Network Trees: A Method for Recursively Partitioning Covariance Structures

Jones, Payton J.; Mair, Patrick; Simon, Thorsten; Zeileis, Achim

doi:10.1007/s11336-020-09731-4

Cited by 41 publications

(67 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This method assesses whether parameters differ between subgroups by checking each parameter for structural differences of the scores (i.e., partial derivative of the log-likelihood with respect to a parameter). It was recently developed for Gaussian graphical models (Jones et al, 2020), but here we report the results for a novel application to Ising models using a Monte-Carlo permutation test approach (Huth et al, 2020). We assessed structural changes instantiated by gender, age, ethnicity, and income.…”

Section: Discussionmentioning

confidence: 99%

Modeling Alcohol Use Disorder as a Set of Interconnected Symptoms - Assessing Differences between Clinical and Population Samples and Across External Factors

Huth¹,

Luigjes²,

Marsman³

et al. 2021

Preprint

View full text Add to dashboard Cite

Alcohol use disorder is argued to be a highly complex disorder influenced by a multitude of factors on different levels. Common research approaches fail to capture this breadth of interconnecting symptoms. To address this gap in theoretical assumptions and methodological approaches, we used a network analysis to assess the interplay of alcohol use disorder symptoms. We applied the analysis to two US-datasets, a population sample with 23,591 individuals and a clinical sample with 483 individuals seeking treatment for alcohol use disorder. First, using a Bayesian framework we investigated differences between clinical and population samples looking at the symptom interactions and underlying structure space. The clinical sample depicts less connections; those connections are additionally weaker. Second, for the population sample we assessed whether the interactions were measurement invariant across subgroups of external factors like age, gender, ethnicity and income. Interactions differed across all external factors. Distinct parameter estimates for subgroups should be considered for better replicable estimates and effective intervention planning.

show abstract

Section: Discussionmentioning

confidence: 99%

Modeling Alcohol Use Disorder as a Set of Interconnected Symptoms - Assessing Differences between Clinical and Population Samples and Across External Factors

Huth¹,

Luigjes²,

Marsman³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Following the same approach as structural equation model trees ( Brandmaier et al 2013 ), the NMT approach combines psychometric network modeling with recursive partitioning techniques to detect significant differences in the network structure based on covariates. That is, the NMT approach assesses how covariates are associated with heterogeneity across the network structure ( Jones et al 2020 ). In this study, we used model-based recursive partitioning (MOB; Zeileis et al 2008 ) to split the network structure of the WJ IV COG subtests based on sex and age groups.…”

Section: Methodsmentioning

confidence: 99%

“…If sex and age groups are associated with statistically significant differences in the network structure, then the MOB algorithm will split the network structure at least once, or more, based on these covariates and create terminal nodes. We estimated network model trees using the MOB algorithm in the networktree package ( Jones et al 2020 ). For the psychometric network analyses, we followed the guidelines of Epskamp et al ( 2018 ) and Jones et al ( 2020 ).…”

Section: Methodsmentioning

confidence: 99%

“…Next, we perform psychometric network modeling to delineate the network structure of the CHC cognitive abilities measured within the WJ IV COG. Finally, we use the network model tree (NMT; Jones et al 2020 ) approach to evaluate the impact of sex and age on the network structure of the WJ IV COG. The NMT approach recursively splits the data based on covariates (i.e., sex and age) to detect significant differences in the network structure.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Age and Sex Invariance of the Woodcock-Johnson IV Tests of Cognitive Abilities: Evidence from Psychometric Network Modeling

et al. 2021

View full text Add to dashboard Cite

The Woodcock-Johnson IV Tests of Cognitive Abilities (WJ IV COG) is a comprehensive assessment battery designed to assess broad and narrow cognitive abilities, as defined by the Cattell-Horn-Carroll (CHC) theory of intelligence. Previous studies examined the invariance of the WJ assessments across sex and age groups using factor analytic methods. Psychometric network modeling is an alternative methodology that can address both direct and indirect relationships among the observed variables. In this study, we employed psychometric network modeling to examine the invariance of the WJ IV COG across sex and age groups. Using a normative sample (n = 4212 participants) representative of the United States population, we tested the extent to which the factorial structure of the WJ IV COG aligned with CHC theory for the school-aged sample. Next, we used psychometric network modeling as a data-driven method to investigate whether the network structure of the WJ IV COG remains similar across different sex and age (age 6 to 19, inclusively) groups. Our results showed that the WJ IV COG maintained the same network structure across all age and sex groups, although the network structure at younger ages indicated weaker relationships among some subtests. Overall, the results provide construct validity evidence for the WJ IV COG, based on both theoretical and data-driven methods.

show abstract

“…Therefore, as a follow-up analysis, we examined whether the relationships between all nodes in our combined network might differ depending upon school. To this end, we conducted modelbased recursive partitioning of our combined network (Jones, Mair, Simon, & Zeileis, 2019) using the networktree package (Jones, Simon, & Zeileis, 2018), which determines whether the goodness-of-fit index for the model is associated with covariates of interest. In this analysis, we included each participant's school as a covariate upon which the network was partitioned.…”

Section: Graphical Lasso)mentioning

confidence: 99%

Loneliness in Kenyan Adolescents: Socio-Cultural Factors and Network Association with Depression and Anxiety Symptoms

Rodríguez¹,

Tl²,

Gan³

et al. 2021

Preprint

View full text Add to dashboard Cite

Objective: Loneliness is associated with negative mental health outcomes and is particularly common among adolescents. Yet, little is known about the dynamics of adolescent loneliness in non-Western, low-income nations. Accordingly, we estimated the severity of loneliness in a sample of Kenyan adolescents and used mixed-effects regression modelling to determine the socio-cultural factors associated with loneliness. We also used network analysis to examine the associations between loneliness, depression, and anxiety at the symptom-level. Method: We analyzed data from a large sample (N = 2,192) of school-attending Kenyan adolescents aged 1219 (58.3% Female, 41.7% Male). Standardized measures of loneliness (ULS-8), depression (PHQ-8), and anxiety (GAD-7) were used. Results: Our mixed-effects model revealed that female and lower-income adolescents felt lonelier. The perception of feeling alone emerged as the aspect of loneliness most strongly linked to depression and anxiety symptoms. Conclusions: Our findings establish an estimate of loneliness levels in Kenyan adolescents, and reveal sociocultural factors associated with loneliness. We found that perceptions of isolation more strongly linked loneliness to psychopathology than did objective measures of isolation or preferences for social contact. Finally, we identify specific aspects of loneliness that could prove to be treatment targets for youth psychopathology; however, further research is needed. Limitations, future directions, and clinical implications are discussed.

show abstract

Network Trees: A Method for Recursively Partitioning Covariance Structures

Cited by 41 publications

References 43 publications

Modeling Alcohol Use Disorder as a Set of Interconnected Symptoms - Assessing Differences between Clinical and Population Samples and Across External Factors

Modeling Alcohol Use Disorder as a Set of Interconnected Symptoms - Assessing Differences between Clinical and Population Samples and Across External Factors

Age and Sex Invariance of the Woodcock-Johnson IV Tests of Cognitive Abilities: Evidence from Psychometric Network Modeling

Loneliness in Kenyan Adolescents: Socio-Cultural Factors and Network Association with Depression and Anxiety Symptoms

Contact Info

Product

Resources

About