Environmental Impact and the Spatial ... - Wiley Online Library

4 downloads 444 Views 277KB Size Report
ized disaster zone (Miami), long-term recovery displaces socially disadvantaged res ... these atmospheric data on local census data to examine how the environ-.
When Nature Pushes Back: Environmental Impact and the Spatial Redistribution of Socially Vulnerable Populations n James R. Elliott, University of Oregon Jeremy Pais, State University of New York at Albany Objectives. This research investigates the spatial redistribution of socially vulnerable subpopulations during long-term recovery from natural disaster. We hypothesize that the local environmental impact of a disaster influences this redistribution process and that how it does so varies by the urban or rural context in which the disaster occurs. Methods. To test these hypotheses, we use a novel research design that combines the natural experiment offered by Hurricane Andrew with GIS technology and local census data. Results. Findings indicate that in a more urbanized disaster zone (Miami), long-term recovery displaces socially disadvantaged residents from harder-hit areas; yet, in a more rural disaster zone (southwestern Louisiana), long-term recovery concentrates socially disadvantaged residents within these harder-hit areas. Conclusion. These findings bridge classic and contemporary research on postdisaster recovery and open new terrain for thinking about how environmental and social forces intersect to transform regions in different settlement contexts.

Over recent years, the study of disasters has shifted away from examining affected communities as monolithic wholes to concentrating instead on the impact of disaster on subpopulations within these communities, as exemplified by Stallings’s (2002) reinterpretation of Moore’s (1958) classic, Tornadoes Over Texas. Whereas Moore’s original account highlights the collective resilience of all involved, Stallings shows how socially disadvantaged members of these affected communities actually became worse off, thereby reconceptualizing disaster recovery not as an act of unified resilience but as a struggle by privileged residents to restore the local social order, with n Direct correspondence, including requests for data and coding materials, to James Elliott, Department of Sociology, 1291 University of Oregon, Eugene, OR 97403 [email protected]. This research benefited from funding from the National Science Foundation (Award 0554818) and from the GIS and Population Workshop held at the Center for Spatially Integrated Social Science organized by Stephen Matthews, Don Janelle, and Michael Goodchild with funding from the National Institute of Child Health and Development (Award HD047744-01). In addition, we thank the editor and anonymous reviewers for their assistance in guiding this article to publication.

SOCIAL SCIENCE QUARTERLY, Volume 91, Number 5, December 2010 r 2010 by the Southwestern Social Science Association

1188

Social Science Quarterly

them on top. The corollary, now widely accepted in disaster studies, is that socially disadvantaged residents are vulnerable not just to disasters but to postdisaster recoveries and that this social fact is growing starker as federal funds for these recoveries continue to increase (e.g., Bolin and Stanford, 1998; Dash et al., 2007; Fothergill and Peek, 2004; Tierney, 2006). These insights have improved our understanding of the social dimensions of ‘‘natural’’ disasters, but they have also come at several methodological costs. One of these costs is the near-universal dismissal of the effects of environmental dimensions of the precipitating hazard on social recovery. In an effort to expose the social underpinnings of natural disasters, researchers have treated the ‘‘environmental side’’ of disasters simply as a trigger for more fundamental social processes of recovery, rather than as a legitimate force that intersects with these social processes to shape who will recover where. A related methodological cost is that contemporary research has tended to sacrifice the significance of local spatial context for the sake of unifying claims about the general types of obstacles faced by disadvantaged subpopulations. Consequently, we continue to know relatively little about the spatial structure of social vulnerability and its transformation during postdisaster recovery. A final methodological cost has been a heavy reliance on case studies to examine social inequities in postdisaster recovery. Although this approach continues to produce important insights, it also lacks an explicitly comparative base from which to formally evaluate how postdisaster recovery might unfold differently in different types of regions (e.g., urban compared with rural). The present study addresses these shortcomings through a novel research design that combines the strength of geographic information software (GIS) with that of a natural experiment. This design begins by using GIS technology and atmospheric data to delineate two very different regions hit by the same environmental hazard—Hurricane Andrew—which struck the urbanized area of South Miami, and then the less densely developed area of Southwest Louisiana, during a three-day span in August 1992. We overlay these atmospheric data on local census data to examine how the environmental impact of the hurricane redistributed vulnerable subpopulations— the low-income, the jobless, single mothers—along the storm’s path during the ensuing eight years of recovery, and how the type of region involved moderated these environmental patterns of population redistribution. In conducting this study, we focus specifically on sources of social vulnerability emphasized by prior sociological research as having negative effects on residents when spatially concentrated. The most prominent type of such place-based vulnerability is what W. J. Wilson (1987) calls ‘‘concentrated disadvantage.’’ Wilson contends, and subsequent studies have affirmed, that when socially disadvantaged residents—specifically, the poor, jobless, and single-mother families—cluster spatially, this clustering reduces access to vital social resources, leaving all residents worse off regardless of their individual status (Sampson, Morenoff, and Gannon-Rowley, 2002).

