Professional Identity, Psychological Resilience, and Adaptive Performance in Junior Nurses Exposed to Workplace Bullying: A Latent Profile and Network Analysis.

Professional Identity, Psychological Resilience, and Adaptive Performance in Junior Nurses Exposed to Workplace Bullying: A Latent Profile and Network Analysis

Aim

To identify resource‐adaptation profiles and profile‐specific network organization among junior nurses exposed to workplace bullying by examining professional identity (PID), psychological resilience (PSR), and adaptive performance (ADP).

Design

A multicenter, cross‐sectional study.

Methods

Survey data were collected from 7231 junior nurses in 90 tertiary hospitals in Sichuan Province, China; 5839 met the workplace bullying exposure criterion and were included. Latent profile analysis was performed using 36 item‐level indicators of PID, PSR, and ADP. Ordered logistic regression examined factors associated with profile membership, and network analysis compared profile‐specific resource organization. Robustness was assessed using network comparison tests, residualized networks, hospital‐cluster robust sensitivity analysis, and bootstrap stability analyses.

Results

Three profiles were identified: the resource‐constrained adaptive profile (n = 1906, 32.64%), moderate resource‐adaptation profile (n = 3152, 53.98%), and high resource‐adaptation profile (n = 781, 13.38%). More frequent night shifts, longer weekly working hours, poorer sleep quality, and lack of regular exercise were associated with greater odds of membership in a more resource‐constrained profile. The high resource‐adaptation profile showed the highest global network strength (16.223) and the largest mean absolute nonzero edge weight (0.086), despite fewer nonzero edges. PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) repeatedly served as bridge nodes. Bootstrap analyses supported centrality stability, with CS coefficients of 0.750 across base and residualized networks.

Conclusion

Junior nurses exposed to workplace bullying showed heterogeneous resource configurations and profile‐specific network organization. Integrating latent profile and network analyses showed how PID, PSR, and ADP may be configured and interconnected under a resource‐threatening work context.

1. Introduction

Workplace bullying has emerged as an important occupational health and workforce problem in healthcare settings [1, 2]. It is commonly understood as repeated exposure to negative acts, including psychological, verbal, or work‐related mistreatment, in situations where the target may have difficulty defending themselves [3]. Unlike violence initiated by patients or family members, workplace bullying is embedded in workplace relationships and organizational routines, making it difficult to recognize, report, and manage [4]. Previous studies have shown that such mistreatment can weaken nurses’ perceived safety, belonging, and professional identity (PID), while increasing anxiety, depression, burnout, and turnover intention among nursing staff [5, 6]. These consequences may threaten workforce stability and patient care quality, particularly in clinical environments where teamwork and continuity of care are essential.

Junior nurses may be especially vulnerable because they have limited clinical experience, fewer coping resources, and professional identities that are still developing [7]. This vulnerability is shaped by the early stage of professional socialization. Junior nurses are still consolidating their clinical confidence, role expectations, team relationships, and sense of organizational belonging. Compared with experienced nurses, they may have fewer accumulated coping strategies, less familiarity with informal workplace norms, and lower bargaining power in hierarchical clinical environments. Exposure to workplace bullying during this period may, therefore, be associated with psychological distress, weakened PID formation, and poorer occupational adaptation [8]. Understanding how junior nurses maintain and organize internal resources after such exposure is relevant to early‐career workforce development and retention.

Research on workplace violence and bullying in nursing has mainly examined adverse psychological and behavioral outcomes, including anxiety, depression, post‐traumatic stress symptoms, burnout, job stress, and turnover intention [9]. Network analysis has also been used to examine how negative symptoms and work‐related outcomes are connected after workplace adversity. For example, previous studies have investigated associations among workplace violence, burnout, depressive symptoms, suicidality, and turnover intention in healthcare workers, as well as symptom networks of anxiety, depression, and post‐traumatic stress among nurses exposed to workplace mistreatment [10, 11]. These studies have improved understanding of maladaptive responses. However, less is known about the positive internal resources that may support continued professional functioning after workplace bullying. Adaptation in this context cannot be fully explained by distress‐related variables alone because nurses may differ in how professional, psychological, and behavioral resources are preserved and organized.

Conservation of Resources (COR) theory provides a suitable framework for examining this issue [12, 13]. The theory proposes that stress arises when valued resources are threatened, lost, or insufficiently replenished. Workplace bullying can, therefore, be viewed as a resource‐threatening work condition in which interpersonal mistreatment challenges resources related to professional belonging, self‐regulation, and adaptive functioning. Under such conditions, junior nurses’ adaptation may depend on the level of available resources and whether these resources remain fragmented, moderately maintained, or more strongly integrated. This perspective directs attention from isolated psychological variables to broader patterns of resource configuration.

Within this resource‐oriented framework, PID, psychological resilience (PSR), and adaptive performance (ADP) represent interrelated domains of occupational adaptation. PID provides an identity anchor through which junior nurses derive professional value, role belonging, social recognition, and commitment to nursing [14]. In the context of workplace bullying, this resource is especially relevant because exclusion, humiliation, unfair criticism, or devaluation may challenge nurses’ sense of professional belonging [15]. PSR reflects the capacity to maintain emotional regulation, perceived control, and problem solving under adversity, thereby supporting the preservation of internal resources under stressful interpersonal conditions [16]. ADP represents the behavioral expression of adaptation, including maintaining composure under pressure, responding flexibly to urgent demands, and adjusting to interpersonal and organizational complexity [17]. Together, these constructs form a resource‐adaptation system that links professional self‐definition, self‐regulatory capacity, and adaptive work behavior.

A resource‐adaptation perspective requires an analytic approach that can capture both heterogeneity between nurses and the organization of resources within nurses. Traditional variable‐centered analyses can estimate average associations among PID, PSR, and ADP, but they provide limited information about whether these resources co‐occur in distinct patterns across individuals. For junior nurses exposed to workplace bullying, an important question is whether internal resources form qualitatively different profiles, such as constrained, moderate, or more favorable resource‐adaptation patterns. Latent profile analysis (LPA) addresses this question by identifying subgroups with distinct configurations across multiple resource indicators, thereby complementing conventional total‐score approaches [18, 19].

Profile membership alone, however, does not indicate how resources are connected within each subgroup. Two profiles may differ in mean resource levels, but they may also differ in the strength, selectivity, and cross‐domain organization of associations among specific indicators. Network analysis can, therefore, extend person‐centered findings by estimating the conditional association structure of PID, PSR, and ADP within each latent profile [20, 21]. From a COR perspective, higher global strength may indicate closer resource integration, bridge expected influence may identify indicators linking identity‐related, resilience‐related, and ADP domains, and negative edges may provide preliminary evidence of possible resource tensions or trade‐offs [22].

To address these issues, this study integrated LPA and network analysis to examine resource‐adaptation patterns among junior nurses exposed to workplace bullying. The aims were to (1) identify latent profiles based on item‐level indicators of PID, PSR, and ADP; (2) examine demographic, work‐related, and health‐related factors associated with profile membership; (3) compare profile‐specific network structures, including global strength, node centrality, bridge expected influence, and network invariance; and (4) evaluate the robustness of the findings through regression diagnostics, hospital‐cluster sensitivity analysis, residualized networks, and bootstrap stability analyses.

2. Methods

2.1. Study Design, Participants, and Ethical Considerations

This multicenter cross‐sectional study was guided by the STROBE reporting standards. Data were collected between September 2021 and October 2022 from junior registered nurses in 90 tertiary‐level hospitals in Sichuan Province, China. Junior nurses were operationally defined as registered nurses with five or fewer years of professional experience and no prior formal employment history before entering their current hospital, excluding clinical internships.

The inclusion criteria were (1) total working duration of 5 years or less; (2) possession of a valid nursing license; (3) voluntary participation with electronic informed consent; (4) age 18 years or older; and (5) exposure to workplace bullying, as determined by the criterion described in Section 2.2.2. Exclusion criteria included major stressful events during the preceding 6 months, such as death of an immediate family member, major surgery, or serious illness, and current use of psychotropic medication or psychotherapy.

A total of 7231 junior registered nurses were recruited. Among them, 5839 nurses met the workplace bullying exposure criterion and were included in the final analytic sample. Participants were required to provide a valid response to each questionnaire item before submission; therefore, no item‐level missing data were retained in the analytic dataset.

This study was approved by the Ethics Committee of the Affiliated Hospital of North Sichuan Medical College (Approval No. 2021ER006‐1). All participants signed electronic informed consent voluntarily before participation. Participant privacy was protected throughout the study. Personal data were collected solely for research purposes and stored securely on password‐protected servers with access restricted to the research team. All personally identifiable information was destroyed upon the study’s completion to ensure long‐term confidentiality. Each research team member signed a formal confidentiality agreement to guarantee ethical handling of sensitive information.

2.2. Measures

2.2.1. The General Information Questionnaire

The general information questionnaire for junior nurses included 21 variables: age, gender, household registration, marital status, number of cohabiting family members, religious affiliation, educational background, hospital level, monthly individual income, monthly per capita family income, years of work experience, weekly day shift frequency, weekly night shift frequency, average daily sleep duration, subjective sleep quality, history of occupational exposure (e.g., needle stick injury or radiation), smoking status, alcohol consumption, regular physical exercise, chronic disease diagnosis in the past year, and history of childhood parental absence due to labor migration.

