Going home: evacuation-migration decisions of hurricane Katrina survivors.

Landry, Craig E. ; Bin, Okmyung ; Hindsley, Paul 等

1. Introduction

Upward of one million residents of the greater metropolitan New Orleans area evacuated on 27 and 28 August 2005, just before Hurricane Katrina struck the Gulf Coast. Evacuees from other parts of Louisiana, Mississippi, and Alabama fled the coast in large numbers, marking Hurricane Katrina as the largest population displacement in the United States since the Dust Bowl of the 1930s (Falk, Hunt, and Hunt 2006). Postdisaster recovery and rebuilding in the Gulf region requires understanding the existing risks, communicating those risks to the public, rethinking land use, deciding on methods to correct deficiencies in public infrastructure, and providing incentives for economic recovery that will give firms and households an opportunity to survive and thrive. In the case of New Orleans, recovery could take up to 11 years or more (Kates et al. 2006). Although many issues remain to be resolved in determining what will become of New Orleans and the Gulf region, the economic, social, and cultural future of the Gulf region will be significantly influenced by who decides to return. In the face of variable but widespread destruction, salient vulnerability, and uncertain prospects, evacuees must choose whether to return to their homes.

As Katrina approached, Alabama, Mississippi, and Louisiana all issued mandatory evacuation orders. In New Orleans, 70,000 people remained, some by choice, but most without means of escape (U.S. Congress 2006b). Many evacuees who sought refuge from Katrina had nowhere to return to after the storm. Immediately after the storm, roughly 275,000 people were forced into group shelters (FEMA 2006a). Between mid-August and mid-November 2005, 250,000 people lost their jobs (U.S. Congress 2006a). Without homes or jobs, many people were forced to decide whether to restock and rebuild their lives along the Gulf coast or to seek out a new location for residence. The National Hurricane Service estimated the total damage losses from Katrina at $81.2 billion (NWS 2006). In the 117 hurricane-affected counties of the Gulf Coast, 40 declined in population between July 1, 2005, and January 1, 2006 (Frey and Singer 2006). The greatest population losses occurred in the parishes and counties holding New Orleans, Louisiana; Gulfport-Biloxi, Mississippi; Lake Charles, Louisiana; Pascagoula, Mississippi; and Mobile, Alabama.

In this paper, we examine the decision to return to the postdisaster Gulf region which we call the "return migration" decision. We review economic models of household migration and build on historical and empirical evidence of migration behavior to postulate on determinants of postdisaster return migration. We identify important research questions that can be examined with return migration data. We explore stated preferred return migration behavior using a number of data sets collected in the wake of Hurricane Katrina and make some inferences about socioeconomic determinants and effects of the return migration decision.

2. Economic Models of Household Migration

Economists have long recognized that economic factors influence the migration patterns of households. Sjaastad (1962) provides a theoretical framework for the decision to migrate, defining the problem in terms of a household's search to maximize the net economic return on human capital. In this framework, migration is viewed as an equilibrating force in the labor market--real wage differences between regions or cities create arbitrage opportunities that can be realized by migration, leading to a redistribution of households across the landscape. Early models focused on interspatial wage differentials, distance between origin and destination, labor market conditions (such as unemployment rate and growth in employment), and household characteristics as factors determining migration flows (Greenwood 1975; Graves 1979, 1980; Greenwood and Hunt 1989).

Models of household migration typically employ a modified gravity modeling structure. Migration flows are assumed to be proportional to origin and destination populations, but inversely related to distance. It has been well documented that migration rates decline with distance, although it is generally believed that out-of-pocket monetary expenses could not alone explain this phenomenon. Moving expenses tend to be a relatively small part of the net returns to migrating. Other explanations include opportunity costs of time, psychic costs of moving (diminution of contact with family and friends, change of environment, etc.), higher search costs associated with greater distances, and uncertainty about destinations (Greenwood 1997). The existence of these potential barriers to migration has created concern about the efficacy of migration in reallocating resources in response to changing market and demographic conditions.

Migration decisions vary across individual households. Economic factors such as worker skills and employment status will influence returns to migration. Life cycle considerations and the availability of information could also influence migration. One would expect some correspondence between migration and changes in life stages--for example, children moving away from home, the completion of school by a family member, marriage, divorce, retirement, etc. Expectations of obtaining gainful employment depend on flow of information of employment opportunities, which might explain why previous-period net migration rates are positively correlated with current migration trends (Greenwood 1969). Social networks could play a role in learning about labor market opportunities and providing support for migration. Especially among race-ethnic minority groups, research suggests that migration patterns tend to follow well-worn pathways and networks (Bean and Tienda 1987; Farley and Allen 1987; Barringer, Gardner, and Levin 1993).

Individuals might also be influenced through learning about amenities in different locations. Sjaastad (1962) considered location-specific amenities (including climate, smog, and congestion) as factors that might affect returns to migration, but characterized them as unimportant in evaluating migration as a redistributive mechanism because they entail no resource cost. This notion does suggest, however, that location-specific amenities might affect the reservation wage of households and, thus, that wage schedules could be conditional on amenity levels. A subsequent branch of literature adopted this perspective, assuming that wages, rents, and the prices of locally produced nontraded goods adjust in response to location-specific exogenous factors, such as local environmental conditions or fiscal considerations, so that utility and profit levels (rather than wages and land rents) are equalized across regions. Under this characterization, persistent differences in wages and rents compensate for amenity levels; they need not equalize across regions or cities in the long run unless the locations have identical amenities.

Roback (1982) shows how wages and land rents are simultaneously determined in an equilibrium setting, conditional on the level of local amenities. In this context, amenities are nonmanufactured attributes that are valued by households--such as temperature, rainfall, and cleanliness of environment--or goods and services that vary in availability spatially--such as professional sports teams, performing arts, cultural resources (i.e., museums), etc. In Roback's model, interregional wages and rent differentials can persist and will reflect the value of location-specific amenities. This formulation of household migration follows the hedonic model formalized by Rosen (1974), in the sense that implicit values of location-specific amenities are reflected in the markets for labor, land, and other locally produced goods and services.

Clark and Cosgrove (1991) examined the persistency of interregional wage differentials. They found evidence that supports both the human capital approach of Sjaastad and the compensating differentials model of Roback. Amenities tend to have a significant negative effect on wages, but wage differentials persist across regions, even when amenities are controlled. Greenwood et al. (1991) provide evidence of disequilibrium in U.S. internal migration between states--real income in amenity-rich states tends to be too high and real income in amenity-poor areas tends to be too low.

