Negative parenting styles and disordered eating: an item-level bridge network and simulation-based intervention analysis

Abstract

Background

Negative parenting styles are known correlates of disordered eating (DE), yet existing studies have largely focused on construct-level associations, leaving the specific item-level pathways between parenting styles and DE unclear. Identifying these fine-grained mechanisms may offer more targeted insights for prevention and intervention.

Methods

A large sample of Chinese college students (N = 2,008) completed the short-form Egna Minnen av Barndoms Uppfostran for Chinese (s-EMBU-C; paternal rejection, maternal rejection, paternal overprotection, and maternal overprotection subscales) and the Eating Attitudes Test–19 (EAT-19). Applying Gaussian Graphical Models, item-level bridge network analyses were conducted to estimate conditional associations (edges) among individual parenting and DE items (nodes), and Bridge Expected Influence was used to identify nodes that link the two communities. A simulation-based intervention analysis was then conducted to examine how constraining individual bridge nodes—and their sequential removal—reduced cross-community connectivity, quantified using the CrossCut (Cross-Community Coupling) metric.

Results

For rejection, bridge nodes included unexplained parental anger, being treated as a scapegoat, and shaming treatment (plus maternal public criticism), along with food avoidance despite hunger and preoccupation with body fat. For overprotection, bridges involved excessive parental worry, exaggerated safety anxiety, and intrusive interference (plus maternal control over appearance), together with desire for an empty stomach and perceived thinness by others. Simulation analyses showed that clamping individual bridge nodes reduced the CrossCut by approximately 13–30%, whereas cumulative clamping produced substantial overall reductions (up to 76–83%), yet displayed clear marginal effects.

Conclusions

These findings highlight that hostile, shaming parental treatments and anxious, intrusive over-involvement, intertwined with restrictive and body-image focused symptoms, constitute the core pathways linking negative parenting styles and DE. Targeting high-impact bridge nodes, rather than the full set, was sufficient to substantially weaken the connection between the parenting and DE communities. These insights provide a useful reference for developing more efficient support strategies for college students.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40337-026-01554-5.

Plain language summary

This study explored how negative parenting styles relate to disordered eating (DE) among 2,008 Chinese college students. Unlike many previous studies, we examined these links at the item level using network analysis, allowing us to identify the specific parenting behaviors and DE symptoms that play key bridging roles. The results revealed that hostile, shaming, or intrusive parenting behaviors (such as unexplained anger and over-control) and restrictive eating or body image concerns form the core pathways connecting negative parenting with DE.

We further tested these key nodes using a simulation-based intervention approach. The results showed that a small number of parenting-related bridge nodes play a disproportionately important role: addressing just these nodes was sufficient to greatly weaken the overall connection between negative parenting and DE. These insights provide a valuable reference for developing more efficient support strategies for college students.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40337-026-01554-5.

Introduction

Disordered eating (DE) refers to maladaptive attitudes and behaviors surrounding food and eating that deviate from healthy norms [1, 2]. Although often considered less severe than clinical eating disorders, DE can still lead to notable impairments in physical and psychological well-being [3–6]. DE has been increasingly recognized as a prevalent concern among college students worldwide [7–9]. In China, epidemiological surveys have documented a steady rise in DE prevalence—from 2.5% in 2010 [10] to 4.5% in 2015 [11], and reaching 10.4% by 2023 [12]—highlighting an alarming upward trend that merits sustained academic and public health attention.

A growing body of research has highlighted the role of family environment—particularly parenting styles—in shaping individual development, with their influence extending into early adulthood [13]. Within this context, negative parenting styles, such as parental rejection and overprotection, have been identified as key psychosocial factors associated with maladaptive eating attitudes and behaviors [14–16]. Such parenting patterns may disrupt emotion regulation and cognitive processing [17, 18], both of which play central roles in the development and maintenance of DE [19]. Despite consistent evidence for their associations at the construct level, little is known about how specific aspects of negative parenting relate to particular eating-related symptoms at the item level. Item-level analyses can reveal the specific parental behaviors and symptom expressions that most directly maintain maladaptive eating patterns; therefore, addressing this limitation is essential for identifying precise targets of intervention.

Traditional variable-centered approaches, such as correlation and regression analyses, tend to treat parenting styles and DE as homogeneous constructs, thereby obscuring the heterogeneity among individual items within each domain. In contrast, network analysis conceptualizes psychological phenomena as systems of interacting components, allowing for the identification of specific item-level associations that may drive co-occurring patterns [20, 21].