2.2.2. The Six‐Item Workplace Bullying Scale

Workplace bullying was assessed using six items adapted from Attell et al. [23]. The items were translated and culturally adapted for the present Chinese nursing sample. The six items assess distinct negative workplace behaviors: excessive work demands, close supervision, teasing or insults, extra workload, being undervalued by supervisors, and unfair humiliation. Items are rated on a 5‐point Likert scale (1 = rarely occurs and 5 = frequently occurs), yielding total scores from 6 to 30. Higher scores reflect greater exposure to workplace bullying. In this study, exposure was defined as a score of ≥ 3 (“sometimes” or more often) on at least one item. This item‐level criterion was used to capture behaviorally diverse bullying experiences and to avoid excluding nurses who face concentrated but meaningful exposure to a specific negative act. It also aligns with prior workplace bullying research that uses exposure to at least one negative act as a classification criterion when no universally accepted cutoff exists [24]. The original study reported a Cronbach’s α of 0.81, and the Cronbach’s α in the present study was 0.812. The scale also showed acceptable construct validity in the present sample: KMO = 0.810, Bartlett’s test χ
2 = 3391.273, and p < 0.001; exploratory factor analysis extracted one factor explaining 53.958% of the variance; and CFA showed χ
2 = 139.488, df = 6, RMSEA = 0.087, CFI = 0.995, TLI = 0.987, and SRMR = 0.023.

2.2.3. The Nine‐Item Professional Identity Scale

The Professional Identity Scale [25] was used to evaluate nurses’ PID. It contains nine items rated on a 5‐point Likert scale, with total scores ranging from 9 to 45. Higher scores show stronger PID. The original study reported a Cronbach’s α of 0.938, and the Cronbach’s α in the present study was 0.908. The scale also showed acceptable construct validity in the present sample: KMO = 0.912, Bartlett’s test χ
2 = 7489.275, and p < 0.001; exploratory factor analysis extracted two factors explaining 69.357% of the variance; and CFA showed CFI = 0.982, TLI = 0.972, and SRMR = 0.031.

2.2.4. The Fourteen‐Item Psychological Resilience Scale

PSR was measured with the Chinese version of the 14‐item Resilience Scale [26]. Each item was rated on a 5‐point Likert scale, and the total score ranged from 14 to 70. Higher scores indicate better PSR. Previous studies reported a Cronbach’s α of 0.928, and the Cronbach’s α in the present study was 0.935. The scale showed acceptable construct validity in the present sample: KMO = 0.951, Bartlett’s test χ
2 = 12,415.856, and p < 0.001; exploratory factor analysis extracted two factors explaining 61.883% of the variance; and CFA showed CFI = 0.976, TLI = 0.971, and SRMR = 0.037.

2.2.5. The Thirteen‐Item Adaptive Performance Scale

ADP was measured using selected dimensions from the scale developed by Tao and Wang [27]. The original scale includes four dimensions: stress and emergency response, innovative problem solving, continuous learning at work, and cultural‐interpersonal facilitation. In the present study, two dimensions were used: stress and emergency response and cultural‐interpersonal facilitation. These two dimensions were selected because the study focused on internal resource‐adaptation configurations among junior nurses exposed to workplace bullying. Stress and emergency response reflect nurses’ ability to regulate emotions and respond effectively to high‐pressure, urgent, or unexpected clinical demands, whereas cultural‐interpersonal facilitation reflects their ability to adapt to interpersonal and team‐based work contexts. These two dimensions are, therefore, closely aligned with the occupational adaptation processes most relevant to workplace bullying exposure, which is embedded in workplace interactions and may be associated with nurses’ emotional, interpersonal, and clinical coping resources. The scale included 13 items rated on a 5‐point Likert scale, with total scores between 13 and 65. Higher scores indicate stronger ADP. The Cronbach’s α in the present study was 0.944. The scale showed acceptable construct validity in the present sample: KMO = 0.926, Bartlett’s test χ
2 = 13,534.472, and p < 0.001; exploratory factor analysis extracted two factors explaining 67.900% of the variance; and CFA showed CFI = 0.974, TLI = 0.967, and SRMR = 0.048.

Because all study variables were measured using self‐report questionnaires, common method bias was assessed using Harman’s single‐factor test. Six factors with eigenvalues greater than 1 were extracted without rotation, and the first factor explained 38.005% of the total variance, suggesting that common method bias was not a severe concern in this study. More details about the questionnaires used in this study and the brief meanings of variables can be found in Table S1.

2.3. Sample Size

Sample size adequacy was evaluated for both LPA and profile‐specific network analysis. Because there is no single universally accepted power formula for exploratory LPA with item‐level indicators, adequacy was assessed using several practical criteria: each retained profile should include at least 50 participants; the smallest profile should account for at least 5% of the analytic sample to support substantive interpretation; and the total sample size should be sufficient relative to the number of observed indicators [28].

To examine the network structure of three key resource‐adaptation domains, namely, PSR, PID, and ADP, among junior nurses exposed to workplace bullying, a total of 36 variables were included in the network model (9 for PID, 14 for PSR, and 13 for ADP). The initial cleaned dataset contained 7231 junior nurses with no missing values in the variables used for this analysis. Of these, 5839 met the workplace bullying exposure criterion and formed the final analytic sample. In the retained three‐profile solution, the smallest profile included 781 participants, representing 13.4% of the analytic sample. Thus, the sample size met the predefined adequacy criteria for LPA and supported the interpretability of all retained profiles.

Network sample size adequacy was further evaluated at the profile level because networks were estimated separately within each latent profile. Following recommendations for psychological network estimation, the number of parameters for a network with k nodes can be approximated as k + k × (k − 1)/2, including node parameters and unique pairwise associations [21]. With 36 nodes, this yielded 666 parameters. Even the smallest latent profile contained 781 participants, exceeding this parameter‐count benchmark for profile‐specific network estimation.

2.4. Statistical Analysis

2.4.1. LPA

LPA was conducted to examine heterogeneity in the resource‐adaptation profiles of junior nurses who had experienced workplace bullying. The model was estimated using Mplus (Version 8.3), with 36 continuous observed indicators. Item‐level indicators were used rather than total scale scores to preserve within‐construct heterogeneity and to ensure consistency with the nodes included in the subsequent network analysis [29, 30]. This approach also avoided the additional assumptions imposed by sum scoring, which may obscure item‐specific variation [31]. Because all indicators used 5‐point Likert‐type response formats, the items were entered on their original metric and treated as approximately continuous. This treatment is consistent with original quantitative applications of workplace bullying measures [23] and with methodological evidence suggesting that ordered variables with five or more categories can often be treated as approximately continuous under appropriate conditions [32]. The implications of this analytic choice were addressed in the limitations section.

Model estimation began with a one‐profile solution and proceeded sequentially to a six‐profile solution. To reduce the risk of local maxima, 200 random starts and 20 optimization iterations were specified. Model fit was evaluated using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted BIC (aBIC), entropy, the Lo–Mendell–Rubin (LMR) adjusted likelihood ratio test, and the Bootstrap Likelihood Ratio Test (BLRT). Lower information criteria indicated better fit, entropy closer to one indicated better classification accuracy, and significant LMR and BLRT values (p < 0.05) suggested improvement over the model with one fewer profile [18]. Average posterior probabilities were also inspected to assess classification certainty. The final profile solution was selected based on statistical fit, classification quality, profile size, substantive interpretability, and consistency with the resource‐adaptation framework.

2.4.2. Demographic Analysis

Baseline characteristics of the three latent profile groups were analyzed using the tableone (Version 0.9.5) Python package [33]. Continuous variables were reported as mean ± standard deviation when normally distributed or as medians with interquartile ranges when nonnormally distributed. Categorical variables were summarized using frequencies and percentages. Normality was assessed using the scipy.stats.normaltest function in the SciPy (Version 1.11.3) Python package [34], with a significance threshold set at α = 0.05. Differences between latent groups were examined using the Kruskal–Wallis test for continuous variables and the chi‐square test for categorical variables.

To identify key factors associated with profile membership, ordinal logistic regression was performed using the OrderedModel method from the Statsmodels (Version 0.14.4) Python package [35]. Continuous predictors were standardized using z-scores before entering the regression model, and categorical predictors were dummy‐coded. The proportional‐odds assumption was evaluated using a global Brant‐style parallelism test. For ordered logistic regression, latent profile membership was reverse‐coded so that higher outcome values represented membership in a more resource‐constrained profile; therefore, odds ratios (ORs) greater than one indicated higher odds of belonging to a more resource‐constrained profile. Results were presented as ORs with 95% confidence intervals. All predictors from the ordered logistic model were displayed in the forest plot, and variables with p < 0.05 were selected as covariates for the residualized network sensitivity analyses.

To examine whether the ordered specification influenced the findings, multinomial logistic regression was conducted as a sensitivity analysis. Because nurses were recruited from multiple hospitals, an additional sensitivity analysis using hospital‐cluster robust standard errors was performed. This analysis was restricted to participants with valid hospital‐cluster information, resulting in a reduced analytic sample of 5682 nurses from 89 hospitals.