Frey and Liaw (2005) identify cultural constraints--such as the need for social support networks, kinship ties, and access to informal employment opportunities--as shaping the migration patterns of race-ethnicity groups. Empirical evidence suggests that minority residence in an ethnically concentrated metropolitan area can inhibit out-migration (Tienda and Wilson 1992). Thus, persistent differentials could reflect cultural constraints in a number of ways: race-ethnic groups might traverse well-worn migration routes with less attention paid to wage differentials at other possible destinations, or connections to place (1) might inhibit out-migration. The implications of this line of reasoning are that migration might not engender complete efficiency in the allocation of labor across space because social and personal constraints could inhibit labor flow. Greenwood et al. (1991) suggest that persistent wage differentials are relatively small, so that efficiency loss could be minor. However, exploration and inference about social connections is something that, to our knowledge, has not been explored. Such an analysis is best pursued with microlevel data.

3. Examining Return Migration

A number of papers have looked at the decision to evacuate before hurricane landfall (Baker 1991; Dow and Cutter 1997; Gladwin and Peacock 1997; Whitehead et al. 2000; Whitehead 2005). Results generally suggest that storm intensity, evacuation orders, perception of flood risk, type of residence, pet ownership, and race/ethnicity influence the likelihood of evacuation. Whitehead (2005) finds some evidence to support the validity of stated evacuation preference data.

Postdisaster migration has been much less researched. A disaster large enough to cause widespread displacement of a population will often cause extensive damage to personal property and infrastructure, limiting the ability of evacuees to return to their homes, businesses, and communities. Depending on the severity of the disaster, return access could be limited for weeks or months. Uncertainty about the timing and composition of return migration can hamper the recovery process because many economic, civic, and social functions are largely population dependent. (2) The nature of return migration also affects reconstruction, in that project prioritization and infrastructure capacity depend on the returning population.

Elliott and Pais (2006) examine evacuation, short-term recovery, emotional stress and support, and likelihood of return for Gulf coast residents in the wake of Hurricane Katrina. They find a high degree of uncertainty regarding the likelihood of return for those households still displaced one month after the storm. They find homeowners are more likely to return than those that do not own property. However, those whose homes were destroyed by the storm are less likely to return. They also find that lower-income households are more likely to return. Falk, Hunt, and Hunt (2006) argue that affluent households should be more likely to return postdisaster because they are likely to be displaced to closer locations and they have better resources to make the return trip. In the case of flooding disasters, affluent households are more likely to own homes in areas less likely to have been flooded and have better resources to rebuild in the event that their home has been damaged. Note that the results of Elliott and Pais (2006) correspond with households that had not returned one month after the disaster. Thus, they are conditioned on their sample selection--those households that did not immediately return. As such, the conjecture of Falk, Hunt, and Hunt (2006) might apply to the general population of evacuees.

Elliott and Pais (2006) also consider the effect of race, gender, age, timing of evacuation, whether the respondents are parents, and employment status on the likelihood of return. They find no statistical support for the significance of these covariates in the return migration decision. Falk, Hunt, and Hunt (2006) speculate on the importance of sense of place as a factor affecting the likelihood of return. They note that sense of place is likely to increase in strength when families or communities exist in an area for an extended period of time, perhaps over a number of generations. Sense of place could keep households in an area through bad times--such as loss of job, economic recession, social turmoil, or natural disaster--even when moving elsewhere could afford better opportunity. As such, sense of place might play a role in persistent wage and land rent differentials identified in the economic migration literature. This notion is related to the psychic costs of moving identified by Sjaastad (1962). Sense of place and a desire to rekindle community and social connections could affect the likelihood of return.

Population displacement because of natural disaster offers an opportunity to examine the importance of sense of place in migration decisions. Displacement creates an exogenous shock that uproots households that might have never chosen to leave their current location, despite differences in wages, prices, or amenities in other areas. How do those households then respond given the current opportunities for employment and quality of life in their displaced location and their connection to the place from which they vacated? This choice likely depends on sense of place and connection to culture. With the right kind of data, one could examine the importance of culture and sense of place in the return migration decision and, by examining contingent wages in the displaced and home locations, could possibly get a sense of the compensating real wage differentials that would affect migration despite connection to place.

Postdisaster perceptions could also affect the likelihood of return. Natural disasters can expose shortcomings of certain locations or the way humans have developed the landscape leading to changing perceptions of vulnerability. Those that perceive areas in which they previously lived as suddenly more vulnerable would be less likely to return. Likewise, mistrust of government to provide risk management and handle emergency services could also influence return migration to high-hazard areas. Finally, expectations of housing and job availability, as well as overall economic outlook, could affect return migration. In the next section, we develop an econometric model of the likelihood of postdisaster return that takes these aspects into account.

4. Return Migration Decision

Consider the return migration decision of a household that has recently evacuated before a natural disaster. We consider this household displaced if they cannot immediately return to their home after the occurrence of the disaster. Inability to return could reflect damage to their home or community, loss of critical infrastructure (such as roads, power, or flood protection), distance traveled for evacuation, uncertainty related to habitability of their home or continuation of employment, or some combination of these factors. We assume household decision making adheres to the tenets of rational choice; thus, the decision to return postdisaster reflects a weighing of benefits (B) and costs (C). Thus, the probability of return is:

Pr(return = 1) = Pr(B < C), (1)

where return is a dummy variable indicating intention to return; B reflects connection to place, perceptions of vulnerability, damage to home and community, likelihood of reengaging in employment, and the likelihood of friends and family returning; and C reflects distance evacuated and wage differentials in the home and host cities. The C vector might also include differences in prices and amenities in the home and host cities.

Thus, quality of life factors and home-specific factors, such as connections to place and individual perceptions and expectations of future conditions, should play a role in the decision to return. Under the assumption that evacuees can find a job in their host and home cities, a cost of returning home is the change in real wages associated with the return. With persistent interregional wage differentials, the loss in real wages stemming from return migration could be significant. On the other hand, wages in the host region could be less than that of the home region, so the wage differential would be a negative cost. The wage differential will reflect economic conditions in the home and host city and labor characteristics of the household.

The household return migration decision has implications for the economic and social recovery of the region affected by natural disaster. The pool of labor that returns (e.g., skilled vs. unskilled) could affect economic activity and industry performance. Although we would expect market adjustments to equilibrate demand and supply of labor over time, shortages or gluts of specific types of labor could cause short-term problems in recovery. The availability of housing could exacerbate labor problems--if unskilled labor tends to rent housing and rental properties are neglected in early recovery efforts, then the return rate of unskilled labor could be relatively low. This could be a problem for New Orleans because the tourism-based economy of the city relies heavily on unskilled labor (Falk, Hunt, and Hunt 2006). Demographics of returning households have implications for the public and private sectors of the economies-are families with school-age children likely to return? How should local school districts plan for their return?

