<p>Persons per household (PpH) is a crucial component for calculating population estimates using the Housing Unit Method. Although decennial census-derived PpH is considered the gold standard in the United States, census values become progressively more outdated over time. The American Community Survey, which measures sociodemographic, economic, and housing characteristics in the United States, provides annual estimates of PpH, but estimates have proven unreliable. Regression modeling offers an alternative method to produce small area estimates of PpH. Most prior research on regression-estimated PpH has focused on counties or metropolitan areas; only a couple of studies have produced regression-based PpH estimates for sub-county geographies, each with its own limitations. This paper seeks to extend the literature on small area estimates of PpH by producing tract-level estimates that can be combined into custom geographies, using New York City as a case study. Three alternative spatial error models were explored, leveraging data from the decennial census, American Community Survey, and local administrative records for 2,218 tracts. To address high levels of uncertainty of estimates at the tract level, the efficacy of averaging was tested by producing composite average estimates from the models. Additionally, accuracy of aggregated estimates for groups of tracts was explored. Results indicate that grouping tracts may produce more accurate estimates of PpH. Although no one method of estimating post-censal PpH may be best for all contexts, this paper proposes a method for producing PpH for flexible, customizable sub-county geographies that may be adaptable to other places in the United States.</p>