2.4.3. Network Analysis

This study applied network analysis using R (Version 4.4.1) to explore the interrelations among PID, PSR, and ADP in junior nurses exposed to workplace bullying. The network comprised 36 nodes, including nine indicators of PID, 14 of PSR, and 13 of ADP. Consistent with the LPA specification, 5‐point Likert‐type items were treated as approximately continuous variables in the primary network models. Network estimation and visualization were conducted separately for each latent profile using the estimateNetwork function and qgraph (Version 1.9.8) packages. The Extended Bayesian Information Criterion Graphical Lasso (EBICglasso) method was employed to estimate the networks [21]. This approach combines the Extended BIC with Lasso regularization to detect partial correlations among variables, allowing for the construction of sparse and stable network structures. Nodes in the resulting graphs represent specific psychological dimensions, and edges denote associations between them. Positive associations were indicated by blue solid lines, whereas negative associations were represented by red dashed lines. The thickness of the edges reflected the strength of the relationships.

To identify influential nodes, two centrality indices were calculated: strength centrality and expected influence. Strength centrality measured the total absolute weight of connections for each node, while expected influence accounted for both the direction and magnitude of these connections. Additionally, bridge centrality indices were calculated using the networktools (Version 1.6.0) package to determine which nodes served as bridges between different psychological domains. These included bridge strength and bridge expected influence.

Network stability was evaluated using two bootstrap procedures implemented in the bootnet package (Version 1.6). First, case‐dropping bootstrap analysis with 1000 iterations was used to assess the stability of centrality indices. The case‐dropping stability coefficient (CS coefficient) was calculated to quantify the maximum proportion of cases that could be dropped while retaining a correlation of at least 0.70 between the original and subset‐based centrality estimates. Second, nonparametric bootstrap analysis with 1000 iterations was used to estimate 95% confidence intervals for edge weights and to evaluate the precision of the estimated network structure.

Network structures were then compared across latent profiles using a permutation‐based framework implemented via the NetworkComparisonTest (Version 2.2.2) package. This approach assessed differences in overall network structure and global strength. Each pairwise comparison used 1000 permutations. For edge‐level comparisons, the Benjamini–Hochberg procedure was applied to control the false‐discovery rate [36].

In addition to the primary networks estimated from raw item scores, residualized network models were estimated as a sensitivity analysis. Each of the 36 psychological indicators was separately regressed on covariates significantly associated with profile membership in the ordered logistic regression, and the resulting residuals were used as adjusted network nodes. This approach was adopted to examine whether the primary network patterns remained stable after accounting for demographic and work‐related covariates.

3. Results

3.1. LPA

Based on the LPA results shown in Table 1, the three‐class solution was identified as the optimal model. Although models with four to six classes showed further reductions in AIC, BIC, and aBIC, these more complex solutions produced smaller and more fragmented subgroups without substantially changing the broad low‐to‐high resource‐adaptation gradient. Therefore, considering model parsimony, classification quality, subgroup size, and substantive interpretability, the three‐profile solution was retained. This model yielded an AIC of 392,451.468, a BIC of 393,425.626, and an aBIC of 392,961.681. It also demonstrated high classification accuracy, with an entropy value of 0.952. Results from the LMR adjusted likelihood ratio test and the BLRT were statistically significant, further supporting the three‐class model over the two‐class alternative. As shown in Table S2, the diagonal average posterior probabilities were 0.975, 0.981, and 0.971 for Classes 1–3, respectively, indicating good classification certainty.

TABLE 1Comparison of latent profile models’ goodness of fit.

Classes AIC BIC aBIC Entropy LMR (p) BLRT (p) Proportion (%)

1 471,466.997 471,947.404 471,718.609 — — — —

2 413,448.043 414,175.325 413,828.954 0.945 < 0.001 < 0.001 49.426/50.574

3 392,451.468 393,425.626 392,961.681 0.952 < 0.001 < 0.001 32.643/53.982/13.375

4 380,360.281 381,581.315 380,999.794 0.944 < 0.001 < 0.001 16.578/35.280/9.214/38.928

5 375,632.994 377,100.904 376,401.807 0.949 < 0.001 < 0.001 6.217/11.663/34.201/38.671/9.248

6 371,463.263 373,178.048 372,361.376 0.955 < 0.001 < 0.001 5.926/11.440/33.413/6.816/34.783/7.622

Figure 1 illustrates the average item‐level scores across subdimensions for the three identified latent profiles. As further summarized in Table 2, the profiles showed a monotonic gradient across PID, PSR, and ADP. Class 1 comprised 1906 nurses (32.64%) and demonstrated consistently lower mean scores in PID, PSR, and ADP than the other two profiles. The mean item scores for this profile were 3.051 for PID, 2.922 for PSR, and 3.314 for ADP, indicating a comparatively constrained internal resource‐adaptation pattern. This class was labeled the “resource‐constrained adaptive” profile. Class 2, which accounted for 3152 participants (53.98%), represented a “moderate resource‐adaptation” profile. Members of this class exhibited intermediate mean scores across all three psychological domains, with mean item scores of 3.742 for PID, 3.620 for PSR, and 3.894 for ADP. As the largest subgroup, it reflected a moderate and relatively balanced resource‐adaptation pattern. Class 3 consisted of nurses with the most favorable profile, labeled “high resource‐adaptation” profile. This group included 781 participants (13.38%) and showed the highest mean scores in PID, PSR, and ADP, with corresponding mean item scores of 4.457, 4.375, and 4.527, respectively.

TABLE 2Mean (SD) PID, PSR, and ADP scores across the three latent profiles.

Class Label N % PID, mean (SD) PSR, mean (SD) ADP, mean (SD)

1 Resource‐constrained adaptive profile 1906 32.643 3.051 (0.468) 2.922 (0.350) 3.314 (0.382)

2 Moderate resource‐adaptation profile 3152 53.982 3.742 (0.380) 3.620 (0.320) 3.894 (0.230)

3 High resource‐adaptation profile 781 13.375 4.457 (0.460) 4.375 (0.370) 4.527 (0.372)

3.2. Factors Associated With Latent Profile Membership

Univariate analyses were conducted to compare characteristics across the three latent profile groups, as shown in Table 3. Significant differences were observed in age (H = 6.275, p = 0.0434), family income (H = 11.300, p = 0.0035), individual income (H = 6.565, p = 0.0375), frequency of night shifts (H = 28.153, p < 0.001), sleep duration (H = 98.866, p < 0.001), gender (χ
2 = 9.886, p = 0.0071), childhood parental absence due to labor migration (χ
2 = 29.736, p < 0.001), marital status (χ
2 = 13.362, p = 0.0096), educational background (χ
2 = 32.913, p < 0.001), occupational exposure (χ
2 = 171.665, p < 0.001), weekly working hours (χ
2 = 72.886, p < 0.001), sleep quality (χ
2 = 352.862, p < 0.001), exercise habits (χ
2 = 299.098, p < 0.001), and chronic disease status (χ
2 = 21.802, p < 0.001). No significant between‐profile differences were observed for the number of cohabiting family members, day‐shift frequency, household registration, religious affiliation, hospital level, alcohol consumption, or smoking status.

TABLE 3Comparison of participant characteristics across latent groups.

Variable Class 1 Class 2 Class 3 Statistic p

Age 23.00 (22.00, 24.00) 23.00 (22.00, 24.00) 23.00 (22.00, 24.00) H = 6.275 0.0434

Family member 4.00 (3.00, 5.00) 4.00 (3.00, 5.00) 4.00 (3.00, 5.00) H = 3.345 0.1878

Family income 3000.00 (2000.00, 4000.00) 3000.00 (2000.00, 4000.00) 3000.00 (2000.00, 5000.00) H = 11.300 0.0035

Income 2200.00 (2000.00, 3000.00) 2300.00 (2000.00, 3000.00) 2300.00 (2000.00, 3000.00) H = 6.565 0.0375

Day shift 3.00 (3.00, 5.00) 3.00 (3.00, 5.00) 3.00 (3.00, 4.00) H = 1.378 0.502

Night shift 2.00 (1.00, 2.00) 2.00 (1.00, 2.00) 2.00 (1.00, 2.00) H = 28.153 < 0.001

Sleep time 7.00 (6.00, 7.00) 7.00 (6.00, 7.50) 7.00 (6.50, 8.00) H = 98.866 < 0.001

Gender       χ2 = 9.886 0.0071

 1 = Male 123 (6.5%) 202 (6.4%) 74 (9.5%)    

 2 = Female 1783 (93.5%) 2950 (93.6%) 707 (90.5%)    

Rural       χ2 = 1.783 0.41

 1 = Rural 1407 (73.8%) 2298 (72.9%) 557 (71.3%)    

 2 = Urban 499 (26.2%) 854 (27.1%) 224 (28.7%)    

LeftBy       χ2 = 29.736 < 0.001

 1 = Yes 1062 (55.7%) 1570 (49.8%) 352 (45.1%)    

 2 = No 844 (44.3%) 1582 (50.2%) 429 (54.9%)    

Religion       χ2 = 0.135 0.9348

 1 = Yes 109 (5.7%) 179 (5.7%) 47 (6.0%)    

 2 = No 1797 (94.3%) 2973 (94.3%) 734 (94.0%)    

Married       χ2 = 13.362 0.0096

 1 = Single 1773 (93.0%) 2884 (91.5%) 694 (88.9%)    

