Assessing the Impact of Hurricanes on Economic Welfare in Jamaica: Implications for Policy Development in the Caribbean

Name : Devin McIntosh ID: 620140404

SECTION ONE

Abstract

This report paper answers the question of the relationship or connection between hurricane intensity and economic well-being of Jamaicans.

Introduction

The Caribbean region is particularly susceptible to storms’ devastation. Several storms have recently hit Jamaica, a small island developing nation in the Caribbean, wreaking havoc on its infrastructure, destroying its property, and taking lives. However, little is known about how hurricanes affect Jamaica’s economy.According to Spencer & strobl (2022), “without wind resistant building material, substantial increases in poverty are likely under most models.” Jamaica, as a developing country with a relatively vulnerable economy is among the countries that this would apply to and is therefore a suitable model to study and develop policy suggestions. Baez & Santos (2007) supports this by stating that weather related phenomena are relevant to understanding household welfare exposure to harmful shocks which is the focus of this report.

Historically, hurricanes in Jamaica have exacted significant tolls to propery, life and economic activity. Based on the Planning Institute of Jamaica estimates, close to 7 percent (6.7) of the Jamaican populace amounting to 179,552 people were affected by Hurricane Dean in 2007 from 169 communties(PIOJ, 2007). This report’s objective is to evaluate how storms affect Jamaicans’ economic well-being using a panel dataset and to highlight the significance of the results for Caribbean policy development.

Data Description

The researcher intends to analyze the impact of Hurricanes on economic well-being. The panel dataset used for this analysis is the Individual Final Assignment, which includes data from different parishes in Jamaica from 1990 to 2005. The following variables are included in the dataset:

year: Data year
parish: Parish number
log_newtotalpercapitacons: Measure of economic welfare
wind10, wind11, and wind12: Hurricane indexes capturing different levels of intensity
share_children: Share of children under 16 years of age in the household
rain1: Average rainfall
heatavg1: Average temperature
cluster: Cluster variable

In this dataset, we have set the dependent variable as:

Dependent Variable

log_newtotalpercapitacons (measure of economic welfare)

The independennt variables wind10, wind11, and wind12 (hurricane indexes capturing different levels of intensity), share_children (share of children under 16 years of age in the household)
rain1 (average rainfall)
heatavg1 (average temperature)

The variable year is a categorical variable representing data collection date (1990 to 2005.
The variable share_children is a continuous variable representing percentage or proportion of children less than 16 years in household.
The variable rain1 represents average rainfall and is a continuous variable.
The variable heatavg1 represents average temperature and it is a continuous variable.
The variable log_newtotalpercapitacons is continuous variable represented on a logarithmic scale.
The variable parish is a categorical variable showing parish number.
Variables wind10, wind11 & wind12: These are hurricane indexes and continuous.

Based on the data, we had 16,074 observations spanning 10 variables in dataset weather_data. Upon data inspection, the summary output showed that there are missing values in the wind10, and wind11 variables, with 602 NAs within each. After cleaning, weather2_clean had 15,472 observations of 10 variables.

New variables were created to help simplify the model such as wind_intense.

Methodology

We will use a panel data regression model to estimate the impact of hurricanes on economic welfare. Specifically, we will estimate the following regression model:

\(\log(newtotalpercapitacons_{it}) = \beta_0 + \beta_1 wind10_{it} + \beta_2 wind11_{it} + \beta_3 wind12_{it} + \beta_4 share\_children_{it} + \beta_5 rain1_{it} + \beta_6 heatavg1_{it} + \epsilon_{it}\)

where log_newtotalpercapitacons_it is the natural logarithm of the indicator of economic welfare for person i in year t, and the hurricane strength is measured by the wind10_it, wind11_it, and wind12_it indices. The proportion of children in the household under the age of 16, the average amount of precipitation, and the average temperature are the control variables.

We will also include an extreme value modeling approach, specifically the Generalized Pareto Distribution (GPD) model, to take the potential effects of extreme weather events into consideration. The tail risk related to extreme hurricane events will be calculated using the GPD model.

Simple fixed effects model for estimating the relationship between economic well being and hurricane intensity:

Simple fixed effects model for estimating the relationship between economic well-being and hurricane intensity: LNTCit = β1HIit + β2EWit + µi + ɛit

LNTCijt is the natural logarithm of new total per capita consumption for parish i and time period t.

