Inequality in Housing Payment Insecurity Across the United States During the COVID-19 Pandemic: Who Was Affected and Where?

Abstract

Widespread job losses and economic disruptions during the COVID-19 pandemic led to significant housing payment insecurity, disproportionately affecting various demographic groups and regions across the United States (US). While previous studies have explored the pandemic’s impact on housing insecurity, they all focused on specific periods, populations or areas. No study has yet provided a comprehensive analysis of inequality in housing insecurity throughout the pandemic, particularly in terms of spatial disparities. Our study addresses this gap by analyzing individual-level and aggregated data from the Household Pulse Survey (HPS) (N = 2,062,005). The findings reveal heightened vulnerability among individuals aged 40–54, those with lower education and income, Black and Hispanic/Latino populations, women, households with children, individuals who experienced job loss, the divorced, and larger households. Renters experienced greater housing insecurity than homeowners. A hotspot analysis identified the southeastern US as a region of acute housing insecurity, revealing that insecurity cannot be solely measured by affordability. The regression results show that poverty is the main reason for differences in housing insecurity across places, and rent burden is also important. The geographically weighted regression (GWR) model shows stronger coefficients in southern states, highlighting that poverty and rent burden are particularly influential factors in these areas. This study shows the multifaceted nature of housing insecurity, calling for targeted group or location policy interventions.

1. Introduction

Housing plays a crucial role in individual and family well-being, social stability, health, and economic participation. Secure and affordable housing allows individuals to maintain a stable environment conducive to physical and mental health, educational success, and participation in the workforce. Conversely, housing insecurity—defined as difficulty in paying rent or mortgages, fear of eviction, or overcrowded conditions—can lead to significant stress and hardship. This, in turn, affects various dimensions of life, reducing quality of life and contributing to broader societal inequality. The negative impact of housing insecurity on individuals’ quality of life has been well documented. Insecure housing creates uncertainty and financial strain, which can heighten stress and anxiety, weaken social networks, and even impact physical health. For families with children, housing instability is linked to disruptions in education, while for the elderly and other vulnerable populations, it can lead to social isolation and poor health outcomes. However, the onset of the COVID-19 pandemic created new layers of economic vulnerability, intensifying housing insecurity across already disadvantaged populations.

The COVID-19 pandemic exacerbated the economic challenges, particularly for vulnerable populations, leading to unprecedented housing insecurity across the United States (US). While earlier studies have provided valuable insights into housing insecurity during the COVID-19 pandemic (e.g., Lake, 2020; Versey, 2021; Ong et al., 2022; Versey & Russell, 2023; Chun et al., 2023; Friedman, 2023; Besbris et al., 2024; Coughenour et al., 2024; Willie et al., 2024; Cornelissen & Hermann, 2023; Reiser et al., 2025), significant gaps remain. To the best of our knowledge, no peer-reviewed research has systematically examined housing insecurity disparities across the entire pandemic, particularly in terms of nationwide spatial disparities. Additionally, there is a lack of comprehensive studies exploring inequality in housing insecurity across different market segments—rental versus owner-occupied housing—making further research necessary to identify which demographic and socioeconomic groups were most affected within these segments.

First, most earlier studies offer a limited snapshot of the housing insecurity situation, particularly during the early stages of the pandemic (Cornelissen & Hermann, 2023). Comprehensive investigations covering the entire pandemic period across the US, especially the later stages, are lacking (Rosenbaum & Friedman, 2024). A study encompassing data from the early to late stages of the pandemic is essential to fully understanding the issue of housing insecurity because of the dynamic nature of the pandemic. The impact of a pandemic might change significantly over time due to variations in public health responses, economic conditions, and the virus itself (such as mutations leading to new variants). This dynamism might alter the circumstances under which individuals, including vulnerable groups, experienced housing insecurity. Government policies and support systems (like unemployment benefits, stimulus checks, eviction moratoriums, and emergency housing assistance) also evolved as the pandemic progressed. These changes might significantly influence housing stability. In addition, individuals or states might adapt differently to the prolonged stresses of a pandemic, including adjustments in spending, saving, and other financial behaviors. These adaptations might have significant implications for housing security. For instance, people might have cut back on non-essential expenses, taken on additional work, moved in with family, or relocated to areas with a lower cost of living. These adaptations could have shifted the distribution of housing insecurity, particularly for vulnerable groups who were initially most affected. Furthermore, as the pandemic evolved, new risk factors for housing insecurity emerged. For example, initial stages marked by immediate job losses, while later stages saw impacts from long-term reductions in income or changes in rental markets. In addition, the pandemic saw significant changes in the housing and rental markets, with rapid increases in housing costs in some regions, alongside a shortage of affordable housing. These shifts likely worsened housing payment insecurity for both renters and new homeowners, especially in high-demand areas. Vulnerable groups, who were already struggling with high housing cost burdens before the pandemic, might have found it even more difficult to maintain secure housing as rents and property prices surged in certain cities and regions during the later stages of the pandemic.

In addition, our current understanding of the spatial dimensions of housing insecurity for geographically targeted interventions in the US is limited. While several studies have explored the spatial dimensions of eviction or housing instability before the pandemic (Shelton, 2018; Laniyonu, 2019; Medina et al., 2020; Kang, 2021; K. Kim et al., 2021; Nelson et al., 2021; Connors & Zhang, 2023), few have examined these dimensions during the pandemic, especially with regard to spatial inequality across the entire US. Jones and Grigsby-Toussaint (2021) highlighted that geographic location is crucial for housing stability and that COVID-19 has exacerbated spatial inequalities. However, there is a clear need for more geographic analyses to understand regional variations, as the pandemic’s impact on housing insecurity might vary significantly based on regional economic conditions, housing markets, and the severity of the pandemic in different states.

Lastly, no studies have investigated the association of housing cost burden or poverty with the spatial disparity of housing insecurity during the pandemic. Understanding these relationships is essential for developing geographically targeted policy interventions that address the unequal distribution of housing instability across the US. On one hand, high housing costs might lead to payment insecurity, especially in areas with expensive living costs. In regions with high-priced housing markets, even middle-income families might struggle, and the pandemic’s impact was likely more severe in regions with high housing costs, particularly for those whose income was reduced. On the other hand, poverty might directly affect a household’s ability to afford basic needs, including housing. In areas with higher poverty rates, residents were more likely to experience housing payment insecurity. The pandemic exacerbated economic inequalities and job volatility, making those already in poverty more vulnerable to income disruptions. Both housing cost burden and poverty might be crucial for understanding spatial disparities in housing insecurity. However, to our knowledge, no study has yet explored the role of housing cost burden or poverty in understanding the housing insecurity disparities during the pandemic.

To address these research gaps, our study conducts a comprehensive analysis of both individual-level and aggregated data from a nationally representative sample of 2,062,005 US households in the Household Pulse Survey (HPS). This analysis systematically examines who experienced housing payment insecurity and where it occurred across the US throughout the entire COVID-19 pandemic. We aim to answer three key questions: (1) Which demographic groups (such as large households, minorities, or the elderly) were particularly vulnerable to housing payment insecurity in different housing market segments (rental and owner-occupied)? (2) To what extent did housing cost burden and poverty contribute to the spatial disparity of housing insecurity? (3) Which regions in the US experienced the highest levels of housing payment insecurity during the pandemic?

