We discuss our empirical results from the fixed-effects models and their implications. It is important to note that our results are based on a long time period (1970–2013). A short-term event such as the U.S. housing bubble during the mid-2000s, for example, would not have an impact on our results.
4.1 Empirical Results
The results of the state-specific and region-specific fixed effects models are presented in Tables
2 and
3. The analysis was conducted in R software (R Core Team
2016). The state/region-by-year fixed effects models were first run with all contemporaneous variables: GDP, economic damages, time trend, state dummies (with Alabama serving as the base/omitted state), and several socioeconomic variables—the political affiliation of the Senate (Republican as the base level), the proportion of the nonwhite population, and people aged over 65.
Table 2
State-specific fixed effects model results for the United States for the period 1970–2013
Intercept | −6.855 | 0.092 | 0.0000*** | NJ | 0.0011 | 0.0045 | 0.8091 |
AZ | 0.0063 | 0.0045 | 0.1592 | NM | 0.0219 | 0.0045 | 0.0000*** |
AR | 0.0127 | 0.0045 | 0.0047** | NY | 0.0199 | 0.0049 | 0.0000*** |
CA | 0.005 | 0.0056 | 0.3711 | NC | −0.0137 | 0.0045 | 0.0024** |
CO | 0.0096 | 0.0045 | 0.0327* | ND | 0.0143 | 0.0045 | 0.0015** |
CT | 0.0296 | 0.0045 | 0.0000*** | OH | −0.0325 | 0.0045 | 0.0000*** |
DE | −0.0146 | 0.0045 | 0.0012** | OK | 0.0177 | 0.0045 | 0.0000*** |
DC | 0.033 | 0.0046 | 0.0000*** | OR | −0.0041 | 0.0045 | 0.3602 |
FL | 0.0354 | 0.0046 | 0.0000*** | PA | −0.0159 | 0.0046 | 0.0000*** |
GA | 0.0097 | 0.0045 | 0.0316* | RI | −0.0105 | 0.0045 | 0.0205* |
ID | 0.0203 | 0.0045 | 0.0000*** | SC | −0.0113 | 0.0045 | 0.0113* |
IL | 0.0001 | 0.0046 | 0.9750 | SD | 0.0415 | 0.0045 | 0.0000*** |
IN | −0.0198 | 0.0045 | 0.0000*** | TN | 0.0057 | 0.0045 | 0.0205 |
IA | −0.0135 | 0.0045 | 0.0027** | TX | 0.0238 | 0.0048 | 0.0000*** |
KS | 0.0031 | 0.0045 | 0.4859 | UT | −0.0024 | 0.0045 | 0.5895 |
KY | −0.004 | 0.0045 | 0.3687 | VT | −0.0086 | 0.0045 | 0.0585 |
LA | 0.0225 | 0.0045 | 0.0000*** | VA | −0.018 | 0.0045 | 0.0000*** |
ME | −0.0187 | 0.0045 | 0.0000*** | WA | −0.0121 | 0.0045 | 0.0074** |
MD | −0.0184 | 0.0045 | 0.0000*** | WV | −0.0263 | 0.0045 | 0.0000*** |
MA | 0.0027 | 0.0045 | 0.5439 | WI | −0.0202 | 0.0045 | 0.0000*** |
MI | −0.0147 | 0.0045 | 0.0012** | WY | 0.0343 | 0.0045 | 0.0000*** |
MN | −0.0079 | 0.0045 | 0.0790 | GDP | 3.18E−08 | 3.68E−09 | 0.0000*** |
MS | 0.0199 | 0.0045 | 0.0000*** | Time | 0.0037 | 0.0001 | 0.0000*** |
MO | −0.0016 | 0.0045 | 0.7166 | Loss | 2.77E−13 | 9.84E−14 | 0.0049** |
MT | 0.0376 | 0.0045 | 0.0000*** | Nonwhite | −2.90E−12 | 1.85E−12 | 0.1161 |
NE | 0.013 | 0.0045 | 0.0039** | Over65 | 0.0051 | 0.0082 | 0.5334 |
NV | 0.0322 | 0.0045 | 0.0000*** | Senate | −.001 | 0.001 | 0.2873 |
NH | −0.0153 | 0.0045 | 0.0000*** | | | | |
Significant codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1 |
R squared | 0.873 | | | Akaike info criterion | −10485.32 | | |
Adj. R squared | 0.870 | | | Schwarz criterion | −10167.47 | | |
F statistic | 266.9 | | |
p value | 2.1E−16 | | |
Table 3
Region-specific fixed effects model results for the United States for the period 1970–2013
Intercept | −6.766 | 0.0949 | 0.0000*** |
Northwest | 0.0039 | 0.0024 | 0.1126 |
South | 0.0168 | 0.0019 | 0.0000*** |
Southeast | 0.0001 | 0.0019 | 0.9924 |
Southwest | 0.0113 | 0.0022 | 0.0000*** |
West | 0.0189 | 0.003 | 0.0000*** |
Northern rockies and plains | 0.0303 | 0.0021 | 0.0000*** |
Upper midwest | −0.012 | 0.0022 | 0.0000*** |
Ohio valley | −0.0098 | 0.0018 | 0.0000*** |
GDP | 3.56E−08 | 0.00E+00 | 0.0000*** |
Senate | −0.0028 | 0.0011 | 0.0137* |