Within this framework, bridge network analysis provides a refined approach for examining how distinct item clusters—here, negative parenting styles and DE symptoms—are interconnected through key bridging nodes [22]. These bridge nodes represent potential pathways of influence or vulnerability that link parental dynamics with maladaptive eating patterns, offering a more granular understanding of their co-development. Building upon this perspective, the present study applied bridge network modeling to identify specific cross-community connections between negative parenting styles and DE. Unlike traditional network studies that remain at a descriptive level by identifying central nodes [23, 24], the current study further employed a simulated intervention analysis to quantify how modifying these bridge nodes may reduce cross-community connectivity, thereby providing a data-driven basis for targeted intervention design.

The present study adopted an item-level network approach to identify the specific cross-community links between negative parenting items and DE symptoms. Four bridge networks were estimated to examine the associations between paternal rejection, maternal rejection, paternal overprotection, and maternal overprotection with DE symptoms, using data from the Short-Form Egna Minnen av Barndoms Uppfostran for Chinese (s-EMBU-C) and the Eating Attitudes Test–19 (EAT-19) scales. A Gaussian Graphical Model (GGM) with EBICglasso regularization [25] was applied, and bridge expected influence (EI) was calculated to identify the most influential nodes linking the two communities. Network stability was examined via case-dropping bootstraps, and sensitivity analyses were performed across different tuning parameters and correlation estimation methods. To move beyond descriptive network analysis, a simulated intervention analysis was conducted to quantify how modifying bridge nodes could reduce cross-community connectivity, indexed by the CrossCut (Cross-Community Coupling) metric, with its absolute form (CrossCut_abs) used in the present study.

The aims of the present study were twofold:

(1) To identify item-level bridge nodes that link specific aspects of negative parenting styles with DE symptoms; and.

(2) To evaluate, through simulated intervention analysis, how targeted constraining of these bridge nodes may reduce the overall cross-community connectivity between parenting and DE.

Method

Participants

A convenience sampling method was employed to recruit undergraduate students from seven colleges in southern China. The survey was administered as part of a routine campus mental-health screening initiative.

Data collection was conducted during scheduled class sessions with the assistance of cooperating faculty members. Students completed the survey anonymously via the Wenjuanxing online platform using their mobile devices. Participation was voluntary, and no exclusion criteria were applied regarding gender, major, or academic year. To protect participant anonymity, specific university affiliations were not collected; therefore, the exact distribution of participants across the seven colleges was not recorded.

Instruments

Negative parenting styles

Parenting styles were measured using the Short-Form Egna Minnen av Barndoms Uppfostran for Chinese (s-EMBU-C), originally developed by Arrindell et al. [26] and later adapted for Chinese populations by Jiang et al. [27]. The Chinese version includes 21 items for each parent (paternal and maternal), sharing identical content across three dimensions: Rejection (6 items), Emotional Warmth (7 items), and Overprotection (8 items). Each item is rated on a four-point Likert scale ranging from 1 (never) to 4 (always).

In the present study, only the negative parenting styles dimensions were included for analysis: paternal rejection (items PR1–PR6), maternal rejection (items MR1–MR6), paternal overprotection (items PO1–PO8), and maternal overprotection (items MO1–MO8) [27]. Higher scores on each subscale reflect a greater perceived use of that parenting behavior by the parents, as recalled by the participants. The Cronbach’s alpha coefficients for the selected dimensions ranged from 0.75 to 0.85, indicating satisfactory internal consistency.

Disordered eating

Disordered eating was assessed using the Eating Attitudes Test (EAT), a self-report measure originally developed by Garner and Garfinkel to evaluate eating-related attitudes and behaviors [28]. The EAT is a six-point forced-choice Likert scale consisting of 26 items, initially rated from 1 (never) to 6 (always). Following standard scoring conventions, responses were recoded into a 0–3 scale: choices 1 through 3 were coded as 0, while choices 4, 5, and 6 were coded as 1, 2, and 3, respectively. Higher scores indicate greater eating-related concerns or symptom severity. The scale has been widely used in both clinical and non-clinical populations [29].

The instrument was later introduced to China by Wang et al. who conducted cultural adaptation and psychometric validation in a non-clinical university sample [30]. Their analysis retained 19 items with the strongest reliability and factor stability, resulting in the EAT-19, an adapted version suitable for Chinese populations. The EAT-19 is designed to assess DE tendencies but does not include an empirically validated cutoff score for diagnostic purposes. In the present study, the EAT-19 demonstrated good internal consistency, with a Cronbach’s alpha of 0.79.

Statistical analysis

Bridge network analysis

The item-level network structure of negative parenting styles and DE was estimated using a Gaussian Graphical Model (GGM). In this model, edges represent regularized partial correlation coefficients between nodes, controlling for the influence of all other nodes in the network.

