Presentation: Group comparison of longitudinal network analysis: An example of eating disorder psychopathology

Modern Method Modeling 2024

Jihong Zhang*, Ph.D; Jinbo He, Ph.D

Educational Statistics and Research Methods (ESRM) Program*

University of Arkansas

2024-06-26

Presentation Outline

  1. General Questions

    • Can eating disorders be considered as a network?

    • Why the longitudinal network is a suitable tool for examining this topic?

    • How to compare groups whithin the framework of longitudinal network?

  2. Measures

  3. Method

  4. Results

  5. Discussion

  6. Future direction

1 General questions

Question 1: Can Eating disorder be considered as a network?

  1. Understanding interrelationship between different components: emotion regulation, interpersonal problems, and eating disorder can be conceptualized as a symtom network via theoretical frameworks
  • The Emotion regulation theory suggests difficulty in emotion regulation issues can results in ED behaviors

  • Interpersonal psychotherapy theory posits that interpersonal problems may exacerbate ED (Murphy et al., 2012)

  • Empirical studies considered these three consitute a “ecosystem” (Ambwani et al., 2014). ER and interpersonal function exibit reciprocal effects on ED maintance.

However, the dynamics of eating disorders has not been well investigated.

Question 1: Can Eating disorder be considered as a network?

  1. Examine why ED have a network-like structure
  • Some considered the development of ED as a longitudinal process (Mallorquí-Bagué et al., 2018)
  • Some argue that the relations between risky factors (emotion, interpersonal relations) and ED are complex (i.e., feedback loop)
  • Additionally, some highlight that boys or girls also exibit gender difference regarding the developmental process of ED

Question 2: Why psychological network can be used for this topic?

Longitudinal network analysis have been widely applied in psychopathology and proven as a suitable tool for addressing the problems mentioned in the first general question.

  1. It considers symptoms related to each other as in a symptom network.
  2. It allows us to identify the most important symptoms (disordered behaviors) in this complex network
  3. It can estimate temporal effects of one symptom on the others controlling effects of other variables

Question 2: Why psychological network can be used for this topic?

We estimated network parameters using graphical vector autoregressive (GVAR; Epskamp, 2020; Wild et al., 2010) algorithm

Three types of network structures are estimated:

  1. Temporal network (temporal effect)

  2. Contemporaneous network (within-person effects controlling for temporal effects)

  3. Between-individual network (individual differences)

Question 3: How to compare groups in longitudinal network

In network analysis, groups can be compared from three aspects:

  1. Network structure (e.g., some nodes connected in group A but not in group B)

  2. Node-level measures: node centrality (importance) or node bridging strength (e.g., some nodes can be more central in group A while not in group B)

  3. Network edge weights (e.g., node 1 and node 2 have strong relationship in group A but weaker relationship in group B)

Things to be considered

  1. Which variables (nodes) should be included in the network?

    • Items?
    • Subscale scores?
    • Latent factor scores?
    • Mixed

There are no clear rules of node types. It depends on theory model.

  1. Are different networks comparable?

The network of sample A differs from the network of sample B in term of network structures or centrality measures. They only have quantative differences in parameter estimates or they are measuring different constructs?

Research Questions:

  1. Are there any gender differences in the network characteristics of longitudinal networks

  2. Are there any gender differences in the network structures of eating disorder longitudinal networks

  3. Are there any gender differences in the node centrality and bridge strength of longitudinal networks

Data

  • 4-wave data collection were conducted over 18 months.

  • For each wave, demographic information and self-reported answers of three questionnaires (emotional regulation, interpersonal problems, and eating disorder) from 1652 high school students in China were collected.

  • After data cleaning, N = 1540 cases left including 53.9% girls and 46.1% boys

  • Age ranges from 11 to 17 years with a mean 15.2 years old

Measures

For network analysis, we used subscales and items as nodes.

  1. Emotion regulation: Difficulties in Emotion Regulation Scale (DERS-18). 6 subscales measures different aspects of emotion dysregualtion: Awareness, Clarity, Goals, Non-acceptance, Impulse, Strategies
  2. Interpersonal problems: Inventory of Interpersonal Problems—Short Circumplex (IIP-SC). 8 subscales were included measuring varied aspects of interpersonal problems: domineering, cold, avoidant etc.
  3. Eating disorder: 12-item short form of Eating Disorder Examination Questionnaire (EDE-QS). 12 items measured different disordered eating behaviors.

We have 26 nodes in initial networks.

Data Analysis

1. Node (item) selection

  • For emotion regulation and interpersonal problem questions, 14 subscales were selected because their measured constructs have been well examined and theory-driven.

