Institutional environment, academic attitude, and misconduct among medical students: a structural equation modeling and cluster analysis Authors: Peng Wang, Fan Li, Ji Li, Jian Yang Published: 2026 ABSTRACT BackgroundAcademic misconduct poses a severe threat to scientific integrity and medical ethics. While gender disparities in misconduct are widely documented, the underlying cognitive mechanisms explaining why males are more susceptible, and whether a supportive institutional environment can buffer the risks associated with academic pressure and negative attitudes, remain underexplored in the context of Chinese medical education.MethodsA multi-center cross-sectional study was conducted across medical universities in Southwest China. Standardized assessments evaluating five core constructs, including academic misconduct, academic attitude, institutional environment, ethical climate, and academic pressure, were administered. Structural Equation Modeling (SEM) and K-means cluster analysis were employed to analyze the mechanisms and student profiles.ResultsFrom an initial cohort of 511 participants, 426 valid questionnaires were retained. The study revealed a robust gender disparity, with male students reporting significantly higher engagement in misconduct (β = 0.29, p < 0.001). SEM analysis demonstrated that Academic Attitude—operationally defined as utilitarian tolerance toward dishonesty—served as a critical mediator, accounting for 50.7% of the total gender effect. Contrary to the strict buffering hypothesis, the institutional environment did not significantly moderate the adverse effects of negative attitudes, partially due to severe range restriction in the highly-regulated sample. Furthermore, cluster analysis identified a distinct "High Misconduct-High Strain" phenotype (11.3% of the sample) characterized by high pressure, negative attitude, and prevalent misconduct. Notably, this group was disproportionately male (60.4%) and exhibited a behavioral phenotype consistent with moral disengagement.ConclusionThe gender gap in academic misconduct is primarily driven by utilitarian cognitive framings rather than behavioral inevitabilities. The limited buffering capacity of environmental factors challenges the reliance on external governance alone. To effectively mitigate misconduct, institutions should avoid over-educating the resilient majority and move toward targeted educational strategies. Interventions should prioritize non-universal cognitive restructuring specifically tailored for the identified high-risk subgroup.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12909-026-09442-x. Introduction Background and significance in the digital age Academic integrity constitutes the ethical bedrock of higher education, yet its significance is profoundly amplified within the context of medical and pharmaceutical training [1]. Unlike general disciplines where academic dishonesty might be viewed solely as a pedagogical failure, misconduct in health professions education—ranging from plagiarism to the falsification of clinical data—carries direct implications for future professional competence and patient safety [2]. Empirical evidence has consistently demonstrated a disturbing correlation between unethical behaviors in medical school and subsequent disciplinary actions by state medical boards. Consequently, academic misconduct in this domain is widely regarded not merely as a breach of institutional regulations but as a potential precursor to professional malpractice [3]. Despite the implementation of rigorous honor codes globally, the prevalence of academic dishonesty remains alarmingly high and is evolving in complexity. Recent multi-institutional studies indicate that a significant proportion of health-professions students admit to engaging in some form of misconduct. Furthermore, this form of dishonesty suggests that current deterrent strategies, which largely rely on external surveillance, are becoming increasingly insufficient in the digital era [4]. Institutional environment and limitations of external control Theoretical frameworks such as Social Cognitive Theory suggest that human behavior is the result of a dynamic interplay between environmental factors, personal cognition, and behavioral feedback. Within the academic setting, the Institutional Environment represents the immediate external factor influencing student behavior. A robust ethical climate acts as a scaffold for moral development. However, empirical research suggests that external controls alone have limited efficacy [5]. For instance, Veríssimo (2020) noted that the severity of penalties does not consistently deter misconduct unless individual psychological traits are also considered [6]. Similarly, Guruswami et al. (2023) highlighted that cultural values and peer norms often override institutional rules, creating a dissonance where students recognize unethical behavior yet still engage in it due to environmental pressures [7]. This inconsistency underscores the need to look beyond external environments and examine how these factors are processed internally by students. Academic pressure and general strain theory Parallel to environmental influences, Academic Pressure has been identified as a potent precipitating factor. Grounded in General Strain Theory, it is posited that when individuals are subjected to strain—such as the intense competition for high GPAs and residency placements typical in pharmacy schools—they experience negative emotions [8]. To alleviate this strain, students may resort to illegitimate means when legitimate avenues seem blocked. However, viewing pressure solely as a negative force is an oversimplification [9]. Not all students under high pressure engage in misconduct; in fact, many thrive. This discrepancy highlights a critical gap: the failure to explain why students exposed to identical stressors exhibit vastly different behavioral responses. This suggests the existence of an internal regulatory mechanism—a "black box"—that mediates the impact of external stimuli. It should be noted that this study does not intend to provide a comprehensive empirical test of General Strain Theory (GST). Rather, we extract its core premise—that structural strain precipitates deviance—and substitute raw 'negative affect' with 'utilitarian academic attitude' as the focal mediator, better capturing the calculated rationalizations typical in high-stakes medical training [10]. The Mediating Role of Academic Attitude Academic Attitude is identified as this critical mediator. It is crucial to clarify that in the present operationalization, "Academic Attitude" does not refer to a broad, abstract moral consciousness. Rather, it specifically denotes students' "cognitive permission" or "utilitarian tolerance" toward concrete scenarios of academic dishonesty. External factors such as institutional environment and pressure likely do not trigger misconduct directly but rather shape students' attitudes toward learning and assessment, which in turn dictate their behavior [11]. Mechanisms such as moral disengagement allow students to rationalize unethical behavior (e.g., "everyone copies homework") without damaging their self-image. A supportive institutional environment may foster a respectful academic attitude, strengthening the student's internal moral compass. Conversely, excessive pressure combined with a perceived unfair environment may erode this attitude [12]. By framing Academic Attitude as a mediator, the present study aims to clarify the psychological pathway from "perception" to "action," offering a more nuanced understanding of how external determinants infiltrate the student's cognitive process. Methodological gaps and heterogeneity