Latent profile analysis in recovery homes: A single quantitative dimension captures most but not all of the important details of the recovery process

Abstract

Background: Our study explored whether latent classes adequately represented the social capital recovery indicators at the resident level and whether latent class membership predicted subsequent exits from the recovery homes. Method: Our sample included about 600 residents in 42 recovery homes. Over a 2-year period of time, every 4 months, data were collected on eight elements of recovery capital. Results: We found 5 latent classes were optimal for representing 8 elements of recovery capital. Representing 79% of the sample, 3 of the 5 latent class profiles of the means of the 8 recovery indicators were roughly parallel and differed only in level, but the remaining 2 latent class profiles, representing 21% of the sample, were not parallel to the first 3, suggesting that a single quantitative dimension of perceived recovery may capture most but not all of the important details of the recovery process. Next, using longitudinal data from homes, the distal outcomes of resident eviction and voluntary exit were found to be related to latent class membership. Resident level pre-existing predictors (e.g., employment status, educational attainment, gender, Latinx ethnicity) and house level pre-existing predictors (e.g., financial health, poverty level of typical population served, new resident acceptance rate) significantly discriminated the classes. In a model that combined both pre-existing predictors and distal outcomes, latent class membership was still the strongest predictor of evictions controlling for the pre-existing predictors. Conclusions: These classes help to clarify the different aspects of the recovery latent score, and point to classes that have different ethnic and gender characteristics as well as outcomes in the recovery homes. For example, the high levels of self-confidence found in class 3 suggest that Latinx might be at higher risk for having some difficulties within these recovery communities.

Keywords:

• Latent profile analysis
• recovery homes
• Oxford Houses
• social networks

Disclosure statement

The funding organization had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Acknowledgments

The authors appreciate the social network help of Ed Stevens. The authors also acknowledge the help of several members of the Oxford House organization, and in particular Paul Molloy, Alex Snowden, Casey Longan, and Howard Wilkins.

Notes

1 One house dropped out completely but another was added after wave 1 bringing the total to 43 houses. However, 42 is accurate in the sense that only 42 houses have 2 or more waves of information available on their residents and a single data point carries no information about change or the probability of an eventual exit.

2 The term single-level, correlation-corrected comes into play because although residents are nested within recovery homes, for simplicity, we opted to treat the correlation among residents within the same home as a nuisance parameter that is important to compensate for in order to avoid under-estimating standard errors but nonetheless remains in the background is not the focus of the analysis. Hence the term, single-level, correlation-corrected. The potential risk to weigh in opting for simplicity this way is the confounding of resident and house level effects of a resident level predictor when the effects are different, not just statistically but in a substantively important way. For example, if the resident and house level effects were statistically significant and had different signs (i.e., positive vs negative) confounding them could result in one overall estimate that is essentially zero. Nothing in our previous work suggests this might occur so radical so we do not believe this is a serious limitation of our initial approach to LPA. To streamline terminology, all references to our LPA models in this article will hereafter drop references to single-level, correlation-corrected.

3 34 entered an OH for the first time at wave 7 and we were not able to determine whether they exited given that the last data collection occurred during wave 7; 15 were forced out because their OH closed before wave 7 and before they filled out a second survey; 15 were missing their reason for leaving and had only filled out 1 survey before leaving.

4 Six participants were observed to leave the recovery homes during the course of the 2 year study but their reason for leaving could not be determined. We right censored these individuals at their penultimate survey as they had at least one additional wave of data beyond their baseline survey. In addition, 34 residents had more than one OH exit, either from two different OH's or from the same OH. To avoid greatly complicating the exit model for a very small number of individuals, we only included their first exit in all analyses.

Additional information

Funding

The present work was financially supported by the National Institute on Alcohol Abuse and Alcoholism [AA022763].