#Set working directory 
setwd("C:/Users/Home/OneDrive/Documents/Econ 3050 Final")
getwd()

[1] "C:/Users/Home/OneDrive/Documents/Econ 3050 Final"

## Clear environment 
rm(list=ls())

## Load the libraries to be used in the report 
library(readxl)
library(ggplot2)
library(dplyr)
library(tidyverse)
library(lubridate)
library(modelr)
library(cowplot)
library(tidyverse)
library(ggthemes)
library(ggtext)
library(extRemes)
library(plm)
library(AER)
library(stargazer)
library(knitr)
library(gtsummary)
library(lmtest)

weather_data <- read_excel("C:/Users/Home/OneDrive/Documents/Econ 3050 Final/IndividualFinalAssignment.xls")

# check for missing values, outliers, and inconsistencies
summary(weather_data)

      year          parish       share_children       wind12           wind10           wind11      
 Min.   :1990   Min.   : 1.000   Min.   :0.0000   Min.   :0.0000   Min.   :0.0000   Min.   :0.0000  
 1st Qu.:1994   1st Qu.: 4.000   1st Qu.:0.0000   1st Qu.:0.0000   1st Qu.:0.0000   1st Qu.:0.0000  
 Median :1997   Median : 9.000   Median :0.2500   Median :0.0000   Median :0.0000   Median :0.0000  
 Mean   :1998   Mean   : 8.374   Mean   :0.2747   Mean   :0.2319   Mean   :0.2302   Mean   :0.2409  
 3rd Qu.:2002   3rd Qu.:13.000   3rd Qu.:0.5000   3rd Qu.:0.0000   3rd Qu.:0.0000   3rd Qu.:0.0000  
 Max.   :2005   Max.   :14.000   Max.   :1.0000   Max.   :8.3260   Max.   :8.3260   Max.   :8.3260  
                                                                   NA's   :602      NA's   :602     
     rain1           heatavg1     log_newtotalpercapitacons    cluster     
 Min.   : 89.21   Min.   :22.82   Min.   : 5.447            Min.   :  1.0  
 1st Qu.:137.18   1st Qu.:24.74   1st Qu.: 8.800            1st Qu.:215.0  
 Median :165.98   Median :25.76   Median : 9.275            Median :444.0  
 Mean   :181.48   Mean   :25.51   Mean   : 9.293            Mean   :439.3  
 3rd Qu.:205.94   3rd Qu.:26.24   3rd Qu.: 9.777            3rd Qu.:648.0  
 Max.   :389.98   Max.   :26.98   Max.   :14.189            Max.   :913.0  
                                                                           

# Missing values
sum(is.na(weather_data))

[1] 1204

# Remove NAs from wind10, and wind11 variables
weather2_clean <- weather_data[complete.cases(weather_data$wind12, weather_data$wind10, weather_data$wind11),]

# View new dataset
kable(head(weather2_clean), align = "c")

year	parish	share_children	wind12	wind10	wind11	rain1	heatavg1	log_newtotalpercapitacons	cluster
1990	1	0.0000000	0	0	0	166.0583	26.10000	10.732485	2
1995	14	0.0000000	0	0	0	154.2917	26.48333	9.983823	202
2004	14	0.5000000	0	0	0	115.6500	26.73333	9.589101	854
1990	1	0.3333333	0	0	0	166.0583	26.10000	10.354423	218
1991	1	0.0000000	0	0	0	126.0583	26.06667	9.213732	426
1996	5	0.6000000	0	0	0	253.6917	23.54167	9.177681	455

tbl_summary(weather2_clean)

Characteristic	N = 15,4721
year	1,997.0 (1,994.0, 2,003.0)
parish	9.0 (4.0, 13.0)
share_children	0.25 (0.00, 0.50)
wind12	0.00 (0.00, 0.00)
wind10	0.00 (0.00, 0.00)
wind11	0.00 (0.00, 0.00)
rain1	166 (138, 215)
heatavg1	25.73 (24.74, 26.24)
log_newtotalpercapitacons	9.27 (8.80, 9.77)
cluster	430 (212, 640)
1 Median (IQR)

str(weather2_clean)

tibble [15,472 × 10] (S3: tbl_df/tbl/data.frame)
 $ year                     : num [1:15472] 1990 1995 2004 1990 1991 ...
 $ parish                   : num [1:15472] 1 14 14 1 1 5 3 14 9 1 ...
 $ share_children           : num [1:15472] 0 0 0.5 0.333 0 ...
 $ wind12                   : num [1:15472] 0 0 0 0 0 ...
 $ wind10                   : num [1:15472] 0 0 0 0 0 ...
 $ wind11                   : num [1:15472] 0 0 0 0 0 ...
 $ rain1                    : num [1:15472] 166 154 116 166 126 ...
 $ heatavg1                 : num [1:15472] 26.1 26.5 26.7 26.1 26.1 ...
 $ log_newtotalpercapitacons: num [1:15472] 10.73 9.98 9.59 10.35 9.21 ...
 $ cluster                  : num [1:15472] 2 202 854 218 426 455 43 854 330 11 ...