Furthermore, existing research is methodologically limited by its predominant reliance on variable-centered approaches (e.g., regression), which assume the student population is homogeneous. This perspective obscures the reality that students comprise diverse subgroups with distinct patterns of integrity. Adopting a person-centered approach, such as Cluster Analysis, allows for the identification of unobserved heterogeneity [13]. By classifying students into distinct profiles—such as "High-Risk" or "High-Integrity" groups—educators can develop targeted strategies tailored to different student archetypes. Control variables and research experience Additionally, while risk factors are well-documented, the role of experiential covariates such as undergraduate research experience warrants careful consideration. Engagement in scientific research is often presumed to cultivate critical thinking and adherence to data protocols [14]. However, it remains conceptually ambiguous whether such participation acts as a genuine ethical buffer, or merely reflects a self-selection bias wherein students with pre-existing positive attitudes are more likely to seek out research opportunities. Therefore, to rigorously isolate the primary psychological pathways and structural moderators, the present study treats research experience, alongside other demographic factors, strictly as a control variable rather than a focal moderator [15]. Objectives of the present study To bridge these gaps, the present study adopts a comprehensive analytical framework combining Structural Equation Modeling (SEM) and Cluster Analysis. The primary objectives are to: (1) classify students into distinct integrity profiles; (2) verify the mediating role of Academic Attitude between environmental stressors and misconduct; and (3) examine the moderating effect of Institutional Environment. By elucidating these complex relationships, this research provides evidence-based insights for building a holistic integrity system in the context of high-stakes medical education. Methods Study design, participants, and ethical considerations A multi-center cross-sectional study was conducted from September to November 2025 across five medical universities in Southwest China, including Kunming Medical University, Sichuan University, and others. The study targeted full-time undergraduate students majoring in pharmacy and related medical disciplines. To ensure the representativeness of the sample, a stratified cluster sampling strategy was employed, stratifying participants by academic year levels. The minimum sample size was calculated using Cochran’s formula for single population proportions: N = {Z2 × p × (1-p)}/{d2}. With a confidence level of 95% (Z = 1.96), an estimated prevalence of academic misconduct set conservatively at 50% (p = 0.5) to maximize sample size, and a margin of error of 5% (d = 0.05), the initial required sample size was calculated as 384. However, considering the requirements for Structural Equation Modeling (SEM), which typically necessitates a sample-to-parameter ratio of at least 10:1, and anticipating a potential invalid response rate of 15%, the target sample size was adjusted to approximately 500 participants. Instrument development and measures The structured questionnaire was developed based on the theoretical frameworks of Social Cognitive Theory and General Strain Theory. To ensure high content validity and cultural applicability, the instrument underwent a rigorous two-round Delphi expert consultation process. A panel of five experts, including senior professors in medical education, psychology, and ethics, reviewed the initial item pool. Based on their feedback, ambiguous items were revised, and the wording was refined to fit the specific context of Chinese pharmaceutical education. The final instrument consisted of two sections measured on a 5-point Likert scale ranging from 1 (Strongly Disagree/Never) to 5 (Strongly Agree/Always). The first section collected demographic information, including gender, grade, major, and Research Experience, which was assessed as a binary variable indicating whether the student had participated in faculty-mentored research projects. The second section measured the core latent constructs. Environmental factors were assessed across two distinct latent constructs based on psychometric validation. Institutional Environment (3 items) evaluated students' perceptions of policy strictness and punishment severity. Ethical Climate (3 items) measured the broader cultural atmosphere, including teacher role models and peer norms. Measurement items were adapted from widely validated academic integrity scales (e.g., the Academic Misconduct Questionnaire [Veríssimo et al.]) and contextualized for Chinese medical education through a two-round Delphi expert consultation (see Appendix A for full items). Prior to formal data collection, a pilot test (N = 30) was conducted to ensure item clarity and face validity. Academic Pressure was measured by three items reflecting the perceived strain derived from competitive ranking and heavy curriculum loads. Academic Attitude, serving as the mediator, was evaluated by four items assessing the student’s intrinsic respect for academic standards and moral stance toward assessment. Finally, Academic Misconduct Behavior, the outcome variable, was measured using a three-item scale that inquired about the frequency of specific dishonest behaviors: cheating in examinations, plagiarism in assignments, and fabrication of experimental data. This concise structure was designed to capture the core dimensions of misconduct while maintaining a high response rate. Data collection and quality control Data were collected via Wenjuanxing (Changsha Ranxing Information Technology Co., Ltd., China), a secure online survey platform widely used in Chinese academic research. The survey link was distributed through institutional networks. Given the digital nature of the administration, a multi-stage quality control protocol was implemented to ensure data integrity. Unique IP addresses and device identifiers were restricted to prevent duplicate submissions from the same user. To identify careless responding, attention-check questions (e.g., "Please select 'Somewhat Agree' for this item") were embedded within the survey. Furthermore, responses with a completion time of less than 120 s were automatically flagged and excluded from the final dataset. Only questionnaires with complete data on all key variables were included in the final analysis. Statistical analysis All statistical analyses were performed using R software (Version 4.5.2). The analytical procedure followed a logical sequence starting with psychometric validation. Confirmatory Factor Analysis (CFA) was conducted using the lavaan package to verify the construct validity of the measurement model. Model fit was assessed using the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR), while convergent validity was established through Average Variance Extracted (AVE) and Composite Reliability (CR). Following validation, descriptive statistics were computed to summarize sample characteristics, and K-means cluster analysis was performed using the factoextra package to identify distinct student integrity profiles based on the measured variables. Differences among these clusters were examined using ANOVA. Subsequently, a Structural Equation Model (SEM) was constructed to test the hypothesized mediating role of Academic Attitude between environmental stressors and misconduct behavior. Finally, hierarchical linear regression and simple slope analyses were employed to examine the moderating effect of Institutional Environment. Crucially, due to methodological concerns