The return migration decision can also be explored from the standpoint of nonmarket valuation. Consider the economic value of returning home, maximum willingness to pay (WTP), with [WTP.sub.i] = [x.sub.i]'[beta] + [[epsilon].sub.i], where [x.sub.i] is a vector of household characteristics and [[epsilon].sub.i] is an independent and identically distributed logistic random error term with zero mean. The conditional probability of return can be rewritten

Pr(return = 1|[x.sub.i]) = Pr[[WTP.sub.i]([x.sub.i], [[epsilon].sub.i]) > C]. (3) (2)

Consider the real wage differential as the primary cost of return: [C.sub.i] = [W.sub.host] - [W.sub.home]. Ignoring other potential costs, (4) we have

Pr(return = 1|[x.sub.i]) = Pr([x'.sub.i][beta] + [[epsilon].sub.i] > [C.sub.i])

= Pr([[epsilon].sub.i] > [C.sub.i] - [x'.sub.i][beta])

= Pr([x'.sub.i][beta] - [C.sub.i] > [[epsilon].sub.i])

= Pr(([x'.sub.i][beta] - [C.sub.i])/[theta] > [z.sub.i]), (3)

where [z.sub.i] is a standard logistic random variate and [theta] = [[sigma].sup.2][[pi].sup.2]/3. (5) As recognized by Cameron and James (1987), this formulation of a dichotomous choice model allows for identification of point estimates of [beta] and calculating fitted values of [WTP.sub.i] because the scale parameter is identified due to the inclusion of a random cost parameter. The parameter estimate on C from the logistic regression is a point estimate of -1/[theta], so [beta] in Equation 3 can be recovered through a simple transformation. In our case, the evacuation location must be exogenously imposed on the household to render [w.sub.host] a random wage offer and, thus, [C.sub.i] exogenous to the household. The expected benefit of return home for the average household is calculated as

WTP = - [bar.x]'[beta]/[[beta].sub.C] (4)

where [bar.x] is a vector of average household characteristics and [[beta].sub.C] is the parameter estimate of the wage difference. (6) Confidence intervals for WTP can be calculated with the Krinsky-Robb Monte Carlo procedure (Krinsky and Robb 1986).

5. Empirical Analysis

The eye of Hurricane Katrina made landfall in southeast Louisiana at 6:10 a.m. on August 29, 2005. At landfall, Katrina had maximum winds of 125 mph, making it the third most intense hurricane on the U.S. record (NWS 2006). Hurricane Katrina devastated the Gulf coast. The National Weather Service (2006) reports that in Mississippi, the storm surge reached 28 feet in certain locations. In Louisiana and Alabama, the storm surge arrived at well above 10 feet. Along the Mississippi coast, the storm surge penetrated at least 6 miles, where preliminary estimates indicated 90% of structures within a half a mile of the coast were destroyed (CBS 2005, NWS 2006). In New Orleans, levee breaches flooded 80% of the city. In all, Hurricane Katrina affected roughly 90,000 square miles (FEMA 2006b).

In response to Hurricane Katrina, the Center for Natural Hazards Research at East Carolina University conducted two separate surveys, each containing questions relevant to the evacuation behavior of individuals living within the affected areas. (7) The two surveys were both random samples of individuals in the affected region, as defined by U.S. Postal Service. (8) In both cases, we used a modified Dillman approach consisting of initial postcards indicating an upcoming survey and multiple waves of mailed surveys and follow-up postcards. We used first class postage to ensure that the U.S. Postal Service would send our postcards and surveys to the household's forwarding address and requested return service so that we could keep track of those households that could not be reached via mail. Survey 1, which focused on the expenditure patterns of evacuees, had two waves of mailed surveys and survey 2, which focused on opinions of and preferences for rebuilding projects in New Orleans, consisted of three waves of mailed surveys. Survey 2 also included additional phone contact to encourage participation. In survey 1, our final targeted sample totaled 2474 individuals within the affected region. Of these 2474 individuals, 597 returned surveys--a 24% response rate. Survey 2 targeted 3532 individuals, of which 730 returned survey--a 21% response rate. Surveys 1 and 2 were then combined to produce the first set of estimates (Mail Survey in Table 1).

The second set of estimates uses data collected by researchers at Rice University. (9) This survey targeted Katrina evacuees in Houston, Texas, and consists of three waves of self-administered questionnaires over a one-year period. The first wave focused on individuals located in evacuation shelters throughout Houston in early September 2005. The second wave occurred in late October through early November of 2005 in motels and apartment complexes in the city. The third wave occurred in July 2006 in apartment complexes. In all, we used 756 observations between the three waves of data. Wilson and Stein (2006) compared descriptive statistics for each wave to other surveys investigating Katrina evacuees in Houston. For a detailed description of the survey methodology, see Wilson and Stein (2006).

We used logistic regressions to analyze evacuees' stated preference decision to return to their predisaster residence after Hurricane Katrina. It is assumed that the probability of return depends on a set of individual and household characteristics according to a logistic cumulative distribution function as follows:

Pr(return = 1) = [LAMBDA](x'[beta]) = exp(x'[beta])/1 + exp(x'[beta])], (5)

where Pr(return = 1) is the probability that an evacuee returns to the pre-Katrina residence given a vector of individual and household characteristics x, and [LAMBDA] represents the logistic cumulative distribution function. The [beta] parameters are estimated by the maximum likelihood method.

The vector x varies across our data sets, but in general includes income level of the household, labor characteristics of the household, indicators of cultural and social connection to the previous place of residence, and demographic characteristics. For the entire population of Hurricane Katrina evacuees, we expect that income will have a positive effect on the likelihood of returning, reflecting access to financial resources to aid in return and recovery. Important labor characteristics could include work history and experience, such as whether members of the household are currently employed and whether they were employed before the disaster. Household social and cultural connection indicators could include length of residence at the home location, intergenerational connections to the home area, and membership in a race-ethnic group that has special significance in the home area. Demographic characteristics that might affect the return migration decision include age, education, marital status, and household size. Finally, the real wage differential ([C.sub.i]) for the household's skill level and job classification associated with the home and host locations could be included in the specification of Equation 5.