#Creating a new variable to represent the overall wind intensity 
weather2_clean$wind_intense <- rowMeans(weather2_clean[, c("wind12", "wind10", "wind11")], na.rm = TRUE)

Data Visualization

ggplot(weather_data, aes(x = log_newtotalpercapitacons)) +
  geom_histogram(binwidth = 0.5, color = "black", fill = "blue") +
  labs(x = "Log Total Consumption per Capita", y = "Count") +
  ggtitle("Histogram of Log Total Consumption per Capita")

ggplot(weather2_clean, aes(x = share_children, y = log_newtotalpercapitacons)) +
  geom_point(alpha = 0.5) +
  labs(x = "Share of Children Under 16 in Household",
       y = "Log Total Consumption per Capita") +
  ggtitle("Scatterplot of Log Total Consumption per Capita vs. Share of Children Under 16")

Based on this scatterplot, no relationship or correlation is suggested between the two variables as the lines are vertical. .

# Scatterplot of log_newtotalpercapitacons vs wind_intense
ggplot(weather2_clean, aes(x = wind_intense, y = log_newtotalpercapitacons)) +
  geom_point() +
  labs(x = "Hurricane Index (wind intense)", y = "Log Per Capita Consumption")

Based on this scatterplot, the data points are gathered closely and mostly towards the left and bottom of the graph indicating a negative relationship between hurricane intnsity and our measure of economic well being.

ggplot(weather2_clean, aes(x = year, y = log_newtotalpercapitacons, group = parish, color = factor(parish))) + 
  geom_line() + 
  labs(title = "Time Series Plot of Log New Total Per Capita Consumption by Parish",
       x = "Year",
       y = "Log New Total Per Capita Consumption")

This graph is done to establish whether there exists any trends in economic well-being. For most of the graph, the trends remain constant indicating not much deviation related to hurricanes with high economic well being exhibited in some parishes than others years such as 1990 and lowest well-being in 2002 & 2005.

The minimum, maximum and averages of three key variables are to be looked at: the year of data, the wind_intense, rain1 and economic welfare measure.

weather2_sub <- weather2_clean[, c("year", "rain1", "log_newtotalpercapitacons", "wind_intense")]
min_values <- apply(weather2_sub, 2, min, na.rm = TRUE)
max_values <- apply(weather2_sub, 2, max, na.rm = TRUE)
mean_values <- apply(weather2_sub, 2, mean, na.rm = TRUE)

summary_table <- data.frame(
  Variable = c("Year", "Rain", "Economic Welfare", "Wind Intensity"),
  Min = min_values,
  Max = max_values,
  Mean = mean_values
)

stargazer(summary_table, type = "text", title = "Summary Statistics for Key Variables", 
          column.labels = c("Variable", "Minimum", "Maximum", "Mean"))

Summary Statistics for Key Variables
============================================
Statistic N  Mean   St. Dev.  Min     Max   
--------------------------------------------
Min       4 521.164 980.075  0.000 1,990.000
Max       4 604.375 950.668  8.326 2,005.000
Mean      4 547.365 970.511  0.237 1,997.703
--------------------------------------------

stargazer(weather2_clean, 
          title = "Breakdown of Key Variables", 
          type = "text", 
          dep.var.caption = "OLS", 
          no.space = TRUE, 
          header = FALSE, 
          digits = 3, 
          covariate.labels = c("Year", "Rain", "Economic Welfare", "Wind Intensity"), 
          report = "vc*s", 
          star.cutoffs = c(0.10, 0.05, 0.01), 
          model.names = FALSE, 
          object.names = FALSE, 
          model.numbers = FALSE, 
          keep.stat = c("rsq", "f"))

Breakdown of Key Variables
=================================
Statistic N Mean St. Dev. Min Max
=================================

Estimate the model using the function plm() to obtain the estimate for β1

fe_model <- plm(log_newtotalpercapitacons ~ rain1 + wind_intense + factor(year), data=weather2_clean, model="within")

Weathermod<-log_newtotalpercapitacons~wind_intense+rain1
Weather2mod<-lm(Weathermod, weather2_clean, subset=year==1990)
summary(Weather2mod)

