Several criteria may be helpful to guide the selection of variables. These are simplicity, ease of measurement, theoretical relevance, minimal measurement overlap, coverage of major domains of interest, and practical usefulness in service matching.
From the perspectives of efficiency and economy, the availability of current instruments is certainly an important practical consideration. Progress would be much faster if one could use existing instruments and data sets than if one had to devise new instruments. However, one should not be guided in the choice of a variable by availability of instruments for that variable. The choice of variables should be determined by its theoretical and practical value. However, one should also consider developing a new instrument for key constructs to the extent they are considered important.
A variety of measurement techniques and standardized assessment procedures have been developed to measure many of the variables relevant to typological formulations. Techniques include self-report questionnaires, personal interview schedules, and administrative data, including demographic characteristics. Assessment procedures include measures of psychopathology, substance use disorders, personal resources, and multiple problem inventories, such as the Addiction Severity Index (McLellan et al., 1992), which covers employment, psychiatric severity, substance abuse, family functioning, and criminal activity. Additional considerations important in the selection of measurement instruments are response burden, administrative load, and the availability of data sets to develop typologies.
A decision point in considering variables is whether to focus on endogenous variables (i.e., characteristics of the homeless families), exogenous variables (i.e., characteristics of the environment of such families), situational variables (i.e., characteristics of the interaction with the environment or of situation in the family’s homelessness history), or all the above.
There has been very little use of available data on the environment of homeless families. Yet such data are of critical importance since homelessness is the result of interactions between persons and their environment (Jahiel, 1992b). There are several readily available sources of data with environment as a unit of analysis that could be used: (1) as a typology of homeless environments, or (2) in a typology of homeless situations (matching homeless persons’ needs and environmental capacity to meet these needs). Environmental data fall into several categories: (1) housing-related; (2) welfare-related (3) employment-related, (4) health-related, (5) mental health and substance abuse related. A partial listing of available secondary data resources that are relevant to critical environmental factors is given in Appendix B.3.
Another decision point is whether to start with “epidemiological type variables” (i.e., variables shared with other environmental problems), or empirically derived variables (variables elicited in qualitative or quantitative empirical studies in the field of homelessness, that are often more complex than the first type, and that have sometimes been used in developing typologies of homeless persons).
To the extent that the goal is to develop a typology that can be justified quantitatively and be useful in quantitative studies, the selection of variables is very important, particularly with regard to the state of disaggregation (to avoid noise), the locus of the variable (the one that best explains), and the specificity or the relevance of the variables for homelessness as it may be revealed by previous qualitative studies or published typologies. Appendix B.4 provides some examples to guide a starting point for an empirically derived typology.
A related consideration in the selection of variables is the availability of data sets that include various measures of homeless families. There are four existing longitudinal data sets on homeless families: the New York City Homeless Family Study (NYC HF, Shinn et al., 1998), the Worcester Family Research Project (WFPR, Bassuk, Buckner, Perloff and Bassuk, 1998)), the Robert Wood Johnson /U.S. Department of Housing and Urban Development data set (RWJ/HUD HF, Rog and Gutman, 1997), and the Substance Abuse and Mental Health Services Administration Homeless Families Program (SAMHSA.HF, SAMHSA, 2004)) and one cross-sectional study (the NSHAPC, Burt et al., 1999). Together the four longitudinal studies have 3,878 subjects. Each of these studies has a set of demographic data (age, race, marital status, work, education, currently pregnant) and certain service needs (health, mental health, substance abuse, trauma, legal history). Three of them have measures of income and foster care history. There are differences in the instruments used to measure these variables but there is enough similarity among them to make it possible to do replication studies, with appropriate correcting factors. There are marked differences in the selection of the study populations. Two of them (NYC-HF and WFPR) have populations of families on welfare and families in shelters. One (the RWJ/HUD HF) has families with multiple needs entering enriched housing. Another (the SAMHSA.HF) has families with mental illness, substance abuse, or both. Thus there are marked selection differences among families in the four studies, including differences in service needs and differences in the stage during the trajectory of homelessness when these families are studied. Furthermore, all studies underselect families that are doubled up (as opposed to literally homeless) and families that are in shelter for battered women, as well as families that have little or no contact with services. Thus, the four studies cannot be considered representative of the homeless family population at large. Further, families with multiple or severe service needs are selected in at least the two largest studies. Nevertheless, the advantage of the large sample sizes of the four combined longitudinal studies cannot be overlooked. They might yield typologies that are robust in the presence of differences in types of populations selected, for instance, typologies reflecting the intensity of service needs.
The only study able to provide good data on families identified before they are homeless and followed longitudinally, including those that remain stably housed and those that do not and have episodes of homelessness, is the National Survey of America’s Families (NSAF) (Abi Habib et al, 2005). NSAF has data from a representative sample of the civilian population with an oversampling of people with low income, with a large sample (n=> 40,000) surveyed in a cross-sectional design every 2 to 3 years. It is the best available source of information on doubling up, since it has a specific question asking whether the family had to move in with another family because of inability to pay mortgage, rent, or utilities. This data set would be very useful in investigating possible typologies of pre-literal homeless trajectories, as well as typologies related to history of doubling up. Further, it has demographic and service need data that might be used in conjunction with the five studies of homeless families.
Eventually studies designed to collect primary data will be necessary to achieve a nationally representative sample of homeless families or families at risk of homelessness and to have sufficient numbers to allow adequate statistical analysis. Ideally, such primary data studies should include longitudinal followup (e.g., 5 years).