Prof_Marcos A only
pzftoresc8h5
Measuringdisasterrecoverybouncing.pdf
Measuring disaster recovery: bouncing back or reaching the counterfactual state?
Shaoming Cheng Associate Professor, Department of Public Administration, Florida International University, United States, Emel Ganapati Associate Professor, Department of Public Administration, Florida International University, United States, and Sukumar Ganapati Associate Professor, Department of Public Administration, Florida International University, United States
How should one measure the recovery of a locale from a disaster? The measurement is crucial from a public policy and administration standpoint to determine which places should receive disaster assistance, and it affects the performance evaluation of disaster recovery programmes. This paper compares two approaches to measuring recovery: (i) bouncing back to pre-disaster conditions; and (ii) attaining the counterfactual state. The former centres on returning to nor- malcy following disaster-induced losses, whereas the latter focuses on attaining the state, using quasi-experimental design, which would have existed if the disaster had not occurred. Both are employed here to assess two housing recovery indicators (total new units and their valuations) in Hurricane Katrina-affected counties (rural and urban). The examination reveals significantly different outcomes for the two approaches: counties have not returned to their pre-disaster housing conditions, but they do exhibit counterfactual recovery. Moreover, rural counties may not be as vulnerable as assumed in the disaster recovery literature.
Keywords: housing, Hurricane Katrina, interrupted time series design, quasi- experimental design, recovery, rural versus urban recovery
Introduction In their seminal work, Human Systems in Extreme Environments, Mileti, Drabek, and Haas (1975) noted that, among the four phases of a disaster—mitigation, preparedness, response, and recovery—the latter is studied the least (see also Berke, Kartez, and Wenger, 1993; Zhang and Peacock, 2009). Disaster recovery is a multi-dimensional process, encompassing the ‘restoring, rebuilding, and reshaping [of ] the physical, social, economic, and natural environment through pre-event planning and post-event actions’ (Smith and Wenger, 2007, p. 237). It is important for public administration research and practice, presenting public administrators with an opportunity to jus- tify the implementation of proactive mitigation strategies and to increase the resil- ience of a locale to disasters (Rubin, 1991; Schwab et al., 1998; Mileti, 1999; Reddy, 2000; Birkland, 2006). Officials at the federal, state, and local government level not only have a significant role to play in the immediate aftermath of a disaster, but also they play a crucial part in the long-term recovery process (McEntire, 2007; National Research Council, 2011).
doi:10.1111/disa.12112
Disasters, 2015, 39(3): 427−446. © 2015 The Author(s). Disasters © Overseas Development Institute, 2015 Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 428
A pivotal issue in the long-drawn process is to conceptualise when a setting has recovered from a disaster. In other words, at any point in time, how should one measure the recovery of a place from a disaster? Although the literature on disaster recovery has grown significantly since the release of Human Systems in Extreme Envi- ronments in the mid-1970s (Mileti, Drabek, and Haas, 1975), theoretical examina- tions of disaster recovery measures remain scant (Rose, 2004; Smith and Wenger, 2007; Chang, 2010). Accordingly, local and state government officials, who are often the principal players in the post-disaster recovery process, have little guidance on how to measure recovery. This paper seeks to address the above question and to fill the gap in the literature. In particular, it compares two approaches to measuring post-disaster recovery:
• bouncing back to pre-disaster conditions; and • attaining the counterfactual state that would have existed if the disaster had not
occurred.
The counterfactual state is arrived at through the employment of a quasi-experimental design, approximated using a comparable matched location that closely resembles the affected area prior to the disaster but that was not damaged in the event. Both approaches are applied in the context of housing recovery following Hurricane Katrina, which hit the Gulf Coast of the United States in August 2005. Since the housing market collapsed nationwide soon after, housing recovery post-Katrina serves as a good case study for assessing how broader macroeconomic and external conditions can impact on recovery measures. The research focuses on housing recovery because it is vital to the long-run recovery of disaster-affected communities (Bolin, 1986; Comerio, 1998; Campanella, 2006). The approaches are applied to rural and urban areas to see if there are any systematic differences between them. Housing recovery is measured in terms of valuations and the number of units. The evaluation reveals significantly different outcomes for the two approaches: counties have not returned to their pre-disaster housing conditions, but they do exhibit counterfactual recovery when compared to matched counties. Given this finding, the paper contends that disaster recovery in general, and housing recovery in particu- lar, need to be viewed in the broader context of changes in the economic environ- ment, where realising pre-disaster conditions may not be feasible. The bouncing-back approach is based on a localised and isolated view of recovery with little or no consideration paid to other external factors that also may influence disaster recovery. By contrast, the counterfactual state approach adopts a more con- textual and broader view of recovery determinants, while controlling the general economic trend and fluctuations. Moreover, the analysis suggests that rural areas may not be as vulnerable as is assumed in the disaster literature. The remainder of the paper is structured as follows: the next section reviews the two approaches to measuring recovery. The subsequent sections set out the research design and present the results of an appraisal of post-Katrina disaster recovery patterns. The final section highlights the policy implications of the two approaches to housing recovery and proposes avenues for future research.
