Table 1. Statistical modeling of the number of freshwater related residential flood insurance claims using a zero-inflated negative binomial model.
| Estimate | Robust Standard Error | [95% Confidence Intervals] | |
|---|---|---|---|
| Minor flooding | 0.921*** | 0.104 | [0.718; 1.125] |
| Moderate flooding | 1.579*** | 0.120 | [1.344; 1.814] |
| Major flooding | 2.493*** | 0.124 | [2.250; 2.736] |
| LnResPol | 0.896*** | 0.044 | [0.809; 0.983] |
| LnHousing | −0.148*** | 0.042 | [−0.231; −0.066] |
| PropLowRisk | −0.183 | 0.336 | [−0.842; 0.476] |
| PropMedRisk | 0.090 | 0.340 | [−0.576; 0.756] |
| PropHighRisk | 0.623* | 0.330 | [−0.024; 1.269] |
| YrDummy2002 | −1.079*** | 0.213 | [−1.496; −0.662] |
| YrDummy2003 | −0.329 | 0.371 | [−1.056; 0.398] |
| YrDummy2004 | −4.058*** | 0.212 | [−4.473; −3.643] |
| YrDummy2005 | −3.304*** | 0.212 | [−3.739; −2.870] |
| YrDummy2006 | −2.884*** | 0.321 | [−3.512; −2.260] |
| YrDummy2007 | −2.475*** | 0.283 | [−3.031; −1.920] |
| YrDummy2008 | −1.446*** | 0.170 | [−1.779; −1.113] |
| YrDummy2011 | −1.333*** | 0.192 | [−1.709; −0.958] |
| YrDummy2012 | −0.464** | 0.194 | [−0.844; −0.085] |
| TS | −2.400*** | 0.148 | [−2.689; −2.110] |
| MajorHurr | 2.298*** | 0.113 | [2.077; 2.518] |
| CoastalState | 0.029 | 0.081 | [−0.129; 0.187] |
| Impervious | 0.700*** | 0.182 | [0.344; 1.055] |
| Miles25-100 | −0.561*** | 0.095 | [−0.748; −0.375] |
| Miles100-500 | −0.593*** | 0.103 | [−0.796; −0.390] |
| Miles500 + | −1.422*** | 0.367 | [−2.146; −0.697] |
| Intercept | −1.248*** | 0.398 | [−2.028; −0.468] |
| Inflate | |||
| MaxFR | −1.436*** | 0.032 | [−1.498; −1.373] |
| LnResPol | −0.120*** | 0.010 | [−0.139; −0.010] |
| Intercept | 2.427*** | 0.063 | [2.303; 2.549] |
| Ln(Dispersion α) | 1.517*** | 0.041 | [−1.437; −1.596] |
| Dispersion α | 4.555*** | 0.185 | [4.207; 4.932] |
There are 150,546 total observations, of which 6,631 are non-zero. The predictors are a community’s distance from the coastline (“Miles25–100”, “Miles100–500”, “Miles500+”, where miles 0 to 25 is the omitted dummy variable category), whether it is located within a coastal state (“CoastalState”), its percentage of impervious surface (based upon 2006 values; “Impervious”), the proportion of the NFIP community in low, medium or high risk areas (“PropLowRisk”, “PropMedRisk”, “PropHighRisk”), the natural log of the number of housing units (based on the 2010 values; “LnHousing”), and the natural log of the number of NFIP residential policies-in-force in the community per year of the event (“LnResPol”). The flood ratio is transformed into a dummy variable, with bins of a flood ratio below 1 is for bankfull conditions; a flood ratio between 1 and 1.5 represents minor flooding conditions, while values between 1.5 and 2.2 and larger than 2.2 are indicative of moderate and major flooding, respectively, where bankfull is the omitted dummy variable category. We also control for any unobserved event-specific fixed effects through event dummy variables represented by the TC intensity at landfall – tropical storm, hurricane, and major hurricane with hurricane the omitted category (“TS”, “MajorHurr”). We also include year variables to control for any unobserved time-specific fixed effects with 2001 the omitted category (“YrDummy”). The zero-inflated part of the model uses as predictors the largest flood ratio for a given community (“MaxFR”) and the natural log of the number of NFIP residential policies-in-force in the community per year of the event (“LnResPol”). The results for the coefficient of dispersion α are also reported, highlighting that the data are overdispersed (α larger than 0). The symbols “***”, “**”, “*” refer to coefficients that are different from zero at the 0.01, 0.05, and 0.10 significance level, respectively. Consult the Methods section for more details on the statistical estimation employed.