Abstract

Among existing research on social vulnerability, virtually no studies have considered homelessness as a variable in their vulnerability assessments. This study identified the relevance of homelessness as a key index in social vulnerability assessment to inform the public, policymakers and the broader body of literature of its impacts on shaping vulnerability patterns in cities. In this study, the 2018 Homeless data for Austin was disaggregated from the council district level to block group level using dasymetric mapping in Geographic Information System (GIS). Principal Component Analysis (PCA) was used to group highly correlated demographic and socioeconomic variables into factors, which were normalized and summed to model social vulnerability with homeless index (SOVI_H) and without homeless index (SOVI) for each Austin Blockgroup. The result revealed significant differences in geographic patterns between SOVI_H and SOVI. SOVI_H showed hotspots of vulnerabilities in Downtown and East-Austin neighborhoods, depicting a slight shift of social vulnerability westwards of the city. This finding differs from past results of social vulnerabilities in Austin where it used to be predominant in the East. This study showed that incorporating homelessness in identifying social vulnerability can help researchers and other associated organizations identify the most vulnerable groups when conducting social vulnerability assessments.

1. Introduction

Natural hazards pose challenges to major cities in the United States. The United States has experienced major transformations in population growth, economic conditions, development patterns and social characteristics. These changes have altered the American hazardscape in profound ways, with more people living in high-hazard areas than ever before as well as increasing frequencies in hazard occurrences ¹. However, some communities are adversely impacted more than others ^{2, 3}. For example, extreme weather in Austin, Texas like the recent Memorial Day flood in 2015 have shown that some population groups, such as the poor, the elderly, female-headed households and/or recent migrants, are generally at greater risk throughout the disaster response. In disaster studies, communities are often characterized by their demography and their resilience to environmental hazards. Social vulnerability refers to how a population's demographic, socioeconomic and cultural characteristics may reflect their capacity to anticipate, respond and recover from a hazardous event ^{4, 5, 6, 7}. In the hope of effective disaster management, vulnerability assessment typically involves: 1) the identification of population groups vulnerable to disasters within affected communities, 2) and the evaluation of their circumstances and needs. In light of global climate change, many cities around the world, such as the City of Austin (COA), are planning to prepare themselves to be climate-resilient that are adaptive to natural hazards. Thus, it is important to conduct a thorough social vulnerability assessment to identify socially at-risk communities thereby ensuring important variables are considered, for the purpose of creating policies that braces the resilience of such communities. In an effort to achieve sustainability and sustainable development, it is important to perform social vulnerability assessments and apply the derived results while planning for cities to ensure improved coping abilities to environmental changes and extreme climatic events.

Homeless population are particularly vulnerable throughout disaster response, relief, recovery and reconstruction. As stated in the Universal Declaration of Human Rights of 1948, homelessness is “the condition of people without a regular dwelling because they are unable to acquire, maintain regular, safe, and adequate housing, or lack fixed, regular, and adequate night-time residence”. Although homelessness is a growing concern in the United States with about 554,000 people being homeless as at 2017, it hasn’t gained much stance in social vulnerability or hazards literature. About half of those experiencing homelessness are in the following five states - California, New York, Florida, Texas and Washington. In the United States, about 65% of the total homeless population can be found in shelters, including emergency shelters, safe havens, and transitional housing. Unsheltered homeless individuals, living in locations like wooded areas, cars, and abandoned buildings, account for 35% of the total homeless population. About 63% of homeless people live alone or not a part of an intact family, with about 67-77% of those single people are men, meaning that single males account for the largest portion of the total homeless population. Homelessness is often a by-product of rapid urbanization, in which the poorest urban dwellers suffer from an increasing living cost that has become unaffordable to them ⁸. Not only is this a socio-economic issue, ethnic minorities in the United States experience homelessness at higher rates than Whites, and therefore make a disproportionate share of the homeless population. During a disaster, vulnerable populations, especially the homeless are subjected to higher risk of displacement, loss of possessions and/or human lives. Despite a dramatic rise in the number of homeless people across the United States since the 1980s, homelessness is hard to quantify given their dynamic mobility, the lack of administration incentive to count them, the unavailability of resources and appropriate measurements. Based on the reports of key informants located in the nation's largest cities, advocates for homeless people have claimed that the number of homeless people in the United States is as high as 2 to 3 million ⁹. However, surveys that try to count people who are currently homeless usually produce much smaller estimates. The underestimation could be a result of inadequate survey planning among other possible political reasons.