Environmental Impact and Spatial Redistribution

1189

This heightened vulnerability stems not just from individual and family deficits in resources, but also from the spatial accumulation of these deficits, which hamper access to opportunities and assets necessary for success. From this perspective, social vulnerability takes on a spatial dynamic, and the more spatially concentrated such vulnerability becomes, the more socially detrimental its effects are presumed to be, not just for individuals but entire communities. By focusing on these types of social disadvantage, we highlight shifts in the spatial structure of social vulnerability during postdisaster recovery and, in the process, further illuminate the unintended consequences of the United States’ current ‘‘free market’’ approach to disaster recovery, in which billions of dollars in public aid are pumped into federally declared disaster zones with minimal regulatory oversight or attention to local social inequalities that shift spatially during recovery. Literature Review

Although research on long-term disaster recovery is limited, notable studies exist. The most systematic, national-level study remains Wright, Rossi, and Wright’s (1979) classic, After the Cleanup. In this analysis, researchers assembled population and housing data from the 1960 and 1970 Censuses for all U.S. counties and Census tracts to determine if those hit by a flood, tornado, or hurricane during the decade experienced different growth trajectories in population and housing than those that were not hit. In short, the authors found no discernible difference between disaster and nondisaster settings; instead, they conclude: ‘‘The comparison of average damages to average resources makes it implausible in the extreme to expect that these disasters would have residual and observable effects. In our studies, none were found’’ (Wright, Rossi, and Wright, 1979:198). Friesema and colleagues (1977) report similar findings from their time-series analysis of city-level indicators of social and economic change before and after natural disasters during the early 1970s. These classic, macro-level studies imply that after a few years, regions hit by an acute environmental hazard tend to achieve a ‘‘functional recovery,’’ defined as ‘‘the replacement of the population and of the functioning equivalent of their needs in homes, jobs, capital stock and urban activities’’ (Haas et al., 1977:3). More recent research on disaster recovery does not deny these patterns; instead, it contends that a focus on aggregate changes in population and housing misses the challenges faced by socially vulnerable subpopulations struggling to reestablish themselves within disaster-torn regions. The basic argument behind this body of research is that social inequalities that divide and structure a region before disaster also divide and structure it after, producing not one but many different recoveries, each reflecting different levels of personal and collective resources available to constituent subpopulations. This perspective can be traced to Bates et al.’s (1963) study of

1190

Social Science Quarterly

Hurricane Audrey, a Category 4 storm that struck southeastern Texas and southwestern Louisiana in 1957. In this longitudinal study of personal and regional recovery, Bates and colleagues found that the lives of working-class victims were disrupted much more seriously than were the lives of people from upper-class backgrounds, who had greater personal, social, and political resources at their disposal. Contemporary case-study research affirms these patterns. For example, in their volume on Hurricane Andrew’s impact on South Miami, Peacock, Morrow, and Gladwin (1997) advance a ‘‘sociopolitical ecology’’ of recovery to explain how preexisting social inequalities were not only exposed by the disaster but also exacerbated by long-term recovery efforts. They find that in the years following Hurricane Andrew, minority households often lacked sufficient insurance to rebuild their housing and businesses; that low-income households had much lower disaster loan approval rates than higher-income households; and, consequently, that low-income and minority families often found themselves moving from place to place after the disaster in search of affordable housing, or leaving the region altogether. Dash and colleagues (2007:20) report similar patterns in their study of long-term recovery in the working-class community of South Miami Heights. They find, for example, that survivors who stayed in the area reported recoveries ‘‘filled with misery and false steps, explained in part by their own lack of knowledge or economic resources, but also by the neglect of authorities, and abuse by those who prey on the weak in times of crisis.’’ Based on these recurrent patterns, Tierney (2006:210) claims: ‘‘To observe disaster aid and recovery in the U.S. is to see the Matthew Effect in action. Benefits accrue to those who possess wealth and social and cultural capital, while larger proportional losses are borne by the poor and marginalized.’’ Indeed, these types of social inequalities now constitute a unifying theme in cross-disciplinary research on New Orleans’s recovery from Hurricane Katrina (e.g., Brunsma, Overfeldt, and Picou, 2007; Elliott and Pais, 2006), as well as follow-up research on long-term recovery from Hurricane Andrew (e.g., Dash et al., 2007). Yet as insightful as this line of research continues to be, it remains inhibited by several methodological limitations. First, its case-study orientation offers no formal comparative basis on which to build more general conclusions about postdisaster recoveries and how they might vary spatially in and across affected regions. Second, by focusing so intently on social vulnerability, contemporary research misses how the ‘‘environmental side’’ of a disaster can intersect with its ‘‘social side’’ to spatially redistribute vulnerable subpopulations during long-term recovery. The present study begins to address these shortcomings by combining the macro-comparative orientation of classic studies of disaster recovery with the social-vulnerability focus of contemporary case studies to examine environmental patterns in the spatial redistribution of social disadvantage during long-term recovery from natural disaster. Next, we review our hypotheses for this examination.