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 429
Two major approaches to measuring recovery The disaster literature contains two major conceptual approaches to measuring recov- ery: (i) returning to pre-disaster conditions; and (ii) attaining the counterfactual state (Chang, 2010). The first approach is conceptually based on a tangible comparison of the state of the locale before the disaster and some years after the recovery process, and it focuses on the place bouncing back as quickly as possible (Wildavsky, 1991; Sherrieb, Norris, and Galea, 2010). In this respect, the pre-disaster conditions are taken to be the normal daily life world of the context. The speedy recovery process aims to minimise interruptions to business operations, to restore damaged proper- ties, and to re-establish broken infrastructure to permit the resumption of normal activities (Alesch et al., 2001). Housing recovery is considered to be a process, whereby housing units follow a trajectory from a damaged state to a recovered state (Rathfon et al., 2012). Since merely returning to pre-disaster conditions could imply that the vulnerability conditions that existed prior to the disaster have not been reduced, the refined strands of this approach include mitigation enhancements to decrease such vulnerability (Blaikie et al., 1994; Enarson and Morrow, 1998; Mileti, 1999; Cutter and Emrich, 2006; Olshansky, 2006; Rubin, 2009). This approach is practically expedient for public officials since it has a tangible end goal, allowing them to measure recovery in relation to pre-disaster conditions. However, the approach is fraught with problems. Notably, it implies a reactive rather than a proactive stance (Manyena, 2006), and there is a tension between the speed and the quality of recovery (Olshansky, 2006). Public participation and deliberation processes, which arguably could improve quality, could be compromised for the sake of speed (Olshansky, 2006; Ganapati and Ganapati, 2009). Olshansky and Chang (2009, p. 207) present this scenario as a conflict between pre-existing and post-disaster plans:
The first plan is that of the pre-existing city. This is the plan in people’s minds, and the pieces are probably still in place: people, maps and human and economic networks. Everyone knows that this plan can work, but only if it is put back quickly while all the pieces are still close at hand. The second plan is the plan for the future. This might be a previous plan or a new recovery plan. It is the conflict between these two plans that must be resolved, and in a short time, so as not to lose the functional capabilities of the first plan and the mitiga- tion and improvement possibilities of the future plan.
The larger methodological issue with the first approach is that it ‘confounds dis- aster effects with exogenous and long-term trends that would have occurred even without the disaster’ (Chang, 2010, p. 308). That is, changes in the broader socio- economic context are not taken into account. Post-disaster recovery could be influ- enced by the broader structural changes that would have affected the area even if the disaster event had not taken place. Pre-disaster conditions may not be attained owing to fluctuations in macroeconomic and structural conditions, or pre-disaster levels may be reached because of the ‘pulling’ effects of upward general economic forces that marginally reflect local recovery dynamics (Chang, 2010).
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 430
The lesser used second approach (attaining the counterfactual state) takes into account the broader impacts of the economic environment that could affect a place. The counterfactual state is the condition of the setting if the disaster had not occurred. Unlike the first approach, there is no tangible pre-existing scenario with which to compare; rather, this is a hypothetical counterfactual comparison between the state of the setting some years after the recovery process and what could otherwise have been the case if the disaster had not happened. The hypothetical counterfactual state is approximated using a comparable location that closely resembles the affected context prior to the disaster but that suffered no damage in the event. Although the approach is well established in the regional science realm for matching geographic locations via a quasi-experimental design (Isserman and Rephann, 1995; Greenbaum and Engberg, 2004; Glasmeier and Farrigan, 2007), only recently has it been applied to disaster recovery (Xiao, 2011). The principal issue with this approach is that it requires forecasting, which ‘introduces errors, uncertainties, and inconsistencies across research- ers, which are compounded for longer-term forecasts that reach five, ten or more years after the event’ (Chang, 2010, p. 308). It is important to acknowledge here the other (more minor) means of considering disaster recovery that exist in the literature. First, some researchers have highlighted how there is a ‘new normal’ after a disaster (Alesch, Arendt, and Holly, 2009; Chang, 2010). The ‘new normal’, though, is more applicable to human attitudes and commu- nal behaviour following an event, indicating the evolution of the collective psyche (such as after the terrorist attacks of 11 September 2001), than it is to physical recov- ery (as in the case of housing recovery). Even if a ‘new normal’ is considered in the sphere of physical housing recovery, reaching the state is potentially comparable in both the first and second approaches to envisage whether or not the stable state could be considered as the recovered state. Second, some recovery indices have emerged to track progress in recovery. Finch, Cutter, and Emrich (2010) use the Social Vulnerability Index to examine the dif- ferential recovery of neighbourhoods of New Orleans, Louisiana, whereas Ward, Leitner, and Pine (2010) use a Spatial Recovery Index (SRI) to assess recovery in communities of New Orleans, Louisiana. The SRI is useful in that it adds a spatial dimension, takes account of recovery in different spheres (for instance, the economy, education, health, municipal services, and social activity), and tracks indicators such as population size, economic growth, quality of life, and the environment (Plyer and Ortiz, 2011). These recovery indices resonate with the refined view of the bouncing- back approach in as much as these dimensions are essential to understand improved conditions after a disaster. Third, innovative modes of tracking recovery exist, including repeat photography, video systems, and geospatial technology (Mills et al., 2008; Curtis et al., 2010; Burton, Mitchell, and Cutter, 2011). These modes track recovery in particular times and locations and are quite localised (Stevenson et al., 2010). In their assessment of housing recovery following Hurricane Charley in Punta Gorda, Florida, in August 2004, Rathfon et al. (2012) combine remote sensing techniques with building permits and property appraiser data to gauge the trajectory of recovery of damaged buildings.
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 431
Although the two major methods of measuring recovery have been highlighted in various works, we are not aware of any research that employs both of them to measure post-disaster recovery. This study aims to fill the gap in the literature, with a specific focus on housing recovery. Housing recovery symbolises the return of a community’s normal daily life world activities, and is one of the key dimensions of the recovery process that affects outcomes along other dimensions (Bolin, 1994; Comerio, 1998). Housing recovery is crucial to revitalising the local economy and social life (Campanella, 2006), yet systematic analysis of post-disaster housing recovery is scant (National Research Council, 2006; Zhang and Peacock, 2009). Furthermore, the existing literature on recovery concentrates primarily on urban areas (see, for exam- ple, Harrigan and Martin, 2002; Godschalk, 2003; Vale and Campanella, 2005; Campanella, 2006). In this respect, rural areas are assumed to be more vulnerable than urban areas (see, for example, Cutter, Boruff, and Shirley, 2003). Hence, this paper assesses recovery in rural and urban areas to see if there are any systematic dif- ferences between them. With regard to housing recovery, the analysis centres on permanent housing since delays in its recovery affects other aspects of recovery in a community (Peacock, Dash, and Zhang, 2007). Permanent housing is a long-range living arrangement, and is one of the four phases of housing recovery defined by Quarantelli (1982)—the others are emergency shelter (immediately following a disaster), temporary shelter (a short-term living arrangement, such as in a tent), and temporary housing (a medium-range living arrangement, such as in a trailer). It is important to note that the four phases of housing recovery are not chronologically distinct (Quarentelli, 1995; Peacock, Dash, and Zhang, 2007).