 2 = Married 125 (6.6%) 257 (8.2%) 83 (10.6%)    

 3 = Other 8 (0.4%) 11 (0.3%) 4 (0.5%)    

Level       χ2 = 3.350 0.1873

 1 = Grade 3B 295 (15.5%) 434 (13.8%) 105 (13.4%)    

 2 = Grade 3A 1611 (84.5%) 2718 (86.2%) 676 (86.6%)    

Education       χ2 = 32.913 < 0.001

 1 = Technical secondary school 59 (3.1%) 74 (2.3%) 18 (2.3%)    

 2 = Junior college 1314 (68.9%) 2342 (74.3%) 616 (78.9%)    

 3 = Bachelor’s degree or above 533 (28.0%) 736 (23.4%) 147 (18.8%)    

Exposure       χ2 = 171.665 < 0.001

 1 = Yes 1155 (60.6%) 1526 (48.4%) 263 (33.7%)    

 2 = No 751 (39.4%) 1626 (51.6%) 518 (66.3%)    

Work Time       χ2 = 72.886 < 0.001

 1 = Less than 41 400 (20.9%) 874 (27.7%) 272 (34.8%)    

 2 = 41–55 1183 (62.1%) 1874 (59.5%) 429 (55.0%)    

 3 = 56–70 251 (13.2%) 324 (10.3%) 61 (7.8%)    

 4 = More than 70 72 (3.8%) 80 (2.5%) 19 (2.4%)    

Sleep quality     χ2 = 352.862 < 0.001  

 1 = Self‐perceived good 218 (11.4%) 668 (21.2%) 264 (33.8%)    

 2 = Self‐perceived average 1147 (60.2%) 2059 (65.3%) 454 (58.1%)    

 3 = Self‐perceived poor 541 (28.4%) 425 (13.5%) 63 (8.1%)    

Alcohol       χ2 = 3.366 0.1858

 1 = Yes 49 (2.6%) 63 (2.0%) 12 (1.5%)    

 2 = No 1857 (97.4%) 3089 (98.0%) 769 (98.5%)    

Smoke       χ2 = 0.788 0.6744

 1 = Yes 29 (1.5%) 56 (1.8%) 11 (1.4%)    

 2 = No 1877 (98.5%) 3096 (98.2%) 770 (98.6%)    

Exercise       χ2 = 299.098 < 0.001

 1 = Yes 403 (21.1%) 1166 (37.0%) 426 (54.5%)    

 2 = No 1503 (78.9%) 1986 (63.0%) 355 (45.5%)    

Chronic disease     χ2 = 21.802 < 0.001

 1 = Yes 92 (4.8%) 94 (3.0%) 12 (1.5%)    

 2 = No 1814 (95.2%) 3058 (97.0%) 769 (98.5%)    

An ordinal logistic regression analysis was performed to further identify factors associated with ordered profile membership. The proportional‐odds assumption was evaluated using a global Brant‐style parallelism test (χ
2 = 18.508, df = 26, p = 0.857), as shown in Table S4. In this model, ORs greater than 1 indicated greater odds of membership in a more resource‐constrained profile. As shown in Figure 2, more frequent night shifts, longer weekly working hours, fair or poor sleep quality, and lack of regular exercise were associated with greater odds of membership in a more resource‐constrained profile. Among these factors, poor sleep quality showed the strongest association (OR = 3.475, 95% CI: 2.883–4.188, p < 0.001), followed by lack of regular exercise (OR = 2.241, 95% CI: 2.005–2.506, p < 0.001). In contrast, higher individual income, being married, working in a higher‐level hospital, higher educational background, absence of occupational exposure, and absence of chronic disease were associated with lower odds of membership in a more resource‐constrained profile. Detailed ordered and multinomial logistic regression results are reported in Tables S3 and S5. Variables significantly associated with profile membership were retained as covariates for the residualized network sensitivity analyses.

3.3. Network Structure Across Different Groups

3.3.1. Network Plot

Figure 3 presents a comparative visualization of the network structures across the three latent profile groups. All networks were plotted using the same layout to ensure comparability. The primary differences were observed in connectivity patterns and edge weights. The strongest connections tended to cluster within the same psychological domain—either PID, PSR, or ADP—rather than across domains. This pattern indicated a high degree of internal cohesion within each construct.

Table 4 summarizes the global characteristics of the class‐specific networks. Although all three networks contained the same 36 nodes, their overall connectivity patterns differed. The resource‐constrained adaptive profile had the largest number of nonzero edges (214) and the highest density (0.340), followed by the moderate resource‐adaptation profile (206 nonzero edges; density = 0.327). The high resource‐adaptation profile had fewer nonzero edges (189) and the lowest density (0.300), but it showed the highest global strength (16.223) and the largest mean absolute nonzero edge weight (0.086). This pattern suggests that the high resource‐adaptation profile was characterized by fewer but stronger contemporaneous associations among PID, PSR, and ADP indicators.

TABLE 4Global characteristics of the class‐specific networks.

Class Nodes Nonzero edges Density Global strength Mean absolute nonzero edge Positive edge proportion Negative edge proportion

1 36 214 0.340 15.854 0.074 0.314 0.025

2 36 206 0.327 14.804 0.072 0.284 0.043

3 36 189 0.300 16.223 0.086 0.284 0.016

Positive edges were more common than negative edges in all three networks. The resource‐constrained adaptive profile showed the highest positive edge proportion (0.314), whereas the moderate resource‐adaptation and high resource‐adaptation profiles had the same positive edge proportion (0.284). Negative edges were relatively infrequent, with the highest negative edge proportion observed in the moderate resource‐adaptation profile (0.043) and the lowest in the high resource‐adaptation profile (0.016). These results indicate that the three profiles differed not only in mean levels of internal resources but also in the density, strength, and valence distribution of their resource‐adaptation networks.

3.3.2. Network Centrality Indices

Figure 4 compares node‐level centrality patterns across the three latent profiles. To facilitate visual comparison, node centrality estimates from each profile‐specific network were displayed in the same graph, with separate lines representing the three profiles. Overall, the centrality patterns showed profile‐specific variation, indicating that the most influential nodes were not uniformly distributed across the three resource‐adaptation profiles.

For node‐level centrality, relatively higher values were mainly observed for nodes reflecting problem‐solving, self‐regulation, calm coping, and adaptive adjustment. These included PSR indicators such as PSR14 (Problem solving), PSR3 (Calm handling), and PSR5 (Control ability), as well as ADP indicators related to emergency response and behavioral regulation, such as ADP4 (Clear thinking) and ADP10 (Behavioral adjustment). Several PID nodes, including PID2 (Social confidence) and PID8 (Social interaction), also showed visible centrality in specific profiles. This pattern indicates that profile‐specific networks were organized not only around resilience‐related capacities but also around interpersonal confidence and adaptive task regulation.

Bridge centrality results are shown in Figure 5, and the top bridge expected influence nodes for each profile are summarized in Table 5. In the resource‐constrained adaptive profile, the leading bridge nodes were PSR2 (Achievement pride), PID8 (Social interaction), PID2 (Social confidence), PSR3 (Calm handling), and PSR1 (Problem solving). In the moderate resource‐adaptation profile, the top bridge nodes were PID2 (Social confidence), PSR3 (Calm handling), PSR2 (Achievement pride), ADP4 (Clear thinking), and PID8 (Social interaction). In the high resource‐adaptation profile, the leading bridge nodes were PID2 (Social confidence), PSR5 (Control ability), PSR3 (Calm handling), PID8 (Social interaction), and PSR2 (Achievement pride).

TABLE 5Top bridge expected influence nodes in each latent profile.

Class Top bridge expected influence nodes

1 PSR2, PID8, PID2, PSR3, PSR1

2 PID2, PSR3, PSR2, ADP4, PID8

3 PID2, PSR5, PSR3, PID8, PSR2

Across the three profiles, PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) repeatedly appeared among the top bridge expected influence nodes. These results suggest that cross‐domain connectivity between PID, PSR, and ADP was primarily organized around interpersonal confidence, social engagement, achievement‐related self‐evaluation, and calm handling of situations. ADP4 (Clear thinking) appeared among the leading bridge nodes only in the moderate resource‐adaptation profile, whereas PSR5 (Control ability) appeared among the leading bridge nodes only in the high resource‐adaptation profile, indicating profile‐specific differences in how adaptive and resilience‐related indicators contributed to cross‐domain associations.

3.3.3. Network Comparison

Network comparison tests (NCTs) were conducted to compare the class‐specific networks in terms of global strength and overall network structure, as shown in Table 6. Consistent with the global network characteristics reported in Table 4, the moderate resource‐adaptation profile showed the lowest global strength (14.804), whereas the high resource‐adaptation profile showed the highest global strength (16.223).

TABLE 6Network comparison test results for class‐specific networks.

Comparison Global strength difference Global strength p
Maximum edge difference Network invariance p

Class 1 vs. Class 2 1.050 0.002 0.107 0.163

Class 1 vs. Class 3 0.369 0.372 0.176 0.031

Class 2 vs. Class 3 1.419 < 0.001 0.134 0.281

Note: Global strength difference refers to the absolute difference in global strength between two networks. Global strength p indicates the permutation‐based significance test for the global strength difference. Maximum edge difference refers to the largest absolute difference in corresponding edge weights between two networks. Network invariance p indicates the permutation‐based test of overall network structure invariance. p < 0.05 was considered statistically significant.