The network was estimated using the EBICglasso algorithm implemented in the R package qgraph (version 1.9.8; [31]). This algorithm applies LASSO regularization in combination with the Extended Bayesian Information Criterion (EBIC) for model selection. Crucially, this method inherently handles variable selection by shrinking trivial or spurious partial correlations to exact zeros. Consequently, all edges retained in the final network represent robust associations, rendering traditional significance testing (e.g., p-values) redundant [25]. The tuning parameter (γ) was set to 0.5 to balance sensitivity and specificity [21]. This model served as the baseline network for subsequent analyses. All analyses were conducted in R (version 4.4.3).

In this study, Bridge Expected Influence (EI) was used as the primary centrality index to quantify the importance of each node in linking the two predefined psychological domains: negative parenting styles (including rejection and overprotection) and DE. Bridge EI represents the sum of all positive and negative edge weights connecting a node to nodes in other communities, thus capturing its overall influence on cross-community connectivity [22]. Bridge centrality values were computed using the R package networktools (version 1.6.0), and the top 20% of nodes with the highest Bridge EI were identified as bridge nodes.

The stability and accuracy of the bridge network were evaluated using nonparametric bootstrapping procedures with the R package bootnet (version 1.6; [21]). Specifically, the correlation stability coefficient (CS-C) was computed to assess the robustness of Bridge EI estimates under case-dropping. A total of 1,000 bootstrap samples were generated, and CS-C values were interpreted according to standard thresholds, with values above 0.25 considered acceptable and values above 0.50 preferred [21].

For sample size adequacy in the network analysis, the largest network in this study contained 27 nodes (8 parental overprotection items and 19 DE items), requiring the estimation of 378 parameters in total . Our sample size exceeded this threshold, indicating that the data were sufficient for obtaining a stable and reliable GGM estimation [21, 32].

Sensitivity analysis

To evaluate the robustness of bridge identification, a series of sensitivity analyses were conducted based on the estimated GGM. Building on the baseline network, Bridge EI values were recalculated across varying EBIC hyperparameters (γ = 0.10, 0.25, 0.50) to examine the stability of node rankings under different levels of network sparsity. Additionally, correlation estimation methods were varied between Spearman and nonparanormal (NPN) correlations to assess the influence of distributional assumptions. For each condition, node-wise Bridge EI values were ranked and compared with the baseline configuration using Spearman rank correlations and the overlap in identified bridge nodes.

In addition to assessing network stability, we also conducted supplementary analyses to explore potential gender heterogeneity. Independent samples t-tests were performed to compare the mean scores of all study variables between male and female participants.

Simulated intervention analysis

To evaluate how specific bridge nodes contribute to cross-community connectivity between negative parenting styles and DE, a simulated intervention analysis was performed on the estimated network W. This approach follows recent developments in network psychometrics that conceptualize targeted changes in network topology as analogs of psychological intervention or system perturbation [22, 33, 34]. All analyses were implemented in R using dplyr (version 1.1.4) for data manipulation and ggplot2 (version 3.5.1) for visualization.

Cross-community connectivity metric

Cross-community connectivity was quantified using a Cross-Community Coupling (CrossCut) index, defined as the total absolute edge weight between the two predefined communities—Negative Parenting Styles (P) and DE:

Whererepresents the weight of the edge (i.e., the regularized partial correlation coefficient) between node i in community P and node j in community DE. Hereafter, “CrossCut” refers to this absolute-value metric.

This index represents the overall magnitude of structural connections between the two systems, computed from absolute edge weights to capture total cross-community strength regardless of direction. In the present GGM, this metric functionally reflects the aggregate structural coupling between the negative parenting styles and DE subnetworks.

Intervention design and evaluation

A clamp procedure was used to simulate node-specific interventions by removing the influence of selected nodes (i.e., setting the corresponding rows and columns in W to zero; [33]).

Candidate intervention targets were the bridge nodes identified based on Bridge EI. For each target node, the post-intervention network was recalculated, and the CrossCut value was recomputed. The relative reduction ratio in cross-community connectivity was then calculated as:

This produced a single-node reduction estimate for each bridge node.

To evaluate cumulative effects, nodes were subsequently clamped sequentially in descending order of their individual reduction ratio, and cumulative decreases in CrossCut were recorded at each step. Accordingly, the CrossCut metric served as the principal outcome indicator, quantifying how targeted removal of bridge nodes progressively weakened the total coupling between the negative parenting styles and DE domains.