Identifying the demographic groups particularly vulnerable to housing insecurity is crucial for developing targeted policies to support these groups in future crises. Understanding where housing insecurity was most prevalent can help us recognize geographical disparities in economic resilience and public health crises. This knowledge may guide more efficient allocation of resources and support to the most affected areas. Determining the association of housing cost burden and poverty with spatial disparity of housing insecurity is also crucial for developing targeted policy interventions. Should housing cost burden emerge as the primary issue, policies could focus on developing affordable housing, implementing rent control, or providing housing subsidies. Conversely, if poverty is identified as the main driver, broader economic strategies such as job creation, raising the minimum wage, or expanding social safety nets might prove to be more effective. Although our analysis focuses primarily on housing insecurity and its spatial disparities, the broader implications of these findings suggest that addressing housing insecurity is critical not only for economic recovery but also for improving the overall well-being of individuals, particularly in future crises.

2. Literature Review

The importance of studying housing insecurity lies in its potential to lead to physical and mental health problems and to have profound social and economic impacts on communities, as evidenced in many studies such as those by Marsh et al. (2000), Diette and Ribar (2018), Tinson and Clair (2020), and Egan et al. (2025). Housing insecurity also affects children’s development, as shown by Bess et al. (2022). Children growing up in insecure housing situations and college students lacking secure, affordable housing often face educational challenges, as explored in research by Broton (2020), Glantsman et al. (2022), and Soria et al. (2023). It is important to understand what causes housing insecurity and its effects so that we can establish better policies to fix problems like a lack of affordable housing, income inequality, and poor social services.

Housing insecurity is a multifaceted issue intersecting with fields such as economics, sociology, urban planning, public health, and law. Scholars from various disciplines have analyzed data on housing conditions, affordability, and homelessness using diverse approaches, as demonstrated in the works of Early (2005), Baker et al. (2016), O’Flaherty (1995, 2019), Power (2017), Seymour and Akers (2021), and DeLuca and Rosen (2022). Although inherently multidimensional, housing insecurity is typically measured in only a few dimensions in national estimates, as observed by Cox et al. (2017, 2019). Some studies suggest a significant intercorrelation among these dimensions, implying that one aspect, such as housing affordability, might capture others due to overlapping categories, as indicated by Cox et al. (2017) and Routhier (2019). Most housing research in the literature tends to examine only one of these issues (Routhier, 2019).

Housing affordability and cost burden are key components of housing insecurity. Housing affordability refers to the cost of housing relative to the income of the household. A common benchmark for affordability is when a household spends no more than 30% of its income on housing costs, including rent or mortgage payments, utilities, and other fees, as noted by Leopold et al. (2016) and Bailey et al. (2016). When housing costs exceed this threshold, they can limit a household’s ability to afford other necessities such as food, healthcare, and transportation, a situation referred to as a ‘cost burden’ (Leopold et al., 2016). When housing is not affordable, it increases the risk of housing insecurity. Households that are cost-burdened may have difficulty maintaining stable housing and may be at greater risk of eviction or foreclosure. Studies have been conducted to understand the interplay of these elements (Mimura, 2008). For example, Medina et al. (2020) and Kang (2021) explored the issue of housing instability from different angles and emphasized the role of socioeconomic and demographic factors in housing instability.

Studying housing insecurity presents challenges, as obtaining reliable data on housing conditions, homelessness, and related factors can be difficult. Privacy concerns, resource constraints, and the transient nature of some populations (such as the homeless) contribute to these data limitations and accessibility. In addition, the housing market and its related policies are continuously evolving. This dynamic and evolving nature makes it difficult to keep up-to-date records. Events such as economic recessions, natural disasters, and pandemics can suddenly and dramatically alter the landscape of housing insecurity, necessitating new data and analysis to understand the patterns and impacts of these external shocks. Several studies have been conducted to investigate the impact of economic recessions on housing insecurity (e.g., Niedt & Martin, 2013; Bayer et al., 2016; H. Kim et al., 2017; Zhang & Lerman, 2019). However, due to data limitations, few peer-reviewed studies have systematically examined disparities in housing insecurity.

A more recent external shock—the COVID-19 pandemic—has significantly intensified concerns about housing insecurity. The pandemic not only exacerbated long-standing shortages of affordable housing but also had far-reaching effects on housing markets and economic stability. Concerns about the shortage of affordable housing units and housing subsidies in the US have been ongoing (Bailey et al., 2016), particularly for the nation’s lowest-income renters, who have long struggled with insufficient affordable housing options. This issue has worsened in recent years due to record-high inflation and the loss of low-cost rental homes, affecting renters nationwide (Aurand et al., 2023). The pandemic further amplified these challenges. Housing prices surged in many US regions during this time (Li & Zhang, 2021; Li & Kao, 2022), reducing the availability of affordable housing and increasing housing insecurity.

Although previous studies have examined the pandemic’s impact on housing insecurity in the US, they all have concentrated on specific timeframes, populations, or geographic regions. These studies tend to explore particular aspects or populations of the crisis, without offering a comprehensive analysis that spans the entire pandemic and populations. For instance, Lake (2020), Cai et al. (2021), and Jones and Grigsby-Toussaint (2021) focused on the early stages, highlighting racial or spatial inequalities but limiting their analysis to the initial months. Versey (2021) investigated renter-focused eviction prevention policies, while Ong et al. (2022) emphasized income and racial disparities in California’s rental-assistance programs based on early pandemic data. Chun et al. (2023) and Cornelissen and Hermann (2023) examined racial disparities through the lens of assets and pre-existing vulnerabilities, but only through the early to mid-pandemic period. Friedman (2023), Manville et al. (2023), and Versey and Russell (2023) analyzed renter distress within specific subpopulations, and Rosenbaum and Friedman (2024) addressed racial differences during the later stages of the pandemic.

While these studies provide valuable insights, a key limitation is their narrow focuses on specific phases, populations, or regions, which prevents them from capturing the full extent of housing insecurity inequality throughout the entire pandemic period across all populations and geographic areas in the US. Additionally, most studies lack a national spatial analysis, limiting their ability to reveal broader patterns of spatial inequality across the US. To address these gaps, our study utilizes both individual-level and aggregated data from a nationally representative sample of 2,062,005 US households from the Household Pulse Survey (HPS), covering the entire duration of the pandemic and the entire country. This comprehensive approach allows us to systematically assess demographic impacts and uncover trends in spatial inequality related to housing insecurity nationwide.

3. Data

The data for this study were obtained from multiple sources. Information regarding housing payment insecurity during the COVID-19 pandemic was sourced from both the individual-level public use files and the aggregated data tables of the Household Pulse Survey (HPS). Housing payment insecurity in this study refers to instances of missed housing payments or expressions of low confidence in making rent or mortgage payments, as reported in the HPS. The HPS, a pioneering and nationally representative survey, offers comprehensive and timely insights into the social and economic consequences of COVID-19. Developed and implemented by the US Census Bureau in collaboration with five federal statistical partner agencies—the Department of Housing and Urban Development, the Bureau of Labor Statistics, the Department of Agriculture’s Economic Research Service, the National Center for Health Statistics, and the National Center for Education Statistics—it gathered a large, nationally representative sample to understand the pandemic’s impacts on American households. This study uses HPS data on housing payment insecurity spanning from survey week 1 (23 April–5 May 2020) to survey week 45 (27 April–9 May 2022).

Table A1. The dates and phases of the HPS data collection cycles.