Existing literature has addressed various hazards and their associations with social vulnerability, including age, race and ethnicity, as well as gender ^{4, 10, 11}. Other indices like income and poverty have been used to study vulnerability in hurricane scenarios ^{12, 13}. Built-up environment indicators like housing, commercial facilities have been used in studies to measure the density of development and to predict areas prone to structural losses in disasters ^{2, 14, 15, 16}. Those affected by the harmful effects of hazards are disproportionately drawn from the segments of society which are chronically marginalized in daily life ⁴. Such people are marginalized geographically as they tend to live in hazardous places; socially and culturally as members of minority groups (e.g. ethnic minorities, people with disabilities); economically because they are poor (e.g. homeless or jobless); and politically because their voice is disregarded by those with political power (e.g. women, gender minorities, children, and elderly) ¹⁷. ² and ¹⁵ used similar variables in their studies to examine social vulnerability of populations living in hazard zones of South Carolina and New Jersey respectively. In contrast, ¹⁴ used only three variables as a proxy to assess social vulnerability. Social vulnerability indices across the literature have been shown to be subjectively selected by researchers regarding the context of their studies. This study uses common criteria generally accepted in social vulnerability literature. Many studies have applied vulnerability assessment in the decision-making process of disaster management (e.g. emergency response and relief, shelter location, routing, evacuation, etc.). But few studies have acknowledged and incorporated homeless people, who are perhaps the most vulnerable population in vulnerability assessment. ⁶ uses U.S. Census data to determine the social vulnerability of every census tract in the U.S. CDC’s Social Vulnerability Index (SOVI) ranks each tract on 15 socioeconomic factors, including poverty, lack of vehicle access, and minority population, and groups them into four related themes; socio-economic status, household composition and disability, housing and transportation, minority status and language. Each tract receives a separate ranking for each of the four themes, as well as an overall ranking. While the literature has explored many variables to assess social vulnerability and presented various methodological approaches and indices to quantify as such, virtually no studies have incorporated homelessness in their studies on social vulnerability Also, up-to-date spatial data of homelessness are rarely used, if any at all, in social vulnerability assessment. Homeless population has special needs that should be accounted for in social vulnerability assessments mainly because of the following:

a. Vulnerability assessment using Census data accounts for only household populations but not homeless population.

b. Homeless population is relatively mobile and unstable, and therefore, they are hard to be quantified and hence have been omitted in most existing framework of vulnerability assessment.

c. Homeless population often locate themselves in hazardous areas (e.g. floodplains, riparian zones, low water crossings, underpasses, transitional homes, etc.).

2. Aim and Objectives

This study intends to quantify homeless population and examine its impact in vulnerability analysis and using Geographic information systems (GIS). This concept attempts to visualize the spatial distribution of vulnerable populations in Austin, Texas and incorporate the homeless populace in the context of vulnerability assessment. Specifically, this study aims to: (1) examine the spatial pattern and distribution of homeless population at the block group level in Austin, Texas and; (2) identify any significant differences between vulnerability assessment with or without homelessness in terms of:

a. spatial pattern and distribution?

b. social vulnerability indices?

Furthermore, this study examines the effectiveness of disaster management to engage the most vulnerable and marginalized group in disaster planning and response.

3. Methodology

To answer the presented research questions, this study utilizes dasymetric modeling—a disaggregation technique, to derive homeless count at a fine spatial resolution, and an additive method of vulnerability assessment to calculate SOVI (Figure 1). Most data for this study was collected in Austin at the block group level to be analyzed at the finest scale possible. By examining the role of homelessness into vulnerability assessment, the results can aid planners and emergency managers in targeting socially vulnerable populations more effectively.