Results

Participant characteristics

A total of 2,417 questionnaires were distributed, and 2,008 valid responses were retained after data screening, yielding an effective response rate of 83.1%. Participants had a mean age of 19.51 years (SD = 1.29), and 46.7% were female. Regarding family structure, among the participants who provided valid responses for this item (n = 1,951), the vast majority (93.3%, n = 1,820) were from intact families, while 6.7% (n = 131) were from single-parent families. The remaining 57 participants were either from other family structures or did not provide this information.

Descriptive statistics

Table 1 presents the internal consistency reliability (Cronbach’s alpha) for the five study dimensions (parenting styles and disordered eating), alongside the item descriptions and descriptive statistics (Mean and SD) for each individual item.

Table 1.

Descriptive statistics and internal consistency of study variables and items

Variable	Description	Mean (SD)	Cronbach’s alpha
Paternal Rejection (PR)		8.82 (2.78)	0.85
PR1	Unexplained parental anger	1.46 (0.62)
PR2	Excessive punishment	1.41 (0.64)
PR3	Public criticism and humiliation	1.66 (0.76)
PR4	Being treated as a scapegoat	1.35 (0.61)
PR5	Shaming treatment	1.44 (0.66)
PR6	Harsh punishment for trifles	1.50 (0.64)
Maternal Rejection (MR)		8.99 (2.83)	0.84
MR1	Unexplained parental anger	1.49 (0.63)
MR2	Excessive punishment	1.41 (0.64)
MR3	Public criticism and humiliation	1.75 (0.79)
MR4	Being treated as a scapegoat	1.38 (0.65)
MR5	Shaming treatment	1.46 (0.65)
MR6	Harsh punishment for trifles	1.51 (0.64)
Paternal Overprotection (PO)		15.98 (3.75)	0.75
PO1	Excessive parental worry	2.59 (0.87)
PO2	Strict monitoring and control	1.83 (0.79)
PO3	Restrictive protection	2.06 (0.82)
PO4	Control over appearance	1.56 (0.73)
PO5	Exaggerated safety anxiety	1.90 (0.77)
PO6	Freedom of movement	2.56 (0.89)
PO7	Intrusive interference	1.63 (0.67)
PO8	Rigid limits and boundaries	1.85 (0.78)
Maternal Overprotection (MO)		16.68 (3.85)	0.75
MO1	Excessive parental worry	2.65 (0.86)
MO2	Strict monitoring and control	1.97 (0.81)
MO3	Restrictive protection	2.16 (0.83)
MO4	Control over appearance	1.77 (0.80)
MO5	Exaggerated safety anxiety	1.99 (0.79)
MO6	Freedom of movement	2.58 (0.87)
MO7	Intrusive interference	1.70 (0.69)
MO8	Rigid limits and boundaries	1.87 (0.78)
Disordered Eating (DE)		4.44 (5.64)	0.79
DE1	Fear of weight gain	0.45 (0.91)
DE2	Food avoidance despite hunger	0.15 (0.51)
DE3	Preoccupation with food	0.52 (0.94)
DE4	Binge eating with loss of control	0.11 (0.48)
DE5	Avoidance of carbohydrates	0.08 (0.40)
DE6	Vomiting after eating	0.04 (0.27)
DE7	Guilt after eating	0.10 (0.47)
DE8	Desire for thinness	0.73 (1.14)
DE9	Exercise for calorie burning	0.33 (0.76)
DE10	Perceived thinness by others	0.55 (1.01)
DE11	Preoccupation with body fat	0.44 (0.90)
DE12	Avoidance of sugar	0.15 (0.54)
DE13	Consumption of diet foods	0.07 (0.35)
DE14	Feeling controlled by food	0.20 (0.65)
DE15	Excessive focus on food	0.20 (0.59)
DE16	Discomfort after eating sweets	0.08 (0.40)
DE17	Engagement in dieting	0.10 (0.44)
DE18	Desire for empty stomach	0.10 (0.44)
DE19	Impulse to vomit	0.04 (0.28)

Gender differences in the mean scores of the study variables are presented in Supplementary Table S1. At the dimension level, results revealed significant gender differences in PR (t = 2.19, p = .029, d = 0.10, very small effect) and DE (t = -8.75, p < .001, d = 0.40, small-to-medium effect). Specifically, male students reported slightly higher levels of PR, while female students reported significantly higher levels of DE symptoms. No significant gender differences were found for MR, PO, or MO. However, at the item level, significant differences were observed in 19 out of 47 items; for detailed test statistics, please refer to Supplementary Table S1 (see columns ‘t-value’, ‘p-value’, and ‘Cohen’s d’ for item-level comparisons). Notably, even within dimensions that showed no overall gender difference (e.g., MR), specific items such as MR1 and MR3 differed significantly between genders.