Survey Week	Date	HPS Phase
Week1	23 April–5 May 2020	1
Week2	7 May–12 May 2020	1
Week3	14 May–19 May 2020	1
Week4	21 May–26 May 2020	1
Week5	28 May–2 June 2020	1
Week6	4 June–9 June 2020	1
Week7	11 June–16 June 2020	1
Week8	18 June–23 June 2020	1
Week9	25 June–30 June 2020	1
Week10	2 July–7 July 2020	1
Week11	9 July–14 July 2020	1
Week12	16 July–21 July 2020	1
Week13	19 August–31 August 2020	2
Week14	2 September–14 September 2020	2
Week15	16 September–28 September 2020	2
Week16	30 September–12 October 2020	2
Week17	14 October–26 October 2020	2
Week18	28 October–9 November 2020	3
Week19	11 November–23 November 2020	3
Week20	25 November–7 December 2020	3
Week21	9 December–21 December 2020	3
Week22	6 January–18 January 2021	3
Week23	20 January–1 February 2021	3
Week24	3 February–15 February 2021	3
Week25	17 February–1 March 2021	3
Week26	3 March–15 March 2021	3
Week27	17 March–29 March 2021	3
Week28	14 April–26 April 2021	3.1
Week29	28 April–10 May 2021	3.1
Week30	12 May–24 May 2021	3.1
Week31	26 May–7 June 2021	3.1
Week32	9 June–21 June 2021	3.1
Week33	23 June–5 July 2021	3.1
Week34	21 July–2 August 2021	3.2
Week35	4 August–16 August 2021	3.2
Week36	18 August–30 August 2021	3.2
Week37	1 September–13 September 2021	3.2
Week38	15 September–27 September 2021	3.2
Week39	29 September–11 October 2021	3.2
Week40	1 December–13 December 2021	3.3
Week41	29 December–10 January 2022	3.3
Week42	26 January–7 February 2022	3.3
Week43	2 March–14 March 2022	3.4
Week44	30 March–11 April 2022	3.4
Week45	27 April–9 May 2022	3.4

in the Appendix A details the dates and phases of the HPS data collection cycles. Each week’s public use file contains individual-level survey responses. We integrated the public use files from all 45 survey weeks for individual-level data analysis and excluded entries marked with −99, which indicate unanswered questions. Following this data cleaning process, over 2 million individual observations remained for the analysis (N = 2,062,005). The aggregated data tables from the 45-week survey period were utilized to study the locations of these housing payment insecurity instances. For clarity, the term “housing insecurity” is used throughout the rest of this paper to specifically denote housing payment insecurity.

Poverty, housing cost burden data, and other socio-economic control variable data—representing demographics, race, housing, education, employment, economic status, and commuting patterns—were acquired from the most recent American Community Survey (ACS) 5-year estimates (2017–2021) by the Census Bureau. Households are considered ‘cost-burdened’ when over 30% of their income is spent on housing costs, a condition explored by Leopold et al. (2016). This measure was employed to represent housing cost burden in our study. The TIGER state boundary shapefiles were downloaded from the Census Bureau website.

Control variable data on the total daily COVID-19 cases came from USA Facts. This comprehensive dataset compiles information from sources including the CDC and state/local public agencies. The accuracy of the COVID-19 data was confirmed through direct referencing of state and local agencies. We calculated the percentage increase in COVID-19 cases for each HPS week as one control variable.

We also used dynamic unemployment data during the pandemic as another control variable, sourced from the Bureau of Labor Statistics. Since this dataset is monthly, we estimated unemployment rates for specific HPS weeks using linear interpolation. Additionally, we used data on state-level changes in unemployment benefit policies as an extra control variable. Considering federal unemployment benefits as a foundational support, we posit that initial state-level modifications inherently yield benefits for economically disadvantaged households. This data, from Katz et al. (2023), includes various policies established during the pandemic.

Table A2. Definitions of the variables used in this study.

Variable	Description	Data Source
PctHouseInsecurity	Percent of House Insecurity	HPS
PctPoverty	Percent of Population whose income in the past 12 months is below poverty level	ACS
PctRentGE30PCT	Percent of Renter Households for whom Gross Rent is 30.0 Percent or More of Household Income	ACS
PctMtgeGE30PCT	Percent of Owner Households with Mortgages whose Monthly Owner Costs are 30.0 Percent or More of Household Income	ACS
PctCovidIncrease	Percent of COVID Case Increase	USA Facts
PctUnemploy	Percent of Unemployment	US Bureau of Labor Statistics
PolicyUnemp	Modifications to Unemployment Benefit Policies	Katz et al. (2023)
Week	HPS Week	HPS Survey
TotalPopulation (Log)	Total Population	ACS
PctMarried	Percent Married	ACS
PctDepend	Percent of Population in Dependent Age Groups (under 18 and 65+)	ACS
MdnAge	Median Age of Total Population	ACS
PctNHWhite	Percent of Population that is White alone, not Hispanic or Latino	ACS
PctBlack	Percent of Population that is Black or African American Alone, Not Hispanic or Latino	ACS
PctAsian	Percent of Population that is Asian Alone, Not Hispanic or Latino	ACS
PctHisp	Percent of Population that is Hispanic or Latino	ACS
PctFB	Percent of Population that is Foreign Born	ACS
PctLTHS	Percent of Population 25 Years and Over whose Highest Education Completed is Less Than High School	ACS
PctBA	Percent of Population 25 Years and Over whose Highest Education Completed is Bachelor’s Degree or Higher	ACS
PctSameHouse	Percent of population 1 year and over who lived in the same house 1 year ago	ACS
PctDiffStat	Percent of population 1 year and over who lived in a different state 1 year ago	ACS
PctNILF	Percent Not in Labor Force	ACS
PctWkHome	Percent of workers who worked at home	ACS
MedianIncome (Log)	Median Household Income in past 12 months (inflation-adjusted dollars to last year of 5-year range)	ACS
PctNoInternet	Percent of Households with No Internet Access	ACS
TotalHous (Log)	Total housing Units	ACS
PctOwnHouse	Percent of Total Housing Units (including Vacant) that are Owner-Occupied	ACS
MdnYearHouse	Median year housing units built	ACS
PctBuild2020	Percent of Housing units built 2020 or later	ACS

and

Table A3. Summary statistics of the variables used in this study.

Variable	Mean	Std. dev.	Min	Max	Category
PctHouseInsecurity	10.67	8.01	0.9	46.4	COVID
PctPoverty	12.55	2.65	7.4	19.4	Economic
PctRentGE30PCT	43.6	3.75	35.5	53.2	Housing
PctMtgeGE30PCT	25.41	4.13	18.5	37.5	Housing
PctCovidIncrease	0.55	0.92	0	7.09	COVID
PctUnemploy	6.42	3.24	1.9	30.6	COVID
PolicyUnemp	0.64	0.48	0	1	COVID
Week	23	12.99	1	45	COVID
TotalPopulation (Log)	6,684,409	7,364,533	576,641	3.95 × 107	Demographic
PctMarried	50.02	3.41	32.6	56.7	Demographic
PctDepend	38.85	1.72	30.5	41.4	Demographic
MdnAge	38.7	2.28	31.3	44.7	Demographic
PctNHWhite	68.02	14.86	35.8	92	Race
PctBlack	11.24	10.37	0.5	43.9	Race
PctAsian	3.58	2.75	0.8	14.7	Race
PctHisp	12.43	10.46	1.7	49.6	Race
PctFB	9.31	6.05	1.6	26.5	Citizenship
PctLTHS	9.73	2.55	5.6	15.8	Education
PctBA	33.27	6.69	21.8	61.4	Education
PctSameHouse	86.1	1.92	80.5	89.7	Mobility
PctDiffStat	2.9	1.15	1.2	8	Mobility
PctNILF	36.16	3.7	28.6	46.7	Employment
PctWkHome	9.25	2.7	3.8	19.8	Employment
MedianIncome (Log)	68,248	11,127	49,111	93,547	Economic
PctNoInternet	10.72	2.78	5.3	18.6	Economic
TotalHous (Log)	2,831,952	2,893,591	271,818	1.43 × 107	Housing
PctOwnHouse	66.53	5.49	41.5	73.9	Housing
MdnYearHouse	1978	9	1955	1995	Housing
PctBuild2020	0.19	0.09	0.05	0.39	Housing

in the Appendix A list the variables used in this study, their data sources, and a summary of the statistics for these variables, respectively.