In 2018, the Ending Community Homelessness Coalition (ECHO) found 2,147 people to be homeless in the City of Austin (COA), a five percent increase than 2017. Despite rapid urbanization and increasing gentrification in Austin, surprisingly, the reported number of homeless populations has remained relatively the same over the past decade. This is probably due to inadequacies in financial resources and practical approaches to counting the homeless. On January 27, 2018, the city conducted its annual "Point-In-Time" count to document the number of people who are unsheltered and homeless in Austin, including people not just on the street but also those inhabiting in cars, tents, parks and under bridges. The derived numbers were combined with the count of people staying in transitional housing. Specifically, the number of people in 2018 sleeping unsheltered on the streets was 1,014— the highest in the last 8 years. As a legacy of the early 20th-century segregation policy and discriminatory practices, Austin’s socio-economically disadvantaged populations are largely concentrated on the east side of the City. Moreover, inequitable housing practices and racial-restrictive covenants persisted beyond the policy, resulting in a geographic isolation of minorities in East Austin ¹⁸. Today, Austin (Figure 2) has one of the nation’s highest levels of income segregation; nearly all census tracts with above-median numbers of families in poverty are situated on the east side ²⁴.

To examine the social vulnerability of COA, 22 relevant socioeconomic data shortlisted from the literature was collected for Austin (Table 1). In multivariate statistics, many socioeconomic and demographic indicators are inter-correlated with one another. Therefore, the variables obtained for this study were grouped into composite factors to mitigate multicollinearity and reduce data redundancy. Data on the state of homelessness in Austin/Travis County is collected mainly by homeless service providers. Due to the sensitivity of this population group, the best available data on homeless population was that of unsheltered persons at the Council District (CD) level. In this study, dasymetric modeling was conducted to disaggregate homeless population at the block group level. This disaggregated result was then presented as a predictor in creating a composite social vulnerability index (SOVI) for the study area. Specific variables from 2013-2017 American Community Survey (ACS) data were acquired from the U.S. Census Bureau to characterize the dimensions of social and physical vulnerability.

The PIT unsheltered homeless count data for 2018 at the CD level was derived from ECHO (1,014 total unsheltered spread over 10 CD’s). As mentioned, dasymetric modeling technique was used to disaggregate the CD level data into Block Group (BG) in consistency with other independent variables to present a composite SOVI index map. Dasymetric mapping has been used by researchers to estimate population distribution using ancillary data like land cover or nighttime light ²⁵. ²⁶ also compared dasymetric and choropleth methods of mapping population distribution in terms of exposure to air pollution. These researchers reported reasonably high accuracies in their results. For the dasymetric process in this study, a 30meter land cover data for 2011 acquired from the National Land Cover Database (NCLD) served as ancillary data. Land cover data are valuable because they serve as a proxy for socioeconomic characteristics through a chain of indirect links that tie together land cover, land use, housing type and density. Each land cover pixel was reclassified into five land use classes: four of which represent potential areas for temporary shelters for the homeless (these are used as related ancillary variables) and one non-homeless class (i.e. water and wetland areas class is unlikely to have any residential/homeless potential). The four land use classes used as related ancillary variables are low and high density residential, agricultural, commercial and industrial (Figure 3).

This study replicates the dasymetric mapping equation from ²⁷ to calculate homeless population for each land cover cell (pixel). The equation below was used:

(1)

Where P is the population of a cell,

-R_A is the relative residential density of a cell with land-cover type A,

-N is the actual population of enumeration unit (i.e., census block group)

-E is the expected population of enumeration unit calculated using the relative densities. E equals the sum of the products of relative density and the proportion of each land-cover type in each enumeration unit.

-A_T is the total number of cells in the enumeration unit.

The relative density (R_A) values used in this study relied solely on tested assumptions for dasymetric mapping and was refined based on the knowledge of social workers familiar with homeless population distribution in COA. The values of R_A for different land-cover types are given in the table below.

Dasymetric model disaggregates the homeless population and allocates each land cover cell a population count value. Using GIS to calculate the population for each land cover cell, the CD homeless polygon data was first converted into raster with homeless count as the input field. Next, to derive ‘E’, which is decided by the proportions of land-cover types in each BG, a raster map of the BGs’ FIPS codes was created, which was tabulated to calculate the areas of different land-cover types present in each BG. The proportions for each land use classes at BG level was then multiplied by its corresponding R_A to solve for ‘E’. The tabulated table was joined to the BG polygon layer and converted into a raster layer to derive the number of homeless populations per cell. To derive A_T, the area of the enumeration units was divided by the cell area (i.e. 900 m²). After deriving all required values, they were evaluated using Equation 1 to derive the cell population raster (Figure 3). Finally, the BG layer was combined with the cell population raster by zonal statistic to derive the final count of unsheltered homeless population for each BG in Austin.