Unlike the linear regression model, the parameter estimates for the logit model are interpreted as the rate of change in the log odds of return as the characteristics change, which is not very intuitive. Therefore, the marginal effects of the individual and household characteristics on the probability of return are also calculated, as follows (Greene 2003):

[partial derivative]Pr/[partial derivative][x.sub.i] = [LAMBDA]([x'.sub.i][beta])[1 - [LAMBDA]([x'.sub.i][beta])][beta]. (6)

The marginal effects are evaluated at the observed mean values, which are reported in Table 2. For dummy variables, marginal effects are computed with the use of the change in the probabilities.

Table 2 reveals striking differences across our two samples. The mail sample corresponds with a higher income, a more highly educated, and an older population. This population also has fewer African Americans than the Houston sample. (10) Almost a third of the mail sample lived in the New Orleans metropolitan area before Hurricane Katrina, whereas the Houston sample is predominantly composed of evacuees from New Orleans (92%). Six percent of mail survey respondents claimed to have Acadian (or "Cajun") heritage. For the subset of mail data for which we had measures of social connection (survey 2), 35% of respondents report that they were born in the parish or county in which they lived before Hurricane Katrina. We construe this as a proxy for connection to place. Sixty-five percent of the Houston sample was engaged in the labor force before Hurricane Katrina. A small proportion, 13%, owned their own home, and the average respondent had lived in the New Orleans area (or some other part of the affected region) for 26 years. Intentions to return across the two populations are significantly different--88% for the mail survey versus 29% for the Houston survey.

We report two sets of estimation results: the first on the basis of the mail surveys conducted by the Center for Natural Hazards Research at East Carolina University and the second on the basis of self-administered questionnaires of Katrina evacuees living in Houston. Table 1 reports the logistic regression estimation results for the mail data. The explanatory variables in the estimated model are jointly significant ([chi square] = 92.15). Results indicate that household income before Katrina, whether the residence was located in the New Orleans metropolitan area, whether the respondent is a senior citizen, and whether the respondent was born in the parish/county in which they lived before the storm, have a statistically significant influence on the evacuee's return decision. The coefficient of household income is positive, indicating that higher income households are more likely to return to their pre-Katrina residence, but the influence diminishes with income (negative quadratic term).

Controlling for the percentage of damage in a county, residents of the New Orleans metropolitan area are less likely to return home, all else being equal. New Orleans residents are 7% less likely to return. Senior citizens are almost 5% less likely to return. The parish-born parameter estimate is negative, indicating that those respondents that were born in the parish or county in which they lived before Katrina are less likely to return. This result is counter to our expectations because we envisioned this covariate as an indicator of social connection to place, which would lead us to expect a positive coefficient. In any event, the marginal effect is not statistically significant. Finally, the economic impact data set (survey 1) exhibited a higher likelihood of return. Unfortunately, because of missing and inconsistent data, we were not able to record wage differentials corresponding with the home and host region for the mail sample.

Table 1 also reports the estimation results for the Houston data set. Results indicate that education level, age, employment status, marital status, and home ownership influence the likelihood of return. Respondents with at least a college level education and those under the age of 30 are less likely to return. Respondents that were working before Katrina are more likely to return home, as are married respondents. Home ownership has a significant influence on the likelihood of return, increasing the probability by 21%. (11)

The Houston model also includes the wage differential. For this data set, the real wage differential ([WD.sub.j]) for j labor classification is defined as

[WD.sub.j] = [W.sup.Host.sub.j] - [CPI.sup.Host]/[CPI.sup.Home] [W.sup.Home.sub.j], (7)

where [W.sup.Host.sub.j] and [W.sup.Home.sub.j] denote an hourly mean wage in Houston and the home location (primarily New Orleans), respectively, for j labor classification in May 2005, and [CPI.sup.Host] and [CPI.sup.Home] denote the consumer price index for Houston and the home location, respectively, as of May 2005. (12) The average real wage differential was $1.55 per hour, indicating that, on average, households in the Houston sample could earn more money by staying in the Houston area. The coefficient on wage differential is negative and statistically significant. A $1.00 increase in the wage differential decreases the likelihood of return by almost 6%. We use Equation 4 to calculate average WTP to return home. Our point estimate is $1.94 per hour (2005 U.S. dollars) with a 95% confidence interval of $1.79 and $2.30 (Krinsky and Robb 1986). The dummy variable indicating the New Orleans metropolitan area captures site-specific amenity affects of return migration for the majority of the sample vis-a-vis other Gulf locations. Beyond this dichotomy for the return location, our WTP measure assumes no intrasite variation in location-specific amenities (i.e., homogeneity of amenities within the Houston and home sites) and homogeneity of amenity perceptions within the population of interest. (13)

6. Discussion

Our results provide insight into the return migration decision of households that have been displaced by natural disaster. The displacement of people can have major social, psychological, and economic implications. Researchers have examined the evacuation decision, the effect that evacuees have on their host region, and the social and psychological effects of the disaster and displacement on evacuees. There has been much less research (14) on an important aspect of recovery--which households will subsequently return and why? Our sense is that many have assumed in the past that all or most evacuees will return, but this is not necessarily so, especially for large disasters that cause mass destruction and highlight the vulnerability of a particular area. Damage from the disaster, perception of vulnerability of the home community, expectation of economic conditions, the behavior of family and friends, and connection to place could all influence the likelihood of return. The magnitude and composition of the returning population has implications for disaster recovery.

We postulate a simple benefit-cost structure on the return decision in order to conduct empirical analysis of two unique data sets. The first corresponds with evacuees from the Gulf region that responded to one of two mail surveys. Although the mail surveys were designed primarily for other purposes (to measure evacuation behavior and expenditures in one case and opinions of rebuilding project in the other), we are able to assess the respondent's intentions of returning to their home after evacuation. The adjusted overall response rate to these two surveys is approximately 22%. We make no claim that this sample is representative of households in the Gulf region. Nonetheless, we can assess what influences the likelihood of return to learn something about the decision-making process.

Our results suggest that household income influences the likelihood of return, although the marginal effect is rather small--a $1000 increase in household income increases the likelihood of return by 0.3%. Residents of metropolitan New Orleans are 7% less likely to return home. The metropolitan area includes counties most heavily damaged by Katrina; however, estimates suggest that the percentage of houses with damage does not significantly affect overall likelihood of return. Given the nonuniform damage distributions within a county, the county-level aggregation in this covariate could be a source of inaccuracy. A particularly vulnerable group, senior citizens, are less likely to return to their home (marginal effect = -5%). This result could reflect heightened perceptions of vulnerability in this population.