4. Methods

Descriptive analyses were conducted on demographic characteristics and housing insecurity. A multivariable logistic regression model was employed to evaluate the associations between various demographic groups and housing insecurity outcomes using individual-level HPS data.

To locate regions where high (or low) housing insecurity clusters and identify areas that diverge from the general trend, we performed a hotspot analysis for the housing insecurity from 45 survey weeks using the Getis–Ord G_i* statistics. The hotspot analysis is valuable because it can help in identifying areas with acute problems.

The Getis–Ord’s $G_i^{*}$ statistic is calculated as follows:

$$G_i^{\mathrm{*}}={\displaystyle \frac{{\sum }_{j=1}^n{w}_{i,j}{x}_j-\overline{X}{\sum }_{j=1}^n{w}_{i,j}}{S\sqrt{\frac{[n{\sum }_{j=1}^n{w}_{i,j}^2-({\sum }_{j=1}^n{w}_{i,j}{)}^2]}{n-1}}}}$$

(1)

where $x_j$ is the housing insecurity percentage for a spatial feature $j$, $w_{i,j}$ is the spatial weight between spatial features $i$ and $j$, $n$ is equal to the total number of spatial features, and

$$\overline{X}={\displaystyle \frac{{\sum }_{j=1}^n{x}_j}{n}}$$

(2)

$$S=\sqrt{{\displaystyle \frac{{\sum }_{j=1}^n{x}_j^2}{n}}-(\overline{X}{)}^2}$$

(3)

The between-effects linear regression model was employed to examine the association of housing cost burden and poverty with the spatial disparity of housing insecurity across different regions. This model is particularly suited to analyzing variations between groups, offering insights into how housing insecurity differs from one location to another. The primary advantage of this model lies in its capacity to account for unobserved heterogeneity across locations. This means that any constant factors that could potentially influence the housing insecurity, such as cultural attributes, natural resources, or inherent location-specific traits, are controlled for, even when these factors are not directly measured in the study. The mathematical formula for the between-effects linear regression model is as follows:

$${\overline{Y}}_i={\beta }_0+{\beta }_1{P}_i+{\beta }_2{C}_i+\theta {\overline{X}}_i+{\delta Z}_i+{\mu }_i+{\overline{ϵ}}_i$$

(4)

where ${\overline{Y}}_i$ represents the averaged dependent variable (PctHouseInsecurity) for location i; $P_i$ denotes the poverty variable (PctPoverty) for location i; $C_i$ shows the housing cost burden variable (PctMtgeGE30PCT or PctRentGE30PCT) for location i; ${\overline{X}}_i$ is the averaged time-variant control variable (e.g., PctCovidIncrease, PctUnemploy, PolicyUnemp) for location i; $Z_i$ is a vector of time-invariant control variables (such as Ln_totalPopulation, PctMarried) for location i; β₀, β₁, β₂, θ, and δ are the coefficients for the intercept, poverty, housing cost burden variables, time-variant and time-invariant control variables, respectively; ${\mu }_i$ is the location-specific effect (unobserved heterogeneity that is constant over time but varies across locations), and ${\overline{ϵ}}_i$ is the averaged error term for location i.

To address the multicollinearity issues among the numerous control variables used in the aggregated data, we employed the exploratory regression mining tool in ArcGIS Pro software to identify an appropriately specified linear regression model. This tool tests all possible combinations of explanatory variables to determine the most suitable model. It operates similarly to stepwise regression, which is found in many statistical software packages. However, a key difference is that the exploratory regression mining tool seeks models that fulfill all requirements and assumptions of the linear regression model, whereas stepwise regression primarily focuses on models with high adjusted R² values. The diagnostic statistics used in the exploratory regression mining tool include adjusted R², coefficient p-values, variance inflation factor (VIF) values, Jarque–Bera (JB) p-values, and spatial autocorrelation (SA) p-values.

Based on the identified explanatory variables from the exploratory regression mining analysis results, we modeled local associations between housing insecurity, poverty, and housing cost burden using a geographically weighted regression (GWR) model. GWR is a local spatial regression model that accommodates spatial dependency and heterogeneity. It deviates from traditional ordinary least square regression models by estimating local parameters instead of global ones (Fotheringham et al., 2003). GWR allows us to explore how relationships between variables and housing insecurity vary across geographic space. The impact of poverty and housing cost burden might be not uniform across geography and it might be more pronounced in some areas compared to other ones. By identifying how relationships vary locally, policymakers could fine-tune interventions to be more effective in specific contexts rather than applying one-size-fits-all solutions.

The GWR model can be expressed as follows:

$${\overline{Y}}_i={\beta }_{0i}+{\beta }_{1i}{P}_i+{\beta }_{2i}{C}_i+{\theta }_i{\overline{X}}_i+{\delta }_i{Z}_i+{ϵ}_i$$

(5)

where β_0i, β_1i, β_2i, ${\theta }_i$, ${\delta }_i$ are the local coefficients for the intercept, poverty, housing cost burden, time-variant and time-invariant control variables at location i, respectively. Other variables are defined the same way as in Equation (4).

5. Results and Analyses

5.1. Which Demographic Groups Were Particularly Vulnerable to Housing Insecurity?

The descriptive statistics presented in Figure A1 offer an initial insight into housing insecurity across different demographic groups.

Table 1. Logistic regression results.

