Using Multilevel Generalized Linear Models to Project the Number of Children Aging out

Background and Purpose: Predicting the number of clients a given program is likely to serve is high on the list of challenges facing policy makers, advocates, and practitioners. Without a firm estimate going forward, it is simply too difficult to know whether the resources allocated are too few, too much, or just right. Moreover, what is true next year may not be true in the years that follow. In this paper, the authors use a multilevel generalized linear model (HGLM) to project the likely demand for independent living programs. The demand in this instance is taken as the number of children between 2002 and 2013 who are be expected to turn age 18 while still in out-of-home care. The years 2009 through 2013 form the projection. The study/approach, which draws inspiration from demography and actuarial statistics, focuses on the long-run entry/exit equilibrium after adjusting for attributes of the child and their placement experience.

Methods: The study is based on all children admitted into out-of-home care in seventeen states over a 7-year period (2002-2008), plus all those children in care at the start of the study (January 1, 2002). Age at admission, placement type, admission year, and duration in care are variables in a Poisson log-linear regression model predicting the proportion of children who will age out of care. Since state child welfare agencies have varying policies with respect to care for children after age 18, we can expect the model coefficients to vary by state. An HGLM with children nested within states accommodates this aspect of the data structure.

Results: The results show the number of children aging out nationally will most likely decline between 2010 and 2013. From a high-level view, the projected decline is tied to an overall decline in admissions to out-of-home care. Within the overall picture, a number of other factors are contributing to the decline in the nearer term. For example, admissions of those children most likely to age out have been declining at a faster rate than children of other ages.

Conclusions and Implications: From a methodological perspective, the results highlight how modern statistical methods can be used to answer fundamentally important contextual questions. At its core, the HGLM is a relatively simple model to exploit, giving ‘users' a window into the future. As the model depends on readily available administrative data, this projection method expands the utility of administrative data. More importantly, a window into the future allows program supporters to operate in a more proactive manner when it comes to managing their piece of the public pie. At a time when resources are increasingly scarce, knowing what you are likely to need in the foreseeable future may be crucial. Finally, projections are only as good as the data on which they are based. Projection models beget the need for more information, a process that often has a positive affect on the development of theory.