To create a composite social vulnerability index, homeless index generated from the dasymetric model was combined with factors generated from the 22 selected variables (Table 1). The literature encounters a crossroad with different approaches of factor weighting. Many researchers, such as ^{2, 24} and ²² have used equal weighting to alleviate the burden of controversial weight assignment. Since weight assignment can greatly impact the resulting vulnerability assessment and there is no consensus for calculating vulnerability index, this study also adopts equal weight as well. Principal Component Analysis (PCA) was used to group highly correlated variables to model SOVI. Based on the inter-correlation among variables, PCA combines the statistically redundant variables into a component to generate a more robust set of social vulnerability factors. The PCA factors were then normalized and summed to obtain the relative measure of social vulnerability for each BG in Austin.

4. Results

The BG homeless count output (Figure 4) from the dasymetric model showed an agreement when compared with the CD homeless distribution (Figure 4). There were relatively higher numbers of homeless population spread across Austin downtown, along the major highways; Interstate-35, northwards along highways 183-North within Mueller park and in east Austin along highway 183-South within Rosewood and Montopolis neighborhoods. There were smaller pockets spread around west Austin. The dasymetric approach to disaggregating homeless data produced some interesting patterns in terms of spatial distribution. A more realistic pattern of the homeless is observed in the BG homeless map. Due to the spatial heterogeneity of land cover/use data used, more precise estimates of population can be derived at smaller census levels (e.g. Census tracts, BG or at pixel level). For instance, CD’s 2 and 8 in the CD homeless map (Figure 4) have the lowest counts (range 4 to 42) whereas these same districts house some BGs having a moderately low homeless count (range 3 to 6). This similar distribution pattern is seen across the other districts when compared with the BG map. This result provides some level of precision and specificity especially for policy makers, social workers as well as shelters to be able to predict locations and concentration of homeless. It also allows for ease and speed in counting homeless population as service providers in Austin can better group volunteers by BGs instead of a more cumbersome and possibly ineffective method at larger census levels. The BG homeless count is incorporated as an index into the social vulnerability assessment in section 4.3.

Using an eigenvalue of one as the threshold, the multicollinearity among all 22 variables was examined using PCA and produced five composite factors.

Factor loading of 0.7 was used as the threshold for grouping and classifying the “loaded” variables for each factor. Factor one depicts race and ethnic minorities and the socio-economically disadvantaged. It explains about 39 percent of the variance with American Indians, Alaska Natives and Hispanic (Spanish speakers), female-headed households, less educational attainment and Spanish-speaking populations having high loadings on this factor. This result corroborates the findings of previous studies that showed these range of traits are often popular among minority populations ^{2, 21}.

The factor scores derived from PCA alongside the homeless index were normalized using the min-max stretching formulae shown in equation 2 where is the summed value of a factor, is the minimum value in the range of a factor and is the maximum value in the range for a factor:

(2)

The additive model equation used to calculate SOVI is shown below, depicts individual factors added together to derive index By using this model, no weight was assigned as all factors were assumed to present equal relevance in the overall vulnerability model:

(3)

Finally, a social vulnerability map without homeless index was created using the five derived factors in the additive model. Likewise, another social vulnerability map with homeless index was created by inserting homeless index into the additive model. These maps are created at the BG level with classes ranging from low to high social vulnerabilities shown below in Figure 5 as well as their respective statistics in Figure 6 and Figure 7. SOVI scores were mapped based on their standard deviations from the mean into five categories to determine the least and most vulnerable BGs respectively.

