Hurricane Katrina: Was There a Political Economy of Death?


An empirical implication of egalitarianism in the provision of public disaster relief services is that the probability of surviving a natural disaster should not be conditioned on a household’s position in the income distribution, or its racial characteristics. In this paper, we utilize data on deaths attributed to Hurricane Katrina in the City of New Orleans to estimate a political economy model of the public provision of disaster rescue services. Parameter estimates reveal that the probability of dying as a result of Hurricane Katrina, at both the census tract and individual level, increased with respect to being black and poor. Our results suggest that there was a departure from egalitarian principles in the provision of public disaster rescue services during Hurricane Katrina, and are consistent with a political economy of race and class governing decisions about the allocation of public resources to ameliorate population environmental risks.

This is a preview of subscription content, log in to check access.


  1. 1.

    For a detailed analysis of the Hurricane Katrina timeline see: Report of the Committee on Homeland Security and Governmental Affairs, USGPO (2006).

  2. 2.

    For a detailed itemization of deceased Katrina victims, see “Vital Statistics of All Bodies at St. Gabriel Morgue”, Louisiana Department of Health and Hospitals (2006).

  3. 3.

    The estimate of the black share of Katrina deaths in New Orleans is based on summary statistics reported in “Vital Statistics of All Bodies at St. Gabriel Morgue”, Louisiana Department of Health and Hospitals (2006).

  4. 4.

    The estimate of the black population share of New Orleans poverty is based on data reported in the Census 2000 Summary File 4 (SF 4) Data.

  5. 5.

    An example of simple yet high profile descriptive analysis of who were the victims of Hurricane Katrina is that of Simerman et al. (2005), who report that the dead neither disproportionately black or poor–only disproportionately elderly.

  6. 6.

    See for details about finding the address of individuals given their name and age at Intellius.

  7. 7.

    Approximately 1% of the victims were less than or equal to 15 years of age.

  8. 8.

    Approximately 40% of the victims were greater than or equal to 75 years of age. Another 25% of the victims were between the ages of 51 and 60, some of which may have also been retired from the labor force, with the result that no active credit history data is available.

  9. 9.

    Draut (2007) reports that in 2004, the fraction of white and black families that had credit cards was 84% and 52% respectively.

  10. 10.

    Let d i be the number n of deaths in census tract i, then for d = 0, 1, 2, ..., N:

    $$ Prob(d_{i} = n \mid \lambda_{i}, \alpha) = \left(\frac{\lambda_{i}}{1 + \alpha\lambda_{i}}\right)^{d_{i}}\left[\frac{ (1 + \alpha d_{i})^{d_{i} - 1}}{d_{i} !}\right]~exp\left[\frac{- \lambda_{i}(1 + \alpha d_{i})}{1 + \alpha \lambda_{i}}\right] $$

    where λ i is the expected value of d i , and α is a dispersion parameter. The variance of d i is λ i (1 + αλ)2 (Famoye 1993). Particular count data regression models result from the value of α, and specifying the mean (λ i ) of d i as a function of exogenous variables with unknown parameter values. When α = 0, a Poisson regression specification results. A Negative Binomial regression specification results when α ≠ 0.

  11. 11.

    See Kiefer (1988) and Van den Berg (2001) for an overview of hazard models. For our specification let T represent the duration of life for a Hurricane Katrina victim. If T is a continuous random variable with realization t, the probability that life will end at time t + Δt conditional on a vector of explanatory X is the hazard rate h l(t |X), which can be specified as:

    $$ h^{l}(t \mid {\bf X}) = h^{l}_{o}(t)exp\left(\sum \beta_{k}X_{k}\right) $$

    where \(h^{l}_{o}(t)\) is the baseline death hazard faced by all Hurricane Katrina victims.

    An estimate of h l(t |X) yields the probability of an individual leaving life as a victim of Hurricane Katrina conditional on the vector of explanatory variables X. This is a standard duration/hazard model. If we are unwilling to make assumptions about the shape or parametric family of baseline hazard \(h^{l}_{o}(t)\), then \(h^{l}_{o}(t \mid {\bf X})\) is a Cox proportional hazards model.

  12. 12.

    For an overview of frailty in hazard/duration models, see Cleves et al. (2004, Chapter 9), and Wienke (2003).

  13. 13.