The comparison between the resource‐constrained adaptive profile and the moderate resource‐adaptation profile showed a significant global strength difference (difference = 1.050, p = 0.002), indicating stronger overall connectivity in the resource‐constrained adaptive profile. However, the overall network structure did not differ significantly between these two profiles (network invariance p = 0.163). Similarly, the comparison between the moderate resource‐adaptation profile and the high resource‐adaptation profile showed a significant global strength difference (difference = 1.419, p < 0.001), with stronger overall connectivity in the high resource‐adaptation profile, whereas the overall network structure was not significantly different (network invariance p = 0.281).

In contrast, the comparison between the resource‐constrained adaptive profile and the high resource‐adaptation profile did not show a significant global strength difference (difference = 0.369, p = 0.372), but the network invariance test was significant (p = 0.031). This finding suggests that these two profiles differed more in the organization of specific associations than in the overall magnitude of connectivity. The maximum edge differences ranged from 0.107 to 0.176 across pairwise comparisons, with the largest value observed between the resource‐constrained adaptive and high resource‐adaptation profiles.

3.4. Robustness Analysis

Several robustness analyses were conducted to examine the stability of the regression and network findings. First, regression diagnostics supported the ordered logistic model, with no evidence of violation of the proportional‐odds assumption. A multinomial logistic regression sensitivity analysis showed a broadly consistent pattern of associations, particularly for income, night‐shift frequency, occupational exposure, weekly working hours, sleep quality, and exercise. In addition, the hospital‐cluster robust sensitivity analysis showed that the direction of ORs remained unchanged and that most statistically significant associations were preserved (Table S6).

Second, residualized network models were estimated after adjusting PID, PSR, and ADP items for covariates significantly associated with profile membership. As shown in Figures S1–S3, the residualized networks illustrate patterns broadly consistent with the base networks. Class 2 retained the weakest overall connectivity, whereas Class 3 retained the highest global strength and the largest mean absolute nonzero edge weight (Table S7). Residualized NCTs also supported the main findings: Class 1 versus Class 2 and Class 2 versus Class 3 differed significantly in global strength, whereas Class 1 versus Class 3 showed evidence of network‐structure difference without a significant global‐strength difference (Table S8). The leading bridge expected‐influence nodes are summarized in Table S9 and Figure S4, and complete centrality comparisons are shown in Figures S5 and S6. These results indicate that the main profile‐specific network patterns were not materially altered after covariate adjustment.

Third, bootstrap stability analyses supported the robustness of centrality estimation. In the base networks, the CS coefficients for strength and expected influence were 0.750 across all three profiles. The same coefficients were observed in the residualized networks, indicating that standard centrality stability was preserved after covariate adjustment. Bridge centrality also showed acceptable stability, with CS coefficients ranging from 0.594 to 0.750 across base and residualized networks. Nonparametric bootstrap analyses indicated wider uncertainty around individual edge weights, supporting our decision to emphasize global strength, network invariance, and bridge‐centrality patterns rather than individual edges. Detailed robustness results are provided in Tables S10–S11 and Figures S7–S24.

4. Discussion

4.1. Principal Findings

This multicenter cross‐sectional study identified three ordered resource‐adaptation profiles among 5839 junior nurses exposed to workplace bullying: the resource‐constrained adaptive profile, the moderate resource‐adaptation profile, and the high resource‐adaptation profile. These profiles differed not only in the mean levels of PID, PSR, and ADP but also in the organization of associations among these resources. The high resource‐adaptation profile showed the highest global strength and the largest mean absolute nonzero edge weight, despite having fewer nonzero edges and lower network density. In contrast, the moderate resource‐adaptation profile showed the weakest overall connectivity. Bridge expected influence analyses further indicated that PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) repeatedly served as cross‐domain connectors between PID, PSR, and ADP. Robustness analyses, including ordered‐model diagnostics, multinomial logistic sensitivity analysis, hospital‐cluster robust adjustment, residualized networks, and bootstrap stability tests, generally supported the stability of the main findings.

Taken together, these findings extend the understanding of junior nurses’ psychological adaptation after workplace bullying by showing that internal resources are not merely high or low in level but are configured and interconnected in profile‐specific ways. This pattern is theoretically compatible with the COR theory, which emphasizes that stress‐related adaptation depends on how individuals conserve, mobilize, and invest available resources under threatening conditions.

4.2. Resource‐Adaptation Profiles Among Junior Nurses Exposed to Workplace Bullying

From the perspective of COR theory, workplace bullying can be conceptualized as a resource‐threatening work context in which interpersonal mistreatment may challenge junior nurses’ identity‐related, emotional‐cognitive, and behavioral adaptation resources. In this study, PID represented an identity‐related resource, PSR represented an emotional‐cognitive self‐regulatory resource, and ADP represented a behavioral adaptation resource. The three latent profiles therefore should not be understood as purely empirical categories; rather, they represent different configurations of internal resources among junior nurses facing workplace bullying. This interpretation is consistent with evidence that workplace harassment and nurse bullying are linked to both individual well‐being and organizational functioning, while also reflecting hostile workplace conditions that require personal and organizational responses [37, 38].

The resource‐constrained adaptive profile was characterized by comparatively lower levels of PID, PSR, and ADP. This pattern may reflect a constrained internal resource configuration in which junior nurses retain some capacity for task adaptation but have fewer identity‐related and resilience‐related resources available to support sustained occupational adjustment. Within a COR framework, such a profile may be consistent with limited resource conservation capacity under continuing interpersonal stress. Previous studies of new graduate and novice nurses have similarly shown that workplace bullying is associated with burnout, reduced work productivity, and impaired early‐career functioning [39, 40].

The moderate resource‐adaptation profile, which represented the largest subgroup, showed intermediate levels across all three domains. This profile may represent a state of functional resource maintenance, in which junior nurses maintain adequate professional functioning but do not show the stronger resource integration observed in the high resource‐adaptation profile. This interpretation is important because moderate resource levels do not necessarily indicate a stable or optimal resource system. Rather, these nurses may occupy an intermediate position in which additional contextual demands, such as shift burden or poor recovery conditions, could be associated with movement toward a more constrained configuration, whereas adequate support may be associated with stronger resource consolidation. This interpretation is also consistent with evidence that burnout and fatigue among nurses are shaped by workload, shift work, recovery opportunities, and organizational conditions [41, 42].

The high resource‐adaptation profile showed the highest mean scores across PID, PSR, and ADP. From the COR perspective, this pattern may reflect stronger resource conservation and resource mobilization, including a more coherent sense of professional value, greater self‐regulatory capacity, and more effective behavioral adaptation. This interpretation aligns with evidence that nurse resilience is a dynamic process supported by personal and contextual resources and that ADP involves adjusting effectively to changing task, interpersonal, and organizational demands [43, 44].

4.3. Network Organization, Bridge Nodes, and Resource Trade‐Offs

The integration of LPA and network analysis provided complementary information. LPA identified person‐centered heterogeneity in internal resource levels, whereas network analysis clarified how PID, PSR, and ADP were organized within each profile. This distinction is central to the contribution of the study: nurses in different profiles differed not only in how many internal resources they reported but also in how strongly and selectively these resources were connected [45, 46].

The high resource‐adaptation profile had fewer nonzero edges and lower density than the other two profiles, yet it showed the highest global strength and the largest mean absolute nonzero edge weight. This pattern suggests a more selective but stronger organization of internal resources. In COR terms, this may be compatible with more efficient resource integration, in which fewer but stronger connections support coordination across identity‐related, resilience‐related, and adaptive domains [47]. Conversely, the moderate resource‐adaptation profile showed the lowest global strength, suggesting that intermediate mean resource levels may coexist with weaker overall resource integration. The resource‐constrained adaptive profile showed relatively high density but lower mean edge strength than the high resource‐adaptation profile, which may indicate a broader but less selectively consolidated association pattern.

NCTs further supported this interpretation. The resource‐constrained adaptive and moderate resource‐adaptation profiles differed significantly in global strength, as did the moderate and high resource‐adaptation profiles. By contrast, the resource‐constrained adaptive and high resource‐adaptation profiles did not differ significantly in global strength but showed evidence of network‐structure difference. This suggests that the distinction between profiles cannot be reduced to overall connectivity alone; the organization of specific associations also matters [48].

Bridge centrality results highlighted several cross‐domain connectors. Across profiles, PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) repeatedly appeared among the leading bridge expected influence nodes. These nodes may represent points at which identity‐related resources, resilience‐related resources, and ADP are linked [49]. PID2 (Social confidence) and PID8 (Social interaction) indicate the interpersonal component of PID, suggesting that confidence in social engagement and willingness to interact may help connect professional self‐perception with resilience and adaptive functioning. PSR2 (Achievement pride) reflects achievement‐related self‐evaluation, whereas PSR3 (Calm handling) reflects emotional and cognitive regulation under challenging conditions. Their repeated bridge roles suggest that resource integration among junior nurses exposed to workplace bullying may rely partly on interpersonal confidence, social engagement, achievement‐related meaning, and calm coping.