Research design The study sample includes the counties affected by Hurricane Katrina in August 2005. According to most estimates, the event wrought damage of nearly USD 200 billion, exceeding that of any other disaster in the US. The hurricane affected in excess of 90,000 square miles of landmass (Levine, Esnard, and Sapat, 2007), claimed the lives of more than 1,700 people, and damaged, destroyed, or made inaccessible some 850,000 housing units overall. Approximately 24 per cent of multi-family housing units were purportedly destroyed in Alabama, Louisiana, and Mississippi (Petterson et al., 2006). According to the Spatial Hazard Events and Losses Database for the United States (SHELDUS),1 compiled by the Hazards and Vulnerability Research Institute of the University of South Carolina, Katrina affected 288 counties (197 rural and 91 urban—as per the classification of the United States Department of Agriculture). Although other databases differ, SHELDUS offers the most comprehensive county- level data on disasters in the US and is the best available source of data on hazard- related losses in the public domain.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 432
To determine recovery following Katrina, two key indicators relating to housing recovery were examined:2
1. The value of new private housing units authorised by building permits, 2000–10 (measured in US dollars, adjusted to 1982–84 prices). The data available were composite for single- and multi-family housing. Hence, the valuation analysis is not separated for single- and multi-family housing.
2. The total number of new private housing units authorised by building permits, 2000–10 (measured as the number of housing units authorised by building permits). The data were available separately for single- and multi-family housing. Hence, the permit counts analysis is separated for single- and multi-family housing.
Building permits commonly serve as the principal indicator of local housing mar- ket strength and are well-established proxies for local housing investment (Dua, Miller, and Smyth, 1999; Hwang and Quigley, 2006; Ghent and Owayang, 2010). As the United States Census Bureau (2010, p. A-48) notes:
Current surveys indicate that construction is undertaken for all but a very small percentage of housing units authorized by building permits. A major portion typically gets under way during the month of permit issuance and most of the remainder begin within the three fol- lowing months.
However, this study acknowledges the data limitation in terms of the time lag between permits and construction in Katrina-affected counties. Furthermore, county- level differences could exist in relation to the time lag between building permit authorisation and buildings actually constructed, and the extent of hurricane damage could influence the number of permits. Although specific county-level informa- tion was unavailable, group-level comparisons cancel out idiosyncratic individual characteristics in relatively large samples. Hence, the focus here is on average recov- ery across the group. Of course, bias may not be fully detected or eliminated, par- ticularly when some of the uncontrolled idiosyncratic county characteristics are not directly observable. The aforementioned housing recovery indicators were measured using the two approaches highlighted earlier. For the first approach (recovery viewed as returning to the pre-disaster level), one-tailed t-tests were conducted to compare the housing recovery indicators for 2004 (the ‘pre-disaster’ condition) with that of each subsequent year up until 2010 (the latest year for which data were available). The t-tests verify whether the means of the groups of affected counties in the years after Katrina are statistically equivalent or different to those in 2004. Statistical difference, if any, in any subsequent year (2005–10) as compared to 2004 would indicate whether or not the affected counties have achieved pre-disaster levels with respect to each housing recovery indicator. The t-tests are carried out at the group level rather than at the individual county level to avoid the idiosyncratic unobserved characteristics that may affect the recovery process. The null hypothesis for the t-test vis-à-vis the bouncing-
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 433
back approach is that there is no statistically significant difference between the two group means of the affected counties in the pre- and post-Katrina periods—that is, H
0 : µ
pre = µpost. The alternative hypothesis is H1: µpre < µpost, which suggests that recov-
ery in the average affected county was better than the pre-disaster condition. Also tested was the other one-tailed alternative hypothesis, H
2 : µ
pre > µ
post , which implies
that housing recovery is not yet accomplished. For the second approach (recovery viewed as reaching the counterfactual state), a two-step process of quasi-experimental design was followed. In the first step, each Katrina-affected county was matched with another closely resembling county in rela- tion to a wide range of pre-selected characteristics prior to the hurricane. All US counties not affected by the event served as the choice set (in order to maximise it) for identifying the matched counties. A matched county is not required to be in the same state as a Katrina-affected county, but the two need to have the same classi- fication of metro or non-metro area as per US Office of Management and Budget guidelines (so as to capture the similarity in the rural–rural and urban–urban economy).
Table 1. Criteria used to identifying matched control counties
Income and earnings
• Income per capita, 2004. • Earnings in agriculture, forestry, fishing, and hunting (NAICS 11), 2004. • Earnings in manufacturing (NAICS 31–33), 2004. • Earnings in finance and insurance (NAICS 52), 2004. • Earnings in real estate and rental and leasing (NAICS 53), 2004.
Rural characteristics
• Metro/rural dummy variable. • Average value of land and buildings per acre, 2002. • Average farm size, 2002. • Percentage of rural resident population, 2000.
Poverty
• Poverty rate, all ages, 2004.
Employment
• Percentage of employment in agriculture, forestry, fishing, and hunting (NAICS 11), 2004. • Percentage of employment in manufacturing (NAICS 31-33), 2004. • Civilian labour force unemployment rate, 2004.
Housing characteristics
• Housing unit estimates—percentage change, 1 April 2000 (base) to 1 July 2004.
Demographic characteristics
• Net domestic migration, 2000. • Median age of resident population, 2000. • Population density, 2004. • Percentage of adult population (25+) with at least a high-school diploma, 2000. • Total resident population, 2004.
Crime
• Number of violent crimes known to police, 2004.