Dependent Variable: House Insecurity	Model 1 All		Model 2 Owner		Model 3 Renter
Age: 18–24 (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
25–39	1.265	***	1.033		1.425	***
	(0.039)		(0.051)		(0.056)
40–54	1.444	***	1.179	***	1.623	***
	(0.045)		(0.059)		(0.066)
55–64	1.131	***	0.926		1.249	***
	(0.038)		(0.049)		(0.058)
65 and above	0.660	***	0.546	***	0.705	***
	(0.025)		(0.031)		(0.041)
Educ: Less than high school (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
High school or GED	0.930		0.922		0.937
	(0.045)		(0.080)		(0.055)
Some college/associates degree	0.853	***	0.878		0.833	***
	(0.041)		(0.075)		(0.049)
Bachelor or high	0.609	***	0.657	***	0.554	***
	(0.030)		(0.057)		(0.033)
Race: White, Alone (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
Black, Alone	1.959	***	2.032	***	1.874	***
	(0.035)		(0.055)		(0.046)
Asian, Alone	1.816	***	1.949	***	1.597	***
	(0.050)		(0.070)		(0.069)
Any other race alone, or race in combination	1.554	***	1.543	***	1.545	***
	(0.041)		(0.060)		(0.057)
Hispanic, Latino, or Spanish origin	1.267	***	1.326	***	1.199	***
	(0.024)		(0.035)		(0.033)
Gender: Male (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
Female	0.974	**	1.023		0.919	***
	(0.012)		(0.017)		(0.018)
Child: No Children (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
With Children	1.242	***	1.141	***	1.361	***
	(0.022)		(0.027)		(0.037)
Work Loss: Yes (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
No	0.340	***	0.342	***	0.342	***
	(0.005)		(0.006)		(0.007)
Marital Status: Married (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
Widowed	1.132	***	1.108	**	1.132	**
	(0.042)		(0.051)		(0.066)
Divorced/separated	1.153	***	1.108	***	1.170	***
	(0.020)		(0.027)		(0.029)
Never married	0.975		0.934	**	0.985
	(0.018)		(0.026)		(0.024)
Income: Less than USD 25,000 (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
USD 25,000–USD 34,999	0.819	***	0.793	***	0.823	***
	(0.018)		(0.029)		(0.022)
USD 35,000–USD 49,999	0.691	***	0.653	***	0.704	***
	(0.015)		(0.023)		(0.019)
USD 50,000–USD 74,999	0.527	***	0.507	***	0.527	***
	(0.011)		(0.017)		(0.015)
USD 75,000–USD 99,999	0.406	***	0.388	***	0.406	***
	(0.009)		(0.013)		(0.017)
USD 100,000–USD 149,999	0.290	***	0.274	***	0.295	***
	(0.007)		(0.010)		(0.014)
USD 150,000–USD 199,999	0.213	***	0.198	***	0.225	***
	(0.008)		(0.010)		(0.017)
USD 200,000 and above	0.162	***	0.142	***	0.228	***
	(0.006)		(0.007)		(0.021)
HouseholdSize: 1 person (Ref)	1	(Ref)	1	(Ref)	1	(Ref)
2 people	0.979		0.869	***	1.033
	(0.019)		(0.026)		(0.027)
3 people	1.110	***	0.984		1.159	***
	(0.025)		(0.032)		(0.038)
4 people	1.211	***	1.111	***	1.215	***
	(0.031)		(0.041)		(0.046)
5 people	1.290	***	1.222	***	1.242	***
	(0.038)		(0.050)		(0.054)
6 people	1.415	***	1.368	***	1.327	***
	(0.051)		(0.067)		(0.069)
7 or more people	1.389	***	1.360	***	1.296	***
	(0.050)		(0.070)		(0.067)
Intercept	0.313	***	0.426	***	0.290	***
	(0.025)		(0.054)		(0.030)
Controlling for time variability (Weeks)	Yes		Yes		Yes
Controlling for location variability (States)	Yes		Yes		Yes
Number of observations	2062005		1386463		675542
Pseudo R2	0.1445		0.1383		0.1285

*** p < 0.01, ** p < 0.05. Robust standard errors in parentheses.

presents the logistic regression results from three distinct models. Model 1 applied to all respondents, Model 2 was specific to respondents from owner-occupied properties, and Model 3 pertained to respondents from rental properties. Overall, the regression results align with those depicted in Figure A1.

These results show that the COVID-19 pandemic exposed significant disparities in housing insecurity across different demographic groups, with notable differences between the rental and owner-occupied housing markets. Renters faced consistently higher levels of housing insecurity compared to homeowners, and this was driven by several key factors. For instance, middle-aged individuals, particularly those aged 40–54, were the most vulnerable group in both housing segments, but renters experienced significantly greater insecurity than homeowners. This age group is often at a stage in life where financial responsibilities, such as supporting a family or managing employment instability, are most pronounced. In contrast, older adults (65+) experienced much lower levels of insecurity, particularly in owner-occupied housing, likely due to having stable sources of income like retirement benefits or the absence of mortgage payments, which often provides homeowners with a stronger financial buffer.

Education also played a critical role in determining housing insecurity, with stark differences observed between renters and homeowners. Individuals with less than a high school education experienced the highest levels of insecurity in both housing segments, but renters in this group were particularly vulnerable. Higher educational attainment was associated with lower levels of insecurity, especially for homeowners, as those with college degrees were more likely to be employed in industries that allowed for remote work or were less affected by the pandemic’s economic disruptions. This correlation between education and housing insecurity highlights the importance of stable, higher-paying jobs, which are more common among homeowners.

Racial and ethnic disparities were pronounced, with Black and Hispanic/Latino individuals facing significantly higher levels of housing insecurity, particularly in the rental market. These groups were disproportionately affected by the pandemic due to their employment in sectors hit hardest by job losses and reduced working hours, such as the service and retail industries. Renters in these racial and ethnic groups lacked the financial flexibility often available to homeowners, such as mortgage forbearance or home equity loans, making them more vulnerable to economic shocks. Conversely, White and Asian individuals experienced lower levels of insecurity, particularly among homeowners, underscoring the racial inequalities in housing stability.

Gender differences in housing insecurity were also significant, with women experiencing more insecurity than men, especially in rental properties. Women were more likely to be employed in sectors severely impacted by the pandemic, such as hospitality and retail, and faced additional challenges related to childcare due to school closures. These pressures made it difficult for many women, particularly renters, to maintain housing stability, reflecting how the pandemic intensified pre-existing gender-based economic vulnerabilities.

Households with children and those that experienced job loss were particularly susceptible to housing insecurity, with renters again bearing the brunt of the challenges. Households with children, which were already facing higher living costs, experienced twice the level of housing insecurity compared to those without children, regardless of housing tenure. Job loss compounded these difficulties, with renters facing immediate financial distress due to having fewer relief options compared to homeowners, who could turn to mortgage forbearance or refinancing. This disparity highlights the limited safety nets available to renters, leaving them more exposed to housing insecurity during economic crises.

Being widowed, divorced, or separated correlates with increased odds of housing insecurity relative to married individuals, likely because of the loss of dual incomes, social support, and the economic advantages of sharing living expenses. Higher income levels correlate with lower odds of housing insecurity, illustrating the protective impact of income on housing stability; the highest income groups face the lowest risk. Generally, larger household sizes are associated with increased odds of housing insecurity, which may be attributable to the greater expenses involved in maintaining suitable housing for more individuals and the potential for overcrowding, a particular concern during the pandemic.

Overall, the pandemic exacerbated housing insecurity for renters far more than for homeowners, particularly among vulnerable demographic groups such as low-income households, racial and ethnic minorities, women, and those with lower educational attainment. The structural inequalities in the rental housing market, where financial resilience is generally lower and fewer relief mechanisms are available, meant that renters were far more exposed to the economic fallout of the pandemic. This analysis underscores the need for targeted policy interventions to address the housing insecurity faced by renters, especially during periods of economic instability.

5.2. Which Regions in the US Experienced the Highest Levels of Housing Insecurity During the Pandemic?

To identify the regions most affected by housing insecurity during the pandemic, we conducted a hotspot analysis. Figure 1a displays the distribution of housing insecurity hotspots throughout the COVID-19 pandemic using the aggregated HPS data. Our findings reveal that the southeastern US faced significant housing insecurity during the pandemic, with Arkansas, Louisiana, Mississippi, Alabama, Tennessee, Georgia, and Florida being the most critical hotspots. Texas, Oklahoma, Kentucky, and South Carolina also showed high insecurity. In contrast, the northwestern US had lower insecurity, with Montana, North Dakota, and South Dakota as cold spots. These findings highlight a stark regional disparity, with the southeastern region hit harder than the northwestern region in terms of housing payment struggles.

