Counting the homeless in Los Angeles County

Over the past two decades, a variety of methods have been used to count the homeless in large metropolitan areas. In this paper, we report on an effort to count the homeless in Los Angeles County, one that employed the sampling of census tracts. A number of complications are discussed, including^{E} the need to impute homeless counts to areas of ^{E}the County^{E} not sampled. We conclude that, despite their imperfections, estimated counts provided useful and credible information to the stakeholders involved.

💡 Research Summary

This paper presents a detailed account of the 2004‑2005 effort to enumerate the homeless population in Los Angeles County, focusing on the statistical design, field implementation, and analytical methods used, as well as the practical challenges encountered. The project was commissioned by the Los Angeles Homeless Services Authority (LAHSA) under pressure from the U.S. Department of Housing and Urban Development (HUD), which required a credible estimate of homelessness to allocate federal funding.

The authors review prior approaches to homeless counting, including Rossi’s 1985 block‑sampling method, capture‑recapture techniques, and regression‑based small‑area estimation using service‑provider data. They note that each of these methods suffers from various limitations such as reliance on enumerator accuracy, strong untestable assumptions, and difficulties in integrating multiple data sources.

For the Los Angeles County count, the basic sampling unit was the census tract (2,054 tracts in total). Based on expert judgment and earlier studies, 211 tracts expected to contain large numbers of homeless individuals were selected with certainty. The remaining tracts were stratified by eight Service Provision Areas (SPAs) – administrative zones that roughly correspond to distinct geographic and political sub‑regions – and a proportionate random sample of 299 tracts was drawn across these strata. This design combined a “certainty” component (to guarantee coverage of high‑density areas) with a stratified random component (to provide unbiased representation of the rest of the county).

Field work took place over three nights (January 25‑27, 2005). Enumerators were paired with “guides” – individuals recruited from the homeless community – to help locate people within each tract. During implementation, three incorporated cities (Long Beach, Pasadena, Glendale) refused to cooperate; consequently, the homeless counts for these jurisdictions had to be imputed using statistical models rather than direct observation. Additionally, in SPA 4 (Metro Los Angeles) the originally drawn random tracts were replaced by adjacent tracts, and the authors treated these replacements as if they had been selected with certainty, effectively reducing the random sample size for that stratum.

Estimation proceeded by computing, for each SPA, a Horvitz‑Thompson‑type expansion estimator:

\