Paternal rejection network analysis

A GGM network was estimated to examine the associations between paternal rejection items and DE symptoms. The resulting network contained 25 nodes and 132 non-zero edges, with a network density of 0.44 (see Fig. 1).

Fig. 1.

Bridge network linking paternal rejection and disordered eating communities. Orange nodes denote the top-20% bridge nodes (by Bridge EI). Blue nodes belong to the Paternal Rejection (PR) community; pink nodes belong to the Disordered Eating (DE) community. Blue and red edges represent positive and negative associations; edge thickness reflects the strength of the relationship. (See Supplementary Table S2 for exact edge weights)

Bridge network analysis identified five nodes (PR1, PR4, PR5, DE2, and DE10) as the top bridge nodes based on Bridge EI. The CS-C for Bridge EI was 0.283, indicating an acceptable level of stability for the bridge estimates. The case-dropping bootstrap plot (see Supplementary Figure S1) and the nonparametric bootstrap confidence intervals for edge weights (see Supplementary Figure S2) provide additional information regarding the robustness of the estimated bridge structure. Across sensitivity analyses, the identified bridge nodes remained among the top-ranked nodes across different γ values (0.10–0.50) and correlation estimation methods, indicating a stable bridge structure between paternal rejection and DE.

A simulated intervention analysis was performed using the five identified bridge nodes. Single-node clamping reduced cross-community connectivity by 14.3%–29.4% (see Fig. 5a), while cumulative clamping of all five nodes produced a 76.2% total reduction in cross-community connectivity (see Fig. 6a). Notably, adding DE2 to the clamped set did not further reduce the CrossCut value, indicating that its cross-community links overlapped with previously clamped nodes.

Fig. 5.

Single-node clamping results for the four parenting dimensions: a Paternal Rejection, b Maternal Rejection, c Paternal Overprotection, and d Maternal Overprotection. Values reflect the relative contribution of individual nodes in sustaining the linkage between each parenting dimension and DE. Higher bars indicate stronger influence. Abbreviations correspond to the items described in Table 1: PR1/MR1 (Unexplained parental anger); PR4/MR4 (Being treated as a scapegoat); PR5/MR5 (Shaming treatment); MR3 (Public criticism and humiliation); PO1/MO1 (Excessive parental worry); PO5/MO5 (Exaggerated safety anxiety); PO7/MO7 (Intrusive interference); MO4 (Control over appearance); DE2 (Food avoidance despite hunger); DE10 (Perceived thinness by others); DE11 (Preoccupation with body fat); DE18 (Desire for empty stomach)

Fig. 6.

Cumulative reduction curves for sequential clamping of bridge nodes across four parenting dimensions. a Paternal Rejection, node order: 1 = PR1 (Unexplained parental anger), 2 = PR4 (Being treated as a scapegoat), 3 = DE2 (Food avoidance despite hunger), 4 = PR5 (Shaming treatment), 5 = DE10 (Perceived thinness by others); b Maternal Rejection, node order: 1 = MR3 (Public criticism and humiliation), 2 = MR1 (Unexplained parental anger), 3 = MR5 (Shaming treatment), 4 = DE11 (Preoccupation with body fat), 5 = MR4 (Being treated as a scapegoat); c Paternal Overprotection, node order: 1 = PO5 (Exaggerated safety anxiety), 2 = PO7 (Intrusive interference), 3 = DE18 (Desire for empty stomach), 4 = PO1 (Excessive parental worry), 5 = DE10 (Perceived thinness by others); d Maternal Overprotection, node order: 1 = MO7 (Intrusive interference), 2 = MO5 (Exaggerated safety anxiety), 3 = DE10 (Perceived thinness by others), 4 = MO4 (Control over appearance), 5 = MO1 (Excessive parental worry). Higher cumulative reductions indicate stronger attenuation of cross-community connectivity as more bridge nodes are clamped

Maternal rejection network analysis

A GGM network was estimated to examine the associations between maternal rejection items and DE symptoms. The resulting network contained 25 nodes and 143 non-zero edges, with a network density of 0.48 (see Fig. 2).

Fig. 2.

Bridge network linking maternal rejection and disordered eating communities. Notes: Orage nodes denote the top-20% bridge nodes (by Bridge EI). Blue nodes belong to the Maternal Rejection (MR) community; pink nodes belong to the Disordered Eating (DE) community. Blue and red edges represent positive and negative associations; edge thickness reflects the strength of the relationship. (See Supplementary Table S3 for exact edge weights)

Bridge network analysis identified five nodes (MR5, MR1, MR3, DE15, and DE11) as the top bridge nodes based on Bridge EI. The CS-C for Bridge EI was 0.283. The case-dropping bootstrap plot (see Supplementary Figure S3) and the nonparametric bootstrap confidence intervals for edge weights (see Supplementary Figure S4) provide additional information regarding the robustness of the estimated bridge structure. Across sensitivity analyses, the same bridge nodes remained among the top-ranked nodes across different γ values (0.10–0.50) and correlation estimation methods, confirming a stable bridge structure linking maternal rejection and DE.