regarding self-selection bias (endogeneity)—as preliminary demographic screening indicated a non-random distribution of research participation across covariates such as year levels and majors—the present study rigorously treated research experience as a confounding control variable. This decision was made to prevent potential over-adjustment and ensure it was not an arbitrary variable exclusion.The optimal number of clusters (k = 3) for the K-means algorithm was determined and validated using the elbow method on the within-cluster sum of squares (see Supplementary Figure S2) and Silhouette analysis. The algorithm was initialized with 25 random starts (nstart = 25) to ensure stability. For the SEM, Maximum Likelihood (ML) estimation was used, with gender coded as a dummy variable (0 = Female, 1 = Male). Finally, given that data were collected through self-reported questionnaires, Common Method Bias (CMB) was assessed using Harman’s single-factor test (Podsakoff et al., 2003). All measurement items were entered into an unrotated Principal Component Analysis (PCA). The results indicated that the first factor accounted for 43.58% of the total variance, which is below the critical threshold of 50%. This result (details provided in Table S1) suggests that common method bias is not a serious concern in this study [16]. Ethical considerations The study was conducted in strict accordance with the principles of the Declaration of Helsinki. The research protocol was reviewed and approved by the Ethics Committee of Kunming Medical University (Approval No. KMMU2025MEC267). Electronic informed consent was obtained from all participants prior to data collection on the first page of the online survey. Participants were explicitly informed of the study's purpose, their right to withdraw at any time without consequence, and the confidentiality of their responses. To minimize social desirability bias and protect participant privacy, the survey was conducted anonymously, and no personally identifiable information (PII) was collected or stored. Results Demographic characteristics and measurement model evaluation Demographic characteristics A total of 511 questionnaires were initially distributed and collected. To ensure data quality, a rigorous screening process was implemented; 85 questionnaires were excluded due to incompleteness, insufficient completion time (less than 120 s), or obvious straight-lining patterns. Consequently, 426 valid responses were retained for the final analysis, yielding an effective response rate of 83.4%. The demographic profile of the final sample (N = 426) is summarized in Table 1. The gender distribution comprised 141 males (33.0%) and 285 females (67.0%). Academically, the majority of respondents were Pharmacy majors (67.0%), followed by Clinical Medicine students (31.0%). The distribution across year levels was relatively balanced, with Juniors (Year 3) constituting the largest subgroup (48.0%), followed by Freshmen (22.0%), Sophomores (20.0%), and Seniors (7.0%). Regarding research involvement, 37.0% of the students reported having prior research experience. Significant differences were observed in research experience across majors (chi^2 = 33.7, p < 0.001) and year levels (chi^2 = 38.6, p < 0.001), whereas no significant gender disparity was found (p = 0.994).Table 1Demographic characteristics of the study participants (N = 426)CharacteristicOverall N = 4261No Research Exp N = 2691With Research Exp N = 1571p-value2Gender0.994 Male141 (33%)89 (33%)52 (33%) Female285 (67%)180 (67%)105 (67%)Major < 0.001 Clinical Medicine134 (31%)65 (24%)69 (44%) Basic Medicine3 (0.7%)2 (0.7%)1 (0.6%) Pharmacy286 (67%)200 (74%)86 (55%) Nursing3 (0.7%)2 (0.7%)1 (0.6%) Public Health0 (0%)0 (0%)0 (0%) Stomatology0 (0%)0 (0%)0 (0%) Forensic Medicine0 (0%)0 (0%)0 (0%) Other0 (0%)0 (0%)0 (0%)Year level < 0.001 Freshman (Year 1)93 (22%)82 (30%)11 (7.0%) Sophomore (Year 2)85 (20%)58 (22%)27 (17%) Junior (Year 3)206 (48%)113 (42%)93 (59%) Senior (Year 4)30 (7.0%)11 (4.1%)19 (12%) Year 512 (2.8%)5 (1.9%)7 (4.5%) Other0 (0%)0 (0%)0 (0%)Academic Ranking (GPA) < 0.001 10%—30% of the major114 (27%)61 (23%)53 (34%) 30%—60% of the major93 (22%)61 (23%)32 (20%) After 60% of the major23 (5.4%)21 (7.8%)2 (1.3%) Top 10% of the major67 (16%)18 (6.7%)49 (31%) Unsure129 (30%)108 (40%)21 (13%)Political status < 0.001 CPC Member40 (9.4%)7 (2.6%)33 (21%) CYLC Member264 (62%)176 (65%)88 (56%) Other3 (0.7%)2 (0.7%)1 (0.6%) The Masses119 (28%)84 (31%)35 (22%)1n (%)2Pearson's Chi-squared test; Fisher's exact test Table 1Demographic characteristics of the study participants (N = 426)CharacteristicOverall N = 4261No Research Exp N = 2691With Research Exp N = 1571p-value2Gender0.994 Male141 (33%)89 (33%)52 (33%) Female285 (67%)180 (67%)105 (67%)Major < 0.001 Clinical Medicine134 (31%)65 (24%)69 (44%) Basic Medicine3 (0.7%)2 (0.7%)1 (0.6%) Pharmacy286 (67%)200 (74%)86 (55%) Nursing3 (0.7%)2 (0.7%)1 (0.6%) Public Health0 (0%)0 (0%)0 (0%) Stomatology0 (0%)0 (0%)0 (0%) Forensic Medicine0 (0%)0 (0%)0 (0%) Other0 (0%)0 (0%)0 (0%)Year level < 0.001 Freshman (Year 1)93 (22%)82 (30%)11 (7.0%) Sophomore (Year 2)85 (20%)58 (22%)27 (17%) Junior (Year 3)206 (48%)113 (42%)93 (59%) Senior (Year 4)30 (7.0%)11 (4.1%)19 (12%) Year 512 (2.8%)5 (1.9%)7 (4.5%) Other0 (0%)0 (0%)0 (0%)Academic Ranking (GPA) < 0.001 10%—30% of the major114 (27%)61 (23%)53 (34%) 30%—60% of the major93 (22%)61 (23%)32 (20%) After 60% of the major23 (5.4%)21 (7.8%)2 (1.3%) Top 10% of the major67 (16%)18 (6.7%)49 (31%) Unsure129 (30%)108 (40%)21 (13%)Political status < 0.001 CPC Member40 (9.4%)7 (2.6%)33 (21%) CYLC Member264 (62%)176 (65%)88 (56%) Other3 (0.7%)2 (0.7%)1 (0.6%) The Masses119 (28%)84 (31%)35 (22%)1n (%)2Pearson's Chi-squared test; Fisher's exact test Demographic characteristics of the study participants (N = 426) 1n (%) 2Pearson's Chi-squared test; Fisher's exact test Measurement model evaluation The psychometric properties of the measurement instruments were rigorously evaluated using Confirmatory Factor Analysis (CFA) prior to structural modeling (see Table 2). Reliability assessment indicated high internal consistency across all latent constructs. Cronbach’s alpha coefficients ranged from 0.790 to 0.961, and Composite Reliability (CR) values ranged from 0.812 to 0.961, all surpassing the recommended threshold of 0.70.Table 2Psychometric properties of the measurement model: reliability, convergent validity, and descriptive statisticsLatent constructItemsFactor loadingCronbach's αCRAVEMean ± SDInstitutional environmentInst_Policy0.866 ns0.9110.9140.7804.39 ± 0.79Inst_Punish0.942***Inst_Fairness0.844***Ethical climateEth_RoleModel0.931 ns0.8900.8930.7364.35 ± 0.77Eth_Integ0.934***Eth_PeerNorm0.731***Academic pressurePress_Diff0.541 ns0.7900.8120.5953.48 ± 0.93Press_GPA0.992***Press_Award0.837***Academic attitudeAtt_Exam0.896 ns0.9560.9570.8482.16 ± 1.14Att_Paper0.954***Att_LabData0.958***Att_Attend0.876***Misconduct behaviorSelf_Exam0.926 ns0.9610.9610.8921.85 ± 1.15Self_Paper0.946***Self_Lab0.960***Model Fit Indices: Chi-square/df = 2.79, CFI = 0.981, TLI = 0.976, RMSEA = 0.057, SRMR = 0.063***p <.001. CR = Composite Reliability; AVE = Average Variance Extracted. All loadings are standardized. Academic Attitude is reverse-scored relative to its normative connotation; higher scores indicate a more negative, utilitarian attitude toward academic misconduct Table 2Psychometric properties of the measurement model: reliability, convergent validity, and descriptive statisticsLatent constructItemsFactor loadingCronbach's αCRAVEMean ± SDInstitutional environmentInst_Policy0.866 ns0.9110.9140.7804.39 ± 0.79Inst_Punish0.942***Inst_Fairness0.844***Ethical climateEth_RoleModel0.931 ns0.8900.8930.7364.35 ± 0.77Eth_Integ0.934***Eth_PeerNorm0.731***Academic pressurePress_Diff0.541 ns0.7900.8120.5953.48 ± 0.93Press_GPA0.992***Press_Award0.837***Academic attitudeAtt_Exam0.896 ns0.9560.9570.8482.16 ± 1.14Att_Paper0.954***Att_LabData0.958***Att_Attend0.876***Misconduct behaviorSelf_Exam0.926 ns0.9610.9610.8921.85 ± 1.15Self_Paper0.946***Self_Lab0.960***Model Fit Indices: Chi-square/df = 2.79, CFI = 0.981, TLI = 0.976, RMSEA = 0.057, SRMR = 0.063***p <.001. CR = Composite Reliability; AVE = Average Variance Extracted. All loadings are standardized. Academic Attitude is reverse-scored relative to its normative connotation; higher scores indicate a more negative, utilitarian attitude toward academic misconduct Psychometric properties of the measurement model: reliability, convergent validity, and descriptive statistics Model Fit Indices: Chi-square/df = 2.79, CFI = 0.981, TLI = 0.976, RMSEA = 0.057, SRMR = 0.063 ***p <.001. CR = Composite Reliability; AVE = Average Variance Extracted. All loadings are standardized. Academic Attitude is reverse-scored relative to its normative connotation; higher scores indicate a more negative, utilitarian attitude toward academic misconduct Convergent validity was established through the examination of standardized factor loadings and Average Variance Extracted (AVE). All measurement items exhibited statistically significant factor loadings (p < 0.001), with values generally exceeding 0.70 (range: 0.541–0.992). The AVE values for the five constructs—Institutional Environment, Ethical Climate, Academic Pressure, Academic Attitude, and Misconduct Behavior—ranged from 0.595 to 0.892, well above the critical benchmark of 0.50. Descriptive analysis further revealed that participants generally perceived a positive Institutional Environment (Mean = 4.39, SD = 0.79) and Ethical Climate (Mean = 4.35, SD = 0.77), while reporting moderate levels of Academic Pressure (Mean = 3.48, SD = 0.93) and relatively low engagement in Misconduct Behavior (Mean = 1.85, SD = 1.15). Bivariate correlations The bivariate correlations among the study variables are visualized in the heatmap (Figure S1 and Table S2). As anticipated, Misconduct Behavior was positively correlated with Academic Pressure (r = 0.26, p < 0.05) and Academic Attitude (r = 0.72, p < 0.001), while exhibiting negative correlations with Institutional Environment (r = −0.21, p < 0.05) and Ethical Climate (r = −0.19, p < 0.05). Importantly, the exceptionally strong positive correlation between Academic Attitude and Misconduct Behavior provides robust preliminary support for the hypothesized mediating mechanism. Regarding the relationships among predictors, a high correlation was observed between Institutional Environment and Ethical Climate (r = 0.79). However, despite this strong association, Variance Inflation Factor (VIF) values in the subsequent regression analyses remained within acceptable limits (VIF < 3.0), indicating that multicollinearity did not distort the structural modeling. All other correlations among independent variables were moderate to weak, establishing a sound foundation for further analyses. Demographic differences in academic misconduct To examine the variations in academic misconduct across different demographic groups, independent sample t-tests and One-Way ANOVA were conducted (Table S3 and Fig. 1).Fig. 1Gender differences in academic attitude and misconduct behavior: A violin plot analysis with mean and standard deviation labels (Welch's t(207.25) = 5.42, p < 0.001) Gender differences in academic attitude and misconduct behavior: A violin plot analysis with mean and standard deviation labels (Welch's t(207.25) = 5.42, p < 0.001) The analysis revealed a significant gender disparity in academic misconduct behavior (t(207.25) = 5.42, p < 0.001, Cohen's d = 0.63). Specifically, male students reported significantly higher engagement in misconduct compared to their female counterparts. However, no statistically significant differences were found based on research experience (t = −1.06, p = 0.288). Similarly, the One-Way ANOVA results indicated that academic misconduct scores did not differ significantly across year levels (F(4, 421) = 1.72, p = 0.145). These findings suggest that gender is the primary demographic factor distinguishing levels of academic misconduct in this sample, warranting further investigation in subsequent multivariate analyses. It is worth noting that while Cohen's d (0.63) captures the unadjusted standardized difference between group means, subsequent hierarchical regression analyses will further isolate the unique predictive power of gender after controlling for relevant covariates. Factors influencing academic misconduct: a hierarchical regression analysis To systematically examine the predictors of academic misconduct and preliminarily validate the mediating role of academic attitude, a three-step hierarchical linear regression analysis was performed (Table 3). This approach allows for the observation of incremental variance explained by adding distinct clusters of variables.Table 3Hierarchical linear regression results predicting misconduct behaviorPredictorsModel 1 (β)Model 2 (β)Model 3 (β)Step 1: Gender (Male)0.29***0.26***0.14***Step 1: Research Experience0.050.060.06Step 2: Institutional Environment-−0.15*−0.02Step 2: Ethical Climate-−0.15*−0.07Step 2: Academic Pressure-0.32***0.12**Step 3: Academic Attitude--0.64***R-squared0.0840.2210.555Adjusted R-squared0.0800.2110.548F-statistic19.423.887.0Delta R-squared0.0840.1360.334Values are standardized coefficients (Beta)*p < 0.05, **p < 0.01, ***p < 0.001 Table 3Hierarchical linear regression results predicting misconduct behaviorPredictorsModel 1 (β)Model 2 (β)Model 3 (β)Step 1: Gender (Male)0.29***0.26***0.14***Step 1: Research Experience0.050.060.06Step 2: Institutional Environment-−0.15*−0.02Step 2: Ethical Climate-−0.15*−0.07Step 2: Academic Pressure-0.32***0.12**Step 3: Academic Attitude--0.64***R-squared0.0840.2210.555Adjusted R-squared0.0800.2110.548F-statistic19.423.887.0Delta R-squared0.0840.1360.334Values are standardized coefficients (Beta)*p < 0.05, **p < 0.01, ***p < 0.001 Hierarchical linear regression results predicting misconduct behavior Values are standardized coefficients (Beta) *p < 0.05, **p < 0.01, ***p < 0.001 Step 1 (Model 1) incorporated gender and research experience as controls. The results indicated that gender was a robust predictor of misconduct (β = 0.29, p < 0.001), with male students reporting significantly higher engagement in academic dishonesty compared to females. In contrast, prior research experience did not show a statistically significant association with misconduct behaviors (β = 0.05, p > 0.05) in this initial model.Step 2 (Model 2) introduced contextual factors, including Institutional Environment, Ethical Climate, and Academic Pressure. The inclusion of these variables significantly enhanced the model's explanatory power, increasing the R2 to 0.221. Specifically, Academic Pressure emerged as a strong positive predictor (β = 0.32, p < 0.001), suggesting that students facing higher competitive pressure are more prone to misconduct. Conversely, both Institutional Environment (β = −0.15, p < 0.05) and Ethical Climate (β = −0.15, p < 0.05) were negatively associated with misconduct, indicating that a supportive and ethical academic environment serves as a protective factor. In this step, the coefficient for gender decreased slightly (β = 0.26, p < 0.001) but remained highly significant.Step 3 (Model 3) added Academic Attitude to the regression equation, which further increased the overall model explanation (R2 = 0.555). This variable proved to be the most potent predictor in the entire model (β = 0.64, p < 