Source: authors.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 434
The housing recovery indicators in the matched counties are used to measure the unobserved counterfactual—what would have happened in the affected counties if Hurricane Katrina had not taken place. The selection of matched counties is crucial. Ideally, the matches should have all of the same features as the affected counties prior to Katrina. However, finding a perfect match in the real world is almost impossible. The alternative is to pinpoint an acceptable or optimal match using specific criteria. The matching criteria employed in this study included income and earnings, rural characteristics, poverty, employment, housing characteristics, demographic charac- teristics, and crime (see Table 1). Housing recovery indicators were intentionally not used as matching criteria to keep them exogenous. Optimal matches were selected using the minimal Mahalanobis distance between the pair of case and control counties. This is given by d(X
T ,X
M )=(XT-XM) -́R
-1 (X T -X
M ),
where X T and X
M are vectors of pre-selected matching criteria (indicated in Table 1)
of affected (treatment) county T and matching county M, and R is the variance– covariance matrix related with variable X. The smaller the Mahalanobis distance, the more alike are a pair of affected and matched counties. The optimal matching algo- rithm provided by Isserman and Rephann (1995) was used to select the best-case con- trol pairs that minimise the sum of the Mahalanobis distances. Table 2 presents summary statistics of matching variables of affected and matched counties. One-tailed t-tests confirm the apparent resemblance between the affected and matched counties. In the second step, one-tailed t-tests also were conducted for the years between 2004 and 2010. The base year of 2004 should indicate no statistical difference, suggesting statistical equivalence or comparability, in that the matched counties on average were indeed statistically similar to the affected counties before the disaster. Statistical dif- ferences, if any, in each subsequent year (2005–10) would suggest annual differences in housing recovery indicators between the affected and control groups, revealing whether or not the affected counties have recovered to counterfactual levels. Similar to the first approach, the t-tests were carried out at the group level to avoid individ- ual idiosyncratic effects. The null hypothesis in the second approach is that there is no statistically significant difference between the two group means of the affected and matched counties, that is, H
0 : µ
matched = µaffected, which implies that Katrina did not
have any impact on the affected counties. The one-tailed alternative hypothesis is that the group mean of the matched counties is smaller than that of the affected counties for a given recovery indicator, that is, H
1 : µ
matched < µaffected. The study also tested the
other one-tailed alternative hypothesis, that is, H 2 : µ
matched > µaffected.
In addition to the above, the interrupted time-series design was used to verify the validity of comparison between affected and counterfactual counties. The interrupted time series design establishes comparability between treatment and control groups based on growth patterns/trends a few years before the treatment is introduced (Shadish, Cook, and Campbell, 2002). Consequently, the interrupted time series design makes the quasi-experimental approach more robust since it eliminates the idiosyncratic influences of particular comparison time points on the matched comparison groups.
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 435
Table 2. Descriptive statistics of affected and matched counties
Variables County Observation Mean Standard deviation
Minimum Maximum
Income and earnings (in US dollars)
Income per capita, 2004 Affected 288 24,303 5,150 14,545 48,422
Matched 288 24,507 5,048 16,107 50,380
Earnings in agriculture, forestry, fishing, and hunting, 2004
Affected 288 4,086 12,117 0 130,605
Matched 288 2,590 5,733 0 69,373
Earnings in manufacturing, 2004
Affected 288 286,301 632,288 0 6,907,272
Matched 288 250,617 457,834 0 4,620,912
Earnings in finance and insurance, 2004
Affected 288 142,695 601,966 0 6,185,010
Matched 288 80,021 265,321 0 2,451,218
Earnings in real estate and rental and leasing, 2004
Affected 288 58,220 239,474 0 2,378,347
Matched 288 31,567 94,922 0 927,326
Rural characteristics
Metro/rural dummy variable (0 = non-metro county; 1=metro county)
Affected 288 0.32 0.47 0 1
Matched 288 0.32 0.47 0 1
Average value of land and buildings per acre, 2002 (in US dollars)
Affected 288 2,619 3,392 817 43,753
Matched 288 2,328 1,944 493 22,852
Average farm size, 2002 (acres) Affected 288 253 252 0 2,053
Matched 288 232 183 0 2,129
Rural population, 2000 (percentage)
Affected 288 0.61 0.30 0.001 1
Matched 288 0.61 0.28 0.005 1
Poverty
Poverty rate, all ages, 2004 (percentage)
Affected 288 17 6 5 36
Matched 288 16 5 4 30
Employment
Employment in agriculture, forestry, fishing, and hunting, 2004 (percent)
Affected 288 0.01 0.02 0 0.11
Matched 288 0.01 0.02 0 0.09
Employment in manufacturing, 2004 (percentage)
Affected 288 0.13 0.09 0 0.44
Matched 288 0.14 0.08 0 0.44
Civilian labour force unemploy- ment rate, 2004 (percentage)
Affected 288 6 2 3 14
Matched 288 6 2 3 11
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 436
Housing characteristics
Change in housing units, 2000–04 (percentage)
Affected 288 6 6 -1 41
Matched 288 6 6 0.3 34
Demographic characteristics
Net domestic migration, 2000 Affected 288 155 3,112 -28,719 23,621
Matched 288 390 2,129 -12,584 12,785
Median age of resident population, 2000 (years)
Affected 288 36 3 25 43
Matched 288 36 3 25 45
Population density, 2004 (population per square mile)
Affected 288 150 285 4 2,599
Matched 288 152 306 7 2,890
Share of adult population (25+) with at least a high-school diploma, 2000 (percentage)
Affected 288 71 8 53 90
Matched 288 72 8 53 97
Total resident population, 2004 Affected 288 95,600 230,508 1,898 2,337,381
Matched 288 74,707 117,880 3,204 832,806
Crime
Number of violent crimes known to police, 2004
Affected 288 533 2,013 0 24,424
Matched 288 276 679 0 8,189
Source: authors.
Table 2. Continued
In addition, the growth patterns of the affected and matched counties during 2000– 04 were compared to check equivalence between the affected and matched counties before Hurricane Katrina. If the affected and matched counties were equivalent pri- or to Katrina, a divergence in the growth trends of the affected and matched counties after the event arguably would denote recovery (or decline) in the affected counties.