Many factors could contribute to the regional differences in housing insecurity experienced during the pandemic. For instance, regions with a higher percentage of poverty might face greater rates of housing insecurity because households have a smaller financial cushion to weather economic shocks, such as job loss or reduced working hours. On the other hand, the burden of housing costs might also influence these regional disparities. Households that allocated a significant portion of their income towards housing expenses were less equipped to handle unexpected costs or a loss of income. Consequently, when income was heavily committed to housing, any interruption could swiftly result in missed payments. During the pandemic, economies that suffered abrupt downturns likely saw that those with high housing cost burdens became more vulnerable to housing insecurity, as their ability to adapt to decreased income or unemployment was limited. Thus, regions where the housing cost burden was greater might also see higher rates of housing insecurity.

To investigate the spatial association among housing insecurity, poverty, and housing cost burden, we also conducted hotspot analyses for poverty and housing cost burden using the most recent ACS 5-year estimates data. Figure 1b displays the hotspots of poverty percentage from 2017 to 2021. A concentration of hotspots is evident in the southern US, particularly in the southeastern states, indicating a higher prevalence of poverty in these areas. In contrast, the northern states, especially the northeastern and some northwestern states, are identified as cold spots, signaling a lower prevalence of poverty. The hotspots of poverty depicted in Figure 1b are similar to those of housing insecurity shown in Figure 1a. This may suggest that the regions identified as having high levels of poverty also tend to be the same regions experiencing high levels of housing insecurity. This implies a correlation between poverty and housing insecurity, which is logical since lower-income households are often more vulnerable to housing-related challenges.

Figure 1c illustrates the hotspots of housing cost burden in terms of PctRentGE30PCT, which represents the percentage of renter households for whom gross rent constitutes 30% or more of household income. Figure 1d presents the hotspots of housing cost burden in terms of PctMtgeGE30PCT, indicating the percentage of owner households with mortgages for whom monthly owner costs exceed 30% of household income. These maps offer valuable insights into the economic pressures faced by different types of households across the country. For renters, as shown in Figure 1c, notable hotspots are found in the Northeast states and Nevada, though these hotspots are less widespread compared to those among homeowners. For homeowners with mortgages, as depicted in Figure 1d, the hotspots are more broadly distributed, with significant concentrations in the Northeast and the West regions. The cold spots on both maps are predominantly situated in the central to north-central region of the country. The differences between the two maps indicate that the experience of housing cost burden varied between renters and homeowners with mortgages, likely due to differences in housing markets, income levels, and the availability of affordable housing across various regions.

In addition, the hotspots of housing cost burden in Figure 1c,d are quite distinct from the hotspots of housing insecurity shown in Figure 1a. This means that the regions where a large proportion of renter households (Figure 1c) or owner households with mortgages (Figure 1d) spend 30% or more of their income on housing costs do not necessarily overlap with the areas experiencing the highest levels of housing insecurity. The different spatial patterns may indicate that housing insecurity is shaped by more than just affordability, suggesting that factors beyond cost burden play a significant role. For instance, a region may have a high housing cost burden but also have robust support systems in place that prevent these costs from translating into housing insecurity. Conversely, areas with lower housing costs could still exhibit high levels of housing insecurity due to factors like job instability, lack of access to affordable housing, or lower median incomes. Housing insecurity might also be exacerbated by temporary factors, such as economic downturns during the pandemic, which may not be fully captured by housing cost burden.

The hotspot analysis reveals a complex relationship between poverty, housing cost burden, and housing insecurity. While housing insecurity is closely tied to poverty, it cannot be solely measured by affordability. The cost of housing does not always directly correlate with housing insecurity, as various mitigating factors also play a significant role.

5.3. How DID Housing Cost Burden and Poverty Contribute to Housing Insecurity?

To determine the relationship between housing cost burden, poverty, and the spatial disparity of housing insecurity across different states, we employed between-effects linear regression models. These models were used to analyze the aggregated HPS and control variable panel data.

Table 2. Results from the between-effects linear regression models.

Dependent Variable PctHouseInsecurity	Model 1		Model 2		Model 3		Model 4
PctPoverty	0.891	***	0.885	**	0.493	***	0.385	***
	(0.290)		(0.358)		(0.075)		(0.101)
PctMtgeGE30PCT			0.041				0.114	**
			(0.082)				(0.047)
PctRentGE30PCT	0.168	**			0.128	***
	(0.081)				(0.046)
Control variables	24		24		6		6
Observations	2205		2205		2205		2205
R2	0.970		0.965		0.939		0.923
Adjusted R2	0.935		0.923		0.927		0.907

*** p < 0.01, ** p < 0.05. Standard errors in parentheses.

outlines the coefficients and standard errors (in parentheses) of the explanatory variables of interest from the between-effects linear regression models. Additionally,

Table A4. Comparison of different between-effects linear regression models.

Dependent Variable PctHouseInsecurity	Model 1		Model 2		Model 3		Model 4
PctPoverty	0.891	***	0.885	**	0.493	***	0.385	***
	(0.290)		(0.358)		(0.075)		(0.101)
PctRentGE30PCT	0.168	**			0.128	***
	(0.081)				(0.046)
PctMtgeGE30PCT			0.041				0.114	**
			(0.082)				(0.047)
Ln_TotalHous	5.430		3.887
	(5.029)		(5.729)
PctOwnHouse	−0.001		0.041
	(0.070)		(0.074)
PctCovidIncrease	−3.260		−4.184	*
	(2.207)		(2.371)
PctUnemploy	0.123		0.168
	(0.179)		(0.195)
PolicyUnemp	−0.129		−0.146
	(0.372)		(0.429)
LN_TotalPop	−5.108		−3.498
	(4.981)		(5.651)
PctMarried	0.466	***	0.361	**	0.184	***	0.210	***
	(0.131)		(0.144)		(0.068)		(0.077)
PctDepend	−0.094		−0.138
	(0.262)		(0.299)
MdnAge	0.044		0.068
	(0.185)		(0.228)
PctNHWhite	−0.093		−0.066
	(0.062)		(0.068)
PctBlack	0.035		0.057		0.128	***	0.171	***
	(0.058)		(0.063)		(0.014)		(0.016)
PctAsian	−0.482	***	−0.472	***	−0.157	*	−0.215	**
	(0.147)		(0.160)		(0.078)		(0.088)
PctHisp	−0.104		−0.081
	(0.069)		(0.074)
PctFB	0.272	***	0.315	***	0.252	***	0.267	***
	(0.086)		(0.092)		(0.042)		(0.047)
PctLTHS	0.210		0.189
	(0.215)		(0.241)
PctBA	−0.188	*	−0.121				−0.201	***
	(0.100)		(0.105)				(0.037)
PctSameHouse	0.300		0.281
	(0.189)		(0.217)
PctDiffStat	0.692		0.568
	(0.413)		(0.452)
PctNILF	−0.269	**	−0.237
	(0.129)		(0.141)
PctWkHome	−0.253		−0.350		−0.426	***
	(0.207)		(0.223)		(0.062)
Ln_MdnIncom	7.440		5.737
	(4.541)		(4.950)
PctNoInternet	0.046		−0.066
	(0.195)		(0.226)
MdnYearHouse	−0.044		−0.025		−0.039	**	−0.077	***
	(0.037)		(0.039)		(0.016)		(0.017)
PctBuild2020	−2.744		−1.369
	(2.821)		(2.981)
Intercept	−29.882		−43.209		67.260	**	148.153	***
	(108.407)		(117.808)		(29.144)		(32.627)
Observations	2205		2205		2205		2205
R2	0.970		0.965		0.939		0.923
Adjusted R2	0.935		0.923		0.927		0.907

*** p < 0.01, ** p < 0.05, * p < 0.1. Standard errors in parentheses.

details the coefficients and standard errors (in parentheses) of all explanatory variables, including both variables of interest and control variables, from the models.