Environmental Impact and Spatial Redistribution

1191

Competing and Moderating Hypotheses Regarding Local Environmental Impact and Recovery Neither classic macro-level research nor contemporary case-study research denies environmental variability in regional recoveries from disaster; instead, they simply ignore it, leaving us with an incomplete understanding of how environmental and social forces intersect to transform regions and their vulnerable subpopulations after disaster strikes. When we focus on this intersection, two competing hypotheses emerge, each equally plausible in the context of contemporary research. On the one hand, there is what we might call the displacement hypothesis, which asserts that where the environmental impact of a hazard is greater, socially vulnerable subpopulations are more likely to become displaced because environmental hardship caused by the hazard mixes with personal hardship to disperse those least able to draw on individual and network resources to recover. Studies have shown, for example, that poorer residents often live in structurally weaker dwellings that are left uninhabitable when disasters strike (Cochrane, 1975) and that these same residents often lack the financial resources necessary to recover ‘‘in place’’ (Bolin and Stanford, 1998; Hewitt, 1997). Research also shows that poorer residents have more difficulty accessing (Dash et al., 2007; Peacock et al., 1997) and navigating (Rovai, 1994; Forthergill, 2004) bureaucratic systems of disaster assistance, leaving more affluent residents better positioned to absorb available housing after disaster, thereby exacerbating housing shortages for less-affluent residents in the area (Quarantelli, 1994; see also Elliott and Pais, 2006). Thus, researchers commonly discover that after a major disaster: ‘‘Low-income families find themselves moving frequently from one place to another (or even leaving the city forever), or in housing they can’t afford’’ (Haas et al., 1977:xxviii). This type of displacement is reminiscent of classic images of urban renewal, in which less-advantaged residents are swept from their neighborhoods by large-scale, federally subsidized redevelopment projects intended largely for the benefit of others. On the other hand, however, there is what we might call the concentration hypothesis, which asserts that where the environmental impact of a hazard is greater, socially vulnerable subpopulations are more likely to increase because environmental hardship caused by the hazard mixes with personal hardship to drain local property values and public infrastructures, thereby decreasing local resettlement by more advantaged residents, who collectively remove themselves from heavily damaged areas. This alternative hypothesis is consistent with Dash and colleagues’ (2007) study of long-term recovery in the multiethnic, working-class community of South Miami Heights after Hurricane Andrew, which severely damaged 70 percent of the town’s housing stock. They find that residents who moved from the area after the disaster had significantly higher incomes than those who stayed, corroborating earlier findings by Smith and McCarty (1996). As one informant told the

1192

Social Science Quarterly

researchers: ‘‘Most all the neighbors left. They sold the houses the way they were or fixed them a little and sold them to go somewhere else they liked better’’ (Dash et al., 2007:18). Another informant added: ‘‘A lot of homes turned into Section 8 [subsidized] housing after Andrew. Because the people bought them cheap, fixed them up cheap, and rented them’’ (Dash et al., 2007:18). This alternative image of disaster recovery is not one of renewal and displacement, but the opposite: uneven repair, transient resettlement, and absentee ownership, much akin to the ‘‘zone of transition’’ in classic models of urban development, in which less-advantaged subpopulations concentrate in rundown areas, as speculative property owners limit repairs and await real estate values to appreciate over time. This scenario is also consistent with the ‘‘minority move-in’’ hypothesis in the environmental justice literature, which asserts that less-advantaged subpopulations typically find themselves concentrated in environmentally undesirable areas because housing prices are cheaper and they can afford few other options. In both scenarios—displacement and concentration—the environmental impact of the respective hazard is conceptualized as having both direct and indirect effects. The direct effects revolve around the acute physical damages inflicted by the hazard—torn roofs, downed trees, flooded structures, and so forth—which influence who returns where, when, and at what rate. By contrast, the indirect effects evolve through a tangled web of private insurance policies and public assistance programs that collectively privilege property ownership and restoration in ways that encourage real estate speculation and upgrading among those who can afford it, while leaving more vulnerable subpopulations to scramble for whatever housing options emerge from this ‘‘trickle-down’’ recovery process. In considering which type of recovery—displacement or concentration—is more likely under these conditions, it is quite possible that the answer depends on the type of region in question; specifically, whether it was already more or less densely developed before the disaster struck. This possibility constitutes a moderating hypothesis because it asserts that the regional context in which a disaster occurs may moderate, or alter, its environmental impact on the spatial redistribution of locally vulnerable subpopulations during recovery. The logic behind this hypothesis is that in more urbanized areas, the number of residents affected by a major disaster is larger; property values are higher; recovery dollars are greater; and local development coalitions, or growth machines (Pais and Elliott, 2008; Logan and Molotch, 1987) are more organized and aggressive in their efforts to boost local exchange values by quickly pursuing new development opportunities presented by the disaster. These factors, in turn, are likely to have one of two effects on recovery. Either they will intensify the recovery patterns evident in less urbanized areas (intensification hypothesis), or they will generate starkly different recovery patterns than in less urbanized areas (divergence hypothesis). A formal test of these moderating hypotheses is useful because prior research offers no definitive guide to thinking about such