A simulated intervention analysis was performed using the five identified bridge nodes. Single-node clamping reduced cross-community connectivity by 14.2%–24.8% (see Fig. 5b), while cumulative clamping of all five nodes produced a 79.6% total reduction in cross-community connectivity (see Fig. 6b). Notably, adding DE11 to the clamped set did not further reduce the CrossCut value, indicating that its cross-community links overlapped with previously clamped nodes.

Paternal overprotection network analysis

A GGM network was estimated to examine the associations between paternal overprotection items and DE symptoms. The resulting network contained 27 nodes and 138 non-zero edges, with a network density of 0.39 (see Fig. 3).

Fig. 3.

Bridge network linking Paternal Overprotection and Disordered Eating communities. Orange nodes denote the top-20% bridge nodes (by Bridge EI). Blue nodes belong to the Paternal Overprotection (PO) community; pink nodes belong to the Disordered Eating (DE) community. Blue and red edges represent positive and negative associations; edge thickness reflects the strength of the relationship. (See Supplementary Table S4 for exact edge weights)

Bridge network analysis identified five nodes (PO5, PO1, PO7, DE2, and DE18) as the top bridge nodes based on Bridge EI. The CS-C for Bridge EI was 0.283, indicating acceptable stability of the bridge estimates. The case-dropping bootstrap plot (see Supplementary Figure S5) and the nonparametric bootstrap confidence intervals for edge weights (see Supplementary Figure S6) provide additional information regarding the robustness of the estimated bridge structure. Across sensitivity analyses, the same bridge nodes remained among the top-ranked nodes across different γ values (0.10–0.50) and correlation estimation methods, confirming a stable bridge structure linking paternal overprotection and DE.

A simulated intervention analysis was then conducted using these five bridge nodes. Single-node clamping reduced cross-community connectivity by 13.9%–29.0% (see Fig. 5c), while cumulative clamping of all five nodes produced an 83.4% total reduction in cross-community connectivity (see Fig. 6c).

Maternal overprotection network analysis

A GGM network was estimated to examine the associations between maternal overprotection items and DE symptoms. The resulting network contained 27 nodes and 148 non-zero edges, with a network density of 0.42 (see Fig. 4).

Fig. 4.

Bridge network linking Maternal Overprotection and Disordered Eating communities. Orange nodes denote the top-20% bridge nodes (by Bridge EI). Blue nodes belong to the Maternal Overprotection (MO) community; pink nodes belong to the Disordered Eating (DE) community. Blue and red edges represent positive and negative associations; edge thickness reflects the strength of the relationship. (See Supplementary Table S5 for exact edge weights)

Bridge network analysis identified five nodes (MO4, MO1, MO5, MO7, and DE10) as the top bridge nodes based on Bridge EI. The CS-C for Bridge EI was 0.283, indicating acceptable stability of the bridge estimates. The case-dropping bootstrap plot (see Supplementary Figure S7) and the nonparametric bootstrap confidence intervals for edge weights (see Supplementary Figure S8) provide additional information regarding the robustness of the estimated bridge structure. Across sensitivity analyses, these bridge nodes consistently ranked among the top positions across γ values (0.10–0.50) and correlation estimation methods, confirming a stable bridge structure linking maternal overprotection and DE.

A simulated intervention analysis was then conducted using these five bridge nodes. Single-node clamping reduced cross-community connectivity by 14.2%–24.1% (see Fig. 5d), while cumulative clamping of all five nodes produced a 76.7% total reduction in cross-community connectivity (see Fig. 6d).

Discussion

The present study employed bridge network analysis to examine how specific components of negative parenting styles are connected to distinct DE symptoms. Across the four parenting dimensions—paternal rejection, maternal rejection, paternal overprotection, and maternal overprotection—the networks revealed clear patterns of cross-community associations and identified a set of bridge nodes that played a central role in linking the two domains. Furthermore, simulated intervention analyses demonstrated that constraining these bridge nodes substantially reduced the overall structural coupling between negative parenting and DE symptoms.

Across the four parenting dimensions, the bridge nodes within the parenting community showed a high degree of convergence. In the rejection-related models, items characterizing unexplained parental anger (PR1/MR1), being treated as a scapegoat (PR4/MR4), shaming treatment (PR5/MR5), and public criticism (MR3) consistently emerged as key connectors to DE symptoms. These indicators describe caregiving marked by hostility, humiliation, and coercive emotional control.

