Readily available data sources on homeless families include four longitudinal studies (Shinn et al., 1998; Bassuk et al., 1998; Rog and Guttman, 1997; SAMHSA, 2004), two in a single city (New York and Washington, DC, respectively) and two in several sites. There is also one cross-sectional study (Burt et al., 1999). The two studies in a single city select single female headed families on welfare. The two longitudinal studies in several sites select families with serious problems, requiring health, substance abuse, and mental health services. The cross sectional study is based on a shelter population. There is considerable heterogeneity in the design and instruments used in these studies. A retrospective meta-analysis would require considerable statistical sophistication. One of the longitudinal studies of low-income housed families is likely to include families that experienced homelessness during the followup period. As a prospective study with good national sampling and little evidence of selection of families, it is a good candidate to study the development of family homelessness, provided there are enough instances of homelessness in the study population.
Clearly the currently available data do not include all types of homeless families with children and, at least in some studies, there has been a tendency to oversample those with mental disorder, substance abuse, and frequent service use. There is an exception in the instance of NSAF. Because its cohort begins as housed families, it is unlikely to select particular pathways or subgroups of homeless families. Thus, a secondary data-based approach to a typology might first find whether there are enough families in NSAF and enough variables relevant to homelessness in that study to warrant using it in clustering studies. Another approach would be to use the five homelessness studies very cautiously, with analysis of resulting clusters for dependence upon the excess categories described above. Along these lines, the following secondary analysis projects are suggested:
- The longitudinal study of low-income housed families should be examined to determine whether it will yield a sufficient number of homeless families to warrant attempts to develop a typology. If so, the careful selection of available severity indicators could be recommended, within the context of the four-celled approach described above.
- Preliminary typological analysis of the five homeless family studies should be performed to assess the extent to which the design and instruments are compatible with pooling their data to develop a typology; and find how much effort would be needed to index the subjects in these studies with environmental data as described in Appendix B.3 and Appendix B.4. If the studies pass both tests, they might be worth further analyses both as pilot projects for the new studies and as a provisional source of data to guide policy and service delivery.
Available data in the five homeless family studies include the approximate dates when the findings were obtained and the localities where the homeless families were situated. Therefore, it should be possible to link demographic and endogenous measures from these studies with data on housing and other environmental variables listed in Appendix B.3 using date and locality information from Federal and local agencies and advocacy sources. Thus, it might be feasible derive a rough four-celled typology model from those linked data.
From a very practical point of view, perhaps it would be best to start with an attempt to create a relatively simple typology using readily available endogenous (e.g., psychopathology/psychiatric severity; substance abuse) and personal history variables (e.g., chronicity of homelessness, minority status) that are particularly relevant to women with children, and to test their interactions with environmental factors as suggested above. This approach could be applied to existing data sets (both longitudinal and cross-sectional) and might lead to a relatively easy way to provide a simple classification into the uncomplicated and complicated subtypes suggested in the literature. If replicated subtypes could be identified, they could provide a basis for some relatively straightforward decisions matching families to the most appropriate levels and types of intervention, including housing, social services, medical services and psychiatric care, with the more severe, chronic subtype perhaps being the subject of additional subtyping analyses to develop a more refined classification into service need categories. New research on primary data sources should also proceed in concert.