Call:
lm(formula = Weathermod, data = weather2_clean, subset = year == 
    1990)

Residuals:
    Min      1Q  Median      3Q     Max 
-2.5670 -0.5677 -0.0378  0.5336  4.7941 

Coefficients: (1 not defined because of singularities)
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)   9.5856390  0.1185481  80.859   <2e-16 ***
wind_intense         NA         NA      NA       NA    
rain1        -0.0011850  0.0006633  -1.786   0.0743 .  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.8748 on 1040 degrees of freedom
Multiple R-squared:  0.003059,  Adjusted R-squared:  0.0021 
F-statistic: 3.191 on 1 and 1040 DF,  p-value: 0.07433

Weathermod2<-log_newtotalpercapitacons~wind_intense+rain1
Weather2mod<-lm(Weathermod2, weather2_clean, subset=year==1998)
summary(Weather2mod)

Call:
lm(formula = Weathermod2, data = weather2_clean, subset = year == 
    1998)

Residuals:
     Min       1Q   Median       3Q      Max 
-2.49479 -0.48524 -0.00278  0.50013  3.05136 

Coefficients: (1 not defined because of singularities)
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)   9.5876581  0.0803759  119.28  < 2e-16 ***
wind_intense         NA         NA      NA       NA    
rain1        -0.0013753  0.0004948   -2.78  0.00551 ** 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.7496 on 1366 degrees of freedom
Multiple R-squared:  0.005625,  Adjusted R-squared:  0.004897 
F-statistic: 7.727 on 1 and 1366 DF,  p-value: 0.005514

Weathermod3<-log_newtotalpercapitacons~wind_intense+rain1
Weather2mod<-lm(Weathermod3, weather2_clean, subset=year==2005)
summary(Weather2mod)

Call:
lm(formula = Weathermod3, data = weather2_clean, subset = year == 
    2005)

Residuals:
    Min      1Q  Median      3Q     Max 
-3.6541 -0.4699 -0.0053  0.4934  2.7018 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)   9.5935658  0.0725973 132.148  < 2e-16 ***
wind_intense -0.0023110  0.0130739  -0.177     0.86    
rain1        -0.0010000  0.0002413  -4.143  3.6e-05 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.7447 on 1592 degrees of freedom
Multiple R-squared:  0.01067,   Adjusted R-squared:  0.009427 
F-statistic: 8.585 on 2 and 1592 DF,  p-value: 0.0001956

PanModel<-plm(Weathermod, weather2_clean, model="pooling")

Warning: duplicate couples (id-time) in resulting pdata.frame
 to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")Warning message:
In is.recursive(.$object) && !is.primitive(.$object) && n > 0 :
  'length(x) = 2 > 1' in coercion to 'logical(1)'

summary(PanModel)

Pooling Model

Call:
plm(formula = Weathermod, data = weather2_clean, model = "pooling")

Balanced Panel: n = 14, T = 14, N = 15472

Residuals:
     Min.   1st Qu.    Median   3rd Qu.      Max. 
-3.759609 -0.493422 -0.018395  0.480816  4.889951 

Coefficients:
                Estimate  Std. Error  t-value  Pr(>|t|)    
(Intercept)   9.42805407  0.02004066 470.4462 < 2.2e-16 ***
wind_intense  0.03365745  0.00780755   4.3109 1.636e-05 ***
rain1        -0.00080179  0.00010789  -7.4314 1.131e-13 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Total Sum of Squares:    8959.2
Residual Sum of Squares: 8925.4
R-Squared:      0.0037675
Adj. R-Squared: 0.0036387
F-statistic: 29.2497 on 2 and 15469 DF, p-value: 2.0941e-13

Weathermod1 <- plm(log_newtotalpercapitacons ~ wind_intense + rain1, 
                   data = weather2_clean,
                   index = c("parish", "year"), 
                   model = "within", 
                   index.allow.duplication = TRUE)

Warning: duplicate couples (id-time) in resulting pdata.frame
 to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")

summary(Weathermod1)$coefficients["rain1", "Estimate"]

[1] 0.0007176655

coeftest(Weathermod1, vcov. = vcovHC, type = "HC1")

t test of coefficients:

                Estimate  Std. Error t value Pr(>|t|)  
wind_intense -0.01064767  0.01118815 -0.9517  0.34127  
rain1         0.00071767  0.00031321  2.2913  0.02196 *
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Weathermod2<-plm(log_newtotalpercapitacons ~ wind_intense + rain1, data=weather2_clean, index=c("year", "parish"), model="within", effect="twoways")

