Understanding the Personal Networks of People Experiencing Homelessness in King County, WA with aggregate Relational Data

This paper analyzes a unique three-year dataset of over 3,000 unhoused individuals in King County, WA, revealing a concerning trend of declining social connectivity and increasing isolation driven by displacement and population growth, which underscores the urgent need for policies that foster community building and reduce social fragmentation.

The Gist

Imagine the homeless community in King County, Washington (which includes Seattle), not as a scattered group of individuals, but as a giant, invisible web. This paper is a deep dive into the threads of that web, trying to understand how people connect, who they know, and how those connections are changing over time.

Here is the story of that web, told in simple terms.

The Big Picture: A Growing Crowd, A Shrinking Web

Think of the homeless population like a party that is getting bigger every year. In 2022, there were a certain number of guests. By 2024, the room was 23% fuller.

Usually, when a party gets bigger, you might expect more people to know each other. But this study found the opposite is happening. The web is getting thinner.

Even though there are more people, they are knowing fewer people.

• The "Acquaintance" Drop: In 2023, the average person knew about 80 other people experiencing homelessness. By 2024, that number dropped to about 40. It's as if the room got twice as crowded, but everyone is suddenly wearing noise-canceling headphones and can't see anyone else.
• The "Close Friend" Stability: The number of people you can really count on (your "close friends") stayed steady at about 2.5. But because the total crowd grew so much, those 2.5 friends now represent a much smaller slice of the whole pie. The community is becoming less "dense" or connected.

The Four Types of Connections

The researchers asked people four different questions to map out their social world, like checking four different layers of an onion:

1. The "Acquaintance" Network (The Crowd): "How many people experiencing homelessness do you know by sight?"
   • Result: This number is crashing. People feel less visible and less connected to the wider community.
2. The "Close Friend" Network (The Inner Circle): "Who are the people you talk to about important things?"
   • Result: This stays steady at about 2 or 3 people. These are the lifelines.
3. The "Family" Network (The Kin): "Who is your family?"
   • Result: This is the only part of the web that is growing. More people are finding that their siblings, parents, or children are also homeless. Families are sticking together tighter because the rest of the world feels so disconnected.
4. The "Referral" Network (The Chain Reaction): "Who would you send a survey to?"
   • Result: This shows how information spreads. It turns out that women are the super-spreaders of information. They are much better at recruiting others into the study (and likely into support networks) than men are.

Why is the Web Breaking?

The authors suggest a few reasons why the community is becoming more "anomic" (a fancy word for feeling disconnected and without rules):

• The "Displacement" Effect: Imagine if the police or city workers kept moving people out of one park and forcing them to sleep in a different park across town. If you get kicked out of your spot every week, you can't build a neighborhood. You lose your neighbors. The study suggests that frequent displacement by local authorities is breaking these social ties.
• The "Newcomer" Effect: With so many new people arriving who have never been homeless before, the "old guard" (people who have been on the streets for years) might not know them yet. The community is growing faster than the friendships can form.

The "Family" Lifeline

Because the wider community is falling apart, people are retreating into their families. The study found that family groups are getting larger. If one person loses their home, their brother, sister, or child often ends up on the street with them. The family unit is becoming the primary safety net because the "community safety net" is fraying.

What Does This Mean for Policy?

The authors argue that we can't just build more shelters; we need to build community.

• Stop the Moving: If you keep moving people around, you break their social webs. Policies should try to keep people in one place so they can build relationships.
• Create "Living Rooms": We need more safe, permanent places for people to gather, hang out, and meet their neighbors.
• Support the Women: Since women are the ones holding the information networks together, policies should specifically support them as key connectors in the community.

The Bottom Line

The homeless community in King County is growing, but it is becoming more isolated. People are losing their neighbors and relying more heavily on their immediate family to survive. To fix this, we need to stop breaking their social connections and start helping them build new ones.

Technical Summary

1. Problem Statement

The social networks of people experiencing homelessness are a critical yet understudied determinant of survival, safety, and access to resources (information, amenities, emotional support). Traditional methods for counting and studying this population, such as the Point-in-Time (PIT) census, often rely on visual counts or shelter records, which may undercount the unsheltered population and fail to capture the complex social dynamics of the community.

With the U.S. homeless population rising by 23% between 2022 and 2023 (reaching 653,100 nationally), there is an urgent need to understand how social connectivity changes as the population grows. Specifically, the authors investigate whether the expanding homeless population is becoming more connected or more fragmented (anomic), and how different types of ties (acquaintances, close friends, kinship) evolve over time.

2. Methodology

Study Design and Sampling:

• Location: King County, Washington (including Seattle metro area), a region with ~2.3 million people and a significant unsheltered population.
• Data Collection: The study utilizes a three-year dataset (2022, 2023, 2024) collected via Respondent-Driven Sampling (RDS), a peer-referral method designed for hard-to-reach populations.
• Sample Size:
  • 2022: 671 respondents (Aggregate data only due to privacy constraints).
  • 2023: 1,106 respondents.
  • 2024: 1,466 respondents.
  • Total: Over 3,000 unique individuals surveyed.
• Procedure: "Seeds" (initial participants) recruited by service workers were given coupons to recruit peers. The process was tracked via "waves" (diffusion chains), with the longest chain in 2024 reaching 20 waves.

