Addressing Disaster Exposure Measurement Issues With Latent Class Analysis

Abstract

Disaster exposure can put survivors at greater risk for subsequent mental health (MH) problems. Within the field of disaster MH research, it is important to understand how the choice of analytic approaches and their implicit assumptions may affect results when using a disaster exposure measure. We compared different analytic strategies for quantifying disaster exposure and included a new analytic approach, latent class analysis (LCA), in a sample of parents and youth. Following exposure to multiple floods in Texas, a sample of 555 parents and 486 youth were recruited. Parents were predominantly female (70.9%) and White (60.8%). Parents were asked to have their oldest child between the ages of 10 and 19 years old participate (M = 13.74 years, SD = 2.57; 52.9% male). Participants completed measures on disaster exposure, posttraumatic stress, depression, and anxiety. The LCA revealed four patterns of exposure in both parents and youth: high exposure (15.5% parent, 9.5% child), moderate exposure (19.8% parent, 28.2% child), community exposure (45.9% parent, 34.4% child), and low exposure (18.8% parent, 27.8% child). In terms of MH, there were similarities across analytic approaches, but the LCA highlighted a threshold effect, with the high exposure class doing worse than all others, d = 1.12. These results have important implications in understanding the different exposure experiences of survivors and the linkage to MH outcomes. The findings are also informative in the development and use of screening tools used in postdisaster contexts in determining who may or may not need MH services.

Why Family Communication Matters: the Role of Co-rumination and Topic Avoidance in Understanding Post-Disaster Mental Health

“…The total sum score represents the total flood exposure items the family experienced. The predictive validity of this measure has been demonstrated through the relationship of increasing levels of exposure to greater levels of PTSS, anxiety, and depression symptoms for both parents and youth (Felix et al, 2019).…”

Subtyping of Internalizing and Externalizing Behaviors in Japanese Community-Based Children: A Latent Class Analysis and Association with Family Activities

“…First, we used a person-based method called latent class analysis (LCA) to explore the potential types of children’s internalizing and externalizing behaviors. LCA can empirically group participants based on represented response patterns across multiple potential factors instead of categorizing them based on cut-off scores [ 33 ], which can be defined through item probabilities and class proportions, or the sample proportion represented in each latent class [ 34 ]. Since no single statistical criterion identified the best-fitting LCA model, we used several fit indices to statistically determine the best model selection.…”

“…The Bayesian information criterion (BIC), Akaike information criteria (AIC), sample-size-adjusted BIC (BIC), and corrected Akaike’s information criterion (CAIC) most often identify the best-fitting model. BIC and CAIC have the best performance in both small and large sample cases [ 35 ]; therefore, BIC was the most reliable measure [ 33 , 36 ]. Smaller values of these fit indices indicate a great model.…”

“…Smaller values of these fit indices indicate a great model. Additionally, entropy is a standardized index of model-selection accuracy, with higher values indicating better classification of individuals into groups and classes that are clearly delineated from one another [ 36 ], where values around 0.8, are considered to have good classification [ 33 ]. Previous studies have shown that when entropy is <0.6, the classification error is more than 20%, and the entropy decreases with an increasing sample size [ 37 , 38 ].…”

Examining the Direction of Effects Between COVID-19 Experiences, General Well-Being, Social Media Engagement, and Insomnia Symptoms Among University Students

“…Scores were summed across all items and coded such that higher scores indicated more negative COVID-19 experiences. Traditional psychometric properties of reliability (e.g., Cronbach's alpha) are inappropriate for event-type scales such as the EPII and were, therefore, not computed (Felix et al, 2019). Recent studies demonstrating significant associations between EPII scores and various indices of poor psychological health (e.g., PTSD, anxiety, depression, and perceived stress) provide evidence of construct validity for the EPII (see Grasso et al, in press).…”