Warning: duplicate couples (id-time) in resulting pdata.frame
 to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")

coeftest(Weathermod2, vcov. = vcovHC, type = "HC1")

t test of coefficients:

                Estimate  Std. Error t value Pr(>|t|)
wind_intense -3.6565e-03  4.0940e-03 -0.8931   0.3718
rain1        -3.6086e-05  5.1370e-04 -0.0702   0.9440

Weathermod3<-plm(log_newtotalpercapitacons~ share_children + rain1 + wind_intense + heatavg1, data=weather2_clean, index=c("year", "parish"), model="within", effect="twoways")

Warning: duplicate couples (id-time) in resulting pdata.frame
 to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")

coeftest(Weathermod3, vcov.= vcovHC, type = "HC1")

Warning in is.recursive(.$object) && !is.primitive(.$object) && n > 0 :
  'length(x) = 2 > 1' in coercion to 'logical(1)'

t test of coefficients:

                  Estimate  Std. Error  t value  Pr(>|t|)    
share_children -0.97429428  0.07123903 -13.6764 < 2.2e-16 ***
rain1           0.00011797  0.00062871   0.1876  0.851160    
wind_intense   -0.01067249  0.00349066  -3.0574  0.002236 ** 
heatavg1        0.17671371  0.11110095   1.5906  0.111727    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

stargazer(Weathermod3, type="text", digit=2,title="Analysis of the Relationship between Hurricane intensity and Economic well-being", out="table2.txt")

Warning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changed

Analysis of the Relationship between Hurricane intensity and Economic well-being
==========================================
                   Dependent variable:    
               ---------------------------
                log_newtotalpercapitacons 
------------------------------------------
share_children          -0.974***         
                         (0.021)          
                                          
rain1                    0.0001           
                        (0.0003)          
                                          
wind_intense             -0.011           
                         (0.012)          
                                          
heatavg1                 0.177**          
                         (0.075)          
                                          
------------------------------------------
Observations             15,472           
R2                        0.127           
Adjusted R2               0.125           
F Statistic    560.528*** (df = 4; 15441) 
==========================================
Note:          *p<0.1; **p<0.05; ***p<0.01

Analysis of the Relationship between Hurricane intensity and Economic well-being
=
2
-

Model using Point over Threshold

library(readxl)
ThreshWeather <- read_xls("IndividualFinalAssignment.xls")

View the data

head(ThreshWeather)

Clean data set once again

Threshweather2 <- na.omit(ThreshWeather)

Get summary of the Dataset

summary(Threshweather2)

      year          parish       share_children       wind12          wind10           wind11      
 Min.   :1990   Min.   : 1.000   Min.   :0.0000   Min.   :0.000   Min.   :0.0000   Min.   :0.0000  
 1st Qu.:1994   1st Qu.: 4.000   1st Qu.:0.0000   1st Qu.:0.000   1st Qu.:0.0000   1st Qu.:0.0000  
 Median :1997   Median : 9.000   Median :0.2500   Median :0.000   Median :0.0000   Median :0.0000  
 Mean   :1998   Mean   : 8.294   Mean   :0.2742   Mean   :0.241   Mean   :0.2302   Mean   :0.2409  
 3rd Qu.:2003   3rd Qu.:13.000   3rd Qu.:0.5000   3rd Qu.:0.000   3rd Qu.:0.0000   3rd Qu.:0.0000  
 Max.   :2005   Max.   :14.000   Max.   :1.0000   Max.   :8.326   Max.   :8.3260   Max.   :8.3260  
     rain1           heatavg1     log_newtotalpercapitacons    cluster     
 Min.   : 89.21   Min.   :22.82   Min.   : 5.447            Min.   :  1.0  
 1st Qu.:137.86   1st Qu.:24.74   1st Qu.: 8.796            1st Qu.:212.0  
 Median :166.06   Median :25.73   Median : 9.272            Median :430.0  
 Mean   :182.23   Mean   :25.50   Mean   : 9.290            Mean   :434.4  
 3rd Qu.:214.88   3rd Qu.:26.24   3rd Qu.: 9.773            3rd Qu.:640.0  
 Max.   :389.98   Max.   :26.98   Max.   :14.189            Max.   :913.0  

plot(Threshweather2$wind12, type = "h", col = "blue", xlab = "Year", ylab = "Wind Speed", main = "Hurricane Intensity in Jamaica (1990-2005)")

The graph demonstrates how Jamaica’s wind patterns change over time, with some years having faster winds than others.

mrlp <- Threshweather2$wind12
mrlplot(mrlp, xlab="Exceedance Probability", main="Mean Residual Life Plot")

In this instance, it looks that the code has already produced a similar plot and determined a 1.5 as the threshold number. This means that for this analysis, extreme wind speeds that are greater than 1.6 times the wind speed distribution’s standard deviation are deemed to be relevant.

