Sample. The basic sample would be a random sample of families requesting shelter for the first time. Depending on resources, the sample could include oversamples of families who come from two-parent families, father-only families, and families who are working to allow greater attention to these understudied groups.
The study could be enhanced by the addition of two other cohorts: families who have previously been homeless at least once, and families who are comparably poor, but domiciled and never homeless. This latter group would need to be selected from a separate sampling frame, such as Temporary Assistance for Needy Families (TANF) rolls.
To achieve a nationally representative sample of shelter requests, a stratified, multistage cluster sample would be used. Similar to the design used in the National Survey of Homeless Assistance Providers and Clients (NSHAPC) (Burt, Aron, Douglas, Valente, Lee, and Iwen, 1999), the first stage of the proposed design would include sampling of metropolitan statistical areas (MSAs) and, for non-MSAs, Community Action Agency (CAA) catchment areas. These sampling units would be clustered according to geography, population size, and economic indicators (e.g., per capita income, percent unemployed). Random samples of MSAs and CAAs within each cluster would then be chosen. All homeless shelters within each MSA or CAA would be identified and, if the number is too large, a random sample of these programs would be chosen. If there are specific subgroups that need to be oversampled, such as two-parent families, shelters could be clustered by type of populations served. Depending on resources, either a complete census of families requesting shelters for the first time or a random sample of these families could be sampled.15, 16
Time Frame. Families would be contacted to participate in the study at the time of the shelter request and would be followed for at least two years, and up to five years, following the shelter request.