Profile‐specific bridge patterns also provided additional insight. ADP4 (Clear thinking) appeared among the leading bridge nodes only in the moderate resource‐adaptation profile, suggesting that cognitive organization during urgent situations may be particularly relevant when resources are present but not highly integrated [50]. PSR5 (Control ability) appeared among the leading bridge nodes only in the high resource‐adaptation profile, suggesting that perceived control may become more central to cross‐domain resource coordination when the overall resource configuration is more favorable. These interpretations should be regarded as hypothesis‐generating and require confirmation in longitudinal or intervention studies.

The negative edges observed in the networks were relatively infrequent but theoretically meaningful. The moderate resource‐adaptation profile showed the highest negative edge proportion, whereas the high resource‐adaptation profile showed the lowest. From a resource‐oriented perspective, negative associations may be compatible with potential resource trade‐offs, in which investment in one adaptive function is associated with lower activation of another resource under certain conditions. For example, behavioral adjustment or interpersonal adaptation may require self‐regulatory effort that is not always aligned with perceived control or achievement‐related self‐evaluation. However, the bootstrap analyses indicated that individual edge weights should be interpreted cautiously. Therefore, these negative associations should be treated as preliminary indicators of possible resource tensions rather than evidence of stable causal trade‐off mechanisms [51].

4.4. Contextual Factors Associated With Profile Membership

The regression findings suggest that latent profile membership was associated with several contextual resource conditions. More frequent night shifts, longer weekly working hours, poorer sleep quality, and lack of regular exercise were associated with greater odds of membership in a more resource‐constrained profile. These factors can be interpreted as workload‐related demands and recovery‐related constraints. Recent nursing research has shown that shift work and night‐shift burden are closely linked to sleep disturbance and poorer recovery, while reviews of nurses’ mental health have also identified shift work, job demands, and insufficient job resources as important contextual risks [52]. From the COR perspective, repeated shift burden and insufficient recovery may be associated with reduced capacity to conserve or replenish internal resources, especially among junior nurses who are still developing professional confidence and workplace adaptation skills.

In contrast, higher individual income, being married, working in a higher‐level hospital, higher educational background, absence of occupational exposure, and absence of chronic disease were associated with lower odds of membership in a more resource‐constrained profile. These variables may reflect socioeconomic, interpersonal, institutional, and health‐related resource conditions. For example, higher income and educational background may indicate greater access to material and developmental resources; marital support may provide emotional resources; higher‐level hospitals may offer more structured training and organizational support; and absence of occupational exposure or chronic disease may indicate fewer additional resource burdens. Recent studies also suggest that perceived social support, PID, organizational support, and resilience are closely linked to nurses’ psychological functioning and intention to stay [50, 53]. Occupational exposure and work‐related injury may add further psychological and safety‐related burdens, particularly when combined with shift‐related sleep disruption or reduced job control [54]. These associations should not be interpreted as causal predictors, but they help contextualize how internal resource profiles may be embedded within broader working and living conditions.

The robustness analyses support a cautious interpretation of these contextual findings. The ordered logistic model satisfied the proportional‐odds assumption, and the multinomial logistic sensitivity analysis showed broadly consistent directions of association. After hospital‐cluster robust adjustment, the direction of ORs remained unchanged, and most statistically significant associations were preserved, although some associations, such as sleep time and hospital level, became nonsignificant. These findings suggest that the main contextual patterns were generally stable, while also indicating that hospital‐level clustering and measurement variation should be considered when interpreting institutional and recovery‐related factors. This caution is consistent with recent multicenter nursing research showing that nurses are often nested within institutional environments, where staffing structures, work environments, and organizational practices may shape workforce outcomes [55, 56]. Therefore, contextual associations should be interpreted as profile‐related correlates within a clustered cross‐sectional design rather than as evidence of individual‐ or hospital‐level causal effects [57].

4.5. Practical Implications

The findings have several implications for nursing management. First, subgroup identification may help nurse managers recognize that junior nurses exposed to workplace bullying are not psychologically homogeneous. Nurses in the resource‐constrained adaptive profile may require earlier and more structured support aimed at reducing resource depletion and strengthening basic recovery conditions. Such support may include workload review, more careful night‐shift scheduling, postexposure support after occupational incidents, sleep and recovery promotion, and accessible psychological consultation.

Second, nurses in the moderate resource‐adaptation profile should not be overlooked. Although this was the largest group and showed intermediate levels of internal resources, it also had the weakest global network strength. This suggests that moderate resource levels may not necessarily correspond to strong resource integration. Managerial strategies for this group may focus on consolidating existing resources through mentoring, peer support, reflective practice, and structured feedback that reinforces professional confidence and adaptive problem handling. Formal transition programs and structured mentorship have been shown to support new nurses’ confidence, clinical competence, communication, and retention, which makes them particularly suitable for nurses whose resources are present but weakly integrated [58]. Peer‐based and reflective approaches may also help normalize help‐seeking, strengthen workplace belonging, and provide opportunities to rehearse adaptive responses to difficult interpersonal or clinical situations [59].

Third, the high resource‐adaptation profile may provide useful information about favorable resource organization. These nurses showed stronger overall network connectivity despite fewer retained edges, suggesting a more selective and integrated resource structure. However, they should not simply be viewed as self‐sufficient. Sustaining their resources may require opportunities for professional development, recognition, and appropriate participation in peer‐support roles without creating additional role overload.

Finally, bridge nodes such as PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) may inform future resource‐oriented intervention design. Programs that strengthen interpersonal confidence, promote constructive social interaction, recognize professional achievements, and support calm coping under pressure may be particularly relevant. For example, interventions could combine communication‐skills training, structured peer interaction, recognition of professional contribution, and stress‐management exercises rather than focusing on a single resource domain [60]. Nevertheless, because the present study used cross‐sectional data, these nodes should be regarded as hypothesis‐generating indicators rather than confirmed intervention targets. Future longitudinal and intervention studies are needed to test whether modifying these domains can produce sustained improvements in resource configuration and occupational adaptation [61, 62].

4.6. Strengths and Limitations

This study has several strengths. First, it focused on a large multicenter sample of junior nurses exposed to workplace bullying, a group that is particularly relevant to workforce stability and early‐career professional development. Second, by combining LPA with network analysis, the study integrated person‐centered and variable‐centered approaches, allowing resource heterogeneity and resource organization to be examined simultaneously. Third, the study strengthened methodological transparency by reporting ordered‐model diagnostics, multinomial logistic sensitivity analysis, hospital‐cluster robust adjustment, residualized network models, NCTs, and bootstrap stability evidence. The CS coefficients for strength and expected influence were stable across both base and residualized networks, supporting the reliability of the centrality patterns reported in this study.

Several limitations should also be acknowledged. First, the cross‐sectional design precludes conclusions about temporal ordering, causal pathways, or resource gain and loss spirals. Therefore, the COR‐informed interpretations should be understood as theoretically plausible explanations rather than evidence of dynamic resource processes. Second, all measures were based on self‐report questionnaires, which may introduce common method variance and response bias, although procedural safeguards and Harman’s single‐factor test suggested that common method bias was not a severe concern. In addition, five‐point Likert‐type items were treated as approximately continuous, and ADP was assessed using selected dimensions rather than the full multidimensional construct. Finally, the generalizability and institutional interpretation of the findings are limited. The sample was drawn from tertiary hospitals in Sichuan Province, and the hospital‐cluster robust analysis was conducted in a slightly reduced sample after cleaning hospital identifiers.

5. Conclusion

This study identified three ordered resource‐adaptation profiles among junior nurses exposed to workplace bullying and showed that these profiles differed not only in the mean levels of PID, PSR, and ADP but also in the structural organization of these resources. The high resource‐adaptation profile was characterized by stronger overall network connectivity and more selective retained associations, whereas the moderate resource‐adaptation profile showed the weakest global strength. Bridge nodes including PID2 (Social confidence), PID8 (Social interaction), PSR2 (Achievement pride), and PSR3 (Calm handling) may represent important cross‐domain connectors among identity‐related, resilience‐related, and adaptive resources.

By integrating LPA with network analysis, this study contributes to a more nuanced COR‐informed understanding of how internal resources may be configured and interconnected among junior nurses facing workplace bullying. Methodologically, the findings demonstrate the value of combining person‐centered and network approaches to examine both resource heterogeneity and resource organization. Practically, the results support tailored managerial strategies that address workload demands, recovery conditions, professional confidence, social interaction, achievement recognition, and calm coping. Given the cross‐sectional and self‐report design, these findings should be interpreted cautiously and regarded as a basis for future longitudinal and intervention research rather than evidence of causal resource gain or loss processes.

Funding

This research was funded by the Humanities and Social Science Fund of Ministry of Education of China (No. 23YJC630130), the Scientific research and development plan of the Affiliated Hospital of North Sichuan Medical College (2023PTRJ001), and the development plan of North Sichuan Medical College (CBY23‐ZDB01).

Disclosure

The authors declare that the funding sources (Humanities and Social Science Fund of Ministry of Education of China, Scientific Research and Development Plan of the Affiliated Hospital of North Sichuan Medical College, and Development Plan of North Sichuan Medical College) had no involvement in study design, data collection, analysis, interpretation, or manuscript preparation.

Ethics Statement

This study was approved by the Ethics Committee of the Affiliated Hospital of North Sichuan Medical College (2021ER006‐1).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting Information

Additional supporting information can be found online in the Supporting Information section.