We were surprised to find that individuals born in the parish or county in which they lived before Katrina were less likely to return, although

the marginal effect for this variable was not statistically significant. We hypothesized that sense of place would be stronger among these individuals, and thus likelihood of return would be greater, but the data do not support this contention. Indicator variables for parents being born in the county (or a nearby county) in which the individual lived proved to have no influence on the pattern of return migration responses. Moreover, those that consider themselves Cajun (Acadian) are no more likely to return than other respondents. Research suggests a possible explanation for this finding the extent of damage tends to cause more distress to people with deep roots in a particular environment (Albrecht 2006). Thus, those with deeper connections to place might be more highly traumatized, leading to a lower likelihood of return.

Our results for the mail sample differ somewhat from those of Elliott and Pais (2006). They examined the return migration decision with interval-scaled data and ordinary least squares, finding that only household income, home ownership, and whether the respondent's home was destroyed influenced the return migration decision. However, they found that income has a negative effect on the likelihood of return, as does loss of home, whereas home ownership has a positive effect. They found no influence of age or place of residence (New Orleans vs. other Gulf Coast communities) on return migration. African Americans were no more or less likely to return in their model; we find similar results with regard to race. Our second data set corresponds with primarily minority Katrina evacuees in Houston. Our logistic regression results suggest that education, age, employment status, marital status, and home ownership influence the likelihood of return. Respondents with at least a college level education are 7% less likely to return home than are the less educated. Those under the age of 30 are 11% less likely to return. Respondents that were working before Katrina are 9% more likely to return home than those that were not working, and married respondents are 11% more likely to return home. Similar to Elliott and Pais (2006), home ownership has a large influence on the likelihood of return, increasing the probability by 21%. Household income has no effect on the likelihood of return for this sample, nor does the number of years that the respondent lived in the area before evacuation. The former coefficient likely reflects the low variability of income in the Houston data; the latter covariate was included as a proxy for connection to place, and again we find little support for this aspect influencing the likelihood of return. Neither of our models finds that the extent of damage in a county influences return migration, but there could be error in this variable (as noted above).

Returning briefly to Table 2, we note that major differences between the two subsamples are along age, racial, and household income lines. In particular, the mail sample exhibits a much higher annual income ($51,000 compared with $19,000), tends to be older, and is more heavily non-African American. It would be remiss were we not to point to these qualitative differences and their possible implications, as intentions for return migration vary significantly across the two subsamples--89% for the mail sample and only 29% for the Houston sample. Although the income effects are small or insignificant in the parametric regression models, we do see a clear pattern in the summary statistics across the two subsamples. A liberal reading of this pattern of results might suggest that wealthier, older, non African American households are more likely to return to the post-Katrina Gulf region.

With the Houston data set, we examine not only the influence of household characteristics, but also individual-specific wage differentials. The economic literature on migration has long recognized that labor market conditions influence migration patterns, as do the prices of location-specific goods and the levels of spatial amenities. In a world of homogeneous agents with perfect information, with no connection to place, and in which moving was costless and could be instantaneously realized, the equilibrium levels of wages and rents should adjust to reflect the value of location-specific amenities (Roback 1982). Under these conditions, utility levels of consumers and profits of firms would be equalized across space. Wages would be higher and land rents lower in areas with poor amenities, whereas amenity-rich locations would pay lower wages and witness higher land rents.

A number of migration studies have found persistent differentials in wages across regions while controlling for amenities (Clark and Cosgrove 1991; Greenwood et al. 1991). Cultural constraints are one factor that could foster persistent wage differentials (Frey and Liaw 2005). Individual need for social support networks, kinship ties, and access to informal employment opportunities could influence migration patterns. Information flows are influenced by social networks, which could inhibit or distort knowledge of prices, wages, and amenities at other locations. Connection to place in which an individual has lived could also inhibit outmigration.

We include a number of proxies for connection to place (which for our purposes could relate to sense of identity, kinship ties, social networks, or other cultural constraints) in our regression models. We find little influence of these factors on the likelihood of return. These results could reflect the unimportance of place in the return migration decision, the poor quality of our proxies, or misspecification of the place phenomenon in our regression models. Nonetheless, we are able to make inferences about the value of returning home with the use of individual-specific wage differentials for the Houston sample.

Real wage differentials are the differences in hourly earnings at home and host locations for a respondent's job class, controlling for home and host region price levels. The average (median) real wage differential is $1.55 ($0.71) per hour, ranging from -$5.74 to $12.78. Less than 5% of the wage differentials were negative, implying that Houston offers higher real wages for the overwhelming majority of the evacuees. Although we are unable to control for amenity levels across the home and host region, we do find the expected negative effect of wage differential on the likelihood of return. Because a larger wage difference implies that the individual faces higher opportunity cost of return, we interpret the wage differential as an implicit price of return. It is an estimate of the amount of hourly income that they must give up to return home.

Our willingness to pay model in Equations 2-4 formalizes the relationship between the economic benefit of returning home and the cost implied by the wage differential. The Houston data suggest that some evacuees choose to return home even though they could earn a higher wage at their host location. In this sense, Hurricane Katrina provides a natural experiment for analyzing migration decisions. Individuals that might have never left their home are suddenly presented with the opportunity to migrate by making their evacuation decision permanent. The natural disaster provides an exogenous shock to the spatial pattern of labor that could allow one to assess the underlying causes of persistent wage differentials.

We employ the WTP formula in Equation 4 to estimate the benefit of returning home. Our results suggest that the average individual is willing to sacrifice $1.94 an hour in higher wages to return home, with a 95% confidence interval of $1.79 and $2.30 (2005 U.S. dollars). For an individual employed full time, this implies an annual willingness to pay of $3954 (95% confidence interval $3651-$4692). Although connection to place as we have defined it might not be the factor motivating return migration, the data suggest that something draws individuals to return home in the face of real and significant economic cost.

7. Conclusion

Natural disasters can unleash widespread death and destruction, displace hundreds of thousands of people, and cause major interruptions in the everyday economic life of still greater populations. Economists have examined evacuation, recovery, and transition but have not looked at the microeconomic decision of displaced households to return home. We explore the evacuation-migration decisions of Hurricane Katrina survivors with two unique data sets that include stated preferences on return migration. For a sample of evacuees in various locations, we find that household income increases the likelihood of returning home. This result is in line with our expectations, in that households with higher income have better resources to make the return trip, are more likely to own homes in areas less likely to have been flooded, and have better resources to rebuild in the event that their home has been damaged. However, this result differs from the only other empirical analysis that we are aware of, which finds a negative relationship between income and likelihood of return (Elliott and Pais 2006). Senior citizens and residents of metropolitan New Orleans are less likely to return home. Percentage of damaged homes in a county does not influence the likelihood of return, but the aggregate level of this measure complicates interpretation.

