Full-text article: https://www.tandfonline.com/doi/full/10.1080/10511482.2025.2570727

🏡 American Housing Shortfalls: Replication Package

This repository contains the full replication code for Changes in Average Household Size and Headship Rates as Indicators of Housing Shortfalls by Peter Hepburn and Lorae Stojanovic (2025). We explain in detail how to replicate the analysis on your personal work environment.

💡 Key Findings

Between 2000 and 2019, the average number of people per U.S. household declined modestly—from 3.47 to 3.37, a 3% decrease. At first glance, this seems to contradict claims of a worsening housing shortage: if household sizes are shrinking, doesn't that imply housing is becoming less scarce?

We take a demographic decomposition approach proposed by Galster (2024) and account for the fact that the U.S. is (among other things) older, more educated, and more racially and ethnically diverse than in 2000. The first two shifts tend to drive down household size, while the latter has potentially mixed effects. We find that:

• Net of demographic change, Americans live in larger households than before. For example, the average 45-49-year-old lives in a household of 3.14 persons in the year 2000; by 2019, that had increased to 3.33. (See Accessory Figure 6.)

  We estimate these patterns using granular IPUMS microdata that classify each individual by age, gender, tenure, income, race/ehnicity, education, birthplace, and geography of residence.

• Even after accounting for all these demographic changes, household size fell less than expected. As shown in Figure 4 below, these measurable characteristics alone would have predicted a 34% larger decline in average household size—to 3.342—than what actually occurred.

  Figure 4: Observed vs. Counterfactual Household Size and Headship Rates

• Headship rates tell a similar story. The headship rate is the share of people counted as "household head" or "person one" in the survey — typically the person who pays the rent or mortgage and serves as the household's primary financial support. For example, if one-third of women age 30-34 is counted as the "person one" in the survey data, then the headship rate among this group is 33%.

  Based on the same set of sociodemographic correlates, we would have expected headship rates to rise between 2000 and 2019, from 38.6% to 40.8%. Instead, they slightly declined, to 38.2%. This implies that, even after adjusting for population changes, Americans today form fewer independent households than in the past.

• Together, these findings point to a shortfall in household formation. Using household size as a benchmark, we estimate a minimum shortfall of 1.9 million households in 2019 relative to expectations based on 2000 patterns. Using headship rates as an alternative metric, the shortfall is 8.4 million households. (For details on why the household-size estimate is a lower bound, and is lower than the headship measurement, as well as proofs of the decomposition and measurement properties, please see Figure 1 and associated explanation in the full text.)

• Our findings reveal that there is a sizable shortfall of households relative to standards from 2000. This could reflect constraints due to housing supply, or it could reflect changing norms around cohabitation. Further research is needed to distinguish between these two explanations.

⚡ How to Replicate: Quick Start

For experienced users who just want to get the project running right away. If you have trouble following these steps, please follow the Detailed Start guide below.

1. Navigate to the directory where you want the project to be saved and clone both required repos side by side

   cd your/path/to/parent/directory

   git clone https://github.com/lorae/american-housing-shortfalls american-housing-shortfalls
   git clone https://github.com/lorae/demographr demographr

2. Enter the main project

   cd american-housing-shortfalls

3. Copy the environment file and edit it with your own IPUMS API key and Census API key

   cp example.Renviron .Renviron
   # (Windows PowerShell: Copy-Item example.Renviron .Renviron)
   # IMPORTANT: open .Renviron and replace "your_ipums_api_key" with your actual key

4. Restore dependencies and run the analysis

   Open american-housing-shortfalls.Rproj in your preferred IDE, then in the R console:

   renv::restore()
   source("run-all.R")

📎 How to Replicate: Detailed Start

Detailed instructions for how to fully install and run this project code on your computer.

Part A: Clone the repo and configure the R project

These steps will allow you to install the code on your computer that runs this project and set up the environment so that it mimics the environment on which the code was developed.

1. Clone the repo: Open a terminal on your computer. Navigate to the directory you would like to be the parent directory of the repo, then clone the repo.

   MacOS/Linux:

   cd your/path/to/parent/directory

   git clone https://github.com/lorae/american-housing-shortfalls american-housing-shortfalls

   Windows:

   cd your\path\to\parent\directory

   git clone https://github.com/lorae/american-housing-shortfalls american-housing-shortfalls

2. Open the R project: Navigate into the directory, now located at your/path/to/parent/directory/american-housing-shortfalls. Open american-housing-shortfalls.Rproj using your preferred IDE for R. (We use R Studio.)

   Every subsequent time you work with the project code, you should always open the american-housing-shortfalls.Rproj file at the beginning of your work session. This will avoid common issues with broken file paths or an incorrect working directory.

3. Initialize R environment: Install all the dependencies (packages) needed to make the code run on your computer. Depending on which packages you may have already installed on your computer, this setup step may take from a few minutes to over an hour.

   First, ensure you have the package manager, renv, installed. Run the following in your R console:

   install.packages("renv") # Safe to run, even if you're not sure if you already have renv

   library("renv")

   Then load all the packages needed for the project:

   renv::activate()
   renv::restore()

   When it asks if you want to proceed, type y for "yes".

Part B: Initialize project for download from IPUMS USA and the U.S. Census

The IPUMS Terms of Use precludes us from directly sharing the raw microdata extract, however, the data used in this analysis is freely available and simple to download after setting up your API key. A section of our manuscript also references results from the McClure-Schwartz analysis. We produced these numbers in src/scripts/mcclure-schwartz-replication-state-surplus.R and they require a Census API key.

4. Copy the file example.Renviron to a new file named .Renviron in the project root directory. You can do this manually or use the following terminal commands:

   MacOS/Linux:

   cp example.Renviron .Renviron

   Windows:

   copy example.Renviron .Renviron

5. Set up your IPUMS USA API key: If you don't already have one, set up a free account on IPUMS USA. Use the new account to login to the IPUMS API Key webpage. Copy your API key from this webpage.

6. Set up your Census API key. Copy your API key from this website

7. Open .Renviron and replace your_ipums_api_key and your_census_api_key with your actual keys. Do not include quotation marks. R will automatically load .Renviron when you start a new session. This keeps your API key private and separate from the codebase.

   🛑 Important: .Renviron is listed in .gitignore, so it will not be tracked or uploaded to GitHub — but example.Renviron is tracked, so do not put your actual API key in the example file.

Part C: Run the analysis scripts

8. Source run-all.R.

   source("run-all.R")

🧾 How to Cite

Hepburn, P., & Stojanovic, L. (2025). Changes in Average Household Size and Headship Rates as Indicators of Housing Shortfalls. Housing Policy Debate, 1–15. https://doi.org/10.1080/10511482.2025.2570727