Supporting information

1Einarsen S. V. , Hoel H. , Zapf D. , and Cooper C. L. , Bullying and Harassment in the Workplace: Theory, Research and Practice , 2020 , 3rd edition, CRC Press.

2Machul M. ,
Krasucka K. N. ,
Pelc D. , and
Dziurka M. , Impact of Workplace Bullying on Nursing Care Quality: A Comprehensive Review , Medical Science Monitor . (2024 ) 30 , 10.12659/msm.944815 .PMC11305058 38915191

3Einarsen S. ,
Hoel H. , and
Notelaers G. , Measuring Exposure to Bullying and Harassment at Work: Validity, Factor Structure and Psychometric Properties of the Negative Acts Questionnaire-Revised , Work & Stress . (2009 ) 23 , no. 1, 24 –44 , 10.1080/02678370902815673 .

4Zhang D. ,
Yu Q. ,
Ning M. et al., Cohort Profile: The Nurses’ Mental Health Study , BMC Nursing . (2025 ) 25 , no. 1, 10.1186/s12912-025-04128-z .PMC12764096 41286936

5Galanis P. ,
Moisoglou I. ,
Katsiroumpa A. , and
Mastrogianni M. , Association Between Workplace Bullying, Job Stress, and Professional Quality of Life in Nurses: A Systematic Review and Meta-Analysis , Healthcare . (2024 ) 12 , no. 6, 10.3390/healthcare12060623 .PMC10970563 38540587

6Peng J. ,
Luo H. ,
Ma Q. et al., Association Between Workplace Bullying and Nurses’ Professional Quality of Life: The Mediating Role of Resilience , Journal of Nursing Management . (2022 ) 30 , no. 6, 1549 –1558 , 10.1111/jonm.13471 .34529304

7Kang K. and
Yu M. , Workplace Bullying, Hardiness, and Occupational Identity Among Nursing Students: Mediation of Academic Burnout , Journal of Nursing Education . (2024 ) 63 , no. 9, 604 –612 , 10.3928/01484834-20240508-04 .39237090

8Cho H. ,
Pavek K. , and
Steege L. , Workplace Verbal Abuse, Nurse-Reported Quality of Care and Patient Safety Outcomes Among Early-Career Hospital Nurses , Journal of Nursing Management . (2020 ) 28 , no. 6, 1250 –1258 .10.1111/jonm.13071 32564407

9Jang S. J. ,
Son Y.-J. , and
Lee H. , Prevalence, Associated Factors and Adverse Outcomes of Workplace Violence Towards Nurses in Psychiatric Settings: A Systematic Review , International Journal of Mental Health Nursing . (2022 ) 31 , no. 3, 450 –468 , 10.1111/inm.12951 .34773361

10Wei H. ,
Liu M. ,
Wang Z. et al., Anxiety, Depression, and Post-Traumatic Stress Disorder in Nurses Exposed to Horizontal Violence: A Network Analysis , BMC Nursing . (2024 ) 23 , no. 1, 10.1186/s12912-024-02408-8 .PMC11472536 39396956

11Yun J.-Y. ,
Myung S. J. , and
Kim K. S. , Associations Among the Workplace Violence, Burnout, Depressive Symptoms, Suicidality, and Turnover Intention in Training Physicians: A Network Analysis of Nationwide Survey , Scientific Reports . (2023 ) 13 , no. 1, 10.1038/s41598-023-44119-1 .PMC10556140 37798353

12Hobfoll S. E. , Conservation of Resources: A New Attempt at Conceptualizing Stress , American Psychologist . (1989 ) 44 , no. 3, 513 –524 , 10.1037/0003-066x.44.3.513 .2648906

13Hobfoll S. E. ,
Halbesleben J. ,
Neveu J.-P. , and
Westman M. , Conservation of Resources in the Organizational Context: The Reality of Resources and Their Consequences , Annual Review of Organizational Psychology and Organizational Behavior . (2018 ) 5 , no. 1, 103 –128 , 10.1146/annurev-orgpsych-032117-104640 .

14Mao A. ,
Lu S. ,
Lin Y. , and
He M. , A Scoping Review on the Influencing Factors and Development Process of Professional Identity Among Nursing Students and Nurses , Journal of Professional Nursing . (2021 ) 37 , no. 2, 391 –398 , 10.1016/j.profnurs.2020.04.018 .33867096

15Tong Y. ,
Wang X. ,
Lei C. et al., Workplace Bullying and Professional Identity Among Male Nurses: The Mediating Role of Work Engagement and the Moderating Effect of Fear of Negative Evaluation , BMC Nursing . (2025 ) 24 , no. 1, 10.1186/s12912-025-03933-w .PMC12534930 41107822

16Chen X. ,
Li Q. ,
Xu F. , and
Han B. , The Mediating Role of Resilience Between Work–family Conflict and Career Development Among Chinese Nurses: A Cross-Sectional Study , Journal of Nursing Management . (2021 ) 29 , no. 6, 1733 –1741 , 10.1111/jonm.13323 .33797812

17Pulakos E. D. ,
Arad S. ,
Donovan M. A. , and
Plamondon K. E. , Adaptability in the Workplace: Development of a Taxonomy of Adaptive Performance , Journal of Applied Psychology . (2000 ) 85 , no. 4, 612 –624 , 10.1037/0021-9010.85.4.612 .10948805

18Spurk D. ,
Hirschi A. ,
Wang M. ,
Valero D. , and
Kauffeld S. , Latent Profile Analysis: A Review and How to Guide of Its Application Within Vocational Behavior Research , Journal of Vocational Behavior . (2020 ) 120 , 10.1016/j.jvb.2020.103445 .

19Woo S. E. ,
Jebb A. T. ,
Tay L. , and
Parrigon S. , Putting the Person in the Center: Review and Synthesis of Person-Centered Approaches and Methods in Organizational Science , Organizational Research Methods . (2018 ) 21 , no. 4, 814 –845 .

20Borsboom D. and
Cramer A. O. , Network Analysis: An Integrative Approach to the Structure of Psychopathology , Annual Review of Clinical Psychology . (2013 ) 9 , no. 1, 91 –121 , 10.1146/annurev-clinpsy-050212-185608 .23537483

21Epskamp S. ,
Borsboom D. , and
Fried E. I. , Estimating Psychological Networks and Their Accuracy: A Tutorial Paper , Behavior Research Methods . (2018 ) 50 , no. 1, 195 –212 , 10.3758/s13428-017-0862-1 .PMC5809547 28342071

22Jones P. J. ,
Ma R. , and
McNally R. J. , Bridge Centrality: A Network Approach to Understanding Comorbidity , Multivariate Behavioral Research . (2021 ) 56 , no. 2, 353 –367 , 10.1080/00273171.2019.1614898 .31179765

23Attell B. K. ,
Brown K. K. , and
Treiber L. A. , Workplace Bullying, Perceived Job Stressors, and Psychological Distress: Gender and Race Differences in the Stress Process , Social Science Research . (2017 ) 65 , 210 –221 , 10.1016/j.ssresearch.2017.02.001 .28599773

24Nielsen M. B. ,
Finne L. B. ,
Parveen S. , and
Einarsen S. V. , Assessing Workplace Bullying and Its Outcomes: The Paradoxical Role of Perceived Power Imbalance Between Target and Perpetrator , Frontiers in Psychology . (2022 ) 13 , 10.3389/fpsyg.2022.907204 .PMC9237549 35774968

25Liu L. ,
Hao Y. F. , and
Liu X. H. , Development of Professional Identity Scale for Nurses (In Chinese) , Nursing Journal of Chinese People’s Liberation Army . (2011 ) 28 , no. 3, 18 –20 , 10.3969/j.issn.1008-9993.2011.03.006 .

26Ni Q. and
Tian J. , Evaluation on Reliability and Validity of 14-Item Resilience Scale (In Chinese) , Chinese Journal of Public Health . (2013 ) 29 , no. 10, 1524 –1527 .

27Tao Q. and
Wang C. , A Structural Analysis of Adaptive Performance in the Context of Management Training (In Chinese) , Psychological Science . (2006 ) 29 , no. 3, 614 –617 .

28Yang Y. ,
Lu J. ,
Dong Y. et al., A Latent Profile and Network Analysis of Social Isolation in Colorectal Cancer Patients Undergoing Chemotherapy , Asia-Pacific Journal of Oncology Nursing . (2025 ) 12 , 10.1016/j.apjon.2025.100731 .PMC12269414 40678770

29Borsboom D. ,
Deserno M. K. ,
Rhemtulla M. et al., Network Analysis of Multivariate Data in Psychological Science , Nature Reviews Methods Primers . (2021 ) 1 , no. 1, 10.1038/s43586-021-00055-w .

