Estimating Homeless Populations through Structural Equation Modeling

This article overviews the results from a test of a model of homeless populations throughout the 3,141 counties of the United States. The data were extracted from the 1990 Census, a Census Bureau survey of its enumerators at completion of the census, and other governmental sources. The model was tested using the generally weighted least squares algorithm, as implemented under the Extended LISREL model. It was found that urbanization, servicetization, McKinney funding, and systematic error arising out of more vigilant enumeration efforts in urban areas, collectively explained 80% of the variation in rates of homelessness. The model was then used to correct for enumeration error and to estimate the actual levels of homelessness in both 1990 and 1995. The 1990 estimates were compared with the results of independent estimates for selected localities. After the adjustment for uneven enumeration efforts, the model suggests that a population of 479 thousand homeless persons in 1990, had declined to 383 thousand by 1995