The threshold value of 1.6 is selected because there are sufficient data points above this level for a valid model to fit, and it appears to be the location on the plot where the distribution of excesses above the threshold becomes somewhat stable.

summary(Threshweather2$wind12)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  0.000   0.000   0.000   0.241   0.000   8.326 

Range of year

yr_rge<-range(Threshweather2$year)
yr_rge

[1] 1990 2005

## Number of years
totalyears<-yr_rge[2]-yr_rge[1]+1
totalyears

[1] 16

Number of events

eventsize<-dim(Threshweather2)
eventsize

[1] 15472    10

number_events<-num.events[1]
number_events/numberofyrs

[1] 967

ThreshWeat1<-fevd(Threshweather2$wind12, Threshweather2, threshold=1.6, type="GP", time.units="967/year")

Estimated Parameters

ci(ThreshWeat1, type="parameter")

fevd(x = Threshweather2$wind12, data = Threshweather2, threshold = 1.6, 
    type = "GP", time.units = "967/year")

[1] "Normal Approx."

      95% lower CI     Estimate 95% upper CI
scale   1.24306701 1.368630e+00   1.49419253
shape  -0.06879412 2.776641e-06   0.06879968

plot(ThreshWeat1, "qq")

The majority of the data points in the plot are in line with the trend line, indicating that the trend line is a good match for the data and that the frequency of extreme weather occurrences has a consistent pattern across time.

plot(ThreshWeat1, "rl")

return.level(ThreshWeat1, c(5,10,15))

fevd(x = Threshweather2$wind12, data = Threshweather2, threshold = 1.6, 
    type = "GP", time.units = "967/year")
get(paste("return.level.fevd.", newcl, sep = ""))(x = x, return.period = return.period)

 GP model fitted to  Threshweather2$wind12 Threshweather2 
Data are assumed to be  stationary 
[1] "Covariate data = Threshweather2"
[1] "Return Levels for period units in years"
 5-year level 10-year level 15-year level 
      9.53428      10.48296      11.03790 

The estimated return levels for the 5-year, 10-year, and 15-year periods are 9.53428, 10.48296, and 11.03790, respectively. These values represent the wind speeds that would be expected to be exceeded on average once every 5, 10, and 15 years, respectively, based on the fitted GP model and the threshold value of 1.5.

## Confidence Bands for individual return levels
ci(ThreshWeat1, type="return.level", return.period=15)

fevd(x = Threshweather2$wind12, data = Threshweather2, threshold = 1.6, 
    type = "GP", time.units = "967/year")

[1] "Normal Approx."

[1] "15-year return level: 11.038"

[1] "95% Confidence Interval: (9.3482, 12.7276)"

This indicates that, with a 95% confidence interval, the true value ranges between 9.3482 and 12.7276, we can anticipate wind speeds exceeding 1.6 times the standard deviation of the wind speed distribution to reach an average maximum speed of 11.038 over a 15-year period.

## Probability of Exceedance
probability<-pextRemes(ThreshWeat1, c(5,10,15), lower.tail=FALSE)
probability

[1] 8.338991e-02 2.160376e-03 5.597075e-05

Results

The analysis found that households with a higher proportion of children under 16 years old tend to have lower economic well-being. However, there is no clear relationship between economic welfare and rainfall or hurricane intensity. While higher hurricane intensity appears to be associated with lower economic well-being, this relationship is not statistically significant.

The model used was able to explain only a small amount of the variation in economic welfare, with an R-squared value of 0.127. Nevertheless, the overall model was deemed to be a good fit based on the significant F-statistic of 560.528.

Discussion

According to the regression results, hurricane intensity, as assessed by the wind_intense variable, appears to have a negative but statistically negligible effect on Jamaican economic welfare, as measured by log_newtotalpercapitacons. This shows that hurricanes may not have a significant influence on Jamaica’s general economic well-being in the short run. On the other hand, the proportion of children in households has a statistically significant negative effect on economic welfare, meaning that programs aimed at assisting families with children may have a positive effect on economic well-being.