Time | 0.0037 | 0.0001 | 0.0000*** |
Loss | 5.09E−13 | 1.14E−13 | 0.0000*** |
Over65 | 0.0438 | 0.0089 | 0.0000*** |
Nonwhite | −2.03E−2 | 2.17E12 | 0.0000*** |
Significant codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1 |
R squared | 0.817 | | |
Adj. R squared | 0.816 | | |
F statistic | 683.5 | | |
Akaike info criterion | −9783.8 | | |
Schwarz criterion | −9693 | | |
p value | 2.2E−16 | | |
The results in Table
2 indicate that the coefficients of state dummies for Arkansas, Colorado, Connecticut, Delaware, District of Columbia, Florida, Georgia, Idaho, Indiana, Iowa, Louisiana, Maine, Maryland, Michigan, Mississippi, Montana, Nebraska, Nevada, New Hampshire, New Mexico, New York, North Carolina, North Dakota, Ohio, Oklahoma, Pennsylvania, Rhode Island, South Carolina, South Dakota, Texas, Virginia, Washington, West Virginia, and Wyoming are all statistically significant. While the coefficients for GDP, time trend, and damages are all statistically significant, the coefficients for proportion of nonwhite people, proportion of people age 65 +, and type of Senate are not statistically significant. In this model, the dependent lagged variable for the Gini coefficient is statistically significant indicating that the Gini coefficient at time
t is highly related to the Gini coefficient at time
t −
1. A positive time trend coefficient suggests that there is an autonomous increase in income inequality over time, with the Gini coefficient increasing by 0.0037 per year, after allowing for the effects of all other predictors in the model. The coefficient for economic damages (both crop and property damages) is 2.77E−13 for the period 1970 to 2013, indicating that an increase in economic damages by USD 100 billion leads to an increase in the Gini coefficient of about 2.77%.
The coefficient for GDP indicates that an increase of annual GDP by USD 1000 billion leads to an increase in the Gini coefficient of about 0.032 or 3.2%. The unit of GDP is million USD. It means that growth in nationwide income leads to a wider spread of income distribution and increases the gap of income inequality, in spite of the fact that the Gini coefficient is relatively inelastic (Miljkovic and Miljkovic
2014). This result is actually consistent with the suggestion Majumdar and Partridge (
2009) made that the increase in economic inequality is coupled with the national economic growth across the United States and that the poor become poorer. The last three coefficients in the model are non-significant. The value of the adjusted
R squared in Table
2 is 0.873, which indicates that the model fits well. The value of the
F statistic indicates that the model is useful for the prediction.
Overall, the results obtained from the state model further confirm the idea that natural hazard-induced disasters increase income inequality not only for hurricane states (Miljkovic and Miljkovic
2014), but also for many other states across the United States. Similar results are also noticeable for the extended time period studied here, so the trend continues.
Table
3 shows the results for the region-specific fixed effects model with the time dependent lag variable. Results imply that between 1970 and 2013, the occurrences of natural hazard-induced disasters in all regions except the Northwest widened the income gap, so the Gini coefficients are larger. Therefore, the coefficients of all regional dummies for these regions are statistically significant except the one for the Northwest region. The Northeast region was chosen by R software to be the omitted region (base level), but any other region could equally serve that purpose, as the results would not be changed qualitatively.