Our second model deals with a data set of evacuees in Houston. The Houston evacuee data represent quite a unique population: the sample has a third of respondents with less than a high school education, is overwhelmingly African American (over 98%), and almost half of the respondents report incomes less than $15,000 per year. For this population, we find that education and youthfulness (being under 30 years of age) decrease the likelihood of return, whereas those who were employed before Katrina, those who are married, and those who own a house are more likely to return. Home ownership has a large influence on the likelihood of return, increasing the probability by 21%. These sets of results are useful in their own right in that they provide insight into the nature of the return migration decision, allow one to make inferences about how the economic and cultural recovery of an area could proceed, and suggest policies that might aid in recovery.

For the Houston sample, we are also capable of exploring the relationship between wage differentials in the home and host region and the likelihood of return. We examine wage differentials in light of the literature on economic migration, in which households are assumed to sort over space according to wages, the prices of location-specific commodities (e.g., land), and spatial amenities. The persistence of significant wage differentials after controlling for land rents and spatial amenities suggests that some component of behavior forestalls spatial arbitrage. Cultural constraints, such as kinship relations or connection to place (Frey and Liaw 2005), could operate to inhibit migration.

Although we find no evidence that proxies for what we call "connection to place" affect the likelihood of return migration in either of our data sets, we do find that households intend to return home in spite of real economic costs in terms of real wage differentials across the home and host location. We exploit individual variation in wage differentials to estimate the effect on the likelihood of return and find a statistically significant and negative effect those that face higher opportunity costs of return in terms of higher relative real wages in Houston tend to stay in Houston, whereas those that face lower or negative opportunity costs tend to return. A signal of value that one could attribute to returning home is that some individuals will accept lower wages to do so. For the sample of Houston evacuees, we estimate that the average household is willing to give up $1.94 per hour to return home. Assuming that the earning individual works full time, this corresponds with an annual WTP of $3954. These numbers are limited in their applicability because of the unique characteristics of the Houston sample, but the results are encouraging and suggest that this approach should be explored further with other data sets.

This research was funded by National Science Foundation (NSF) grants "SGER: Collecting Economic Impact Data: Implications for Disaster Areas and Host Regions" (CMS 0553108), "SGER: The "New" New Orleans: Evaluating Preferences for Rebuilding Plans after Hurricane Katrina" (SES 0554987), and "SGER: Cooperation among Evacuees in the Aftermath of Hurricane Katrina" (SES 0552439). The NSF bears no responsibility for the comments or conclusions reached within this study. Our sincere gratitude goes to Jamie Kruse for providing the economic impact data and to Rick Wilson for providing the Houston evacuee data.

References

Albrecht, Glen. 2006. Solastalgia. Alternatives Journal 32:33-6.

Baker, E. J. 1991. Hurricane evacuation behavior. International Journal of Mass Emergencies and Disasters 9:287-310.

Barringer, Herbert R., Robert W. Gardner, and Michael J. Levin. 1993. Asians and Pacific Islanders in the United States: A 1980 census monograph. New York: Russell Sage Foundation.

Bean, Frank D., and Marta Tienda. 1987. The Hispanic population of the United States: A 1980 census monograph. New York: Russell Sage Foundation.

Bowden, M., C. Pijawka, G. S. Roboff, K. J. Gelman, and D. Amaral. 1977. Reconstruction following disaster, edited by J. E. Haas, R. W. Kates, and M. J. Bowden. Cambridge, MA: MIT Press.

Bureau of Labor Statistics (BLS). 2005. "May 2005 Occupational Employment and Wage Estimates of the Department of Labor: United States." Accessed 5 February 2007. Available http://www.bls.gov/oes/2005/May/oes_nat.htm.

Cameron, Trudy Ann, and Michelle D. James. 1987. Efficient estimation methods for 'closed-ended' contingent valuation surveys. Review of Economics and Statistics 69:269-76.

CBS News, "Mississippi Coast Areas Wiped Out." Accessed 1 September 2005. Available http://www.cbsnews.com/stories/2005/09/01/katrina/main810916.shtml.

Clark, David E., and James C. Cosgrove. 1991. Amenities versus labor market opportunities: Choosing the optimal distance to move. Journal of Regional Science 31:311-28.

Dow, K.. and S. L. Cutter. 1997. Crying wolf: Repeat responses to hurricane evacuation orders. Coastal Management 26:237-51.

Elliott, James R., and Jeremy Pals. 2006. Race, class, and Hurricane Katrina: Social differences in human responses to disaster. Social Science Research 35:295-321.

Falk, William W., Matthew O. Hunt, and Larry L. Hunt. 2006. Hurricane Katrina and New Orleanians' sense of place: Return and reconstruction or 'gone with the wind'? Du Bois Review 3:115-28.

Farley, Reynolds, and Walter R. Allen. 1987. The color line and the quality of life in America: A 1980 census monograph. New York: Russell Sage Foundation.

Federal Emergency Management Agency (FEMA). "By the Numbers One Year Later FEMA Recovery Update for Hurricanes Katrina." Accessed 22 August 2006a. Available www.fema.gov/news/newsrelease.fema?id-29109.

Federal Emergency Management Agency (FEMA). "Hurricane Katrina, One-Year Later." Accessed 22 August 2006b. Available www.fema.gov/news/newsrelease.fema?id=29108.

Frey, William H., and Kao-Lee Liaw. 2005. Migration within the United States: Role of race-ethnicity. Brookings-Wharton Papers on Urban Affairs 2005:207-62.

Frey, William H., and Audrey Singer. 2006. Katrina and Rita impacts on Gulf Coast populations: First census findings. In Brookings census 2000 series. Washington, DC: Brookings Institution Press, pp. 1-20.

Gieryn, Thomas. 2000. A space for place in sociology. Annual Review of Sociology 26:463-96.

Gladwin, H., and W. G. Peacock. 1997. Warning and evacuation: A night of hard choices. In Hurricane Andrew: Ethnicity, gender, and the sociology of disasters, edited by W. G. Peacock, B. H. Morrow, and H. Gladwin. London: Routledge, pp. 52-73.

Graves. Philip E. 1979. A life-cycle empirical analysis of migration and climate, by race. Journal of Urban Economics 6:135-47.

