Repeated-Measures and Multilevel Latent Class Analysis in Child Welfare Research

Purpose: This paper presents examples of the use of two recently-developed extensions of LCA to studies of public child welfare. In the first example (pathways to residential care study), repeated-measure LCA is used to examine the sequence and timing of different placement events among a sample of adolescents entering residential care for the first time. The goal of this study is to develop a typology of pre-residential-care placement pathways based on the interrelationship between different placement events. Given the nature of children's placement histories, an important criteria for these analysis is that the measurement model allows for the specification of non-linear, discontinuous pathways. In the second example (supervisory relationship study), multilevel LCA is used to develop a typology of child welfare supervisory climate, and to examine the interrelationships between individual caseworker-supervisor relationships, unit-level supervisory climate, and the broader organizational contexts in which they are situated. The defining methodological feature of this study is that the primary construct of interest (i.e., supervisory climate) is an aggregate-level latent construct that is itself a manifestation of a latent dyadic-level construct (i.e., worker-supervisory relationship).

Methods: The sample for the pathways to residential care study includes youth, age 13 to 16, who entered residential care for the first time under the auspices of a Midwestern state public child welfare agency (N=7,187). Based on administrative data records of placement moves, runaway episodes, placements in detention, and psychiatric hospitalizations during the year preceding placement into residential care, repeated-measures latent class analysis (Lanza & Collins, 2006) is used to develop a typology of pre-residential care placement pathways.

The sample for the supervisory relationship study includes child welfare caseworkers (N = 1,460) from 56 private and public agencies in Illinois who completed a web-based survey between August and October, 2010 (91% response rate). Respondents were queried about their experiences as child welfare workers, the characteristics of the organizational and resource environments in which they worked, and the nature of their relationships with their immediate supervisors. Using a variant of latent class analysis (LCA) suitable for multilevel data (Vermunt, 2003), we develop a typology of supervisory climate based on the nature of individual caseworker-supervisor relationships.

Results: Results of the pathways to residential care study suggest the existence of placement distinct pathways characterized by notable discontinuities and strong interrelationships between the sequence and timing of different pre-residential placement events. Results of the supervisory relationship study suggest the existence of distinct supervisory climates composed of distinct worker-supervisor relationship types.

Implications: These studies illustrate the utility of using LCA when examining phenomena of interest to child welfare researchers. These phenomena, which are often complex and not directly observable (latent), are also often poorly conceptualized and measured. However, recent advances in mixture models (including LCA), now afford researchers the opportunity to faithfully summarize this complexity.