2 - (SYM 102) Locally Weighted Structural Equation Modeling as a Tool to Evaluate Model Fit Across a Continuous Moderator (Age): Application to the Anxiety Sensitivity index-3

Introduction: The Anxiety Sensitivity Index-3 (ASI-3) measures anxiety sensitivity (AS), a transdiagnostic mechanism for psychopathology recently linked to age-related issues. Research supports a bifactor structure for the ASI-3 including general AS and three specific factors: physical, social, and cognitive concerns. To date, work studying invariance of the ASI-3 across age has used artificial and inconsistent age categories. Alternatively, locally weighted structural equation modeling (LSEM) can estimate model parameters at each point of a continuous moderator such as age. All subjects are used to estimate parameters at all age points, effectively increasing sample size, with each subject’s data weighted according to their proximity to a given focal point. The current study employed both traditional and LSEM approaches to evaluate measurement invariance of the ASI-3 across age.

Method: Participants (N = 1128) were drawn from an outpatient anxiety clinic (N = 993; M/SD age = 33.7/15.5, 56.4% Female, 67.5% White) and a study on older age and sleep (N = 135; M/SD age = 65.6/8.9 years, 68.9% Female, 91.1% White). Traditional invariance analyses were run on participants both when grouped into two (< 50 and 50+ years) and three (18-29, 30-50, and 50+ years) age groups. LSEM analyses estimated model fit statistics and factor loadings continuously across age.

Results: A bifactor structure fit the data well in the full sample, c² = 593.98, RMSEA = .060, CFA = 0.96, TLI = 0.95, and SRMR = .03. Traditional invariance testing did not support metric invariance across age groups, either when categorized into two or three age groups. However, LSEM analyses revealed that RMSEA stayed within the range of approximately .06-.07, though it generally increased with age. CFI and TLI estimates also remained above .95 across all ages, though both indices exhibited a curvilinear pattern. LSEM analyses of factor loadings revealed that some items became better indicators of a given factor as age increased, while some became worse. These trends will be reviewed in greater detail during the presentation.

Conclusion: Results provide mixed support for the stability of the ASI-3 bifactor structure across age. Although traditional invariance testing would suggest that the ASI-3 is not equivalent across age groups, LSEM results provide more nuanced insight into how fit statistics and factor loadings change, including non-linearly, with increasing age. Implications for both techniques will be discussed, including their application to dimensional models of psychopathology.