The Vulnerable: How Race, Age and Poverty Contribute to Tornado Casualties

Transcription

1 The Vulnerable: How Race, Age and Poverty Contribute to Tornado Casualties Tyler Fricker Department of Geography, Florida State University February 28, 2018

2 Who Survives? Credit: Dayton Daily News

3 Take-Home Points A 10 mobile home increase under the path increases the casualty rate by 4.2% A 100 person increase in the population over 65 under the path increases the casualty rate by 1.4% A percentage increase in the white population under the path decreases the casualty rate by 25%

4 Objectives Establish statistical estimates (and margins of error) on how sensitive casualties are to: Changes in population and, Changes in tornado strength Examine these sensitivities across space

5 Casualty-Producing Tornadoes

6 Tornado Paths

7 Casualties 100 Number of Tornadoes Casualties

8 Population Density A 200 Number of Tornadoes Population Density

9 Energy Dissipation The equation for energy dissipation is E = A p ρ J j=0 w j v 3 j, (1) where A p is the area of the approximate path (width times length), ρ is the air density (assumed to be 1 kg/m 3 at the surface), v j is the midpoint wind speed for each damage rating j, and w j is the corresponding fraction of path area.

10 Schematic of Path Area EF0 EF1 EF2 EF3

11 Percent of Path Area by EF Rating Wind Speed Maximum EF rating (w j ) [m s 1 ] EF0 EF1 EF2 EF3 EF4 EF5 EF EF EF EF EF EF

12 Energy Dissipation B 150 Number of Tornadoes Energy Dissipation (W)

13 May 22, 2011 Casualty-Producing Tornadoes A Joplin, MO Energy Dissipation [GW] Tornado Casualties Population Density [people/km 2 ] B

14 2011 Casualty-Producing Tornadoes 100,000 10,000 Joplin, MO Energy Dissipation [GW] Tornado Casualties Population Density [people/km 2 ]

15 All Casualty-Producing Tornadoes 100,000 10,000 Joplin, MO Energy Dissipation [GW] Tornado Casualties ,000 Population Density [people/km 2 ]

16 Additive Model 1, Energy Dissipation [GW] ,000 Population Density [people per sq. km]

17 Additive Model Results A doubling of the population under the path of a tornado leads to a 21% increase in the casualty rate A doubling of the energy dissipated by the tornado leads to a 33% increase in the casualty rate

18 Interactive Model 1, Energy Dissipation [GW] ,000 Population Density [people per sq. km]

19 Casualty Rates Casualty Rate [No. of Casualties Per Casualty Producing Tornado] Population Density [people/km 2 ] ,000 Energy Dissipation [GW]

20 Interactive Model Results The percentage increase in casualties with increasing energy dissipation increases with population density The percentage increase in casualties with increasing population density increases with energy dissipation

21 Slopes by State Rank State Magnitude of Slope 1 Arkansas Tennessee Missouri Kentucky Illinois Oklahoma Alabama Mississippi Texas North Carolina Louisiana Georgia 0.128

22 Drilling-Down Establish estimates of socioeconomic and demographic variables at the tornado level Evaluate the relationship of socioeconomic and demographic variables between: Individual tornadoes Damage characteristic (e.g. casualties)

23 Overlay Method Household Median Income (USD) < $10,000 $10,000 $20,000 $20,001 $30,000 $30,001 $40,000 > $40,000

24 Tornado-Level Estimates Date State Total Population Male Population Female Population FL TN OK KY AL MO MA TX

25 Validation A 75 Estimated Male Deaths Observed Male Deaths

26 Validation B 75 Estimated Female Deaths Observed Female Deaths

27 Summary of Estimates Variable Mean 25 th Percentile Median 75 th Percentile Total Population Population Density Number of Males Number of Females White Population Black Population Household Median Income $48,500 $38,700 $46,000 $55,400 Number of Mobile Homes Not shown: Population by age (under 17, 18-44, 45-64, and over 65)

28 Estimates: Total Population Location/Tornado Date Number of Casualties Total Population Detroit, MI St. Louis, MO Pittsburgh, PA Tuscaloosa Birmingham, AL Springfield, MA Bridge Creek Moore, OK St. Louis, MO Minneapolis, MN Nashville, TN Hackleburg Phil Campbell, AL

29 Estimates: Mobile Homes Location/Tornado Date Number of Casualties Mobile Homes Hackleburg Phil Campbell, AL Vilonia, AR Shoal Creek Valley Ohatchee, AL Cordova, AL Tuscaloosa Birmingham, AL Wichita, KS Bridge Creek Moore, OK Auburn, AL Tallulah Yazoo City Durant, LA Little Rock, AR

30 Estimates: Household Median Income Location/Tornado Date Number of Casualties Median Income Denver, CO $161,152 Los Altos, CA $149,597 Fairfax, VA $139,458 Dunwoody, GA $119,549 Sonoma, CA $118,608 Westchester County, NY $117,882 Centreville, VA $116,493 Blue Ash, OH $115,559 Livingston-Genesee County, MI $114,426 Lake Travis, TX $112,936

31 Correlation Between Casualties and Estimates Variable Correlation Total Population.33 Population Density.02 Number of Males.21 Number of Females.20 White Population.32 Black Population.05 Household Median Income.00 Number of Mobile Homes.44 Not shown: Population by age (under 17, 18-44, 45-64, and over 65)

32 Estimated White Casualties by Tornado Estimated White Casualties ,200

33 Estimated Black Casualties by Tornado Estimated Black Casualties ,200

34 Casualty Model Estimates can be added to previous models (additive or interactive) to better understand the impact demographic and socioeconomic variables have on tornado casualties Estimates of interest include: Young and old population White and black population Household median income Mobile homes

35 Casualty Model Results A 10 mobile home increase under the path increases the casualty rate by 4.2% A 100 person increase in the population over 65 under the path increases the casualty rate by 1.4% A percentage increase in the white population under the path decreases the casualty rate by 25%

36 Casualty Model Results When controlling for the number of mobile homes, household median income IS NOT a significant factor Older populations (over 65) are a stronger predictor than younger populations (under 17) White populations are a stronger predictor than black populations

37 Future work The casualty model will improve with more variables Additional variables of interest include: Disability Educational attainment Ethnicity Language

38 Impact Credit: NY Times

39 What questions do you have?

40 Addititve Model The model is given by: C NegBin(ˆµ, n) (2) log(ˆµ) = ˆα log(p) + ˆβ log(e) + ˆν, (3) where NegBin(ˆµ, n) indicates that the conditional casualty counts are described by negative binomial distributions with mean (rate) ˆµ and size n. The coefficient ˆα is the population elasticity, the coefficient ˆβ is the energy elasticity and ˆν is the intercept parameter.

41 Interactive Model The model is given by: C NegBin(ˆµ, n) (4) ˆµ = ˆβ 0 P ˆβ P E ˆβ E (E P) ˆβ P E, (5) where the coefficient ˆβ P is the population elasticity, the coefficient ˆβ E is the energy elasticity and ˆβ P E is the interactive term.

42 Casualty Model The casualty model is given by: log(ĉ) = ˆβ 0 + ˆβ 1 log(p) + ˆβ 2 log(e) + ˆβ 3 log(p) log(e) + ˆβ 4 (V 1 ) + ˆβ 5 (V 2 ) + ˆβ 6 (V 3 ) +... where the coefficients ˆβ 1, ˆβ 2, and ˆβ 3 are the population elasticity, energy elasticity, and interaction term, respectively. The coefficients ˆβ 4, ˆβ 5, and ˆβ 6 are variable estimate terms, respectively. (6)