  • For eating disorder, we want to makre sure each item represent one unique eating disorder problem. Thus, we used the goldbricker algorithm to drop overalapping (dupplicated) symptoms, which give rise to 8 items included in the analyzed network.

We have 22 nodes in further network analysis.

2. Network estimation

  • Multi-group GVAR was applied to the data to estimate boys’ and girls’ temporal/contemporaneous/between-subject network.

  • Furthermore, we pruned the networks and identify the most important nodes and edges using prune function in psychonetric package in R.

Data Analysis (Cont.)

3. Overall network evaluation

  • Network stability measured by correlation stability (CS) coefficients (Epskamp et al., 2018), which estimates the accuracy of centrality measures and node strength using bootstrapping

4. Test group differences

  • Using likelihood ratio test (LRT) to examine network structure differences by gender

    • Model H0 (the model with all edge weights constrained to be equal)

    • Model H1 (the model with all edge weights freely estimated)

    • Calculate the likelihood ratio between Model H0 and H1 and perform significance test

  • Examine gender differences in node centrality and bridging strength

    • Compare estimated node centrality and bridge strength by gender

    • Accuracy: test the accuracy of node centrality differences using bootstrapping sampling

2 Results

Overall Network Stability

  1. The multigroup network stability statistics were all acceptable for contemporaneous/between-subject networks according to the criteria of CS coefficients above 0.7
  2. The temporal network was less stable with CS coefficients ranging from .51 to .56

Group differences - Temporal Network

Boys’ temporal network

Girls’ temporal network

Summary: Group differences - Temporal Network

Boys

  1. Sparsity: 8.06% of non-zero edges (temporal)
  2. Strength: Mean(SD) of edge weights is 0.127(0.094)
  3. Node weight/shape preoccupation (EDE-WP) exhibited the highest InStrength
  4. Node weight/shape dissatisfaction (EDE-WD) exhibited the highest OutStrength
  5. Node Awareness (Awr) exhibited the highest bridge strength

Girls

  1. Sparsity: 10.60% of non-zero edges (temporal)
  2. Strength: Mean(SD) of edge weights is 0.128(0.102)
  3. Node weight/shape preoccupation (EDE-WP) exhibited the highest InStrength
  4. Node weight/shape dissatisfaction (EDE-WD) exhibited the highest OutStrength
  5. Node weight/shape dissatisfaction (EDE-WD) exhibited the highest bridge strength

Accuracy test: Node Strength Difference

Are the node differences due to the sampling error? We used bootstrapping method to test that:

  1. Long periods without eating (EDE-WE) and Food preoccupation (EDE-FP) have significant gender differences in node Outstrength
  2. Weight/shape control by vomiting or taking laxatives (EDE-VT) have significant gender differences in node InStrength

Accuracy test: Bridge Strength Difference

Are the node differences due to the sampling error? We used bootstrapping method to test that:

  1. Awareness (Awr) and Goals (Gls) have higher bridge strength in boys than girls
  2. Weight/shape preoccupation (EDE-WP) and Binge eating episode (EDE-BE) have higher bridge strength in boys than girls.

Target nodes for intervention on comorbidity.

Discussion: Group Commonality

  1. Edge weight strength of temporal network are similar for boys and girls, suggesting symptoms have similars impact on other symptoms.

  2. Emotion dysregulation have interconnections with eating disorders and interpersonal problems.

  3. Disordered eating behaviors also closely relate to each others within the eating disorder community. One disorder eating problem is likely to activte other problems.

  4. For both groups, nodes related to overvaluation of weight/shape (preoccupation or dissatisifaction) are the most influential factors in ED networks, which implys the symptoms should be main focus for eating disorder interventions.

Discussion: Group Differences

  1. Network structures of boys and girls are significantly different in term of likelihood ratio test (LRT).
  2. Nodes of girls’ network are more densely related than boys’, suggesting higher likelihood of comorbidity for girls.
  3. Lack of emotional awareness is an potential maintenance factor of boys, which is a novel finding.

Further direction

There are more things to do to examine group differences in the longitudinal network framework:

Global:

  1. More simulation studys to validate the LRT method in examining group differences of global network structure

  2. Groups may differ in network density and average edge weights. What it means in application research need more investigation?

Node-level:

  1. Groups may differ in most important nodes but also differ in less important nodes. What that mean and how we interpret that?

  2. Are group’s edge weights comparable? For example, the partial correlation between node A with node B differ by groups but how to interpret that?

Thank you.

Let me know if you have any questions.

You can also contact me via jzhang@uark.edu