    For j = 1, 2, ⋯ n, groups of individuals, let the unobserved random frailty be α j where the frailty varies across individuals—a random effect—and has a mean of unity and a variance of θ, then a Cox proportional hazards specification with frailty is:

    $$ h^{l}(t \mid {\bf X}, \alpha_{j}) = h^{l}_{o}(t)exp\left(\sum \beta_{k}X_{k} + \nu_{j} \right) $$

    where ν j = log α j . Frailty is shared within groups, and is assumed to have a gamma distribution. An interpretation of frailty is afforded by the magnitude of ν j : if it is greater than unity then an individual in group j has unobserved attributes that make him predisposed toward the hazard—the converse is true if the frailty is less than unity.

  14. 14.

    This is the official definition of a high poverty census tract. See 20 U.S.C Sec. 6177 (2005).

  15. 15.

    See www.gnodc.dc.

  16. 16.

    Our measure of flood risk is binary, and is zero or one to the extent that a census tract has a body of water, or not. Such a measure of flood risk may not capture the type of variation in flood risk across census tracts that are important for death probabilities given a flood.

  17. 17.

    Given f(b i , |λ i , α) a Zero-Inflated Generalized Poisson (ZIGP) regression specification is:

    $$ g(b_{i} \mid x_{i}, z_{i}) = \left\{ \begin{array}{cc} \varphi_{i} + (1 - \varphi_{i})f(b_{i} \mid \lambda_{i}, \alpha) & \mbox{if $b_{i} = 0$} \\[3pt] (1 - \varphi_{i})f(b_{i} \mid \lambda_{i}, \alpha) & \mbox{if $b_{i} > 0$}\\ \end{array} \right. $$

    where λ i  = λ i (x i ) and ϕ i  = ϕ i (z i ) satisfy ln λ i  = ∑ β i x i and \(logit(\varphi_{i}) = ln(\varphi_{i}[1 - \varphi_{i}])^{-1} = \sum \beta_{i} z_{i}\). The x i are the covariates that determine λ i , and the z i are covariates that determine the zero observations in the two distinct states governed by probabilities ϕ i and (1 − ϕ i ) respectively. As the z i condition the distribution and type of zero realizations (e.g. sampling zeros with probability (1 − ϕ i ) and structural zeros with probability ϕ i .) we can call the z i zero inflators. The mean and variance of generalized Poisson random variable in a ZIGP regression model is E(b i ) + (1 − ϕ i )λ i , and \(Var(b_i) = E(b_i)[(1 + \alpha \lambda_i)^2 + \varphi_i\lambda_i ]\).

  18. 18.

    Greene (2003) notes that when 0 ≤ v ≤ 2, it is generally not possible to discriminate between the ZIP/ZINB and their simple Poisson/Negative Binomial alternatives.

  19. 19.

    In particular, the two frailty variables are measured as: (1) Percent of Census Tract population between the ages of 21–64 with a disability: Source: Census 2000, File DP-2. Profile of Selected Social Characteristics, and (2) Percent of Census Tract population 65 years of age and older with a disability: Source: Census 2000, File DP-2. Profile of Selected Social Characteristics.

  20. 20.

    If θ = 0 cannot be rejected, the standard Cox specification is adequate, and there are no group-specific frailties that explain the hazard under consideration.

  21. 21.

    The central idea here is that if a Cox Proportional Hazards specification fits the data, then the true cumulative hazard function conditional on the covariates has an exponential distribution with a hazard rate of 1 (Cleves et al. 2004, p. 190). As this implies that cumulative hazard of the Cox-Snell residuals is a 45-degree line when plotted against the Nelson-Aalen cumulative hazard, we measure this by computing the correlation between the Cox-Snell residuals and the Nelson-Aalen cumulative hazard.

  22. 22.

    An attempt was made to estimate the parameters of a Cox regression specification where the variables included variables for of being black,living in a high poverty census tract, and their interaction. The parameter estimates for being black and the interaction between being black and living in a high poverty census tract were insignificant, whereas living in a high poverty census tract was significant. On the basis of the \(R^{2}_{h}\), the Cox specifications reported in Table 2 are higher. This suggests that the race and poverty are jointly significant, and relative to their separate effects, better explain the likelihood of individual deaths due to Hurricane Katrina.


  1. Alesina A, Glaeser E, Sacerdote B. Why doesn’t the United States have a European-style welfare state? Brookings Pap Econ Act. 2001;2:187–254.

    Article  Google Scholar 

  2. Anbarci N, Escaleras M, Register CA. Earthquake fatalities: the interaction of nature and political economy. J Publ Econ. 2005;89:1907–33.