0.001). Crucially, the inclusion of attitude led to substantial changes in the coefficients of other variables. The protective effects of Institutional Environment (β = −0.02) and Ethical Climate (β = −0.07) became non-significant, and the adverse effect of Academic Pressure diminished from 0.32 to 0.12 (p < 0.01). Furthermore, the coefficient for gender dropped markedly from 0.26 in Model 2 to 0.14 (p < 0.001). This pattern of attenuation—where the direct effects of distal factors (Gender, Environment) are absorbed by a proximal factor (Attitude)—strongly supports the hypothesis that Academic Attitude acts as a critical mediator in the mechanism of academic misconduct. However, given the cross-sectional design, the possibility of reverse causality cannot be definitively ruled out—such as students adjusting their expressed attitudes to reduce cognitive dissonance after engaging in misconduct. Step 1 (Model 1) incorporated gender and research experience as controls. The results indicated that gender was a robust predictor of misconduct (β = 0.29, p < 0.001), with male students reporting significantly higher engagement in academic dishonesty compared to females. In contrast, prior research experience did not show a statistically significant association with misconduct behaviors (β = 0.05, p > 0.05) in this initial model. Step 1 (Model 1) incorporated gender and research experience as controls. The results indicated that gender was a robust predictor of misconduct (β = 0.29, p < 0.001), with male students reporting significantly higher engagement in academic dishonesty compared to females. In contrast, prior research experience did not show a statistically significant association with misconduct behaviors (β = 0.05, p > 0.05) in this initial model. Step 2 (Model 2) introduced contextual factors, including Institutional Environment, Ethical Climate, and Academic Pressure. The inclusion of these variables significantly enhanced the model's explanatory power, increasing the R2 to 0.221. Specifically, Academic Pressure emerged as a strong positive predictor (β = 0.32, p < 0.001), suggesting that students facing higher competitive pressure are more prone to misconduct. Conversely, both Institutional Environment (β = −0.15, p < 0.05) and Ethical Climate (β = −0.15, p < 0.05) were negatively associated with misconduct, indicating that a supportive and ethical academic environment serves as a protective factor. In this step, the coefficient for gender decreased slightly (β = 0.26, p < 0.001) but remained highly significant. Step 2 (Model 2) introduced contextual factors, including Institutional Environment, Ethical Climate, and Academic Pressure. The inclusion of these variables significantly enhanced the model's explanatory power, increasing the R2 to 0.221. Specifically, Academic Pressure emerged as a strong positive predictor (β = 0.32, p < 0.001), suggesting that students facing higher competitive pressure are more prone to misconduct. Conversely, both Institutional Environment (β = −0.15, p < 0.05) and Ethical Climate (β = −0.15, p < 0.05) were negatively associated with misconduct, indicating that a supportive and ethical academic environment serves as a protective factor. In this step, the coefficient for gender decreased slightly (β = 0.26, p < 0.001) but remained highly significant. Step 3 (Model 3) added Academic Attitude to the regression equation, which further increased the overall model explanation (R2 = 0.555). This variable proved to be the most potent predictor in the entire model (β = 0.64, p < 0.001). Crucially, the inclusion of attitude led to substantial changes in the coefficients of other variables. The protective effects of Institutional Environment (β = −0.02) and Ethical Climate (β = −0.07) became non-significant, and the adverse effect of Academic Pressure diminished from 0.32 to 0.12 (p < 0.01). Furthermore, the coefficient for gender dropped markedly from 0.26 in Model 2 to 0.14 (p < 0.001). This pattern of attenuation—where the direct effects of distal factors (Gender, Environment) are absorbed by a proximal factor (Attitude)—strongly supports the hypothesis that Academic Attitude acts as a critical mediator in the mechanism of academic misconduct. However, given the cross-sectional design, the possibility of reverse causality cannot be definitively ruled out—such as students adjusting their expressed attitudes to reduce cognitive dissonance after engaging in misconduct. Step 3 (Model 3) added Academic Attitude to the regression equation, which further increased the overall model explanation (R2 = 0.555). This variable proved to be the most potent predictor in the entire model (β = 0.64, p < 0.001). Crucially, the inclusion of attitude led to substantial changes in the coefficients of other variables. The protective effects of Institutional Environment (β = −0.02) and Ethical Climate (β = −0.07) became non-significant, and the adverse effect of Academic Pressure diminished from 0.32 to 0.12 (p < 0.01). Furthermore, the coefficient for gender dropped markedly from 0.26 in Model 2 to 0.14 (p < 0.001). This pattern of attenuation—where the direct effects of distal factors (Gender, Environment) are absorbed by a proximal factor (Attitude)—strongly supports the hypothesis that Academic Attitude acts as a critical mediator in the mechanism of academic misconduct. However, given the cross-sectional design, the possibility of reverse causality cannot be definitively ruled out—such as students adjusting their expressed attitudes to reduce cognitive dissonance after engaging in misconduct. Structural equation modeling: the mediating role of academic attitude To further elucidate the mechanisms underlying academic misconduct, Structural Equation Modeling (SEM) was employed to test the hypothesized mediation model. The standardized path coefficients are illustrated in the path diagram (Fig. 2, Table S5), while the detailed decomposition of direct, indirect, and total effects is visualized in the forest plot (Fig. 3) and summarized in Table 4.Fig. 2Structural equation modeling (SEM) path diagram illustrating the mediating role of academic attitude between gender and misconduct behaviorFig. 3Decomposition of standardized effects on academic misconductTable 4Decomposition of direct, indirect, and total effects (1000 Bootstrap Samples)PredictorEffect_TypeStandardized BetaBoot SEBootstrap 95% CIP_valueAcademic pressureTotal Effect0.3220.054[0.222, 0.430] < 0.001Direct Effect0.1230.040[0.044, 0.203]0.002Indirect Effect0.1990.046[0.119, 0.294] < 0.001Gender (Male)Total Effect0.2700.050[0.172, 0.367] < 0.001Direct Effect0.1330.038[0.060, 0.212] < 0.001Indirect Effect0.1370.036[0.072, 0.208] < 0.001Inst. environmentTotal Effect−0.2620.054[−0.367, −0.152] < 0.001Direct Effect−0.0660.045[−0.159, 0.018]0.140Indirect Effect−0.1960.038[−0.273, −0.124] < 0.001Data represent standardized coefficients (β). Bootstrap sample size = 1,000. The proportion of mediation for Gender (Male) is 50.7% Structural equation modeling (SEM) path diagram illustrating the mediating role of academic attitude between gender and misconduct behavior Decomposition of standardized effects on academic misconduct Table 4Decomposition of direct, indirect, and total effects (1000 Bootstrap Samples)PredictorEffect_TypeStandardized BetaBoot SEBootstrap 95% CIP_valueAcademic pressureTotal Effect0.3220.054[0.222, 0.430] < 0.001Direct Effect0.1230.040[0.044, 0.203]0.002Indirect Effect0.1990.046[0.119, 0.294] < 0.001Gender (Male)Total Effect0.2700.050[0.172, 0.367] < 0.001Direct Effect0.1330.038[0.060, 0.212] < 0.001Indirect Effect0.1370.036[0.072, 0.208] < 0.001Inst. environmentTotal Effect−0.2620.054[−0.367, −0.152] < 0.001Direct Effect−0.0660.045[−0.159, 