Level of Agreement Between Problem Gamblers' and Collaterals' Reports: A Bayesian Random-Effects Two-Part Model

Area of Science:

• Psychology
• Behavioral Science
• Addiction Research

Background:

• Problem gambling significantly impacts individuals and their social circles.
• Accurate assessment of financial losses is crucial for understanding gambling severity and treatment effectiveness.
• Discrepancies in reported financial losses between problem gamblers and their concerned significant others (CSOs) are common.

Purpose of the Study:

• To quantify the agreement level between problem gamblers and their CSOs regarding gambling-related financial losses.
• To compare agreement levels across different CSO-gambler dyad types (e.g., partner vs. parent).
• To evaluate the suitability of statistical models for analyzing skewed financial data in addiction research.

Main Methods:

• Analysis of self-reported and CSO-reported gambling losses over the past 30 days from 266 participants (133 dyads).
• Calculation of intraclass correlation coefficients (ICCs) to measure agreement.
• Application of a two-part generalized linear mixed-effects model to handle skewed data and zero-loss reports, comparing Gaussian, two-part gamma, and two-part lognormal distributions.

Main Results:

• A fair level of agreement (ICC = .57) was found between gamblers and CSOs regarding money lost.
• Partner CSOs demonstrated better agreement than parent CSOs (ICC_diff = .20).
• Agreement estimates became inconclusive when zero loss reports were included (ICC_diff = .16).
• A simulation confirmed the two-part model's efficacy and highlighted the unreliability of standard ICC calculations for skewed gambling loss data.

Conclusions:

• Problem gamblers and their CSOs exhibit a moderate level of agreement on financial losses, suggesting shared perceptions but also room for discrepancy.
• Dyad type influences agreement, with partners showing higher concordance than parents.
• Standard statistical assumptions (e.g., Gaussian distribution) are inappropriate for analyzing gambling loss data, necessitating advanced modeling techniques like the two-part generalized linear mixed-effects model for accurate assessment.