Environmental Impact and Spatial Redistribution

1193

contextual variation and the common, and perhaps incorrect, intellectual temptation is to take lessons learned from heavily studied urban disasters and apply them directly to rural contexts, where relevant processes and outcomes may differ greatly by comparison. Next, we discuss the research design and data used to examine these hypotheses. Data and Research Design

We use a natural experiment initiated by Hurricane Andrew—the second costliest federally declared disaster in U.S. history—to examine how environmental impacts from such a disaster intersect with social processes to redistribute vulnerable subpopulations during long-term recovery. This unintended experiment began abruptly at the end of August 1992, when Hurricane Andrew made landfall as a Category 4 storm that cut a 25-milewide, 60-mile-long swath across the southern Miami metro area, damaging more than 100,000 private homes and rendering approximately half these structures uninhabitable (Peacock, Morrow, and Gladwin, 1997:2–7). The experiment continued two days later, as Hurricane Andrew—reenergized by warm waters of the Gulf of Mexico—made landfall again over Southwest Louisiana, with peak wind speeds estimated at more than 115 miles per hour. In the two Louisiana parishes most directly affected by the storm, more than 95 percent of homes were damaged, and recovery loans totaled more than $39 million (at the time), making it one of the worst storms ever to hit the region. Yet, despite sharing the same environmental hazard, the Miami and Louisiana areas hit by Hurricane Andrew could hardly have differed more in their population size, density, and physical orientation toward the hazard. In the Miami area, properties and people concentrated heavily near the shoreline, as part of a large urban-tourist complex, whereas in the Louisiana area, smaller numbers of people and businesses concentrated off the relatively uninhabited shoreline because of soggy wetland soils that thwarted development. Consequently, the two regions hit by Hurricane Andrew offer two ideal-types of settlement areas in which to examine long-term recovery, while controlling for the historical period and length of recovery under investigation. To draw comparisons across these two areas, we use atmospheric data from the Hazards U.S. Multi-Hazard (HAZUS-MH) database to establish consistent spatial boundaries for each disaster zone. The HAZUS-MH is a federally sponsored program developed under contract with the National Institute of Building Sciences (NIBS), which devised a wind modeling technology to estimate hurricane intensities across affected regions. This technology has been validated using historical records for all major hurricanes that struck the United States between 1886 and 2001. In the present study, we employ HAZUS-MH to designate each respective disaster zone as

1194

Social Science Quarterly

consisting of all contiguous Census tracts—our primary unit of analysis— that experienced at least tropical-storm-force winds (over 50 miles per hour) during Hurricane Andrew. This approach provides a more consistent and refined delineation of environmental impact than larger city or county boundaries can offer, and it provides a means of examining local variability within each disaster zone because each Census tract is assigned an average estimated wind speed from the storm. To estimate hurricane paths and local wind speeds, the HAZUS database uses mathematical models first tested by Russell (1971) and most recently refined by Vickery et al. (2000). Wind estimates are calculated using known information about the storm that includes central pressure, speed of the system, storm heading, and distance from the eye to hurricane force winds. With these data, HAZUS generates an accurate representation of a hurricane wind field that engineers use to approximate the level of structural damage posed by hurricane winds. The estimated wind speeds for this study are an average between the peak gusts and maximum sustained wind speeds for each Census tract hit by Hurricane Andrew. These wind speeds serve as our chief variable of interest in statistical analyses of environmental impact and yield a disaster zone of 627 Census tracts in the Miami area, with a total population of 3.2 million in 1990 and a density of 581 residents per square mile. The disaster zone in the Louisiana area consists of 268 Census tracts with a 1990 population of 1.1 million and a density of 125 residents per square mile. Maps of these two disaster zones, central storm tracks, and local wind speeds appear in Figure 1, along with basic demographic data for each disaster zone. To analyze local demographic change, we merge these environmental data with Census tract data from the 1990 (prestorm) and 2000 (poststorm) population censuses, using Geoltyics’ Neighborhood Change Database (NCDB). Through sponsorship of the Urban Institute and compiled by GeoLytics, the NCDB standardizes tract boundaries across decennial censuses, which means that our analyses of tract-level changes are computed for fixed spatial units over time, using 2000 boundaries (see NCBD Data Users Guide Long Form Release for a detailed discussion of this methodology). The implied recovery period of approximately seven years (1992–1999) is consistent with recent studies of long-term disaster recovery in California and Florida (Dash et al., 2007; Webb, Tierney, and Dahlmer, 2002). To examine changes in socially vulnerable subpopulations at the tract level, we take two, overlapping approaches. The first estimates a common index of concentrated disadvantage, or vulnerability, at the tract level using four highly correlated variables: the poverty rate; median household income (reverse coded); percentage of families that are female headed with children under 18 years old; and the rate of joblessness among men ages 18–64. We standardize each of these indicators separately for each zone and Census-year of observation using Z scores and then sum these scores to compute a single Tract Disadvantage Index (TDI). Cronbach’s alpha for this index is 0.89 in