Data Instruments (Aggregate Relational Data - ARD):
The study collected four distinct network measures:

1. Peer Referral Network: The number of people a respondent successfully recruited (capped at 3).
2. Acquaintance Network: An aggregate count of the total number of homeless individuals the respondent knows (by sleep location).
3. Close Friendship Network: A list-based generator of specific close friends who are homeless.
4. Kinship Network: A list of family members (including extended kin) who are homeless.

Statistical Modeling:

• Estimation: Population-level statistics were estimated using the RDS-II estimator to correct for sampling biases (e.g., high-degree individuals are more likely to be sampled).
• Regression Analysis: The authors employed Generalized Linear Models (GLM) with count data distributions to model network size (degree) based on covariates (age, gender, race, health status, chronic homelessness).
• Model Selection: Models tested included Poisson, Negative Binomial (NB), Zero-Inflated Poisson (ZIP), and Zero-Inflated Negative Binomial (ZINB). Selection was based on AICc (corrected Akaike Information Criterion) to handle overdispersion and zero-inflation.
  • Acquaintance Network: Best fit by Negative Binomial.
  • Close Friendship & Kinship: Best fit by ZINB.
  • Referral Network: Best fit by ZIP.

3. Key Contributions

• Largest Network Study: This represents the most extensive dataset on the social networks of the U.S. homeless population to date, covering over 3,000 individuals across three years.
• Methodological Innovation: It successfully applies Aggregate Relational Data (ARD) and RDS to estimate not just population size, but the structural properties (degree distribution, mixing rates) of the homeless social network.
• Longitudinal Insight: By analyzing data across three waves, the study captures the temporal dynamics of network fragmentation as the homeless population grows.
• Policy-Relevant Metrics: The study moves beyond simple counts to measure "connectivity density," providing actionable data for emergency response (e.g., heat warnings) and social service delivery.

4. Key Results

A. Decline in Interconnectedness (Anomie):

• Acquaintance Network: The mean degree (average number of homeless people known) dropped significantly from ~80 in 2023 to ~36 in 2024.
• Close Friendship Network: The mean number of close friends remained stable at ~2.5 (2.6 in 2023, 2.5 in 2024).
• Implication: Since the total homeless population grew by ~23% while the average number of acquaintances halved, the density of the network has decreased. The community is becoming less connected; the probability of two homeless individuals knowing each other has declined.

B. Shift Toward Kinship:

• Kinship Network: The mean degree for family groups increased from 1.76 (2023) to 1.98 (2024).
• Family Structure: There is a growth in the size of family units living together. The data suggests that as non-familial social connections weaken, the extended family unit is becoming the primary source of support.
• Demographics: Black, Asian, and Native Hawaiian populations showed statistically larger family groups compared to White populations. Chronically homeless individuals also had larger family groups.

C. Demographic Variations:

• Age: Older individuals reported larger acquaintance networks, while younger individuals reported fewer close friends but larger kinship networks.
• Gender: Women had slightly larger close friendship networks than men. In the referral network, women were significantly more effective recruiters than men.
• Race/Ethnicity: Hispanic/Latino and Multiracial groups reported larger acquaintance networks compared to White respondents.

D. Diffusion and Mixing:

• Recruitment Trees: The RDS diffusion process showed many short chains and a few "super-spreader" chains, similar to online social networks.
• Homophily: High levels of homophily (preference for similar others) were observed in both gender and race mixing. For example, White respondents were most likely to recruit other White respondents (59% in 2024).

5. Significance and Implications

Theoretical Significance:
The study provides empirical evidence that rapid population growth in marginalized communities can lead to social fragmentation (anomie). As the population expands and displacement events increase, the "weak ties" (acquaintances) that facilitate information flow and resource sharing are dissolving, leaving individuals more isolated.

Policy Implications:

1. Emergency Response: The decline in network density suggests that traditional "word-of-mouth" emergency alerts (e.g., heat domes) may fail. New strategies must target "hubs" or well-connected individuals to cascade information effectively.
2. Service Delivery: As non-familial support networks shrink, the family unit is becoming the critical safety net. Policies should support family-based housing solutions rather than solely individual-focused interventions.
3. Displacement Mitigation: The authors hypothesize that frequent encampment displacements and legal changes (e.g., Grants Pass decision) are disrupting social networks. Reducing displacement and providing stable congregating spaces are essential to rebuilding community cohesion.
4. Data Collection: The success of RDS and ARD in this context validates these methods as superior to traditional visual counts for understanding the social fabric of the homeless population.

Limitations:
The study acknowledges standard RDS limitations, including potential convergence issues and recall bias in network reporting. Additionally, the lack of "alter-alter" tie data limits the ability to model full network clustering.

Conclusion:
The homeless population in King County is becoming more numerous but less connected. While the "close friend" network remains stable, the broader community network is fracturing, forcing a reliance on kinship. Future policy must prioritize rebuilding social infrastructure to prevent isolation and improve the efficacy of support systems.