Graves, Phillip E. 1980. Migration and climate. Journal of Regional Science 20:227-37.

Greene, W. 2003. Econometric analysis. 5th edition. Upper Saddle, NJ: Prentice Hall.

Greenwood. Michael J. 1969. An analysis of the determinants of geographic labor mobility in the United States. Review of Economics and Statistics 2:189-94.

Greenwood, Michael J. 1975. Research on internal migration in the United States: A survey. Journal of Economic Literature 13:397-433.

Greenwood, Michael J. 1997. Internal migration in developed countries. In Handbook of population and family, economies, edited by Mark R. Rosenzweig and Oded Stark. New York: Elsevier, pp. 647-720.

Greenwood, Michael J., and Gary L. Hunt. 1989. Jobs versus amenities in the analysis of metropolitan migration. Journal of Urban Economics 25:1-16.

Greenwood, Michael J., Gary L. Hunt, Dan S. Rickman, and George I. Treyz. 1991. Migration, regional equilibrium and the estimation of compensating differentials. American Economic Review 81:1382-90.

Haab, Timothy C., and Kenneth E. McConnell. 2002. Valuing environmental and natural resources: The econometrics of. non-market valuation. Cheltenham, UK: Edward Elgar.

Kates, R. W.. C. E. Colten, S. Laska, and S. P. Leatherman. 2006. Reconstruction of New Orleans after Hurricane Katrina: A research perspective. Proceedings o/' the National Academy of Sciences 103:14653-60.

Krinsky, L., and A. L. Robb. 1986. On approximating the statistical properties of elasticities. Review of Economics and Statistics 68:715-9.

National Weather Service (NWS). June 2006. Service assessment: Hurricane Katrina August 23-31, 2005. Silver Spring, MD: National Oceanic and Atmospheric Administration, U.S. Department of Commerce.

Roback, Jennifer. 1982. Wages, rents, and the quality of life. Journal of Political Economy 90:1257-78.

Rosen, Sherwin. 1974. Hedonic prices and implicit markets: Product differentiation in perfect competition. Journal of Political Economy 82:34-55.

Sjaastad, Larry A. 1962. The costs and returns of human migration. Journal of Political Economy 70:80-93.

Tienda, Marta, and Franklin D. Wilson. 1992. Migration and the earnings of Hispanic men. American Sociological Review 57:661-90.

Whitehead, John C. 2005. Environmental risk and averting behavior: Predictive validity of jointly estimated revealed and stated behavior data. Environmental and Resource Economics 32:301-16.

Whitehead, John C., Bob Edwards, Marieke Van Willigen, John R. Maiolo, Kenneth Wilson, and Kevin T. Smith. 2000. Heading for higher ground: Factors affecting real and hypothetical hurricane evacuation behavior. Environmental Hazards 2:133-42.

Wilson, R. K., and R. M. Stein. 2006. Katrina evacuees in Houston: One-year out. Rice University Division of Social Sciences Working Paper.

U.S. Congress. 2006a. The budget and economic outlook: Fiscal years 2007 to 2016. Washington, DC: Congressional Budget Office.

U.S. Congress. 2006b. A failure of initiative: The final report of the Select Bipartisan Committee to Investigate the Preparation for and Response to Hurricane Katrina. 109th Cong., 2nd sess. H. Rpt:109-377. Washington, DC: U.S. House of Representatives Committee on Government Reform.

(1) "Place" is defined as a geographical unit in which identity is grounded (Gieryn 2000).

(2) For example, a survey of previous residents one year after a devastating earthquake revealed that 74% of unskilled workers had not returned to the area, whereas only 40% of skilled workers did not return (Bowden et al. 1977).

(3) Haab and McConnell (2002) illustrate that the willingness to pay function approach is equivalent to a utility difference model (the basis of most discrete choice models) if utility is linear in parameters and the marginal utility of income is constant across the discrete choice states (in our case, going home or remaining in the host city).

(4) Because they are likely to be very small relative to the present value of the wage differential and will only be incurred once, we ignore the pecuniary and time costs of return.

(5) Line 3 of Equation 3 only holds for symmetric distribution of [epsilon]. The logistic distribution is symmetric.

(6) WTP measure assumes constant marginal utility of income.

(7) These surveys were the result of two National Science Foundation grants: "SGER: Collecting Economic Impact Data: Implications for Disaster Areas and Host Regions" (CMS 0553108); and "SGER: The 'New' New Orleans: Evaluating Preferences for Rebuilding Plans after Hurricane Katrina" (SES 0554987).

(8) These samples were purchased from Survey Sampling of Fairfield, CT.

(9) The Houston evacuee study was sponsored by the National Science Foundation "SGER: Cooperation among Evacuees in the Aftermath of Hurricane Katrina" (SES 0552439). The grant was awarded to Dr. Rick Wilson, chair of the Department of Political Science and the Herbert S. Autrey Professor of Political Science and Professor of Statistics and Psychology at Rice University.

(10) Although summary statistics for race are not provided with the Houston data, most of the respondents to this survey were African Americans.

(11) For the Houston data set, we also estimated an ordered logit regression using the dependent variable with the values of very unlikely, somewhat unlikely, somewhat likely, and highly likely categories. The sign and significance of most coefficients are the same as the logit regression. We only report the results from the logit regression to compare the results with the mail survey.

(12) The wage data come from the Bureau of Labor Statistics (BLS 2005) website. The data provided wage estimates for over 800 occupations by geographic area. The website states "these estimates are calculated with data collected from employers in all industry sectors in metropolitan and non-metropolitan areas in every State and the District of Columbia."

(13) Variation in amenities within sites is largely unobserved because of data limitations.

(14) Elliott and Pais (2006) are the only authors that we are aware of to examine the return migration decision in a quantitative framework. Falk. Hunt. and Hunt (2006) speculate on how the demographics of New Orleans might change in the wake of Hurricane Katrina.

Craig E. Landry, * Okmyung Bin, ([dagger]) Paul Hindsley, ([double dagger]) John C. Whitehead, ([section]) and Kenneth Wilson ([parallel])

* Department of Economics and Center for Natural Hazards Research, East Carolina University, A-433 Brewster Building, Greenville, NC 27858, USA: E-mail landryc@ecu.edu; corresponding author.

([dagger]) Department of Economics, East Carolina University, A-435 Brewster Building, Greenville, NC 27858, USA.

([double dagger]) Coastal Resources Management, Center for Natural Hazard Research, East Carolina University, Greenville, NC 27858, USA.

([section]) Department of Economics, Appalachian State University, Boone, NC 28608, USA.