Environmental Impact and Spatial Redistribution

1195

FIGURE 1 Disaster Zone, Central Storm Track, Estimated Wind Speeds, and Basic Demographic Changes in Two Regions Hit by Hurricane Andrew, 1992 Miami Strike Zone (627 census tracts) 1990 % Change 1990-2000 Population 3,250 18.9 (000s) Density 581 19.5 (per square mile) Housing Units 1,465 13.3 (000s) 40,714 5.3 Median HH income % Single-Mother 26.0 9.2 HHs % Jobless Men 21.0 4.2 % in Poverty 14.6 7.9

Louisiana Strike Zone (268 census tracts) 1990 % Change 1990-2000 Population 1,111 9.2 (000s) Density 125 8.9 (per square mile) Housing Units 449 11.1 (000s) 30,751 10.4 Median HH income % Single-Mother 27.5 12.7 HHs % Jobless Men 28.1 6.5 % in Poverty 24.7 -13.5

1990 and 0.90 in 2000, indicating a high degree of statistical reliability in the measurement of a single, unidimensional latent construct of social vulnerability. The advantages of using this index lie in its multidimensionality and ubiquity in studies of neighborhood disadvantage; however, by standardizing and combining indicators in this way it becomes difficult to determine which types of vulnerable subpopulations are most at risk of spatial redistribution and whether respective patterns apply to all groups involved. To address these shortcomings and increase the robustness of our analyses, we also estimate tract-level changes in each subpopulation separately. To test hypotheses regarding variation in recovery along the storm’s track within and across disaster zones, we estimate a series of linear regression equations of the following general form, where all variables are measured for

1196

Social Science Quarterly

the ith Census tract: DV 19902000;ij ¼ a þ b1½wind speedij  þ b2½regionj  þ b2½wind speedij regionj  þ bX ½controlsij : For each indicator of vulnerability, V, we compute a change score, DV1990–2000, i, by subtracting the 1990 value from the 2000 value in each tract. (For a review of using change scores in panel models, see Halaby, 2004.) We then regress this score on the tract’s estimated wind speed at time of impact (in miles per hour), regional location (Miami or Louisiana), and an interaction term to test for moderating effects between these two factors. In addition to these variables, we include a number of statistical controls. One is a dummy indicator for location on the coast (1 5 touching the coast; 0 5 not touching the coast) because property values and use patterns are likely to vary by coastal and inland location, and because storm surge associated with coastal location can significantly influence the cost and extent of long-term recovery. Another set of controls include those commonly used in prior macro-level analyses of postdisaster recovery (e.g., Friesema et al., 1977; Wright, Rossi, and Wright, 1979): population size (logged), population density (measured as persons per square mile), and minority presence (measured as percent non-Hispanic white). Prior studies indicate that these factors influence the relative speed with which a tract’s demographic composition changes over time, as well as racial inequalities that might influence this change. Each of these control variables is measured in 1990, prior to Hurricane Andrew, to establish temporal priority in estimation of change in the dependent variable. All continuous covariates are zeromean centered to facilitate interpretation. In fitting these models of change, temporal regression to the mean and spatial autocorrelation are both concerns. To adjust for the former, we include the lagged value of our dependent variable in 1990 as another control variable. To adjust for spatial autocorrelation, we construct a spatially lagged dependent variable for each tract in 1990 from a row-standardized spatial weights matrix that uses rook-based contiguity to define the connectivity among neighboring Census tracts. Inclusion of this spatially lagged dependent variable in our regression equations reduces estimation bias and helps control for the influence of a tract’s spatial context on its estimated change over the observed period. Results of our analyses follow. Results

Background statistics in Figure 1 indicate that both disaster zones under review grew substantially in total population and housing units during the