    Article  Google Scholar 

  3. Baugh J. It aint about race: some lingering (linquistic) consequences of the African slave trade and their relevance to your personal historical hardship index. Du Bois Rev. 2006;3:145–59.

    Article  Google Scholar 

  4. Besley T, Burgess R. The political economy of government responsiveness. Rev Econ Stat. 2002;117:1415–52.

    Google Scholar 

  5. Cleves MA, Gould WW, Gutierrez RG. An introduction to survival analysis using STATA. Revised ed. College Station, Texas: Stata Press, STATA Corp; 2004.

    Google Scholar 

  6. Darity W Jr. Stratification economics: the role of intergroup inequality. J Econ Finance. 2005;29:144–53.

    Article  Google Scholar 

  7. de Souza Briggs X. After Katrina: rebuilding lives and places. City Commun. 2006;5:119–28.

    Article  Google Scholar 

  8. Deaton A, Muellbauer J. Economics and consumer behavior. New York: Cambridge University Press; 1980.

    Google Scholar 

  9. Draut T. Testimony before the United States senate commmittee on banking, housing and urban affairs. In: Examining the billing, marketing, and disclosure practices of the credit card industry and their impact on consumers. Hearing. Accessed 25 January 2007.

  10. Famoye F. Restricted generalized poisson regression model. Commun Stat, Theory Methods. 1993;22:1335–54.

    Article  Google Scholar 

  11. Famoye F, Singh KP. Zero-inflated generalized poisson regression model with an application to domestic violence data. J Data Sci. 2006;4:117–30.

    Google Scholar 

  12. Garrett T, Sobel R. The political economy of FEMA disaster payments. Econ Inq. 2003;41:496–507.

    Article  Google Scholar 

  13. Greene WH. Econometric analysis. 5th ed. Upper Saddle River: Prentice Hall; 2003.

    Google Scholar 

  14. Horwich G. Disasters and market response. Cato J. 1990;9:531–5.

    Google Scholar 

  15. Horwich G. Economic lessons from the Kobe earthquake. Econ Dev Cult Change. 2000;48:521–42.

    Article  Google Scholar 

  16. Kahn ME. The death toll from natural disasters: the role of income, geography, and institutions. Rev Econ Stat. 2005;87:271–84.

    Article  Google Scholar 

  17. Kiefer NM. Economic duration data and hazard functions. J Econ Lit. 1988:26:646–79.

    Google Scholar 

  18. Louisiana Department of Health and Hospitals. Vital statistics of all bodies at St. Gabriel Morgue. Accessed 23 February 2006.

  19. McFadden D. The measurement of urban travel demand. J Public Econ. 1974;3:303–28.

    Article  Google Scholar 

  20. Sharkey P. Survival and death in New Orleans: an empirical look at the human impact of Katrina. J Black Stud. 2007;37:482–501.

    Article  Google Scholar 

  21. Simerman J, Ott D, Melnick T. Assumptions about Katrina victims may be incorrect. Knight Ridder Newspapers; December 29, 2005.

  22. Toya H, Skidmore M. Economic development and the impacts of natural disasters. Econ Lett. 2007;94:20–5.

    Article  Google Scholar 

  23. United States Government Printing Office. Hurricane Katrina: a nation still unprepared. Special Report of the Committee on Homeland Security and Governmental Affairs (S. Rept. 109–322), United States Senate. (2006).

  24. Van den Berg JG. Duration models: specification, identification, and multiple durations. In: Heckman JJ, Leamer E, editors. Handbook of econometrics, vol. 5, chapter 55. Amsterdam: Elsevier/North-Holland; 2001.

    Google Scholar 

  25. Wienke A. Frailty models. MPIDR Working Paper WP 2003 - 032. Rostock Germany: Max Planck Institute for Demographic Research; 2003.

    Google Scholar 

Download references


The author would like to thank participants of the Michigan State University Department of Economics Theory Seminar, William Darity Jr., and other participants of the 2007 University of North Carolina Institute for African American Research Conference on “Race and the Environment” for critical yet helpful comments on earlier versions of this paper.

Author information



Corresponding author

Correspondence to Gregory N. Price.

About this article

Cite this article

Price, G.N. Hurricane Katrina: Was There a Political Economy of Death?. Rev Black Polit Econ 35, 163–180 (2008).

Download citation


  • Egalitarianism
  • Political economy
  • Hurricane Katrina