Poster #99 - Using Machine Learning to Identify Linguistic Markers of Parents’ Anxiety Using Baseline Data from an Early Intervention for Inhibited Preschoolers

• In Event: Poster Session 01

Abstract

Behavioral markers of parental anxiety grant valuable insight into the manifestation and severity of symptomology that can inform etiological models and support the development of personalized interventions. While existing research has primarily focused on parenting behaviors (e.g., intrusiveness, overinvolvement) given the overwhelming evidence that they moderate risk for social anxiety in behaviorally inhibited children, additional evidence suggests that parental language may hold particular significance for parent anxiety. Studies examining parents’ language, including syntactical and morphological structures, have isolated categories of word use (e.g., mental state language) that are associated with increased parental anxiety and risk for internalizing symptoms in young children. However, despite the substantial body of literature demonstrating associations between parents’ language and children’s social and emotional development, very little is known about specific linguistic markers of parental anxiety. Additional research gaps persist due, in part, to limitations of common statistical approaches to integrate data on parents' language across many, and sometimes hundreds, of variables, and the necessary reliance on a priori hypotheses specifying potentially influential linguistic variables. Addressing this gap will inform etiological models of anxiety and will provide insight into whether and how parents' language specifically might serve as a treatment target across clinical and nonclinical work with parent-child dyads.
The current study addresses these gaps by leveraging machine learning algorithms to examine parenting styles and maternal linguistic patterns during a parent-child interaction task as unique factors that are markers of parental anxiety. Our transformative analytic framework allows us to automate the splitting of predictors and capture both linear and nonlinear main effects, as well as potential interactions between risk factors to optimize the prediction of parental anxiety. Specifically, we utilized Random Forest models in R (gbm package, Greenwall et al., 2022) to elucidate parenting and linguistic factors that demonstrate the strongest prediction of parental anxiety using variables from an expansive set of observed maternal language and behaviors (total predictors n =73). Using a sample of 151 mothers of children (Mage = 4.21y, SDage = 1.02y) recruited for an intervention study for behaviorally inhibited preschoolers, Random Forest models with boosting revealed that attention-focused (e.g., look, watch), insightful (e.g., know, think, feel), and future-focused (e.g., will, going to) language markers had the greatest relative influence on parental anxiety while parenting styles characterized by participation and nurturance had the greatest relative influence. Post-hoc path models assessing the directionality of predictive relations between identified language variables, parenting styles, and maternal anxiety revealed that heightened anxiety was significantly associated with decreased future-focused language, β= -.19, p = 007, and increased number of words used during the interaction, β= .20, p =.018. Findings support machine learning methods as a useful statistical approach for identifying markers of risk for anxiety across a large number of predictors, and study results suggest that parental language, in addition to parenting strategies, may hold promise for better understanding the intergenerational transmission of risk for anxiety.

Authors

• Deaven Winebrake, Boston University
  Presenting Author

• Kathy Sem, Boston University

• Nila Shakiba, Boston University

• Hong N. T. Bui, University of Maryland - College Park

• Danielle Novick, University of Maryland - College Park

• Christina Danko

• Andrea Chronis-Tuscano, University of Maryland, College Park

• Kenneth H Rubin, University of Maryland - College Park

• Nicholas James Wagner, Boston University