Environmental Impact and Spatial Redistribution

1197

observed recovery period, despite being hit by a major environmental disaster. The Miami zone added approximately 19 percent to its total population and 13 percent to its total housing stock between 1990 and 2000, while the Louisiana zone added approximately 9 percent and 11 percent, respectively, over the same period. Thus, the general demographic context in which long-term recoveries from Hurricane Andrew unfolded is one of growth and development, not decline and abandonment. Next, to test hypotheses regarding environmental impact and changes in socially vulnerable subpopulations, we estimated our regression model two ways: once with, and once without, the region-by-wind-speed interaction term, which tests for the moderating influence of regional context. We found that the interactive (i.e., moderating) model provides a statistically better fit than the additive model for all measures of social vulnerability except the percentage of residents living in poverty. However, all interaction effects in Table 1 point in a negative direction, which means that the effect of hurricane force winds on changes in vulnerable subpopulations at the Census-tract level varies consistently by region for all indicators. For example, in the Miami area, a one-mile per hour increase in hurricane winds significantly reduced the level of neighborhood disadvantage by 0.005 standardized units (0.0041  0.009), whereas in Louisiana, the effect of wind damage (0.004) is not significantly different from zero. In Miami, this is a nontrivial effect that translates into roughly a 10 percent greater decline in neighborhood disadvantage for Census tracts that experienced at least Category 2 force winds relative to Census tracts that experienced only Category 1 force winds. Additionally, Table 1 shows patterns of change in median household income and in the percentage of households headed by single mothers that parallel changes in our standardized index of neighborhood disadvantage. However, results for male joblessness differ in one sense: over the recovery period, Miami experienced approximately a 2 percent greater increase in male joblessness than southern Louisiana (2.152). However, the negative interaction effect (  0.068) indicates that the hardest-hit tracts in Miami experienced a smaller rate of increase in joblessness. As mentioned above, the fifth indicator—change in the percent of residents living in poverty—does not vary significantly along the storm’s path (although the negative interaction coefficient points in a consistent direction with other indicators,  0.037). This finding underscores the point that strict reliance on poverty statistics to assess changes in vulnerable subpopulations may miss significant changes in related forms of social vulnerability at the local level. Overall, these changes indicate significant shifts in socially vulnerable subpopulations along Hurricane Andrew’s path and that these shifts occur in different directions in different regional contexts. To illustrate these divergent patterns, we graph the results of Table 1 in Figure 2, holding control variables constant at their sample means. In short, the negative and statistically significant coefficients for the region-by-wind-

0.244 (0.031) n n n

0.249 (0.023) n n n 0.027 (0.058) 2.21 0.38 895

 0.853 (0.095) n n n  0.660 (0.021) n n n

 0.890 (0.062) n n n  0.593 (0.017) n n n

 0.018 (0.035) 3.33 0.22 895

 1.968 (0.853) n 2.764 (0.627) n n n  0.002 (0.079)

 0.217 (0.062) n n n 0.306 (0.046) n n n 0.009 (0.006)

 0.115 (0.037) n n 0.140 (0.027) n n n 0.004 (0.003)

4.330 (0.798) n n n 3.11 0.26 895

0.282 (0.032) n n n

 9.868 (1.487) n n n  0.603 (0.023) n n n

 3.015 (0.876) n n  0.167 (0.048) n n n

(s.e.)

 0.177 (0.064) n n  0.014 (0.003) n n n

b

 0.153 (0.038) n n n  0.009 (0.002) n n n

(s.e.) 0.049 (0.045)

b 0.006 (0.003)

(s.e.)

% Single-Mother Households

0.004 (0.002)

b

Median Household Incomeb

n

a

(s.e.)

2.943 (0.589) n n n 2.44 0.43 895

0.276 (0.036) n n n

 14.954 (0.984) n n n  0.571 (0.024) n n n

 0.816 (0.631)  0.060 (0.455) 0.060 (0.057)

2.152 (0.693) n n  0.068 (0.034) n

0.055 (0.032)

b

% Male Joblessness

po0.05; n npo0.01; n n npo0.001. Models are estimated via seemingly unrelated regression (SUR). See Greene (2005:340–51) for additional information. b Median family income is inverse coded so that higher values signify greater levels of disadvantage for all indicators. c Highest variance inflation factors (VIF) among the control variables (i.e., the noninteractive covariates).

Independent Variables Average peak wind speed (in miles per hour) Region (1 5 Miami; 0 5 Louisiana) Region  Average peak wind speed Control Variables On coast (1 5 yes; 0 5 no) Ln(population) 1990 Population density (persons per sq. mile)1990 Percent non-Hispanic white1990 Temporally lagged dependent variable1990 Spatially lagged dependent variable1990 Constant Highest VIFc R2 N

Tract Disadvantage Index

Dependent Variables: Change Scores (2000–1990)

(s.e.)

 0.720 (0.610) 3.79 0.21 895

0.221 (0.030) n n n

 8.608 (1.062) n n n  0.531 (0.023) n n n

0.172 (0.603) 0.830 (0.443) 0.033 (0.055)

0.063 (0.722)  0.037 (0.033)

 0.004 (0.031)

b

% in Poverty

Regression Results Predicting Subregional Changes of Social Vulnerability During the Recovery Period from Hurricane Andrew in 1992a

TABLE 1

1198 Social Science Quarterly

Environmental Impact and Spatial Redistribution

1199

FIGURE 2 Subregional Change of Socially Vulnerable Subpopulations from 1990 to 2000 by Wind Speed and Hurricane Region B 8

.6 % Change in Male Joblessness

Change in Neighborhood Disadvantage (Standardized Units)

A .4 .2 0 –.2 –.4

2 0

–.6 60

100 80 120 140 Average Wind Speed (mph)

60

120 80 100 140 Average Wind Speed (mph)

60

80 100 120 140 Average Wind Speed (mph)

D Change in Median Family Income (Inverse Coded Standardized Units)

C % Change in Single-Mothered Households

6 4

10 8 6 4 2 0 –2 –4 –6 60

80 100 120 140 Average Wind Speed (mph)

.6 .4 .2 0 –.2 –.4 –.6

speed interaction terms in Table 1 support the displacement hypothesis in the more urbanized Miami disaster zone but the concentration hypothesis in the less urbanized Louisiana disaster zone. These divergent patterns are clearly evident in the inverted graph lines in Figure 2, which show increases in vulnerable subpopulations along the storm’s path in Louisiana but the opposite pattern in Miami. These findings support the hypothesis that changes in vulnerable subpopulations vary along the storm’s path and the moderating hypothesis that this variation differs by the type of disaster zone under investigation.