The statistically significant coefficients are reported for GDP, economic damages, proportion of people aged 65 and over, the political makeup of the U.S. Senate (Republican versus Democrat), and time trend. The coefficient for the proportion of nonwhite people is not statistically significant.
The coefficient for GDP indicates that an increase of annual GDP by USD 1000 billion will lead to an increase in the Gini coefficient of about 0.036 or 3.6%. This result is in line with the result reported for the state-level fixed effects model. The region-specific fixed effects model suggests that a 1% increase in the proportion of the population aged over 65 leads to a 4.4% increase in the Gini coefficient. This result is consistent with the suggestion made by Anbarci et al. (
2005) that the losses from natural hazard-induced disasters influence the existing income inequalities more pronouncedly, especially for women, the young and elderly, and people of ethnic or racial minorities. Hence, the growth of the elder population proportion will automatically lead to a further increase in income inequality. In the region-specific model, income inequality seems not to be directly related with the proportion of nonwhite people.
The makeup of the Senate seems to be an important predictor of the Gini coefficient in the region-specific model. The base level for the Senate variable is Republican. Some people argue that although it seems that the political balance in the U.S. Senate has important implications on income distribution, in reality, there is a minimal relationship between the rise in income inequality and class-based voting (Gelman et al.
2010). The
p value for this variable is 0.0137, indicating a significant relationship between political party and income inequality. Holding other variables constant, the Gini coefficient decreases by 0.003 or 0.3%, if the majority of people in certain states vote for a Democratic Senate. The region-specific fixed effects model is a good fit based on the reported
R
2 being equal to 0.817. The model is also useful in the prediction as
F = 683.5 with
p value = 2.2E−16.
4.2 Discussion and Implication
The objective of this article is to confirm that economic damages caused by all kinds of catastrophic events further increase income inequality, not only in hurricane-affected areas, but also across the United States as a whole. The results of our study have important implications for the insurance industry and policymakers who stand in a position to provide compensation to those in affected areas.
By looking at Fig.
1, relatively small changes in Gini coefficients are shown for each state in 1970, indicating small variations in income distribution across the country. The only state that stands out with a Gini coefficient above 0.5 is Idaho. However, two decades later, more states have relatively greater income inequality. States such as California, Texas, and Florida had Gini coefficients above 0.6. In 2013, states with more income inequality are mostly located along the two coasts. States such as California, Florida, Nevada, and Wyoming all had Gini coefficients that were roughly larger than 0.65. By looking at the transitions between the three maps, an overall increase in Gini coefficients is expected. It is interesting to see the result because it affirms that as the nation develops, chances are technological innovations made by certain individuals or companies would further increase their own wealth, thus increase overall income inequality.
A vital takeaway from this article is that economic damages due to all kinds of natural hazards, together with GDP, time trend, and political parties serve to predict income distribution across the United States. It is normal that many people link income equality to how our economy performs, which is the GDP value for each year. But because the two major parties can respond to natural hazards differently, it is interesting to see income inequality slightly related to whether the government is primarily represented by one or the other party. In Fig.
2, there is an overall increasing trend for Gini coefficients for both parties, indicating an overall increase in income inequality. Also, there is not much difference between Gini coefficients for Republican versus Democrat before 2000. However, there has been a staggering increase in terms of Gini index whenever the majority of people in certain states have voted Republican since the early 2000s, and the gap reached its maximum in the mid-2000s. Since a higher Gini index represents larger income inequality, Fig.
2 (right) shows that on average the prevalence of income inequality is lower in Democratic states.
Natural hazard-induced disasters cause property damages, which in turn increase income inequality. The insurance industry, together with the government, could think of methods to help ease the pressures people face in impacted areas. It is likely that consequences associated with income inequality, such as high unemployment rate and crime rate, would in turn be alleviated.
By analyzing the past, we are able to have a broad understanding as to how income distribution has been affected by extreme weather conditions across the United States. The result will be critical for all parties, including policymakers, local industries, insurance companies, and residents.