Findings Tables 3 and 4 summarise the one-tailed t-test results derived using the two approaches to measure disaster recovery in all counties affected by Hurricane Katrina. The tables depict the two housing recovery indicators of valuations and number of building permits for all counties, rural counties, and urban counties from 2004–10. The inten- tion of the separation of rural and urban is to see if there are systematic differences between the two subgroups in terms of recovery. The t-tests indicate whether or not there is any statistically significant difference between the group means of the post- and pre-disaster counties within the bouncing-back approach and the group means of the affected and control counties within the counterfactual approach. All of the
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 437
t-test results for 2004 suggest that there is no statistically significant difference between the group means of the affected and matched control counties at the 10 per cent sig- nificance level, thus validating statistical comparability between the two groups and the appropriateness of the matching criteria.
Table 3. Housing recovery as returning to the pre-disaster state: a comparison of the
group means in pre- and post-disaster periods
One-tailed t test: H 0 : µ
pre = µ
post ; H
1 : µ
pre < µ
post ; H
2 : µ
pre > µ
post
Year Valuation of total new permits (USD thousands, 1982–84 prices)
New permits, total (counts)
New permits, single-family housing units (counts)
New permits, multi-family housing units (counts)
All rural and urban counties (288)
2004 56,339 803 600 185
2005 61,756* 821 619 186
2006 56,641* 754 555 182
2007 37,683† 534† 373† 146
2008 20,306† 312† 201† 99†
2009 13,088† 186† 146† 32†
2010 13,612† 186† 144† 36†
All rural counties (197)
2004 8,616 130 110 15
2005 10,057* 144* 124* 17
2006 10,416* 142 126* 12
2007 7,856 116 92† 21
2008 4,312† 72† 55† 13
2009 2,794† 45† 35† 6
2010 2,685† 41† 31† 8
All urban counties (91)
2004 159,651 2,262 1,660 555
2005 173,675* 2,286 1,692 551
2006 156,712 2,077† 1,483† 549
2007 102,255† 1,438† 983† 418
2008 54,928† 831† 518† 285†
2009 35,372† 494† 389† 89†
2010 37,267† 499† 388† 98†
Notes: * Statistically significant difference at the one-tailed 95 per cent confidence level. † Alternative hypothesis H
2 : µ
pre > µ
post is supported at the one-tailed 95 per cent confidence level.
Source: authors.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 438
The first approach to housing recovery (returning to pre-disaster conditions) re- vealed that Katrina-affected counties did not return to pre-disaster levels. As Table 3 shows, for all counties (rural and urban), the valuations of new building permits increased considerably after Katrina in the first two years (H
1 : µ
pre < µpost is statistically
significant for 2005 and 2006), but fell considerably below 2004 valuations subse- quently (H
2 : µ
pre > µ
post is statistically significant for 2007–10). The number of new
building permits (both single- and multi-family housing units) did not experience a similar rise in the first two years (H
1 : µ
pre < µpost is not statistically significant for 2005
and 2006), but there was an appreciable decline from 2004 valuations subsequently (H
2 : µ
pre > µpost is statistically significant for 2007–10, except for multi-family building
permits in 2007). The surge in valuations, but not in the number of building permits, during the first two years is interesting, as it explains the immediate rise in value and costly repairs owing to hurricane damage. The decrease after the first two years in both valuations and the number of permits arguably could indicate the waning off of initial demand, and the inability of the counties to attain pre-disaster levels in the long term. Of course, such an explanation does not capture trends in the broader national housing market. The overall pattern of housing recovery not achieving pre-disaster levels is reflected broadly in rural and urban counties as well, although the time periods for the initial surge and for subsequent dips are somewhat different. The main exception to this is that of multi-family building permits in rural areas—the numbers are not statistically different for all years (perhaps because the demand for multi-family units is much less than that for single-family units). The initial rise in valuations is significant for two years after Katrina in rural counties, but for only one year in urban counties. The number of single-family permits also increased for two years in rural counties. Statistically significant falls in valuations and in the number of single-family permits in rural counties began one year later than was the case in urban counties. The major decrease in total permits in rural counties began two years after it did in urban coun- ties. So, if at all, the dip in the two housing recovery indicators in rural areas happened somewhat later than it did in urban areas. Interestingly, the t-test results for the second approach to housing recovery (attain- ing the counterfactual state) demonstrate a distinctively different pattern. As Table 4 demonstrates, there is no statistically significant difference between the group means of the affected and the matched counties and there is very little difference between them. That is, housing recovery has occurred in as much as the indicators have reached the counterfactual level. Although valuations and the number of housing permits may have decreased from the year before Katrina in the hurricane-affected counties (as reflected in the first approach), these indicators are comparable to those of the matched counties, implying that the affected counties are on par with other similar counties in the country. The decline may not be because of the hurricane, but due to broader factors in the overall national housing market. A comparison of affected and matched counties in rural and urban areas also re- veals no statistically significant differences. In rural counties, the differences between
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 439
Table 4. Housing recovery as reaching the counterfactual state: a comparison of the
group means of affected and matched counties
One-tailed t test: H 0 : µ
matched = µ
affected ; H
1 : µ
matched < µ
affected
Year Valuation of total new permits (USD thousands, 1982–84 prices)
New permits, total (counts)
New permits, single- family housing units (counts)
New permits, multi- family housing units (counts)
All rural and urban counties (288)
Affected Matched Affected Matched Affected Matched Affected Matched
2004 56,339 45,393 803 617 600 506 185 90
2005 61,756 51,543 821 660 619 547 186 94
2006 56,641 44,332 754 560 555 455 182 85
2007 37,683 32,522 534 408 373 322 146* 73
2008 20,306 18,773 312 250 201 186 99* 56
2009 13,088 13,401 186 188 146 137 32 45
2010 13,612 14,030 186 189 144 141 36 42
All rural counties (197)
Affected Matched Affected Matched Affected Matched Affected Matched
2004 8,616 5,523 130 130 110 108 15 16
2005 10,057 5,290 144 130 124 115 17 11
2006 10,416 6,034 142 121 126 107 12 9
2007 7,856 6,378 116 114 92 97 21 13
2008 4,312 7,620 72 66 55 56 13 7
2009 2,794 8,035 45 48 35 35 6 10
2010 2,685 8,065 41 39 31 31 8 5
All urban counties (91)
Affected Matched Affected Matched Affected Matched Affected Matched
2004 159,651 127,165 2,262 1,672 1,660 1,368 555 250
2005 173,675 145,731 2,286 1,808 1,692 1,481 551 274
2006 156,712 122,845 2,077 1,509 1,483 1,209 549 250
2007 102,255 86,039 1,438* 1,046 983 811 418* 202
2008 54,928 50,147 831 647 518 467 285* 164
2009 35,372 35,896 494 491 389 361 89 119
2010 37,267 38,668 499 515 388 380 98 120
Note: * Statistically significant difference at the one-tailed five per cent level.