0.018]0.140Indirect Effect−0.1960.038[−0.273, −0.124] < 0.001Data represent standardized coefficients (β). Bootstrap sample size = 1,000. The proportion of mediation for Gender (Male) is 50.7% Decomposition of direct, indirect, and total effects (1000 Bootstrap Samples) Data represent standardized coefficients (β). Bootstrap sample size = 1,000. The proportion of mediation for Gender (Male) is 50.7% Direct effects The path analysis (Fig. 2) revealed that gender had a significant positive effect on Academic Attitude (β = 0.21, p < 0.001), indicating that male students tended to hold more negative or utilitarian attitudes toward academic integrity. Furthermore, Academic Attitude was identified as a powerful proximal predictor of Misconduct Behavior (β = 0.66, p < 0.001), suggesting that a compromised academic attitude strongly predisposes students to engage in dishonest behaviors. In terms of contextual factors, Institutional Environment (β = −0.30, p < 0.001) and Academic Pressure (β = 0.30, p < 0.001) significantly influenced Academic Attitude, although their direct effects on Misconduct Behavior were less pronounced (β = 0.12, p < 0.01 for Pressure) or non-significant (β = −0.07, p > 0.05 for Environment) in the full model. Mediation analysis and effect decomposition The bootstrapping analysis (1,000 samples) supported the significant mediating role of Academic Attitude. As visually depicted in Fig. 3, the total effect of gender on misconduct (β = 0.27, p < 0.001) was significantly decomposed. The indirect effect of gender on misconduct through attitude was estimated at 0.14 (p < 0.001), accounting for 50.7% of the total effect. This substantial indirect path indicates that the gender disparity in academic misconduct is largely transmitted through differences in students’ attitudes toward academic integrity. In comparison, the direct effect of gender on misconduct remained significant (β = 0.13, p < 0.001) but was notably smaller than the total effect. Figure 3 clearly illustrates this attenuation, highlighting that while male students have a residual direct tendency toward misconduct, approximately half of their risk profile is statistically attributable to their academic attitudes, which serves as a modifiable target for educational intervention. Moderation analysis: the role of institutional environment To test whether a supportive institutional environment buffers the risks of negative attitudes and high pressure, moderation analyses were conducted (Table 5). The results revealed that while the main effects of Academic Attitude (β = 0.70, p < 0.001) and Academic Pressure (β = 0.37, p < 0.001) were highly significant, their respective interaction terms with Institutional Environment were not statistically significant (Attitude × Environment: β = −0.034, p > 0.05; Pressure × Environment: β = 0.074, p > 0.05).Table 5Moderation analysis results (Interaction Effects)Model AModel BAttitude (Centered)0.702 ***(0.038)Pressure (Centered)0.366 ***(0.056)Environment (Centered)−0.044−0.355 ***(0.052)(0.069)Interaction: Att x Env−0.034(0.053)Interaction: Press x Env0.074(0.058)Gender (Male)0.331 ***0.625 ***(0.083)(0.106)N426426R-squared0.5400.212*** p < 0.001; ** p < 0.01; * p < 0.05 Table 5Moderation analysis results (Interaction Effects)Model AModel BAttitude (Centered)0.702 ***(0.038)Pressure (Centered)0.366 ***(0.056)Environment (Centered)−0.044−0.355 ***(0.052)(0.069)Interaction: Att x Env−0.034(0.053)Interaction: Press x Env0.074(0.058)Gender (Male)0.331 ***0.625 ***(0.083)(0.106)N426426R-squared0.5400.212*** p < 0.001; ** p < 0.01; * p < 0.05 Moderation analysis results (Interaction Effects) *** p < 0.001; ** p < 0.01; * p < 0.05 These non-significant interactions are visually corroborated by Fig. 4, which incorporates 95% confidence bands. Both Fig. 4a and b display wide and substantially overlapping confidence intervals across different environmental levels (+ 1 SD, Mean, −1 SD). Rather than definitively proving the absolute absence of a buffering effect, this overlap largely reflects the severe range restriction in environment scores (Mean = 4.39 out of 5), which likely attenuated the statistical power to detect a significant interaction.Fig. 4Moderation analysis of Institutional Environment on the relationship between academic attitude and misconduct behavior. Shaded areas represent 95% confidence bands. The wide and overlapping confidence intervals visually corroborate the non-significant interaction term, reflecting the range restriction in environment scores Moderation analysis of Institutional Environment on the relationship between academic attitude and misconduct behavior. Shaded areas represent 95% confidence bands. The wide and overlapping confidence intervals visually corroborate the non-significant interaction term, reflecting the range restriction in environment scores Consequently, while Academic Attitude and Pressure operate as dominant predictors in this highly-regulated sample, the failure to detect a significant interaction merely implies that the environment's buffering capacity is modest in this specific population, rather than suggesting institutional climates are universally irrelevant. Identification of student profiles: a cluster analysis To move beyond variable-level relationships and identify distinct subgroups of students with similar behavioral and psychological patterns, a K-means cluster analysis was performed. Based on standardized scores of Academic Attitude, Academic Pressure, Institutional Environment, and Misconduct Behavior, three distinct student profiles were identified. The characteristics of these clusters are detailed in Table S4 and visualized in the radar chart (Fig. 5).Fig. 5Identification of student risk phenotypes via K-means cluster analysis. a Radar chart depicting the mean scores of the three clusters across key dimensions using neutral descriptive labels. b Principal Component Analysis (PCA) visualizing the cluster separation and distribution on the first two principal components Identification of student risk phenotypes via K-means cluster analysis. a Radar chart depicting the mean scores of the three clusters across key dimensions using neutral descriptive labels. b Principal Component Analysis (PCA) visualizing the cluster separation and distribution on the first two principal components Cluster 1: Low misconduct-low strain group This cluster represents the majority of the sample (N = 254, 59.6%). Students in this group exhibited the lowest levels of Misconduct Behavior (Mean = 1.29) and the most positive Academic Attitude (Mean = 1.52, where lower scores indicate better attitude). They also perceived the Institutional Environment as highly supportive (Mean = 4.80). Demographically, this group had the lowest proportion of male students (27.2%), suggesting that female students are more likely to fall into this low-risk category. Cluster 2: moderate risk group Comprising 29.1% of the sample (N = 124), this group is characterized by moderate scores across all dimensions. Notably, they reported the lowest perception of the Institutional Environment (Mean = 3.44) and the lowest Academic Pressure (Mean = 3.11). This pattern suggests a group of students who may be disengaged from the academic system—neither highly pressured nor highly committed to ethical standards. Cluster 3: high misconduct-high strain group Although the smallest group (N = 48, 11.3%), this cluster is of greatest concern. These students reported the highest levels of Misconduct Behavior (Mean = 4.14) and the most negative/utilitarian Academic Attitude (Mean = 4.26). Crucially, they also experienced the highest Academic Pressure (Mean = 4.58). A striking demographic feature of this high-risk profile is