Conclusion

Our comparative study of long-term recovery from Hurricane Andrew in two different regions contributes to understanding of the social dimensions of natural disasters in several ways. First, it shows that disasters do not simply expose socially vulnerable subpopulations but also redistribute them during long-term recovery. Second, it shows that this redistribution bears the environmental imprint of the actual hazard involved, as residents with highly unequal social and personal resources jockey to reestablish themselves in recovering regions over time. Third, it shows that how this redistribution occurs differs greatly in more- and less-developed coastal areas. In more

1200

Social Science Quarterly

developed, or urbanized, areas, long-term recovery works to displace socially vulnerable residents from harder-hit areas; yet, in less-developed regions, long-term recovery works to concentrate socially vulnerable residents in these harder-hit areas. Although our analyses cannot fully determine why these regional differences occur, our reasoned assessment is that disaster recovery in more urbanized areas is similar to traditional patterns of urban renewal in areas of valued real estate: it uses public dollars to leverage private investment by those who can afford it, which ends up displacing those who cannot. Where wind speeds (and presumably damage) are greater, this displacement of socially vulnerable subpopulations increases significantly, as indicated by the negative slopes for the Miami area in Figure 2. In more sparsely populated disaster zones—with less property, fewer people, and smaller pro-growth coalitions—the opposite pattern occurs: the greater the local impact of the storm, the greater the local increase in socially vulnerable subpopulations. This concentration of social disadvantage along the storm’s path is similar to patterns of collective disinvestment and speculative real estate practices in classic ‘‘zones of transition’’ awaiting future development. More generally, these findings help bridge classic and contemporary research on disaster recovery and begin to open new terrain for thinking about how environmental and social forces intersect, often unintentionally, to transform affected regions during postdisaster recovery. The broader implication is that to ignore either side of the recovery equation—social vulnerability or environmental impact—is to develop an incomplete view of challenges facing less-advantaged residents in the wake of natural disaster. As future research explores these dynamics, several limitations of the present study are worth considering. First, our focus on net demographic changes at the tract level offers little direct insight into the micro-level processes driving these changes. Consequently, we cannot tell if socially disadvantaged groups shifted along the storm’s path because prestorm residents in these areas changed status or because they were replaced by newcomers who themselves differed in status from prestorm residents. We suspect that both processes are at work and that future research could examine these dynamics more closely by using individual-level data to explore which types of residents tend to enter, exit, and stay in harder-hit areas during long-term recovery. Another limitation of the present study is that it relied on an indirect measure of environmental impact—wind speed (controlling for coastal proximity)—to assess its influence on regional recovery. Our working assumption was that this measure provides a reliable proxy for direct and indirect ‘‘effects’’ associated with a disaster’s physical impact on local recovery. We believe this assumption is reasonable, but also recognize that more research on this issue would be useful. One way to proceed would be to develop more direct measures of local destruction (e.g., average repair costs; average length of time to repairs; average change in housing values) and receipt of recovery capital in various

Environmental Impact and Spatial Redistribution

1201

forms (e.g., from private insurance claims; federal residential assistance; small business loans). Such information would be labor intensive to collect and analyze, but it would help distinguish direct versus indirect effects of environmental impact on subsequent recoveries. Finally, it is important to remember that these data limitations are not just academic but shared by actual communities trying to recover from disasters. In this respect, it is somewhat surprising that the federal government has not implemented formal mechanisms or protocols to ascertain basic population estimates after federally declared disasters. In the face of a potential catastrophe in a major U.S. metropolitan area, this deficiency is troubling, to say the least. Knowledge about affected subpopulations and their spatial redistribution is essential for the efficient and effective distribution of disaster aid; it is fundamental for determining how basic social services (healthcare, education, childcare) should be redeveloped and where; and it is critical for ensuring basic equities during government-sponsored recovery efforts. Our recommendation is for a federal program to be established that monitors and reports changing population characteristics in affected communities throughout the recovery process whenever a federal disaster is declared. Moreover, this information should be shared with affected residents, as well as with local and state officials and researchers. Without this basic information, improvements in federal disaster policies will be long in coming, and the possibility of remedying unequal recoveries remains unlikely.