Source: authors.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 440
affected and matched counties are not statistically significant across the board, suggest- ing that the rural counties were similar to other matched rural counties. In urban areas, the differences are significant for one year for total permit counts, and for two years for multi-family housing permit counts. That is, there is very limited differ- ence between the affected and matched counties in urban areas, too. This finding raises the question of whether or not rural areas are indeed as vulnerable as they are assumed to be in the disaster literature. To verify if the distinction between the two approaches is robust, an interrupted times series analysis was conducted to examine the growth patterns of affected and matched counties before and after Hurricane Katrina. Figure 1 shows the valuations and the number of total permits in affected and matched counties in all rural and urban counties from 2000–10 (five years before and five years after the disaster). The affected and matched counties display a similar growth trajectory between 2000 and 2004 for all counties in terms of valuations and the number of building permits, illustrating that they were comparable. The downward trajectory between 2006 and 2010 also is similar, highlighting again the comparability of the affected and matched counties. The curves of the affected counties are higher than those of the matched counties from 2000–07 (although year by year numerical differences are not statis- tically significant), and converge from 2008–10. Thus, the affected and matched counties exhibit very similar patterns throughout the period 2000–10, for the two housing recovery indicators. In rural and urban counties, the recovery indicators maintain quite similar growth patterns before Katrina, and then decline subsequently. The only exception is for rural counties for the valuation of building permits in 2005 (see Figure 1): while the affected counties maintained growth in 2005–06, the matched counties experienced moderate or no growth; however, both declined similarly after 2006. The convergence between the affected and the matched counties in 2008–10 implies that external economic forces, such as economic recession and housing mar- ket meltdown, have been powerful enough to erase any idiosyncratic differences between the two groups. Figure 2 shows new single- and multi-family housing permits from 2000–10. The trajectories of single-family housing permits in affected and matched counties are similar overall, and converge from 2008–10. Indeed, the similarity of trajectories and convergence from 2008–10 in affected and matched counties is applicable to single- family housing permits in rural and urban areas. The trajectories of multi-family housing permits also are similar, although there are two years (2007 and 2008) when there are statistically significant differences between affected and matched counties (such a difference is not present in rural counties, though). The analysis reveals that the overall trajectories of total building permits are closely reflected in the compo- nent parts of single- and multi-family housing permits. In sum, the time-series data confirm that the impact of Hurricane Katrina on the housing market in affected counties may be very limited in comparison to general economic forces, such as the 2007 housing market collapse. The affected counties may not be able to attain pre-disaster level conditions because of such broader market forces.
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 441
Figure 1. Valuations and the number of total new private housing units authorised by
building permits, 2000–10
Note: all valuations are in 1982–84 prices.
Source: authors.
Figure 2. New private housing units authorised by building permits, 2000–10
Source: authors.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 442
The inability of the bouncing-back approach to tease out the influences of other eco- nomic factors hindering (or facilitating, in some other cases, such as economic booms) the return of affected counties to their pre-disaster levels may make this approach problematic and certainly less accurate in measuring recovery progress. The counter- factual state approach more aptly captures external broader economic forces in meas- uring recovery. The close resemblance of rural and urban post-Katrina recovery patterns may raise additional questions regarding the conventional view that rural areas are less resilient, more vulnerable, and slow to recover (see, for example, Cutter, Boruff, and Shirley, 2003).
Conclusion This paper has compared two approaches to measuring post-disaster recovery: attain- ing the pre-disaster state and the counterfactual state, as applied to measuring hous- ing recovery following Hurricane Katrina. The first approach has the tangible end goal of bouncing back to pre-disaster conditions, whereas the second approach uses a quasi-experimental design to realise the hypothetical objective of matched coun- ties. The most salient difference between them is that the latter is more capable of taking into account broader economic changes (at least as far as building permits are concerned, in the case of this study) that either hinder or facilitate a return to pre- disaster conditions. The implications of the two approaches are significantly different in the case of housing recovery in the wake of Katrina. With regard to the first approach, the affected counties, either in their totality or in rural–urban subgroups, have declined further from the pre-disaster state in terms of valuations and the number of new building permits, implying that the detrimental effects of the event might be severe and long-lasting. Continued governmental and/or non-governmental assistance and policy tools targeted at specific Katrina-affected counties might be required to battle the adverse impacts. In contrast, the second approach indicates that affected counties have maintained fairly similar growth trajectories with matched locations. There is convergence between affected and matched counties in valuations and the number of building permits. Instead of localised, isolated decline of Katrina-affected counties alone, as depicted by the first approach, the second approach suggests that affected counties have fol- lowed the general downward trend of matched counties. The impact of Hurricane Katrina on the housing market in the affected counties may be very limited as com- pared to that of general economic forces—the decline in the recovery indicators may have been because of the collapse of the housing market that happened subsequent to Katrina. Utilisation of the second approach would imply policy options with a broader geographic orientation and a more intense focus on turning the economy around. This study contributes to the disaster literature by raising the question of whether or not rural areas are indeed more vulnerable than urban areas, as assumed in existing
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 443
works. The findings show that the dip in the two housing recovery indicators in rural areas occurred somewhat later than that in urban areas (using the first approach). There is a very limited difference between the affected and the matched counties in rural and urban areas. Rural areas are not more vulnerable than urban areas, at least as far as the two indicators of housing recovery are concerned. Even though this paper concentrates on housing recovery, the analytical approach can be used to measure disaster recovery with respect to other economic and social variables, such as personal income, poverty, and unemployment. Since some eco- nomic indicators are available in finer time intervals, such as a quarterly or monthly unemployment rate, the quasi-experimental and interrupted time-series designs would be more robust in evaluating patterns in both affected and matched locations, before and after a disaster, and in measuring recovery progress. Another future research endeavour is to conduct a similar research design at the sub-county level (such as city, community, or neighbourhood) rather than at the county level. The sub-county level analysis could yield more geographically fine- grained insights into the recovery process. In addition, future research could con- sider a longer time frame for studying post-disaster recovery and focus on disaster events other than Hurricane Katrina. As indicated, disaster recovery is a very long- drawn process, lasting for many years. Lastly, although the approaches employed here provide important information on recovery, there is a need for contextual qualitative case studies with a ‘thick descrip- tion’ that can capture better complex recovery processes on the ground.