A complementary configuration was observed in the overprotection-related networks. Bridge nodes clustered around intrusive interference (PO7/MO7), excessive parental worry (PO1/MO1), and exaggerated safety anxiety (PO5/MO5), with the specific addition of control over appearance (MO4) in the maternal network. These indicators capture specific facets of anxiety-driven intrusiveness and behavioral control.

Taken together, the item-level findings indicate that two specific relational processes constituted the primary pathways linking negative parenting to DE: (1) hostility, humiliation, and coercive emotional control, and (2) anxiety-driven intrusiveness. Emotionally dysregulated family climates impede the development of effective emotion regulation [35, 36]. From an emotion-regulation perspective, exposure to unexplained anger, public criticism, or being treated as a scapegoat may deprive individuals of opportunities to learn adaptive strategies, increasing reliance on behavioral regulation—specifically food avoidance and strict dietary control—to downregulate distress [37]. Likewise, parenting characterized by exaggerated safety anxiety and intrusive interference undermines autonomy and self-efficacy [38, 39], making control over appearance or maintaining an empty stomach more appealing as compensatory strategies to regain a sense of control [40].

This pattern was clearly reflected in the DE community. Across the parenting models, four symptoms served as bridge nodes: perceived thinness by others (DE10), food avoidance despite hunger (DE2), desire for an empty stomach (DE18), and, less consistently, preoccupation with body fat (DE11). Among these, DE10 was the most stable, appearing in three of the four networks. Notably, the core bridge symptoms—DE2, DE10, and DE18—shared a clear common feature: they reflected rigid physiological denial and sensitivity to external body evaluation. These patterns align naturally with the kinds of hostility and intrusive control identified on the parenting side. Rather than generalized dietary concerns, the emergence of these specific physiological suppression nodes suggests that the compensatory pursuit of emotional stability and personal agency is fought primarily through the direct conquest of bodily sensations [37, 41].

Beyond descriptive identification of bridge nodes, the present study applied a simulation-based intervention analysis to evaluate their structural influence on cross-community connectivity. In all four networks, clamping individual bridge nodes produced meaningful reductions in cross-community connection strength. The cumulative clamping results further showed that constraining these bridge nodes substantially reduced overall cross-community connectivity, indicating that a relatively small subset of nodes accounted for the majority of the coupling between negative parenting styles and DE.

A notable pattern in the cumulative curves was that the marginal reduction plateaued once the main parenting-side bridges had been clamped. Methodologically, this does not imply that DE nodes are unimportant or that parenting nodes serve as causal drivers of the system. Rather, it suggests that many DE-side bridges share overlapping cross-community connections with the parenting bridges. Once the key parenting nodes—and the edges associated with them—are removed, the remaining DE nodes have little unique cross-cluster connectivity left to disrupt, producing the observed flattening of the cumulative reduction curves. In this sense, the simulation reveals a degree of redundancy among bridge nodes. From an applied perspective, the results point to a clear pattern of diminishing marginal returns: once the most influential bridges are constrained, additional nodes contribute very little further reduction in cross-community connectivity. This pattern offers a practical heuristic for intervention planning, emphasizing that it is unnecessary to target every bridge node; the structurally dominant bridges account for most of the modifiable linkage, whereas later interventions yield progressively smaller benefits.

Implications

The findings of this study offer several implications for prevention and early intervention efforts targeting DE among college students.

First, the study identified two specific relational processes—hostile, shaming rejection and anxiety-driven intrusiveness—as the primary relational pathways linking negative parenting styles to DE. These results suggest that parenting interventions should move beyond broad calls to “improve parenting quality” and instead focus on modifying these concrete interaction patterns. In practice, parent-focused programs may incorporate brief, skills-based components such as helping parents regulate hostility and eliminate shaming interactions, manage their own exaggerated safety anxiety, and replace intrusive interference with more autonomy-supportive guidance; these targeted strategies can be delivered through campus-based psychoeducation, online modules, or community workshops to provide parents with concrete and accessible tools.

Second, the bridge nodes on the DE side primarily reflected rigid physiological suppression and body-image anxiety—specifically food avoidance despite hunger, desire for an empty stomach, and sensitivity to external evaluation. This pattern highlights the importance of addressing interoceptive disconnection in university-based psychological services. Counseling and health-promotion programs may incorporate modules that help students identify early signs of suppressing bodily signals (e.g., ignoring hunger), understand their links to stress regulation, and learn alternative strategies for managing distress or regaining a sense of control. Skills-based approaches—such as interoceptive awareness training, emotion-regulation training, and flexible goal-setting—may be particularly helpful for students who rely on controlling physiological states as a coping mechanism.

