To examine the two proposed pathways, a cross-sectional study was conducted from January to July 2023 using a multistage cluster sampling method. In the first stage, four provinces spanning the southern, central, and northern China were selected: Sichuan, Henan, Shanxi and Beijing. In the second stage, one city was randomly selected from each division, specifically Luzhou, Zhengzhou, Taiyuan and Beijing. In the third stage, two primary schools were selected within each sampled city. In the final stage, one to two classes were randomly sampled from each of the three grade levels (Grades 1–2, 3–4, and 5–6) at each school. Data were collected using a structured health literacy questionnaire (described below), which was administered to all students. Ethical procedures followed a strict dual-consent process: written informed consent was obtained from parents or legal guardians, and assent was obtained from the students themselves prior to the survey. Inclusion criteria required students to be in the selected classes with valid consent, while exclusion criteria included students who were unwilling to participate or unable to complete the questionnaire independently due to severe cognitive impairments.

To account for variations in cognitive development and comprehension abilities across different age groups, we developed the Chiese Rapid Health Literacy Questionnaire (CRHLQ) comprising three sub-questionnaires: CRHLQ-12 for grades 1–2, CRHLQ-34 for grades 3–4, and CRHLQ-56 for grades 5–6. Although specific items varied to ensure developmental appropriateness and avoid floor/ceiling effects, all sub-questionnaires were strictly designed to measure the same four dimensions within HL framework: Health Knowledge, Health Motivation, Health Skills, and Health Participation. The sub-questionnaires share a uniform structure comprising two sections: (1) a general information section, which collects basic characteristics on height, weight, self-assessments of health status, academic performance, and character, and (2) a measurement section assessing four HL dimensions. Specifically, Health Knowledge and Health Skills measurements consist of objective judgment or choice questions scored based on correctness (correct = 1, incorrect = 0). Health Motivation and Health Participation measurements utilize Likert-type scales—a 3-point scale for Grades 1–4 and a 5-point scale for Grades 5–6—to capture attitudes and behavioral frequencies. Final scores were calculated by summing item responses, with negative items reverse-coded prior to aggregation. Missing data were handled using listwise deletion to ensure data quality. Detailed item wording, scoring protocols, and dimensional mapping for the CRHLQ are provided in Supplementary Tables 1–3.

The total score for each sub-questionnaires is 100. Dimension weights were assigned based on importance ratings provided by a Delphi expert panel, with scores allocated accordingly. The validity and reliability of the CRHLQ was evaluated using the Rasch model of Item Response Theory (IRT). As shown in Appendix Table 6, the item reliability coefficients for all sub-questionnaires were exceptional (ranging from 0.98 to 1.00), and the item separation indices ranged from 7.72 to 15.43, far exceeding the recommended threshold of 3.0. This indicates a highly stable item hierarchy. In terms of construct validity, the Rasch dimension explained 30.2% to 59.7% of the variance, supporting the assumption of unidimensionality (see Appendix Table 6). Furthermore, the model fit was well; the mean Infit and Outfit Mean Square (MNSQ) values for items were highly consistent with the expected value of 1.0 (ranging from 0.87 to 1.00), with ZSTD scores generally falling within the ± 2.0 acceptable range (see Appendix Table 5). Although the person separation indices (1.08–1.20) and reliability coefficients (0.54–0.59) were relatively modest, the exceptional item fit statistics and high item reliability provide robust evidence of the instrument’s construct validity and measurement stability.

Descriptive statistics and Pearson correlation analyses were conducted using IBM SPSS V.29 to explore the relationships among Health Knowledge, Health Motivation, Health Skills, and Health Participation. To rigorously address the potential clustering effects arising from the multistage sampling design (students nested within regions), we first calculated the Intraclass Correlation Coefficients (ICCs) for the outcome variable (Health Participation) using linear mixed models. The results indicated that the clustering effects were minimal to negligible across all grade levels (ICC = 0.003 for Grades 1–2, 0.029 for Grades 3–4, and < 0.001 for Grades 5–6). Although the design effect was small, to ensure the most robust statistical inference, we adopted a fixed-effects approach by including regional dummy variables as covariates in all regression models to control for unobserved regional heterogeneity. Furthermore, we utilized HC3 heteroscedasticity-consistent standard errors to strictly adjust for any potential non-independence of error terms. The chain-mediated effects were tested using the PROCESS macro for SPSS (Version 4.1) developed by Andrew F. Hayes. Specifically, Model 6 was selected because it is explicitly designed to test serial (chain) mediation models. This allowed us to examine the sequential theoretical pathways and determine whether the influence of the independent variable is transmitted through a specific sequence of mediators. The significance of indirect effects was assessed using a bias-corrected non-parametric percentile bootstrap method with 5,000 replications. Effects were considered statistically significant if the 95% confidence intervals (CI) did not include zero. To account for developmental differences, all analyses were conducted separately for three grade levels (Grades 1–2, 3–4, and 5–6). Gender, academic performance, and personality traits were included as covariates in all models alongside the regional fixed effects to control for potential confounding factors. Finally, a sensitivity analysis was performed by re-estimating the models without regional fixed effects to confirm that the directions and significance of the core mediation findings remained consistent regardless of model specification.