Correspondence Shaoming Cheng, 11120 SW 8th Street, Miami, FL 33199, United States. E-mail: [email protected]
Endnotes 1 See http://hvri.geog.sc.edu/SHELDUS/ (last accessed on 13 October 2014). 2 The data were compiled from the United States Census Bureau’s ‘USA Counties Data File Downloads’.
See http://www.census.gov/support/USACdataDownloads.html#HSG (last accessed on 13 Octo- ber 2014).
References Aldrich, D.P. and K. Crook (2008) ‘Strong civil society as a double-edged sword siting trailers in
post-Katrina New Orleans’. Political Research Quarterly. 61(3). pp. 379–389. Alesch, D., J.N. Holly, E. Mittler, and R. Nagy (2001) Organizations at Risk: What Happens when
Small Businesses and Not-for-Profits Encounter Natural Disasters. Public Entity Risk Institute Press, Fairfax, VA.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 444
Alesch, D., L.A. Arendt, and J.N. Holly (2009) Managing for Long-term Community Recovery in the Aftermath of Disaster. Public Entity Risk Institute Press, Fairfax, VA.
Berke, P.R., J. Kartez, and D. Wenger (1993) ‘Recovery after disaster: achieving sustainable develop- ment, mitigation, and equity’. Disasters. 17(2). pp. 93–109.
Birkland, T.A. (2006) Lessons of Disaster: Policy Change after Catastrophic Events. Georgetown University Press, Washington, DC.
Blaikie, P., T. Cannon, I. Davis, and B. Wisner (1994) At Risk: Natural Hazards, Peoples Vulnerability, and Disasters. Routledge, New York, NY.
Bolin, R. (1986) Disaster Impact and Recovery: A Comparison of Black and White Victims. International Journal of Mass Emergencies and Disasters. 4(1). pp. 35–50.
Bolin, R. (1994) Household and Community Recovery after Earthquakes. University of Colorado Press, Boulder, CO.
Burton, C., J.T. Mitchell, and S.L. Cutter (2011) ‘Evaluating post-Katrina recovery in Mississippi with repeat photography’. Disasters. 35(3). pp. 488–509.
Campanella, T.J. (2006) ‘Urban resilience and the recovery of New Orleans’. Journal of the American Planning Association. 72(2). pp. 141–146.
Chang, S.E. (2010) ‘Urban disaster recovery: a measurement framework and its application to the 1995 Kobe earthquake’. Disasters. 34(2). pp. 303–327.
Comerio, M.C. (1998) Disaster Hits Home: New Policy for Urban Housing Recovery. University of California Press, Berkeley, CA.
Curtis, A.J., J.W. Mills, T. McCarthy, A.S. Fotheringham, and W.F. Fagan (2010) ‘Space and time changes in neighborhood recovery after a disaster using a spatial video acquisition system’. In P.S. Showalter and Y. Lu (eds.) Geospatial Techniques in Urban Hazard and Disaster Analysis. Springer, New York, NY. pp. 373–392.
Cutter, S.L., B.J. Boruff, and W.L .Shirley (2003) ‘Social vulnerability to environmental hazards’. Social Science Quarterly. 84(2). pp. 242–261.
Cutter, S.L. and C.T. Emrich (2006) ‘Moral hazard, social catastrophe: the changing face of vulnera- bility along the hurricane coasts’. The ANNALS of the American Academy of Political and Social Science. 604(1). pp. 102–112.
Dua, P., S.M. Miller, and D.J. Smyth (1999) ‘Using leading indicators to forecast U.S. home sales in a Bayesian vector autoregressive framework’. The Journal of Real Estate Finance and Economics. 18(2). pp. 191–205.
Enarson, E., and B.H. Morrow (eds.) (1998) The Gendered Terrain of Disaster: Through Women’s Eyes. Praeger Publishers, Westport, CT.
Finch, C., S.L. Cutter, and C.T. Emrich (2010) ‘Disaster disparities and differential recovery in New Orleans’. Population and Environment. 31(4). pp. 179–202.
Ganapati, N.E. and S. Ganapati (2009) ‘Enabling participatory planning in post-disaster contexts: a case study of World Bank’s housing reconstruction in Turkey’. Journal of the American Planning Association. 75(1). pp. 41–59.
Ghent, A.C. and M.T. Owyang (2010) ‘Is housing the business cycle? Evidence from US cities’. Journal of Urban Economics. 67(3). pp. 336–351.
Glasmeier, A.K. and T. Farrigan (2007) ‘The economic impacts of the prison development boom on persistently poor rural places.’ International Regional Science Review. 30(3). pp. 274–299.
Godschalk, D.R. (2003) ‘Urban hazard mitigation: creating resilient cities’. Natural Hazards Review. 4(3). pp. 136–143.
Greenbaum, R.T. and J.B. Engberg (2004) ‘The impact of state enterprise zones on urban manufac- turing establishments’. Journal of Policy Analysis and Management. 23(2). pp. 315–339.
Harrigan, J. and P. Martin (2002) ‘Terrorism and the resilience of cities’. Federal Reserve Bank of New York Economic Policy Review. 8(2). pp. 97–116.
Measuring disaster recovery: bouncing back or reaching the counterfactual state? 445
Hwang, M. and J. M. Quigley (2006) ‘Economic fundamentals in local housing markets: evidence from U.S. metropolitan regions’. Journal of Regional Science. 46(3). pp. 425-453.