Third, the simulation-based intervention results revealed clear marginal effects: targeting only a small subset of structurally influential bridge nodes produced most of the attainable reduction in cross-community connectivity. This pattern suggests that, in practice, intervention efforts do not need to address every identified bridge symptom. Instead, prioritizing the most central bridge features—such as hostile, shaming rejection and anxiety-driven intrusiveness on the parenting side and rigid physiological suppression on the DE side—may yield efficient and meaningful improvements. This provides a useful heuristic for resource allocation in university mental-health services, where time and personnel are often limited: focusing on the most impactful relational and symptomatic targets may offer a pragmatic path for early prevention and selective intervention.

Taken together, these implications highlight the value of item-level network approaches for identifying precise leverage points in the prevention of DE. By clarifying the specific relational mechanisms that connect negative parenting to maladaptive eating behaviors, this study provides actionable guidance for parents, counselors, and university mental-health practitioners seeking to support students at risk of developing problematic eating patterns.

Limitations and prospects

Several limitations of the present study should be acknowledged. First, the study relied on cross-sectional self-report data, which may introduce response biases and does not permit causal inference regarding the relations between negative parenting styles and DE symptoms. Although the bridge network and simulation analyses provide structural insights, the identified associations remain correlational. Future longitudinal, experimental, or experience-sampling studies should validate the temporal sequencing and bidirectional influences between parenting and eating behaviors.

Second, the GGM with EBICglasso regularization assumes linear and relatively stable item associations, which may not fully capture the complexity of interactions between parenting behaviors and eating-related symptoms. Future studies may consider using non-linear, dynamic, or time-varying network models to better capture potential nonlinearities and fluctuations in cross-community associations.

Third, the bridge network utilized a mixed-gender sample. Although supplementary analyses indicated significant mean-level gender differences in specific items, our primary goal was to identify the general backbone of symptom interactions and stable intervention targets in a large-scale population. Prioritizing the combined sample maximized statistical power; splitting the data for gender-specific comparisons (e.g., via Network Comparison Test) would reduce precision and potentially obscure robust, trans-diagnostic pathways. Future studies with stratified samples are recommended to investigate structural gender differences.

Fourth, the simulated intervention analysis reflects in-silico perturbations rather than real behavioral change; therefore, the identified bridge nodes and reduction patterns should be interpreted as heuristic rather than prescriptive. Subsequent work could test these structural leverage points using experimental micro-interventions, laboratory-based perturbation tasks, or longitudinal intervention trials to evaluate their practical relevance.

Fifth, the EAT-19 is not designed to distinguish clinical subtypes (e.g., restrictive vs. binge patterns) and lacks empirically validated cutoff scores for Chinese samples. Consequently, no diagnostic classification was possible, and caution is warranted when generalizing these findings to specific clinical eating pathology. Future research should utilize multi-method assessments (e.g., clinical interviews) to develop culturally validated cutoff scores and enhance the clinical interpretability of the results.

Finally, this study utilized a convenience sampling method within a single cultural context (Chinese), and the sample was characterized by high demographic homogeneity (93.3% from intact families). These factors may limit the generalizability of our findings to other cultural backgrounds, age cohorts, or diverse family structures (e.g., single-parent families). Future research should employ more representative sampling strategies across cross-cultural and diverse demographic contexts—including adolescents and individuals from non-intact families—to verify the robustness of the identified bridge structures.

Conclusion

This study applied bridge network analysis and simulation-based interventions to elucidate how specific components of negative parenting styles connect to DE symptoms at the item level. The results identified two relational processes—hostile, shaming rejection and anxiety-driven intrusiveness—as the key parenting-side bridge nodes. On the DE side, the dominant bridge nodes involved rigid physiological suppression and body-image anxiety, specifically food avoidance despite hunger and desire for an empty stomach. Simulation findings further showed that targeting these bridge nodes substantially reduced cross-community connection, confirming a pattern of diminishing marginal returns where a small subset of influential nodes accounts for the majority of the risk.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

CY conceptualized the study, designed the methodology, curated the data, performed the formal analyses, and drafted the original manuscript. YY provided supervision, contributed to the conceptualization and methodological design, and revised the manuscript critically for important intellectual content. All authors read and approved the final manuscript.

Funding

The authors received no funding from an external source.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The project was approved by the Guangdong Business and Technology University Ethics Committee (NO. 2022007) and endorsed by all partner universities. All participants participated in this study with fully informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.