([parallel]) Department of Sociology, East Carolina University, A-403 Brewster Building, Greenville, NC 27858, USA.

Table 1. Logistic Regression Results for the Likelihood of Return

 Mail Survey

Variable Coeff. SE

CONSTANT 1.005 0.649
INCOME 0.040 ** 0.016
INCOME2 -3.0 x 1.2 x
 [10.sup.-4] ** [10.sup.-4]
COLLEGE 0.297 0.294
UNDER30 0.113 0.351
SENIOR -0.607 * 0.331
NOMA -0.976 ** 0.346
PERCDAM -0.31 0.738
MALE 0.265 0.256
BLACK 0.229 0.409
CAJUN -0.134 0.57
PARISH -0.649 * (b) 0.34
IMPACT 1.428 ** 0.374
WORKING
MARRIED
CHILDREN
OWNHOME
LIVEDYR
WAGEDIFF
Obs. 746
Pseudo-
 [R.sup.2] 0.176
Log L -216.458

 Mail Survey

Variable Marginal
 effects (a) SE

CONSTANT
INCOME 0.003 ** 0.001
INCOME2 -2.0 x 1.0 x
 [10.sup.-5] ** [10.sup.-5]
COLLEGE 0.02 0.019
UNDER30 0.007 0.022
SENIOR -0.047 ** 0.029
NOMA -0.078 ** 0.032
PERCDAM -0.021 0.05
MALE 0.018 0.018
BLACK 0.014 0.024
CAJUN -0.009 0.042
PARISH -0.055 0.035
IMPACT 0.124 ** 0.04
WORKING
MARRIED
CHILDREN
OWNHOME
LIVEDYR
WAGEDIFF
Obs.
Pseudo-
 [R.sup.2]
Log L

 Houston Survey

Variable Coeff. SE

CONSTANT -2.239 0.595
INCOME 0.003 0.015
INCOME2 -7.1 x 1.9 x
 [10.sup.-6] [10.sup.-4]
COLLEGE -0.397 * 0.203
UNDER30 -0.561 ** 0.180
SENIOR 0.329 0.861
NOMA 1.054 ** 0.400
PERCDAM 0.473 0.402
MALE -0.052 0.176
BLACK
CAJUN
PARISH
IMPACT
WORKING 0.672 ** 0.219
MARRIED 0.544 ** 0.222
CHILDREN 0.037 0.049
OWNHOME 0.962 ** 0.254
LIVEDYR 0.009 0.010
WAGEDIFF -0.287 ** 0.059
Obs. 756
Pseudo-
 [R.sup.2] 0.086
Log L -415.679

 Houston Survey

Variable Marginal
 effects (a) SE

CONSTANT -0.440 0.114
INCOME 0.000 0.003
INCOME2 -1.4 x 3.8 x
 [10.sup.-6] [10.sup.-5]
COLLEGE -0.075 ** 0.037
UNDER30 -0.114 ** 0.037
SENIOR 0.069 0.192
NOMA 0.163 ** 0.045
PERCDAM 0.093 0.079
MALE -0.010 0.035
BLACK
CAJUN
PARISH
IMPACT
WORKING 0.125 ** 0.039
MARRIED 0.115 ** 0.050
CHILDREN 0.007 0.010
OWNHOME 0.214 ** 0.061
LIVEDYR 0.002 0.002
WAGEDIFF -0.056 ** 0.012
Obs.
Pseudo-
 [R.sup.2]
Log L
Coeff., coefficient; Obs., observed.

(a) Marginal effects of the dummy variables are computed with the
changes in the probabilities; otherwise, marginal effects are
evaluated at those observed means.

(b) The PARISH variable is set to 0 for the IMPACT sample.

* Significance at 10% level.

** Significance at 5% level.

Table 2. Variable Definitions and Summary Statistics

 Mail Survey

Variable Description Mean SD

RETURN Returning to pre-Katrina residence (=1) 0.887 0.316
INCOME Household annual income in thousands of
 dollars 51.434 32.560
COLLEGE Attended college (=1) 0.430 0.495
UNDER30 Age under 30 (=1) 0.208 0.406
SENIOR Age over 63 (=1) 0.256 0.437
NOMA Residence located within the New Orleans
 Metropolitan area 0.316 0.465
PERCDAM Percentage of damaged property in county 0.449 0.232
MALE Gender answered as male (=1) 0.540 0.499
BLACK Race-ethnic group answered as black (=1) 0.129 0.335
CAJUN Race-ethnic group answered as Cajun (=1) 0.067 0.250
IMPACT Observation from Economic Impact survey
 (survey 1 of mail portion) (a) (=1) 0.677 0.468
PARISH Born in parish/county of residence
 (b) (=1) 0.348 0.478
WORKING Employed before Katrina (=1)
MARRIED Married (=1)
CHILDREN Number of children
OWNHOME Own home residence (=1)
LIVEDYR Number of years lived in New Orleans
WAGEDIFF Real wage difference by labor class
 (Houston wage--NOLA wage)

 Houston Survey

Variable Description Mean SD

RETURN Returning to pre-Katrina residence (=1) 0.290 0.454
INCOME Household annual income in thousands of
 dollars 18.704 15.887
COLLEGE Attended college (=1) 0.328 0.470
UNDER30 Age under 30 (=1) 0.640 0.480
SENIOR Age over 63 (=1) 0.008 0.089
NOMA Residence located within the New Orleans
 Metropolitan area 0.923 0.266
PERCDAM Percentage of damaged property in county 0.452 0.214
MALE Gender answered as male (=1) 0.508 0.500
BLACK Race-ethnic group answered as black (=1)
CAJUN Race-ethnic group answered as Cajun (=1)
IMPACT Observation from Economic Impact survey
 (survey 1 of mail portion) (a) (=1)
PARISH Born in parish/county of residence
 (b) (=1)
WORKING Employed before Katrina (=1) 0.652 0.477
MARRIED Married (=1) 0.171 0.376
CHILDREN Number of children 2.015 1.803
OWNHOME Own home residence (=1) 0.128 0.335
LIVEDYR Number of years lived in New Orleans 25.737 8.963
WAGEDIFF Real wage difference by labor class
 (Houston wage--NOLA wage) 1.553 2.049

The summary statistics for the mail survey is based on 746
observations. The sample size for the Houston survey is 756.

(a) IMPACT data did not record information on social/family
connections to the home location.

(b) Descriptive statistics for PARISH correspond with the subset of
the mail data that recorded social/family connections (n = 241).