the gender composition: 60.4% were male, which is disproportionately higher than the overall male representation in the sample (33.0%). This finding reinforces the earlier regression results, pinpointing male students under high pressure and with poor attitudes as the primary "at-risk" population for academic misconduct. Discussion The mechanism of gender disparity: attitude as the key mediator The present study confirms a robust gender disparity in academic misconduct, with male students demonstrating significantly higher engagement in dishonest behaviors compared to females. To ensure interpretative clarity, it is important to distinguish the magnitude of this effect: the unadjusted group difference was substantial (Cohen's d = 0.63 in the bivariate T-test), and the unique predictive effect of being male remained robust even after controlling for demographic and environmental factors (regression β = 0.29, p < 0.001). While previous literature has frequently documented this gap, often attributing it to innate biological tendencies toward risk-taking, Structural Equation Modeling (SEM) results provide a more nuanced cognitive explanation [17]. The analysis revealed that Academic Attitude partially mediates the relationship between gender and misconduct, accounting for 50.7% of the total effect (Fig. 3). However, the operationalization of this mediator requires precise theoretical bounding. In the current measurement model, "Academic Attitude" does not reflect a broad, stable moral consciousness; rather, it specifically captures students' "cognitive permission" or "utilitarian tolerance" toward concrete scenarios of dishonesty (e.g., fabricating data or copying assignments) [18]. While this pathway aligns with Social Role Theory—which suggests that males are often socialized to prioritize "agency" (achievement, competition, and status) over "communion" (compliance and community harmony)—it must be acknowledged that the conceptual homology between "tolerance for a behavior" and "engagement in that behavior" may have artificially inflated the statistical magnitude of this mediation effect. In high-stakes academic environments, this male socialization may manifest as highly outcome-oriented rationalizations, leading male students to view rule-breaking not as a moral failing but as a calculated risk to achieve competitive success [19]. Therefore, interventions addressing gender disparities should move beyond generic punitive measures, which may elicit reactance. Instead, educators should focus on reshaping the cognitive value systems of male students, specifically utilizing cognitive dissonance strategies to challenge their utilitarian definitions of "academic success." It must be underscored that the present study did not directly assess gender-role constructs; rather, the observed cognitive pathway is merely consistent with theoretical predictions derived from Social Role Theory [20]. The primary role of individual attitude over institutional environment Contrary to the "Buffering Hypothesis" derived from Person-Environment Fit theory, the moderation analysis (Table 5) revealed that a supportive institutional environment failed to mitigate the adverse effects of negative academic attitudes or high academic pressure (Interaction p > 0.05). As illustrated by the parallel slopes in Fig. 4, the strong positive relationship between negative attitude and misconduct remains constant regardless of environmental quality. Rather than viewing attitude and environment as entirely independent forces, it is highly plausible that students' utilitarian cognitive framing is itself a product of prolonged adaptation and learning within high-pressure, heavily regulated environments. Once a student internalizes this "success at all costs" mindset, the protective power of external institutional rules becomes limited, leading to "Ethical Fading" where self-interest overrides organizational culture [21]. However, alternative methodological explanations must also be acknowledged. Specifically, the present sample exhibited severe range restriction regarding the institutional environment (Mean = 4.39, SD = 0.79 out of 5), as all surveyed institutions enforce stringent governance. The limited variability at the lower end of the environment scale may have significantly attenuated the statistical power to detect an interaction effect. Therefore, this non-significant finding merely demonstrates that the buffering capacity of the environment is modest within this restricted population, rather than proving that institutional climates are universally ineffective in contexts with greater environmental heterogeneity [22]. This finding challenges the traditional administrative reliance on "building better systems" as a universal solution. It suggests that without addressing students' internal moral agency, simply optimizing the external environment—such as increasing surveillance or publishing stricter guidelines—is insufficient. Educators must prioritize "value internalization" over mere "rule compliance," fostering an internal moral compass that remains stable even in permissive environments [23]. Precision identification of "at-risk" students Moving beyond variable-level analysis, the cluster analysis (Table S4) identified a distinct "High Misconduct-High Strain" profile, comprising 11.3% of the sample. This subgroup is characterized by compounded risks: high misconduct (Mean = 4.14), high pressure (Mean = 4.58), and an extremely negative attitude (Mean = 4.26). Demographically, this group is disproportionately male (60.4%), reinforcing the regression findings regarding male vulnerability. The identification of this phenotype is critical because it reveals that the majority of students (Cluster 1, 59.6%) remain resilient. This granular profiling necessitates a shift from generic educational approaches to targeted interventions. Given finite educational resources, institutions should avoid over-educating the low-risk majority. Instead, efforts must be reallocated to design non-universal, precision interventions specifically tailored for the 11.3% high-risk subgroup [24]. For these specific students, particularly males, interventions must center on "cognitive restructuring" rather than generic ethics training, aiming to dismantle their utilitarian rationalizations before misconduct becomes habitual [25]. From awareness to internalization: patterns of moral disengagement A paradoxical finding from the cluster profiling is that the "High Misconduct-High Strain" group reported a remarkably high perception of the institutional environment (Mean = 4.66), which was even higher than the "Moderate Risk" group. This indicates that these students are fully aware of the rigorous institutional rules and ethical climate, yet they consciously choose to engage in misconduct [26]. It must be noted that the identification of this "high pressure-negative attitude-high misconduct" co-occurrence merely represents a behavioral phenotype. While this pattern is highly consistent with theoretical expectations of what Bandura terms "Moral Disengagement [27]", the present study did not directly measure this internal psychological mechanism; therefore, this interpretation remains an inferential speculation [28]. Within this theoretical framework, such students may employ specific cognitive restructuring mechanisms to rationalize their behavior. In the context of medical education, this may manifest as moral justification (e.g., believing that fabricating basic science data is victimless and does not harm future patients), displacement of responsibility (e.g., blaming the hyper-competitive residency placement system for forcing them to cheat), or advantageous comparison (e.g., rationalizing that copying homework is trivial compared to clinical malpractice) [29]. It implies that current integrity education succeeds in knowledge transmission (students know the rules exist) but fails in moral internalization (students do not accept the rules as binding). Future curriculum reform should incorporate dilemma-based learning and active mentorship programs. By discussing real-world gray areas, educators can help students bridge this gap, transforming "external rules" from perceived obstacles into "internal convictions" [30]. Contextualizing pressure: a general strain theory perspective Academic pressure emerged as a robust and consistent predictor of misconduct across both correlation (r = 0.26) and regression analyses (β = 0.32 in Model 2). Interpreted through the lens of General Strain Theory (GST), academic misconduct can be viewed as a maladaptive coping strategy for students facing structural strain in a highly competitive academic ecosystem. When legitimate means are perceived as blocked or insufficient to achieve high-stakes goals (e.g., scholarships, graduation requirements), students experience strain that propels them toward illegitimate means [31]. This supports the notion of "Institutional Anomie," where the emphasis on goals (winning) overwhelms the emphasis on means (integrity). However, as the present mediation and cluster analyses demonstrate, structural pressure alone does not inevitably lead to misconduct for the majority of students [32]. Rather, it is the lethal combination of high pressure and a utilitarian academic attitude—present primarily in the 11.3% high-risk subgroup—that catalyzes dishonest behavior [33]. Therefore, instead of attempting the daunting and often impractical task of eliminating academic pressure entirely or overhauling macro-level evaluation metrics, educational interventions should focus on equipping this specific vulnerable subgroup with psychological resilience and cognitive reframing strategies to cope with strain ethically [34]. Limitations Several limitations of this study should be acknowledged. First, the cross-sectional design precludes causal inferences; specifically, the bidirectional relationship between attitude and behavior warrants longitudinal investigation to determine if bad attitudes cause misconduct or if misconduct rationalizes bad attitudes. Furthermore, the conceptual homology between the operationalization of attitude (measuring tolerance) and behavior (measuring actual frequency) may have artificially inflated the explanatory power of the mediation model. Second, although Harman’s single-factor test (43.58%) indicated no serious common method bias, the reliance on self-reported data may still be subject to social desirability effects, particularly regarding sensitive misconduct behaviors [35]. Additionally, the inherent limitations of Harman's test in definitively ruling out common method variance must be acknowledged, which optimally requires common method factor CFA approaches. Finally, while this study involved a multi-center survey across exactly five medical universities, the sample was primarily drawn from Southwest China. As a developing region with rapidly expanding educational infrastructure, the academic pressures here may differ from those in established educational hubs in Eastern or Northern China. Moreover, the sample exhibited a skewed disciplinary distribution (67% pharmacy vs. 31% clinical medicine). Given the substantial differences in curricula, stress sources, and professional ethics foci between these fields, this imbalance may restrict the broader applicability of the results. Consequently, the generalizability of findings to other cultural, economic, or disciplinary contexts requires further verification [36]. Conclusion This study elucidates the role of cognitive framing in academic misconduct, demonstrating that utilitarian academic attitudes are key drivers of the observed gender disparity. The findings suggest that a compromised internal moral compass can override perceived supportive institutional climates. Consequently, addressing the misconduct observed in high-risk students requires moving beyond generic rule enforcement toward targeted educational strategies. Specifically, medical schools are advised to: (1) avoid universally applied, generic ethics courses that risk over-educating the resilient majority; (2) design non-universal, precision interventions centered on cognitive restructuring specifically tailored for the identified 11.3% high-risk subgroup (predominantly male); and (3) utilize dilemma-based reasoning and targeted psychological support to directly dismantle the compounded risks of high pressure and negative attitudes within this specific behavioral phenotype. Supplementary Information Supplementary Material 1. Table S1. Common Method Bias Test Results. Supplementary Material 2. Table S2. Descriptive statistics and bivariate correlations of the study variables. Supplementary Material 3. Table S3. Demographic Differences in Academic Misconduct (Unified). Supplementary Material 4. Table S4. Characteristics of Identified Student Clusters. Supplementary Material 5. Table S5. Standardized Path Coefficients and Hypothesis Testing for the SEM Model. Supplementary Material 6. Figure S1. Correlation heatmap of key study variables. Note: Values represent Pearson correlation coefficients (r). Colors indicate the direction and strength of the associations (red = positive correlation, blue = negative correlation). All displayed correlations are statistically significant at p < 0.05. Notably, Misconduct Behavior exhibits a strong positive correlation with Academic Attitude (r = 0.72) and a moderate positive correlation with Academic Pressure (r = 0.26), providing robust preliminary support for the hypothesized theoretical pathways. Supplementary Material 7. Figure S2. Determination of the optimal number of clusters via the elbow method. Note: The line graph displays the within-cluster sum of squares across different numbers of clusters (k). The distinct "elbow" (inflection point) occurs at k = 3, empirically justifying the selection of the three-cluster solution for identifying student risk phenotypes in the K-means analysis. Supplementary Material 8. Supplementary Material 1. Table S1. Common Method Bias Test Results. Supplementary Material 2. Table S2. Descriptive statistics and bivariate correlations of the study variables. Supplementary Material 3. Table S3. Demographic Differences in Academic Misconduct (Unified). Supplementary Material 4. Table S4. Characteristics of Identified Student Clusters. Supplementary Material 5. Table S5. Standardized Path Coefficients and Hypothesis Testing for the SEM Model. Supplementary Material 6. Figure S1. Correlation heatmap of key study variables. Note: Values represent Pearson correlation coefficients (r). Colors indicate the direction and strength of the associations (red = positive correlation, blue = negative correlation). All displayed correlations are statistically significant at p < 0.05. Notably, Misconduct Behavior exhibits a strong positive correlation with Academic Attitude (r = 0.72) and a moderate positive correlation with Academic Pressure (r = 0.26), providing robust preliminary support for the hypothesized theoretical pathways. Supplementary Material 7. Figure S2. Determination of the optimal number of clusters via the elbow method. Note: The line graph displays the within-cluster sum of squares across different numbers of clusters (k). The distinct "elbow" (inflection point) occurs at k = 3, empirically justifying the selection of the three-cluster solution for identifying student risk phenotypes in the K-means analysis. Supplementary Material 8.