REFERENCES Bates, Frederick L., C. W. Fogleman, V. J. Parenton, R. H. Pittman, and G. S. Tracy. 1963. The Social and Psychological Consequences of a Natural Disaster: A Longitudinal Study of Hurricane Audrey. Washington, DC: National Academy of Sciences, National Research Council. Bolin, Robert, and Lois Stanford. 1998. The Northridge Earthquake: Vulnerability and Disaster. London: Routledge. Brunsma, David L., David Overfeldt, and J. Steven Picou. 2007. The Sociology of Katrina: Perspectives on a Modern Catastrophe. Lanham, MA: Rowman and Littlefield. Cochrane, Harold C. 1975. Natural Hazards and Their Distributive Effects. Monograph #NSF-RA-E-75-003. Boulder, CO: Institute of Behavior Science. Dash, Nicole, Betty Hearn Morrow, Juanita Mainster, and Lilia Cunningham. 2007. ‘‘Lasting Effects of Hurricane Andrew on a Working-Class Community.’’ Natural Hazards Review February:13–21. Elliott, James, and Jeremy Pais. 2006. ‘‘Race, Class and Hurricane Katrina: Social Differences in Human Response to Disaster.’’ Social Science Research 35:295–321. Fothergill, Alice, and Lori A. Peek. 2004. ‘‘Poverty and Disaster in the United States: A Review of Recent Sociological Findings.’’ Natural Hazards 32:89–110.

1202

Social Science Quarterly

Friesema, H. Paul, J. Caporaso, G. Goldstein, R. Lineberry, and R. McMcleary. 1977. Community Impacts of Natural Disasters. Evanston, IL: Northwestern University Press. Greene, William H. 2005. Econometric Analysis, 5th ed. Upper Saddle River: NJ Pearson Education. Haas, J. Eugene, Robert W. Kates, and Martyn J. Bowden. 1977. Reconstruction Following a Disaster. Boston, MA: MIT Press. Halaby, Charles. 2004. ‘‘Panel Models in Sociological Research: Theory into Practice.’’ Annual Review of Sociology 30:507–44. Hewitt, K. 1997. Regions of Risk: A Geographic Introduction to Disasters. London: Longman. Logan, John, and Harvey L. Molotch. 1987. Urban Fortunes. Los Angeles, CA: University of California Press. Moore, Harry Estill. 1958. Tornadoes Over Texas: A Study of Waco and San Angelo in Disaster. Austin, TX: University of Texas Press. Pais, Jeremy, and James R. Elliott. 2008. ‘‘Places as Recovery Machines: Vulnerability and Neighborhood Change After Major Hurricanes.’’ Social Forces 86(4):1415–53. Peacock, Walter Gillis, Betty Hearn Morrow, and Hugh Gladwin. 1997. Hurricane Andrew: Ethnicity, Gender and the Sociology of Disasters. New York: Routledge. Quarantelli, E. 1994. Draft of a Sociological Disaster Agenda for the Future: Theoretical, Methodological and Empirical Issues. University of Delaware Disaster Research Center Preliminary Papers 228. Available at hhttp://www.udel.edu/DR/preliminary/228.pdfi. Rovai, E. 1994. ‘‘The Social Geography of Disaster Recovery: Differential Community Response to North Coast Earthquakes.’’ Association of Pacific Coast Geographers Yearbook 56. Russell, L. R. 1971. ‘‘Probability Distributions for Hurricane Effects.’’ Journal of Waterways, Harbors, and Coastal Engineering Division 1:139–54. Sampson, Robert J., Jeffrey D. Morenoff, and Thomas Gannon-Rowley. 2002. ‘‘Assessing Neighborhood Effects: Social Processes and New Directions in Research.’’ Annual Review of Sociology 28:443–78. Smith, Stanley K., and Christopher McCarty. 1996. ‘‘Demographic Effects of Natural Disasters: A Case Study of Hurricane Andrew.’’ Demography 33:265–75. Stallings, Robert A. 2002. ‘‘Weberian Political Sociology and Sociological Disaster Studies.’’ Sociological Forum 17(2):281–305. Tierney, Kathleen. 2006. ‘‘Foreshadowing Katrina: Recent Sociological Contributions to Vulnerability Science.’’ Contemporary Sociology 35:207–12. Vickery, P. J., P. F. Skerjl, A. C. Steckley, and L. A. Twisdale. 2000. ‘‘Hurricane Wind Field Model for Use in Hurricane Simulations.’’ Journal of Structural Engineering 126(10):1203– 21. Webb, Gary R., Kathleen J. Tierney, and James M. Dahlmer. 2002. ‘‘Predicting Long-Term Business Recovery from Disaster: A Comparison of the Loma Prieta Earthquake and Hurricane Andrew.’’ Global Environmental Change Part B: Environmental Hazards 4(2/3):45–58. Wilson, William Julius. 1987. The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy. Chicago, IL: University of Chicago Press. Wright, James D., Peter H. Rossi, and Sonia R. Wright. 1979. After the Clean Up: LongRange Effects of Natural Disasters. New York: Sage.