Meanwhile, average temperature has a statistically significant positive influence on economic well-being, implying that policies focused at mitigating the negative effects of heat and developing climate resilience may have a positive impact on economic well-being. With an R-squared value of 0.127, the employed model was only able to account for a small portion of the variation in economic welfare. Nevertheless, based on the significant F-statistic of 560.528, it was determined that the overall model was a good fit.

The model’s output shows that it approximated normally and calculates a 15-year return level of 11.038 for wind speeds greater than the threshold of 1.6. (9.3482 to 12.7276) is the 95% confidence interval for this estimate. This indicates that, with a 95% confidence interval, the true value ranges between 9.3482 and 12.7276, we can anticipate wind speeds exceeding 1.6 times the standard deviation of the wind speed distribution to reach an average maximum speed of 11.038 over a 15-year period. Policymakers and other decision-makers can use these figures to better understand the possible effects of extreme wind events on the economy and make informed plans.

These results imply that programs intended to increase the financial security of households with a high proportion of children may be successful. It’s crucial to take into account how hurricanes can harm the economy and to put policies in place that could lessen such effects.

References

Nekeisha, Spencer and Eric Strobl. “Poverty and Hurricane Risk Exposure in Jamaica.” Geneva Risk and Insurance Review 47, no. 1 (03, 2022): 141-157. doi:https://doi.org/10.1057/s10713-021-00063-6. https://www.proquest.com/scholarly-journals/poverty-hurricane-risk-exposure-jamaica/docview/2635109071/se-2.

Baez, J.E., and I.V. Santos. 2007. Children’s vulnerability to weather shocks: A natural disaster as a natural experiment. New York: Social Science Research Network

Damage and loss assessments: 2007 Pioj report hurricane dean (2019) The Planning Institute of Jamaica. Available at: https://www.pioj.gov.jm/product/damage-and-loss-assessments-2007-pioj-report-hurricane-dean/ (Accessed: April 16, 2023).

PART 2

Title: An examination of the relationship between Weather Phenomena and Electoral Behaviour

INTRODUCTION

Weather and other environmental phenomena are generally consequential in their impact on industry, commerce and the decisions and behaviors of individuals in society. Likewise, politics plays a tremendous role in determining economic success among other factors within any given country. Thus, due to the fundamental nature of these variables, it is integral to uncover the nature of their connection if it so exists. According to Gomez et al, “when compared to normal conditions, rain significantly reduces voter participation by rate of less than 1% per inch(2007).” This information is quite alarming as a propensity of elections have been decided within that range. This behaviors however, also extends to other behavior such as voting choices where individuals have become influenced to support environmental policies or environmentally aligned candidates based on inclement or adverse. weather. In the paper, Brody et. al (2007), it is stated the perception of risk to climate change is influenced by one’s vulnerability to floods etc. This has clear implications for one’s voting behaviour as it pertains to willingness to support climate change policy initiatives. This research therefore, is motivated by the willingness to investigate the relationship between voting patterns, behaviour and support being impacted by severe weather conditions or events.

The research problem that is to be addressed is the impact of weather on electoral behavior. There is a large body of research looking at predictor variables for electoral decisions but not enough looking at this external factor and thus, there exists lack of concensus or agreement into the role of the weather on the influence of people’s voting choices. There exists a large body of data and research left to be done in this area including at the identification of the causation or correlation connection between the variables. Therefore, the research questions to be postulated and answered in regards to the topic include:

1. Does there exist a significant relationship between weather, to include temperature, precipitation or extreme phenomena and voter turnout pr participation?

2. What is the statistical relationship between weather and voting choices as it relates to candidates picked or party affiliations?

This topic offers unparalleled potential in the field of economics, social sciences and meteorology. If done correctly, it can encourage policy development by governments as well as strategies to improve the turnout of voters around election time and in particularly vulnerable areas. This would also provide benefits to political candidates who may or may not be adversely affected by these adverse conditions at the ballot box so as to boost their campaign strategies.

METHODOLOGY

In this study, the researcher will collect data on electoral behavior as well as weather phenomena. Data will be publicly sourced from the Electoral Commission database regarding voter turnout, Meteorological Institute of Jamaica for weather data among others that may be useful.

This would be compiled to create a panel data set looking at various variables over several time periods. In this research, we would be looking at employing a regression model since it is used to analyze the relationship between the weather phenomena and electoral behavior. To account for their effects, additional pertinent variables, such as socioeconomic and demographic ones, might be added to the model.