Likewise, Figure 5 shows an east-west divide commonly reported in the social studies conducted in Austin. Inferring from these maps above, it is noticeable that the most vulnerable populations based on the selected variables are concentrated on the east side of the city where greater ethnic and racial inequalities as well as rapid population growth is prevalent. A particularly interesting observation from the SOVI map (Figure 5) shows a significant concentration of vulnerable populations in BGs around downtown Austin. SOVI with homelessness showed 115 BGs (22%) have a medium high to high (> 0.5 standard deviations (S.D.)) social vulnerability, while SOVI without homelessness showed only 58 BGs (11%). The least vulnerable BGs (> -2 S.D.) are seen to be in West Austin. Statistically, BGs at (> ± 2 S.D.) have p value of < 5% assuming normal distribution. These patterns signify the impact of incorporating homelessness as an index in calculating vulnerability and can help direct the attention of COA toward the above identified vulnerable block group locations either with or without homelessness. The highly vulnerable BGs include a geographic mix of highly urbanized BGs, large minority and socially dependent populations, including those in poverty and lacking in educational attainments. The highly vulnerable BGs are spread across Austin downtown, along the major highways; Interstate-35, northwards along highways 183-North within Mueller park, Rosewood and Montopolis neighborhoods in the Eastern Austin neighborhoods. The least vulnerable BGs are seen in West Austin having Davenport Ranch, Tarrytown and Northwest Hills neighborhoods. It is observed that more BGs in SOVI_H distribution fall into the medium vulnerable and highly vulnerable categories when compared with the frequency distribution for SOVI. Additionally, to spatially compare the derived SOVI maps with or without homelessness, a difference map and their mean centers and directional distribution ellipses were derived (Figure 8). Based on the difference map, there are more highly vulnerable BGs around downtown and not predominantly in the east as SOVI pattern shows. Also, the frequency distribution plot for SOVI_H in Figure 7 shows higher numbers of vulnerable BGs when compared with the distribution for SOVI (Figure 6). The directional distribution (Figure 8) was used to observe the pattern of SOVI and SOVI_H based on their scores. The pattern shows that SOVI_H shifts a bit more western than SOVI alone.

A paired t-test was performed to observe for any differences in the means of social vulnerability with homelessness (SOVI_H) or social vulnerability without homelessness (SOVI) in terms of their social vulnerability indices (part b of research question two) based on 534 BGs in Austin. From the t-test, the mean and standard deviation of SOVI and SOVI_H scores were (0.478 and ±0.072) and (0.393 and ±0.062) respectively. Furthermore, the paired t-test conducted revealed that there was a significant difference between both indexes (t = 92, p < 0.05, n = 534). This result also confirms the result derived from the spatial pattern observation between both indexes for RQ2a.

5. Discussion

In many cities, Austin included, homeless population data are typically aggregated to the level of administrative units for many reasons (e.g. privacy, ease of administration). However, detailed information on the spatial distribution of the population within these units is masked. In this study, dasymetric mapping techniques was used to disaggregate population to a finer spatial scale using ancillary data (i.e. land cover/use data). Thus, the dasymetric model used in this study reveals a successful method for disaggregating population data at a desired scale. The results from the social vulnerability assessments reveal the differences between the two social vulnerability assessments.

The geographic patterns observed in the result for this study suggests a key to improving social vulnerability assessment. As seen in the SOVI map (Figure 5), there were only a few BGs around Austin Downtown with medium high to high SOVI, with more of the concentration in the East while a high concentration of vulnerable population is seen both in BGs around Downtown and East Austin. Beside the spatial distribution observed (Figure 4 and Figure 5), it is interesting to note that more BGs in SOVI_H are categorized as being vulnerable when compared with BGs in SOVI. This finding stresses the relevance and importance of considering homelessness as one of the many social factors when evaluating social vulnerability indices for cities and considering the appropriate disaster management planning. Downtown Austin is a hub for many homeless individuals because of the presence of welfare and temporary shelter providers such as Salvation Army and Foundation for the Homeless. Most homeless population flock around to receive meals, donated clothing and other welfare resources. Homeless individuals are also known to be concentrated around major highways in Austin especially at road intersections and traffic lights (e.g. Mueller Park in North-east Austin).

Often, homeless people roam the streets begging for alms and putting themselves at high risk of being assaulted or hit by moving vehicles. Their whereabouts at road intersections, especially low water crossings, may also be susceptible to various natural hazards. Figure 4 and 5 reveal the pattern of social vulnerabilities in Austin which suggests that the most vulnerable BGs should be prioritized in disaster management. These neighborhoods (Mueller Park, Rosewood and Montopolis) have high potentials for losses during natural disasters and should therefore serve as priorities for disaster management officials during disaster emergencies. Hence, incorporating homeless distribution can better help researchers to identify the most vulnerable groups when conducting social vulnerability assessments. More importantly, a noticeable pattern in those maps (Figure 4 and Figure 5) suggest that using SOVI variables alone without homeless would have underestimated the vulnerability distribution and thereby under-prepare for the severe disaster to hit those communities.

