured rigorous construct validation and optimized item selection, an approach consistent with contemporary measurement practices (30, 31).

Theoretically, the present study the results provide empirical support for the sequential mechanism posited by the KAP model, knowledge shapes attitudes, which in turn influence behavioral practices, clarifying how health education affects preventive behavior. From a public health perspective, this questionnaire provides a reliable and evidence-based tool for assessing youth cognition and behavior toward e-cigarettes. It can be used to: (1) evaluate the effectiveness of health education interventions, (2) identify high-risk groups with low awareness or misperceptions, and (3) inform school- and community-based tobacco control strategies.

Several limitations should be acknowledged. First, the participants were young students from two cities, which limits generalizability to younger young students or non-student groups. Future research should expand sampling to secondary school students and working young adults to capture broader behavioral diversity. Second, although SEM provides comprehensive structural verification, temporal stability (test–retest reliability) and predictive validity were not assessed; future longitudinal research is warranted. Third, this study did not formally assess cross-cultural equivalence. Although the instrument showed acceptable psychometric properties, measurement invariance across cultures has not been established. Cross-cultural adaptation and comparability will be addressed in a separate forthcoming study. Finally, as with most self-reported surveys, responses may be influenced by social desirability bias. Future work ought to implement this questionnaire in intervention and surveillance contexts to monitor cognitive and behavioral trends over a period of time. Additional validation of the instrument should be conducted across a greater variety of countries with larger sample sizes, in order to enhance its cross-cultural applicability and underpin evidence-based global efforts in e-cigarette control.