The model can be specified as follows:

\(\ Electoral Behavior = \beta_0 + \beta_1 Hurricane{it} + \beta_2 Rainfall{it} + \beta_3 Temperature{it} + \beta_4 Age{it} + \beta_5 Income{it} + \epsilon_{it}\)

Where:

β0 represents the intercept of the model, which represents the expected value of the dependent variable when all the independent variables are equal to zero. β1 to β3 are the coefficients of the weather variables, representing the effect of each weather variable on the dependent variable. The term “error” represents the variance of the dependent variable that cannot be accounted for by the independent variables.

Because they can also affect voting behavior, demographic and socioeconomic factors should be included in the model.

An innovative idea for a variable that can be included in the model could be social media engagement related to weather events. Social media is known to influence discourse and therefore has an impact on people’s opinions. The effect of public perception and sentiment toward weather occurrences on election behavior can be captured by this variable.

Finding the coefficient values that best fit the data is the estimating process in a regression model. This is often accomplished by reducing the sum of the squared errors between the dependent variable’s actual values and its anticipated values as determined by the model.

Depending on the particular model being used, the estimating process might be iterative and may incorporate different techniques such maximum likelihood estimation, ordinary least squares, or generalized least squares.

The statistical program will offer data such as the estimated coefficients, standard errors, t-statistics, and p-values once the model has been estimated.

A critical stage in ensuring the validity and dependability of the model’s results is model evaluation. The model can be evaluated in a variety of ways, such as by examining the quality of fit and statistical significance.

DISCUSSION

Voting is considered a corerstone of democracy. In the literature, scientists and researchers, however, have found difficulty determining the rationale behind the patterns seen in voting and how it varies extensively by geography and socially(Fujiwara et. al, 2016). The findings indicate that weather events, notably temperature, precipitation, and hurricanes, have a major impact on electoral behavior. This finding is consistent with previous research published in the literature. The literature in Cohen (2012) speaks to temperature having impact on abstentions of legislators in their political duties. Similarly, the effect would be similar to those experienced by constituents who would elect them.

The regression analysis would find that an increase in temperature is associated with a decrease in voter turnout, with a unit increases in temperature leading to a decrease in turnout. This finding supports prior research that has found a negative relationship between temperature and voter turnout (Gomez et al, 2007).

The findings of the regression analysis will shed light on how election behavior and meteorological occurrences are related.

The findings are anticipated to reveal a statistically significant correlation between particular meteorological occurrences and electoral behavior. For instance, prior studies have demonstrated that severe weather conditions, such as hurricanes, can have a detrimental effect on voter attendance. As a result, it’s likely that the correlation between hurricane frequency and voter turnout will be negative, showing that hurricane frequency is positively correlated with voter turnout.

Overall, the findings of this study add to the body of evidence supporting the link between meteorological events and voting patterns. The findings have significant ramifications for decision-makers and election officials who must take weather into account while organizing and conducting elections. Election administrators might, for instance, take into account alternate voting procedures like mail-in ballots during severe weather to prevent a detrimental influence on voter turnout. Similar to this, campaigns might utilize customized messaging to boost voter turnout and support for their candidates during specific weather conditions.

CONCLUSION

In conclusion, this study offers proof that there is a meaningful connection between weather occurrences and voting behavior. The findings show the importance of taking the influence of weather on voter turnout and candidate choice into consideration for political campaigns and policymakers. Future studies should delve more into the ways in which the weather influences voting behavior, including the function of media coverage or voter views.

References

Brody, Samuel & Zahran, Sammy & Vedlitz, Arnold & Grover, Himanshu. (2008). Examining the Relationship Between Physical Vulnerability and Public Perceptions of Global Climate Change in the United States. Environment and Behavior. 40. 72-95. 10.1177/0013916506298800.

Cohen, Alexander. 2012. “Sweating the Vote: Heat and Abstention in the US House of Representatives.” PS, Political Science & Politics 45 (1): 74–77. https://doi.org/10.1017/S1049096511001727.

Fujiwara, Thomas, Kyle Meng, and Tom Vogl. 2016. “Habit Formation in Voting: Evidence from Rainy Elections.” American Economic Journal. Applied Economics 8 (4): 160–88. https://doi.org/10.1257/app.20140533.

Gomez, Brad T., Thomas G. Hansford, and George A. Krause. 2007. “The Republicans Should Pray for Rain: Weather, Turnout, and Voting in U.S. Presidential Elections.” The Journal of Politics 69 (3): 649–63. https://doi.org/10.1111/j.1468-2508.2007.00565.x.