Characteristics and Dynamics of Homeless Families with Children. Review of the Literature on Use of Typologies


Although there has been some limited attention to typologies for homeless families (e.g., Danesco and Holden, 1998) the literature that is most helpful involves efforts to develop typologies and classification systems for a range of populations, including individuals who abuse substances (e.g., Epstein et al., 2002; German and Sterk, 2002), individuals with chronic mental illness (Braucht and Kirby, 1986), individuals in the criminal justice system (e.g., Harris and Jones, 1999), homeless single adults (Kuhn and Culhane, 1998), homeless and runaway youth (Mallett, Rosenthal, Myers, Milburn, and Rotheram-Borus, 2004; Zide and Cherry, 1992), families involved in Head Start (Ramey, Ramey, and Lanzi, 1998), and children referred to mental health treatment (Hodges and Wotring, 2000).

There are numerous dimensions along which typologies vary, including whether the typology is based on a theoretical scheme or developed empirically; whether it is developed on one variable or dimension, or multiple dimensions and variables; the nature and measurement of the variables used; and whether the variables include only risk factors or strengths as well. In addition, some typologies are developed using qualitative data (e.g., German and Sterk, 2002), while others involve quantitative data, often using cluster analysis (e.g., Babor et al., 1992). The variations often relate to the purposes of the typology, as well as to the state of the knowledge in an area.

Although typologies based on theory are found in the literature, the majority of typologies are developed through various statistical approaches (e.g., Epstein et al., 2002). Braucht and Kirby (1986) demonstrated the value of a step-wise statistical approach to developing a typology of individuals with chronic mental illness. As a first step, the authors examined 49 variables and used cluster analysis to identify subsets of the variables that correlated along a similar dimension (Tryon and Bailey, 1966). Four dimensions resulted, involving 17 of the measures. The four dimensions, or homogeneous subsets of variables, resulted in little loss of information over the use of the 49 original individual variables and became the core building blocks for the typology. For step two, a score on each dimension was computed for each client in the sample. With these four scores on each client, another cluster analysis was conducted to identify subsets or clusters of clients. Twenty-one distinct subsets or types of clients resulted. As a third step, the authors followed an iterative process to systematically condense the 21 types into a smaller number of groups. Five groups resulted and the four average dimension scores were computed for each. For the fourth and final step, to determine whether the division into the five groups was meaningful, the authors examined additional clinical and psychosocial variables to see if they distinguished the groups from each other. The pattern of statistical differences across these variables for the five groups verified that they were distinct and provided a much more complete characterization of the types.

This example illustrates the value of a multivariate approach to typology development. Although some typologies are developed using a single measure, especially those involved in classification systems that need to be used by practitioners, multidimensional typologies appear to hold the most promise for delineation of meaningful groups. Epstein and colleagues (2002) demonstrated the value of a multivariate approach to typology development of individuals with alcohol use disorders. The authors compared four prevailing typologies and examined the extent to which they overlapped using baseline data from five treatment outcome studies. Two of the typologies were multidimensional and two were single-variable, dichotomous typologies. The comparative approach the authors used was instructive in revealing the strengths and problems with each typology. The authors found that the dichotomous typologies (single variable, two groups) were not complex enough to be clinically useful and often described only a portion of the population.

As in these other areas, a multidimensional strategy appears most promising for the typology of homeless families. Past research, as reviewed earlier, has revealed that understanding the complexity of demographic, background, family composition, service need, human and social capital, and other variables is critical to fully understanding families and how their needs may best be met. In particular, understanding families at various stages of vulnerability for homelessness will be important to a more complete understanding of when and how to intervene.

Typology development, however, is sensitive to variable measurement (e.g., type of measure, cutoff points such as age cutoffs, and extent of missing data). In particular, multidimensional typologies can be sensitive to the existence of outliers and can be temporally less stable if current status variables are used in their development (Epstein et al., 2002). In the homeless families' area, it is important to understand the operationalization of the variables and how they vary across different data sets. Different measures of stability have been used across research studies, as have varying measures of mental health and other areas of service need. In addition, for any database or panel surveys that are candidates for use in this project, it will be important to understand the extent to which there are any artifacts to the data that will challenge its usefulness, such as missing data on particular variables or on subsets of the population.

Because of the many subjective decisions made in developing a typology, the strategy of developing more than one possible typology, as well as investigating multiple data sets and conducting concordance analyses, is also a useful idea for reconciling differences and developing the best, most parsimonious, and most feasible approach (Epstein et al., 2002). This strategy allows cross-validation and testing the universality of the typology.

Criteria for Evaluating a Typology. In evaluating the usefulness of a typology, several criteria can be used (Babor et al., 1992; Epstein et al., 2002; Harris and Jones, 1999). The typology can be examined to determine whether it satisfies the following conditions:

  • Results in subgroups that have homogeneity within them;
  • Results in subgroups that are nonoverlapping and have distinct nontypology characteristics (i.e., has discriminant validity);
  • Is comprehensive in its coverage of the overall population;
  • Demonstrates construct validity by having the theoretical constructs empirically supported; and
  • Has predictive validity in that members of different subgroups show different patterns of homelessness and different responses to treatments (i.e., has clinical utility).

Developing distinct homogeneous subgroups is aided by techniques such as cluster analysis and the use of rich data systems that cover the complexities of the population. One of the challenges in the study of homeless families, however, is to identify data systems that provide for comprehensive coverage of the population. Each of the typology efforts reviewed concentrated on developing the typology in one data system.

Many of the existing homeless families' data systems involve a subset of the population, such as first-time homeless families or families with multiple problems. Others are limited geographically and would have questionable external validity given the context-dependent nature of homelessness. Still others, such as NSHAPC, provide greater external validity and a less selective population but lack the richness of inquiry needed to fully understand the complexity of the individual groups. Similarly, few data sets currently available provide the longitudinal perspective needed to examine the predictive validity of the typology. Given the status of the research, it may be useful to develop a limited number of typologies in the most comprehensive data set and test them in several other candidate data sets. This would provide a greater test of the generalizability of the typology.

View full report


"report.pdf" (pdf, 4.18Mb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®