In Table 2 and Table A4, Models 1 and 2 include a comprehensive set of control variables (twenty-four in each model), while Models 3 and 4 employ a more concise set of six control variables, chosen based on an exploratory regression analysis conducted using ArcGIS Pro. Models 1 and 2 experienced multicollinearity due to the large number of control variables. To address this issue, the exploratory regression analysis was used to identify the most suitable controls.

Table A5. The diagnostic statistics for Models 3 and 4 (the linear regression models with fewer control variables) derived from the exploratory regression analysis.

Adj-R2	AICc	JB	K(BP)	VIF	SA	Model	Notes
0.93	121.31	0.88	0.88	6.4	0.25	+PctPoverty *** +PctRentGE30PCT *** +PctMarried *** +PctBlack *** -PctAsian * +PctFB *** -PctWkHome ***-MdnYearHouse **	Variables are used in Model 3 in Table 2 and Table A4
0.91	133.04	0.44	0.61	6.32	0.9	+ PctPoverty *** +PctMtgeGE30PCT ** +PctMarried *** +PctBlack *** -PctAsian ** +PctFB *** -PctBA *** -MdnYearHouse ***	Variables are used in Model 4 in Table 2 and Table A4

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1. The high adjusted R² values (0.93, 0.91) reported in Table A5 suggest that Models 3 and 4 account for a significant portion of the variation in explaining the spatial disparities of housing insecurity (93% and 91%, respectively). The low AICc values (121.31, 133.04) indicate a strong fit of the models to the observed data. The Jarque–Bera (JB) statistic probabilities (0.88, 0.44), being well above 0.05, suggest that the residuals are normally distributed, which indicates the unbiased nature of these models. Moreover, the Koenker (BP) statistic probabilities (0.88, 0.61), being substantially higher than 0.05, denote a consistent relationship between the explanatory and dependent variables across both geographic and data spaces. Additionally, the VIF values (6.4, 6.32) fall below 7.5, denoting a lack of redundancy among the explanatory variables. Also, the probabilities of the SA statistics (0.25, 0.9), being significantly greater than 0.05, indicate no substantial clustering of high and/or low residuals, which points to spatial randomness.

presents the diagnostic statistics for Models 3 and 4, confirming that these models are properly specified linear regressions, which are free from multicollinearity.

The positive and statistically significant coefficients for PctPoverty across Models 1, 2, 3, and 4 in Table 2 demonstrate that higher percentages of poverty are associated with increased levels of housing insecurity. The magnitude of these coefficients is substantial across the models, indicating that the percentage of poverty is a robust predictor of housing insecurity. This association remains strong even with the inclusion of a larger number of control variables. The coefficient for housing cost burden PctMtgeGE30PCT is significant at the 0.05 level only in Model 4 and not significant in Model 2, suggesting a weaker and less consistent relationship with housing insecurity. Its significance is pronounced only when fewer control variables are included in the model. The coefficients for PctRentGE30PCT show a positive and significant relationship with PctHouseInsecurity in both Models 1 and 3, signifying its consistent and robust link to housing insecurity. This suggests that for renter households, the share of income devoted to rent is a crucial determinant of housing insecurity.

Considering both the significance levels and the sizes of the coefficients, PctPoverty emerges as a more influential factor in explaining spatial disparity of housing insecurity compared to housing cost burden variables. This is evidenced by its consistent significance across various model specifications and its relatively large coefficients. PctRentGE30PCT also proves to be a significant factor. While housing cost burden PctMtgeGE30PCT shows a notable association, it is contingent on specific conditions. This indicates that the cost burden for homeowners with mortgages is not as consistently predictive of housing insecurity as the other explanatory variables of interest.

The high R² values, ranging from 92.3% to 97%, suggest that the variables of interest, along with the control variables, account for a significant proportion of the variance in housing insecurity across different regions. The adjusted R² values are also high, ranging from 90.7% to 93.5%, indicating that the models provide a good fit to the data even after adjusting for the number of predictors. Overall, the results imply that both poverty and rent burden are crucial predictors of housing insecurity. Nevertheless, poverty is highlighted as playing a more significant role than housing cost burden in explaining the spatial disparity of housing insecurity across different states.

The between-effects linear regression models reveal the overarching relationship between housing cost burden, poverty, and the spatial disparity of housing insecurity. However, the impact of poverty and housing cost burden on housing insecurity might vary across different geographic locations, influenced by a variety of factors including demographic characteristics, local economic conditions, the availability of affordable housing, local social support programs, and regional policies. This array of regional factors highlights the importance of localized analysis in exploring the association between poverty, housing cost burden, and the spatial disparity of housing insecurity. To understand and visualize these spatial variations, we conducted spatial analysis using the GWR model.

Figure 2 presents the outcomes of the GWR model, which includes the same explanatory variables selected for the between-effects Model 3. Figure 2a depicts the observed data on the percentage of housing insecurity during the COVID-19 pandemic, while Figure 2b displays the predicted percentage of housing insecurity for the same period using the GWR model. Both maps show a broadly similar geographic distribution of housing insecurity, with higher levels predominantly observed in the southern states and lower levels in the northern and some western states. This pattern suggests that the GWR model used for prediction mirrors the general trend captured in the actual data, indicating that the model’s parameters and variables are effectively grasping the factors that drive housing insecurity. However, discrepancies in a few regions between the observed and predicted levels of housing insecurity might result from the model’s inability to fully encapsulate all local determinants or from unforeseen events not considered in the model.

Figure 2c illustrates the standardized residuals of the GWR model. Areas depicted in lighter colors, where the standardized residuals are close to zero, demonstrate a strong consistency between the model’s predictions and the observed data, signifying a precise fit in these locations. The random distribution of the standardized residuals across the map suggests that the model is effectively capturing the local variations in the relationship between the dependent and independent variables and there is no systematic bias in the model predictions. We also calculated the Global Moran’s I statistic for the raw residuals of the GWR model to check if the model has properly accounted for spatial autocorrelation (spatial dependence) in the data.

Table 3. Summary of global Moran’s I statistics for GWR model residuals.

Moran’s Index	−0.1548
Expected Index	−0.0208
Variance	0.0036
z-score	−2.2398
p-value	0.0251

summarizes the results. The Moran’s I analysis indicates a significantly dispersed spatial pattern in the residuals. Specifically, the z-score of −2.2398 (p < 0.05) suggests there is less than a 5% likelihood that this dispersion is due to random chance. This finding supports the conclusion that the GWR model adequately accounts for spatial structure in the data.

Figure 2d presents the local R² values from the GWR model. The consistently high local R² values (ranging from 0.926 to 0.984) across the map suggest that the GWR model generally provides a robust fit for the data, indicative of its effectiveness in explaining the variation in housing insecurity at the local level.

