Point-in-time Trend Exploration
\[\\[1in]\] This trend analysis explores King County point-in-time estimates of homelessness, vacancy rates, and rental prices from 2019 to 2025. Each dataset is described individually, followed by an exploration of relationships between them.
Datasets
Point-in-time counts
Annual point-in-time counts estimate the number of both sheltered and unsheltered people. From 2019 to 2025 both counts rose, but the unsheltered count grew by 87%, far outpacing the sheltered growth of 19%.
The figure below shows sheltered status as a proportion of the total. In 2019 the unsheltered population was 47% of the total population, and by 2025 was the majority of the population at 60%.
Vacancy
To measure the rental vacancies in King County from 2019 to 2025, I
used the American Community Survey (ACS) which is a component of the
Census. Specifically the variable B25004 contains details
on different types of vacancies including rentals, units for sale, and
units of seasonal/recreational housing. Here I’m using only
For Rent vacancies. Note: The ACS is released
annually, generally in September, so vacancy data are not yet available
for 2024 or 2025.
\[\\[0.25in]\]
From 2019 to 2023 rental vacancies in King County increased from 14,222 to 31,552 vacant units, a 45% increase.
Rent prices
To explore rent prices I looked at 3 distinct datasets:
- Zillow Observed Rent Index.
- UW’s Washington Center for Real Estate Research (WCRER) dashboards.
- HUD’s Fair Market Rents (FMRs).
Of these three data sources, only the HUD FMR data provided detail on rentals by the number of bedrooms available, so I used the FMR data for subsequent trend analysis. I used the Zillow and WCRER data as a consistency check.
\[\\[0.25in]\]
\[\\[0.25in]\]
\[\\[0.25in]\]
All three data sources are in rough agreement in terms of their general upward trend. All three likewise show a market shock in 2020/2021 due to the COVID pandemic. For a short time there was a slowing in the growth of rental costs which resumed by 2023.
To measure the agreement between the rental costs I measured the difference between the minimum and maximum costs (plotted below). In 2019, the 3 sources of rent costs were in strong agreement, only diverging by $70 from the minimum cost estimate to the maximum. From 2020 to 2025 the estimates begin to diverge with the HUD FMR2 estimate generally higher than the other estimates. This is not wholly unexpected as the Zillow and WCRER estimates include studio and 1-bedroom rentals that are lower cost on average compared with 2-bedroom rentals. The Zillow mean and medians are in strong agreement.
Exploring trends
The correlation matrix below displays the relationships between, sheltered and unsheltered point-in-time counts, vacancies, and rental prices. Across the top are the x-axis labels, and down the left hand side are the y-axis labels. The lower half shows scatter plots of every combination of variable, and the upper half shows Pearson correlation coefficients for each pairwise variable. The asterisks and dots on the correlation coefficients indicate the level of statistical significance between two variables and there are 3 significant correlations. First the the correlation between the unsheltered PIT count and the total PIT count are highly correlated (0.992, p < 0.01), which is to be expected. There is also a strong correlation between the monthly rent cost and the Sheltered PIT count (0.856, p < 0.05) and monthly rent and the total PIT count (0933, p < 0.10).
Basic modeling
The data are sparse, but I wanted to quantify the relationship between sheltered PIT counts, vacancies and rents that could generate preliminary hypotheses to explore on a larger dataset in the future. I fit the following simple regression:
\(Sheltered\ PIT = \beta_0 + \beta_1(vacant\ units) + \beta_2(monthly\ rent)\)
The model outputs are in the table below. Notably, none of the coefficients achieved statistical significance.
| term | estimate | std.error | statistic | p.value |
|---|---|---|---|---|
| (Intercept) | 1280.55 | 1800.24 | 0.71 | 0.55 |
| vacancy_for_rent | -0.05 | 0.04 | -1.29 | 0.33 |
| hud_fmr_2 | 2.74 | 1.13 | 2.43 | 0.14 |
Model interpretation
The equation describing the PIT count is below. At a high level it shows a negative relationship between vacancies and PIT counts, i.e. as vacancies rise, PIT counts fall suggesting that increased rental capacity reduces sheltered PIT counts. The model shows a positive relationship between rent cost and PIT count, so as rents rise, PIT count also rises.
\(PIT = 1280 - 0.05(vacant\ units) + 2.74(monthly\ rent)\)
- The first term is the intercept, which is this case doesn’t have much meaning. If monthly vacancies where at 0 units and monthly rent was $0, the sheltered PIT count would be roughly 1,280 people.
- The coefficient of 0.05 on vacancy rate means that, all else equal, every 20 additional units of rental inventory results in a reduction of 1 person in the sheltered PIT count.
- The coefficient of 2.74 on the monthly rent means that, all else equal, every additional dollar of rent, increases the sheltered PIT count by nearly 3.
Limitations
Missing data is a major limitation in this analysis. We have incomplete Census data in 2024 and 2025 and while we have annual sheltered PIT counts, we only biannual unsheltered making matched analysis impossible.
The model is also limited by the depth of data. There are a few technical details that I omitted in this example, but would apply to a larger dataset. One is that the outcome is a count, so a generalized linear model such as a Poisson regression would be more appropriate. Also, this is time series data, which means that the errors are likely correlated over time, and failure to account for correlation can lead to unstable estimates of the standard error.