30Epskamp S. and
Fried E. I. , A Tutorial on Regularized Partial Correlation Networks , Psychological Methods . (2018 ) 23 , no. 4, 617 –634 , 10.1037/met0000167 .29595293

31McNeish D. and
Wolf M. G. , Thinking Twice About Sum Scores , Behavior Research Methods . (2020 ) 52 , no. 6, 2287 –2305 , 10.3758/s13428-020-01398-0 .32323277

32Rhemtulla M. ,
Brosseau-Liard P. É. , and
Savalei V. , When Can Categorical Variables Be Treated as Continuous? A Comparison of Robust Continuous and Categorical SEM Estimation Methods Under Suboptimal Conditions , Psychological Methods . (2012 ) 17 , no. 3, 354 –373 , 10.1037/a0029315 .22799625

33Pollard T. J. ,
Johnson A. E. ,
Raffa J. D. , and
Mark R. G. , Tableone: An Open Source Python Package for Producing Summary Statistics for Research Papers , JAMIA Open . (2018 ) 1 , no. 1, 26 –31 , 10.1093/jamiaopen/ooy012 .PMC6951995 31984317

34Virtanen P. ,
Gommers R. ,
Oliphant T. E. et al., Scipy 1.0: Fundamental Algorithms for Scientific Computing in Python , Nature Methods . (2020 ) 17 , no. 3, 261 –272 , 10.1038/s41592-019-0686-2 .PMC7056644 32015543

35Seabold S. and
Perktold J. , Statsmodels: Econometric and Statistical Modeling With Python , Proceedings of the 9th Python in Science Conference, 2010, 92 –96 .

36Benjamini Y. and
Hochberg Y. , Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing , Journal of the Royal Statistical Society: Series B . (1995 ) 57 , no. 1, 289 –300 , 10.1111/j.2517-6161.1995.tb02031.x .

37Bowling N. A. and
Beehr T. A. , Workplace Harassment From the Victim’s Perspective: A Theoretical Model and Meta-Analysis , Journal of Applied Psychology . (2006 ) 91 , no. 5, 998 –1012 , 10.1037/0021-9010.91.5.998 .16953764

38Vessey J. A. ,
DeMarco R. F. ,
Gaffney D. A. , and
Budin W. C. , Bullying of Staff Registered Nurses in the Workplace: A Preliminary Study for Developing Personal and Organizational Strategies for the Transformation of Hostile to Healthy Workplace Environments , Journal of Professional Nursing . (2009 ) 25 , no. 5, 299 –306 , 10.1016/j.profnurs.2009.01.022 .19751935

39Berry P. A. ,
Gillespie G. L. ,
Gates D. , and
Schafer J. , Novice Nurse Productivity Following Workplace Bullying , Journal of Nursing Scholarship . (2012 ) 44 , no. 1, 80 –87 , 10.1111/j.1547-5069.2011.01436.x .PMC12126179 22339938

40Laschinger H. K. S. ,
Grau A. L. ,
Finegan J. , and
Wilk P. , New Graduate Nurses’ Experiences of Bullying and Burnout in Hospital Settings , Journal of Advanced Nursing . (2010 ) 66 , no. 12, 2732 –2742 , 10.1111/j.1365-2648.2010.05420.x .20722806

41Caruso C. C. , Negative Impacts of Shiftwork and Long Work Hours , Rehabilitation Nursing Journal . (2014 ) 39 , no. 1, 16 –25 , 10.1002/rnj.107 .PMC4629843 23780784

42Dall’Ora C. ,
Ball J. ,
Reinius M. , and
Griffiths P. , Burnout in Nursing: A Theoretical Review , Human Resources for Health . (2020 ) 18 , no. 1, 10.1186/s12960-020-00469-9 .PMC7273381 32503559

43Cooper A. L. ,
Brown J. A. ,
Rees C. S. , and
Leslie G. D. , Nurse Resilience: A Concept Analysis , International Journal of Mental Health Nursing . (2020 ) 29 , no. 4, 553 –575 , 10.1111/inm.12721 .32227411

44Jundt D. K. ,
Shoss M. K. , and
Huang J. L. , Individual Adaptive Performance in Organizations: A Review , Journal of Organizational Behavior . (2015 ) 36 , no. S1, S53 –S71 , 10.1002/job.1955 .

45Cramer A. O. ,
Waldorp L. J. ,
Van Der Maas H. L. , and
Borsboom D. , Comorbidity: A Network Perspective , Behavioral and Brain Sciences . (2010 ) 33 , no. 2–3, 137 –150 , 10.1017/s0140525x09991567 .20584369

46Howard M. C. and
Hoffman M. E. , Variable-Centered, Person-Centered, and Person-Specific Approaches: Where Theory Meets the Method , Organizational Research Methods . (2018 ) 21 , no. 4, 846 –876 , 10.1177/1094428117744021 .

47Halbesleben J. R. ,
Neveu J.-P. ,
Paustian-Underdahl S. C. , and
Westman M. , Getting to the COR: Understanding the Role of Resources in Conservation of Resources Theory , Journal of Management . (2014 ) 40 , no. 5, 1334 –1364 .

48Van Borkulo C. D. ,
van Bork R. ,
Boschloo L. et al., Comparing Network Structures on Three Aspects: A Permutation Test , Psychological Methods . (2023 ) 28 , no. 6, 1273 –1285 , 10.1037/met0000476 .35404628

49Castro D. ,
Gysi D. ,
Ferreira F. ,
Ferreira-Santos F. , and
Ferreira T. B. , Centrality Measures in Psychological Networks: A Simulation Study on Identifying Effective Treatment Targets , PLoS One . (2024 ) 19 , no. 2, 10.1371/journal.pone.0297058 .PMC10903921 38422083

50Tao J. ,
Wang Y. ,
Chen M. ,
Yan Z. , and
Huang P. , Chain Mediating Effect of Perceived Organizational Support and Resilience on Workplace Violence and Perceived Professional Identity of Emergency Nurses: A Structural Equation Model Analysis , BMC Nursing . (2025 ) 24 , no. 1, 10.1186/s12912-025-03350-z .PMC12211311 40598342

51Schumacher L. ,
Burger J. ,
Echterhoff J. , and
Kriston L. , Methodological and Statistical Practices of Using Symptom Networks to Evaluate Mental Health Interventions: A Review and Reflections , Multivariate Behavioral Research . (2024 ) 59 , no. 4, 663 –676 , 10.1080/00273171.2024.2335401 .38733300

52Yang J. ,
Chen Y. ,
Tian Y. et al., Risk Factors and Consequences of Mental Health Problems in Nurses: A Scoping Review of Cohort Studies , International Journal of Mental Health Nursing . (2024 ) 33 , no. 5, 1197 –1211 , 10.1111/inm.13337 .38622945

53Pu J. ,
Wang W. ,
Li G. et al., Psychological Resilience and Intention to Stay Among Nurses: The Mediating Role of Perceived Organizational Support , Frontiers in Psychology . (2024 ) 15 , 10.3389/fpsyg.2024.1407206 .PMC11464349 39391853

54Davis J. ,
Granger S. , and
Turner N. , Assessing Psychological Distress of Healthcare Workers With and Without Work Injuries: The Role of Job Control , Journal of Safety Research . (2024 ) 90 , 333 –340 , 10.1016/j.jsr.2024.07.001 .39251290

55van Kraaij J. ,
de Vries N. ,
Wessel H. et al., Enhancing Work Environments and Reducing Turnover Intention: A Multicenter Longitudinal Cohort Study on Differentiated Nursing Practices in Dutch Hospitals , BMC Nursing . (2025 ) 24 , no. 1, 10.1186/s12912-024-02681-7 .PMC11721496 39794770

56Zhang G. ,
Zou T. , and
Zhang M. , Horizontal Violence or Workplace Bullying Among Intensive Care Unit Nurses: A Mixed Systematic Review , Journal of Advanced Nursing . (2025 ) 81 , no. 12, 8394 –8408 , 10.1111/jan.16849 .PMC12623697 40008631

57Mahfodz M. M. ,
Siddiqui S. , and
Zainal H. , Knowledge, Attitude and Practices About Pharmacovigilance Activities Among Hospital Nurses: A Multicentre Cross-Sectional Survey , Journal of Research in Nursing . (2024 ) 29 , no. 3, 259 –274 , 10.1177/17449871241232340 .PMC11179598 38883248

58Rush K. L. ,
Janke R. ,
Duchscher J. E. ,
Phillips R. , and
Kaur S. , Best Practices of Formal New Graduate Transition Programs: An Integrative Review , International Journal of Nursing Studies . (2019 ) 94 , 139 –158 , 10.1016/j.ijnurstu.2019.02.010 .30965203

59Armstrong N. , Management of Nursing Workplace Incivility in the Health Care Settings: A Systematic Review , Workplace Health & Safety . (2018 ) 66 , no. 8, 403 –410 , 10.1177/2165079918771106 .29792131

60Spiva L. ,
Davis S. ,
Case-Wirth J. et al., The Effectiveness of Charge Nurse Training on Leadership Style and Resiliency , Journal of Nursing Administration . (2020 ) 50 , no. 2, 95 –103 , 10.1097/nna.0000000000000848 .31977947

61Bringmann L. F. ,
Elmer T. ,
Epskamp S. et al., What Do Centrality Measures Measure in Psychological Networks? , Journal of Abnormal Psychology . (2019 ) 128 , no. 8, 892 –903 , 10.1037/abn0000446 .31318245

62Fried E. I. and
Cramer A. O. , Moving Forward: Challenges and Directions for Psychopathological Network Theory and Methodology , Perspectives on Psychological Science . (2017 ) 12 , no. 6, 999 –1020 , 10.1177/1745691617705892 .28873325