A second set of analyses were performed to answer the questions:
- What are the risk and protective factors that differentiate homeless families from all others?
- What are the risk and protective factors that differentiate residentially stable families from all others?
To answer these questions, statistical procedures (logistic regressions) were used that could test for the effects of all relevant variables at one time (rather than one at a time, as in the descriptive analyses). By looking at all variables simultaneously, it is possible to identify variables that are relatively more important in distinguishing residentially stable families from all others or those that are relatively more important in distinguishing homeless families from all others. The variables that set residentially stable families apart from others may be considered "protective" factors for homelessness and residential risk, while the factors that distinguish homeless families from all others can be considered potential "risk" factors for homelessness.
Logistic regressions were computed for Year 1 groups, Year 3 groups, and the combined residential groups. Only those variables that showed substantial variation between housing groups (e.g., there were statistically significant differences between stably housed or homeless and at least two of the three other groups) or were considered important background and demographic characteristics were included in these analyses. Each logistic model began by entering all of the variables in the model, and then removing non-significant variables4. Tables 5-4 and 5-5, which show the results from these logistic analyses, list all of the variables that were initially included in the model (e.g., non-shaded variables), but parameter estimates are only shown for those variables that were statistically significant at the .05 level in the final models.
Three models examined the factors that related to a family experiencing recent homelessness at Year 1, Year 3, and at either time-point. Three additional models examined the factors that related to a family remaining residentially stable at Year 1, Year 3, and at both time-points.
Homelessness. Table 5-3 presents the results of the three homeless models (Year 1, Year 1 and 3, and Year 3). The Nagelkerke R2 (a pseudo- R2 statistic that measures the amount of variance explained by the model) for the Year 1 and Year 1-3 models are both less than .2, indicating that neither model is doing a very good job of fitting the data. The Year 3 model has a Nagelkerke R2 of .333, however, indicating that this is a better fitting, more powerful model (closer to Cohen's definition of a medium effect).
Only one variable, income, is significant in all three models. Families with relatively higher household incomes were consistently less likely to experience homelessness, an effect that was strongest for the Year 3 model (parameter estimate of -.303).
A few variables were significant in two of the three models. Receiving housing assistance (local, state, or Federal) appears to be a protective factor. People who reported receiving housing assistance at baseline or Year 1, as well as those who obtained housing assistance during the followup period, (having a negative coefficient for the change score) were also less likely to experience homelessness.
Mental health issues, substance abuse issues, and reports of domestic violence were also somewhat related to a greater likelihood of experiencing homelessness. Finally, receipt of TANF was positively related to the likelihood of becoming homeless, but was likely a proxy for need and lack of income rather than a predictor of homelessness.
Stably Housed. The descriptive analyses showed that it was often the Stably Housed group that differed the most from the other residential groups. Table 5-4 presents models that examine factors to predict who was residentially stable at Year 1, at Year 3, as well as Year 1 AND Year 3. The overall fit of all three models is fairly consistent and low; Nagelkerke R2 of .221 for the Year 1 model, .183 for the Year 3 model, and .197 for the Year 1-3 model (all would be considered small effects). Table 5-4 presents the results for the stably housed group.
Looking for results that were significant in more than one model showed that living with a partner/spouse, at least at baseline, made it more likely that a mother would be residentially stable. Changes in this relationship, however, had contradictory effects in different models; in the Year 1 model, having a partner join the household was associated with greater likelihood of being stable, whereas in Year 3, the household was less stable if a partner joined (or more stable if the partner left). Perhaps this was due, in part, to the decrease in partner employment noted earlier in Year 3.
The more adults there are living in the household, and having a spouse/partner who is working or who has found employment, all make it more likely that a mother will be residentially stable. Living in public housing was also frequently associated with being stably housed, while obtaining public housing was significant only for the combined Year 1/Year 3 outcome.
Factors that made it less likely that someone would be residentially stable somewhat mirror the results of the homeless analyses. Reported substance use and mental health issues made it less likely that a woman would be residentially stable in all three models. Poorer reported physical health was also associated with a decreased risk of residential stability in the combined model, and reported domestic violence was significant in two of the models.