Two competing pathway models were constructed to examine the mechanisms driving Health Participation (Figure 1):

Pathway 1 (Knowledge-Driven): Health Knowledge → Health Motivation → Health Skill → Health Participation.

Model I tested the effect of HK on HM.

Model II tested the effects of HK and HM on HS.

Model III tested the combined effects of HK, HM, and HS on HP.

Pathway 2 (Motivation-Driven): Health motivation → Health knowledge → Health skill → Health Participation.

Model I tested the effect of HM on HK.

Model II tested the effects of HM and HK on HS.

Model III tested the combined effects of HM, HK, and HS on HP.

In this survey, 1,070 students were in grades 1–2, 1,190 in grades 3–4, and 1,065 in grades 5–6. The gender ratio of boys to girls was nearly equal: 47.38% girls and 52.62% boys in grades 1–2; 46.05% girls and 53.95% boys in grades 3–4; and 48.26% girls and 51.74% boys in grades 5–6. For grades 1–2 and grades 3–4, there were significant differences in HL across academic performance and character (p < 0.05). For grades 5--6, there were significant differences in HL by region, gender, academic performance, and character (p < 0.05). Details showed in Table 1.

For students in Grades 1–2, the regression analysis indicated that Health Knowledge significantly predicted Health Motivation (β = 0.289, p < 0.001). Both Health Knowledge (β = 0.440, p < 0.001) and Health Motivation (β = 0.311, p < 0.001) significantly predicted Health Skill. In Model III, Health Motivation (β = 0.180, p = 0.002) and Health Skill (β = 0.104, p = 0.006) significantly predicted Health Participation, whereas the direct effect of Health Knowledge was not significant (β = −0.039, p = 0.644). Bootstrap analysis confirmed a significant total indirect effect (Effect = 0.107, 95% CI [0.053, 0.172]), and all specific indirect paths were significant (Figure 2A).

For students in Grades 3–4, the regression analysis showed that Health Knowledge did not significantly predicted Health Motivation (β = 0.143, p = 0.077). While Health Knowledge strongly predicted Health Skill (β = 0.290, p < 0.001), Health Skills did not significantly predict Health Participation (β = 0.076, p = 0.122). Instead, Health Knowledge exerted a strong and significant direct effect on Health Participation (β = 0.260, p < 0.001). Consequently, the chain mediation path was not significant [Effect = 0.000, 95% CI (−0.001, 0.000)] (Figure 3A).

For students in Grades 5–6, the regression analysis showed Health Knowledge strongly predicted both Health Motivation (β = 0.409, p < 0.001) and Health Skill (β = 0.507, p < 0.001). Health Motivation significantly predicted Health Participation (β = 0.115, p < 0.001). The direct effect of Health Knowledge on Health Participation was not significant, but the indirect path through Motivation alone (Ind1) was significant [Effect = 0.047, 95% CI (0.028, 0.070)] (Figure 4A; Details see Tables 2, 3).

For students in Grades 1–2, the regression analysis showed that Health Motivation significantly positively predicted both Health Knowledge (β = 0.115, p < 0.001) and Health Skill (β = 0.311, p < 0.001). Health Knowledge also significantly predicted Health Skill (β = 0.440, p < 0.001). Both Health Motivation (β = 0.180, p = 0.002) and Health Skill (β = 0.104, p = 0.006) significant predicted Health Participation. The bootstrap analysis confirmed the chain mediation path Health Motivation- Health Knowledge- Health Skill-Health Participation was significant [Effect = 0.005, 95% CI (0.001, 0.011)] (Figure 2B).

For students in Grades 3–4, Health Motivation did not significantly predict Health Knowledge (β = 0.038, p = 0.073). Although Health Motivation had a significant direct effect on Health Participation (β = 0.070, p = 0.009). The path Health Motivation- Health Knowledge- Health Participation was not significant [Effect = 0.001, 95% CI (0.000, 0.0003)] (Figure 3B).

For students in Grades 5–6, the regression analysis showed that Health Motivation strongly predicted Health Knowledge (β = 0.273, p < 0.001) and Health Knowledge strongly predicted Health Skill (β = 0.507, p < 0.001). Neither Health Knowledge (β = 0.023, p = 0.541) nor Health Skill (β = −0.014, p = 0.433) significantly predicted Health Participation in Model III. The impact of Health Motivation on Health Participation was primarily driven by a significant direct effect (β = 0.115, p < 0.001). All indirect paths, including the chain mediation, were not significant (Figure 4B; Details see Tables 4, 5).