Figure 3 displays the local coefficients for PctPoverty and PctRentGE30PCT derived from the GWR model (note: Figure A2 shows the local coefficients for all explanatory variables from the GWR model). In Figure 3a, the local association of PctPoverty with housing insecurity is depicted. Areas shaded in darker red signify higher positive coefficients, suggesting that in these regions, poverty correlates more strongly with the dependent variable, housing insecurity. Conversely, lighter shades indicate a weaker relationship. The variety in color intensity across different regions implies that poverty’s impact on housing insecurity is heterogeneous and location-dependent. Specifically, states in the South and Midwest, marked by darker reds, exhibit a more pronounced relationship between poverty and housing insecurity. Figure 3b illustrates the local association between PctRentGE30PCT and housing insecurity. Once more, darker red shades indicate a stronger positive relationship between rent burden and housing insecurity, while lighter shades represent a weaker relationship. The darker tones prevalent in some southwestern and southeastern states point to a greater predictive power of rent burden for housing insecurity in these locations. Regions depicted in very light colors suggest either an insignificant relationship or coefficients near zero.

Both coefficient maps offer an intricate view of how the connections of poverty or housing rent cost burden with housing insecurity vary across geographic regions. The presence of stronger coefficients in the southern states on both maps signals that poverty and rent burden are substantial issues contributing to housing insecurity in these regions. The variety of coefficients indicates a complex relationship between these factors and housing insecurity, influenced locally by factors such as economic conditions, housing markets, and social policies. The differences observed between the coefficient maps indicate that while poverty and rent burden both impact housing insecurity, their effects are not uniform across geographic locations. These maps may be useful for policymakers seeking to understand and address regional disparities in housing insecurity. The variability in coefficients emphasizes the necessity of tailoring interventions to local conditions and considerations.

6. Discussions and Conclusions

Housing insecurity significantly impacts individuals’ quality of life and subjective well-being. Stable housing is foundational to overall life satisfaction, mental and physical health, and economic stability. When housing becomes uncertain, individuals face increased psychological stress, which can manifest in anxiety, depression, and decreased productivity. The lack of a secure home environment also disrupts children’s education, access to healthcare, and social connections, further diminishing well-being.

Given the profound impact of housing insecurity on well-being, it is essential to understand how these effects played out over the course of the COVID-19 pandemic, which caused widespread job losses and economic disruptions. Although earlier studies explored the pandemic’s impact on housing insecurity in the US, they all typically focused on specific periods, populations, or local regions. No comprehensive analysis has yet covered the full duration of the pandemic or addressed spatial inequality across the entire country. To address this gap, we utilized both individual-level and aggregated data from the Household Pulse Survey to explore which demographic groups experienced housing insecurity and where these insecurity occurrences were located in the US during the COVID-19 pandemic.

Our results show high housing insecurity among the 40–54 age group, Black and Hispanic/Latino individuals, females, households with children or job loss, and divorced/separated individuals. In contrast, those aged 65 and older, with higher education, White individuals, widowed or married individuals, and higher-income households faced less insecurity. Renters experienced significantly greater insecurity than homeowners. These findings suggest that policies could be designed to provide targeted financial and housing assistance to the groups identified as most vulnerable. This could include increased access to affordable housing, rental assistance programs, and direct financial support. Given that renters faced greater challenges than homeowners, policies could be developed to strengthen renter protections. This might include measures such as rent control, eviction moratoriums, and legal assistance for renters facing housing instability.

A hotspot analysis revealed that the southeastern region of the US experienced the highest levels of housing insecurity during the pandemic, with more people struggling, compared to the northwestern region. The overlap between poverty and housing insecurity hotspots indicates a strong link, while differences between cost burden and insecurity hotspots suggest other contributing factors and that insecurity cannot be solely measured by affordability. These findings highlight the need for region-specific interventions in the southeastern US. Policies could be directed at increasing housing affordability, reducing poverty, and improving job security in these areas.

Understanding the link between housing cost burden, poverty, and payment insecurity offers insights into housing market dynamics. Using between-effects linear regression, we found that both poverty and rent burden significantly affect spatial disparities in housing insecurity, with poverty being the more influential factor. Consequently, policies should prioritize poverty reduction, which can be achieved through measures such as increasing access to well-paying jobs, enhancing social security benefits, or implementing income support programs. While addressing housing costs remains important, it should form part of a more comprehensive strategy that addresses the underlying causes of poverty. Such a broader approach could include initiatives like education and training programs, improved healthcare access, and affordable childcare, all of which could indirectly alleviate housing insecurity by bolstering overall financial stability.

The GWR model shows that the impact of poverty and rent burden on housing insecurity varies by region, with stronger effects in southern states, highlighting these as key factors driving insecurity in those areas. Policies could be introduced to cap the percentage of income that can be charged as rent or to subsidize rents for low-income families, thereby reducing the rent burden. The local coefficient maps from the GWR model might be useful for targeted policymaking, as they can help to identify areas where interventions are most needed or most effective. They also highlight the importance of localized analysis and the potential limitations of one-size-fits-all solutions when dealing with complex issues such as housing insecurity.

These findings align with prior research that highlights the effectiveness of demographic targeting, region-specific interventions, structural anti-poverty strategies, and localized policymaking in addressing housing instability (D. Kim, 2021; Benfer et al., 2021; Shanks & Danziger, 2010; Michener, 2023). In addition, while this study centers on the COVID-19 pandemic, the insights gained may have broader relevance for future economic or public health disruptions. The patterns identified—such as who was most affected and where—reveal structural vulnerabilities that persist beyond the pandemic. These findings can offer critical lessons for building long-term housing system resilience and preparing for future shocks. Policymakers should institutionalize flexible support mechanisms, including emergency rental assistance, automatic mortgage forbearance, and unemployment benefits that can be rapidly deployed in times of crisis. Governments should also consider establishing permanent housing support infrastructure, such as emergency housing funds or pre-designed relief packages. Financial institutions can contribute by developing hardship assistance programs and expanding financial literacy resources for borrowers, while regulatory agencies should invest in early-warning data systems. Community organizations should also be engaged in delivering targeted support, particularly to renters and low-income households—especially in high-risk regions like the southeastern US. By adopting these forward-looking strategies, stakeholders—including residents, banks, landlords, and public agencies—can ensure a more proactive and equitable response when future crises emerge.

Although this study focuses on the US, the findings can help other countries as well. Many regions around the world faced similar housing challenges during the COVID-19 pandemic, especially among renters, low-income families, and marginalized groups. The policy ideas in this paper—such as rental assistance, poverty reduction, and targeted support—can be useful in countries with similar issues. The mapping methods we used can also help researchers and governments identify where housing assistance is most needed.

While our spatial analysis offers some insights into the housing insecurity disparities among demographic groups and geographic regions during the pandemic, our study is still limited, and a more comprehensive analysis is essential for a complete understanding of the situation. For example, despite examining some factors, the lack of available data (e.g., various policy data) hindered a more exhaustive analysis in this study. Future research should consider including more variables related to state and federal policy interventions to assess their effectiveness in addressing the disparity of housing insecurity. It is also important to recognize that housing insecurity is a complex issue encompassing challenges like difficulties in paying rent or mortgages, overcrowded living conditions, eviction, foreclosure, and homelessness. However, this study primarily focused on housing payment insecurity. Future research would benefit from a broader approach, evaluating various aspects of housing insecurity.