For both vulnerability assessments with or without homeless (i.e. SOVI_H and SOVI respectively), the most vulnerable BGs are still predominantly in east Austin (Figure 5). The difference map (Figure 8) shows that the BGs with the most difference between SOVI and SOVI_H are in Downtown Austin, with a positive difference being mostly in the west (i.e. SOVI_H > SOVI) and the negative difference in the east (i.e. SOVI_H < SOVI). This result means western BGs could have been overestimated using the SOVI framework. For disaster management, this may not necessarily mean reversing the trend and investing more effort and resources in West Austin than East Austin (because the most vulnerable group are indeed in East Austin as indicated in Figure 5), but disaster managers may want to do targeted disaster planning in West Austin and consider the homeless population that are “hidden” in West Austin so that they are not being overlooked. The t-test results also confirm the importance of incorporating homeless as a variable in assessing vulnerability with a high significance observed when paired with SOVI without homelessness. Incorporating this key index, homelessness, in vulnerability studies will go a long way in aiding COA and Travis County managers in their effort to implement effective strategies and programs targeted at improving living conditions and overall social capital of vulnerable populations within their jurisdictions.

6. Strength and Limitation

The results from this study have shown differences in spatial patterns when compared to the results from past studies. The spatial distribution and orientation of the overall vulnerable populations take a slight shift to the West (Downtown Austin) unlike previous studies that have reported it being predominantly in East Austin. This study provides useful insights for identifying the neighborhoods that can benefit most from direct resources to aid social and economic development. Also, future studies in hazards and social vulnerabilities should consider adding homelessness in their works to create a more socially significant and realistic interpretation of the spatial distribution of social vulnerability. The sensitivity of homeless population presented some drawbacks in data availability. Since the best available homeless count data for COA was at the CD level, a finer scale level would have been preferred to validate the dasymetric modeling of homelessness at the BG level, and to aid the analysis with fewer uncertainties. A future direction of this study considers sampling homeless population and identifying salient factors that define pathways into homelessness. Furthermore, the rather complex and lengthy dasymetric approach could be further refined by developers and incorporated into a simple toolbox in mapping software for efficiency.

7. Conclusion

This study has introduced the application and relevance of homelessness as a factor in social vulnerability literature and applies it to Austin, Texas as a case study. This study has also presented social vulnerability as a multidimensional concept that helps in identifying those characteristics and experiences of communities that enable them to respond to and recover from hazards. The major dimensions of the social vulnerability of the study area are clustered into specific locations, East and downtown Austin. At a general consideration, economic welfare, age, and ethnicity are the major social attributes affecting the residents of those locations. In contrast with many studies that report social vulnerability in Austin as being solely as a result of a classic divide, this study presents a slight change in perspective showing that not only is East-Austin predominantly vulnerable, its Downtown region is highly vulnerable as well. This shows the impact of homelessness in computing social vulnerability indices. The methodology applied in this study has shown that incorporating homelessness into the broader range of social variables of vulnerability presents a significant difference when compared with SOVI with commonly used variables. This study also presents a framework for potential improvement and adaptation in the existing framework of social vulnerability assessment that have been widely adopted at various government levels in the U.S ². Besides mapping the SOVI with GIS, PCA was used to further explore the indicators with respect to their ranges of contribution to the overall SOVI. The factors identified in the statistical analysis are consistent with the broader hazard’s literature ^{2, 4, 29} which reveals the geographic variability in social vulnerability and the fundamental causes of vulnerability. While the methods used in this study can be replicated in future studies of social vulnerability and risk assessment, the results obtained can also be useful for decision making and prioritizing plans and strategies with regards to building effective coping capacity in those areas with higher social vulnerabilities. Ultimately, treating social vulnerability as a planning tool and applying it to various planning practices can motivate sustainable planning; promote sustainable development and human life. Handling social vulnerability as a planning tool is crucial for increasing resilience to natural hazards ^{28, 29}.

References