Isserman, A.M. and T. Rephann (1995) ‘The economic effects of the Appalachian Regional Commission: an empirical assessment of 26 years of regional development planning’. Journal of the American Planning Association. 61(3). pp. 345–364.
Levine, J., A-M. Esnard, and A. Sapat (2007) ‘Population displacement and housing dilemmas due to catastrophic disasters’. Journal of Planning Literature. 22(1). pp. 3–15.
Manyena, S.B. (2006) ‘The concept of resilience revisited’. Disasters. 30(4). pp. 434–450. McEntire, D.A. (2007) Disaster Response and Recovery. John Wiley and Sons, Hoboken, NJ. Mileti, D.S. (1999) Disasters by Design: A Reassessment of Natural Hazards in the United States. Joseph
Henry Press, Washington, DC. Mileti, D.S., T.E. Drabek, and J.E. Haas (1975) Human Systems in Extreme Environments. Institute of
Behavioural Science, University of Colorado, Boulder, CO. Mills, J.W. et al. (2008) ‘The clearinghouse concept: a model for geospatial data centralization and
dissemination in a disaster’. Disasters. 32(3). pp. 467–479. National Research Council (2011) Building Community Disaster Resilience through Private–Public Collaboration.
The National Academies Press, Washington, DC. National Research Council (2006) Facing Hazards and Disasters: Understanding Human Dimensions. The
National Academies Press, Washington, DC. Olshansky, R.B. (2006) ‘Planning after Hurricane Katrina’. Journal of the American Planning Association.
72(2). pp. 147–153. Olshansky, R.B. and S.E. Chang (2009) ‘Planning for disaster recovery: emerging research needs
and challenges’. Progress in Planning. 72(4). pp. 200–209. Peacock, W.G., N. Dash, and Y. Zhang (2007) ‘Sheltering and housing recovery following disaster’.
In H. Rodríguez, E.L. Quarantelli, and R.R. Dynes (eds.) Handbook of Disaster Research. Springer, New York, NY. pp. 258–274.
Petterson, J.S., L.D. Stanley, E. Glazier, and J. Philipp (2006) ‘A preliminary assessment of social and economic impacts associated with Hurricane Katrina’. American Anthropologist. 108(4). pp. 643–670.
Plyer, A. and E. Ortiz (2011) The New Orleans Index at Six: Measuring Greater New Orleans’ Progress toward Prosperity. Greater New Orleans Community Data Center, New Orleans, LA.
Quarantelli, E.L. (1982) Sheltering and Housing after Major Community Disasters: Case Studies and General Observations. Ohio State University, Columbus, OH.
Quarentelli, E.L. (1995) ‘Patterns of shelter and housing in U.S. disasters’. Disaster Prevention and Management. 4(3). pp. 43–53.
Rathfon, D., R. Davidson, J. Bevington, A. Vicini, and A. Hill (2012) ‘Quantitative assessment of post-disaster housing recovery: a case study of Punta Gorda, Florida, after Hurricane Charley’. Disasters. DOI: 10.1111/j.1467-7717.2012.01305.x.
Reddy, S.D. (2000) ‘Factors influencing the incorporation of hazard mitigation during recovery from disaster’. Natural Hazards. 22(2). pp. 185–201.
Rose, A. (2004) ‘Defining and measuring economic resilience to disasters’. Disaster Prevention and Management. 13(4). pp. 307–314.
Rubin, C.B. (1991) ‘Recovery from disaster’. In T.E. Drabek and G.J. Hoetmer (eds.) Emergency Management: Principles and Practice for Local Government. International City/County Management Association, Washington, DC. pp. 224–262.
Rubin, C.B. (2009) ‘Long term recovery from disasters—The neglected component of emergency management’. Journal of Homeland Security and Emergency Management. 6(1). pp. 1–17.
Schwab, J., K.C. Topping, C.D. Eadie, R.E. Deyle, and R.A. Smith (1998) Planning for Post-Disaster Recovery and Reconstruction. PAS Report 483/484. American Planning Association, Chicago, IL.
Shaoming Cheng, Emel Ganapati, and Sukumar Ganapati 446
Shadish, W.R., T.D Cook, and D. T. Campbell. (2002). Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin Company, New York, NY.
Sherrieb, K., F.H. Norris, and S. Galea (2010) ‘Measuring capacities for community resilience’. Social Indicators Research. 99(2). pp. 227–247.
Smith, G. and D. Wenger (2007) ‘Sustainable disaster recovery: operationalizing an existing agenda’. In R. Havidán, E.L. Quarantelli, and R.R. Dynes (eds.) Handbook of Disaster Research. Springer, New York, NY. pp. 234–257.
Stevenson, J.R., C.T. Emrich, J.T. Mitchell, and S.L. Cutter (2010) ‘Using building permits to monitor disaster recovery: a spatio-temporal case study of coastal Mississippi following Hurricane Katrina’. Cartography and Geographic Information Science. 37(1). pp. 57–68.
United States Census Bureau (2010) State and Metropolitan Area Data Book. Seventh edition. United States Census Bureau, Washington, DC.
Vale, L. and T.J. Campanella (eds.) (2005) The Resilient City: How Modern Cities Recover from Disaster. Oxford University Press, New York, NY.
Ward, S.M., M. Leitner, and J. Pine (2010) ‘Investigating recovery patterns in post disaster urban settings: utilizing geospatial technology to understand post-Hurricane Katrina recovery in New Orleans, Louisiana’. In P.S. Showalter and Y. Lu (eds.) Geospatial Techniques in Urban Hazard and Disaster Analysis. Springer, New York, NY. pp. 355–372.
Wildavsky, A. (1991) Searching for Safety. Transaction Publishers, New Brunswick, NJ. Xiao, Y. (2011) ‘Local economic impacts of natural disasters’. Journal of Regional Science. 51(4). pp. 804–820. Zhang, Y. and W.G. Peacock (2009) ‘Planning for housing recovery? Lessons learned from Hurricane
Andrew’. Journal of the American Planning Association. 76(1). pp. 5–24.
Copyright of Disasters is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
