Also, there were several factors which made a straightforward north-south relationship between storm zone intensity and proportional economic loss risk elusive to identify. First, the storm zones, although generally aligned north to south, are not static. For Instance, zones 2 and 3 curve around peninsular Florida, intersecting with part of the southern coast in western Florida. Therefore, parts of northwestern Florida are subject to greater proportional economic loss risk compared areas of the same latitude in northeastern Florida.

         Regarding the histogram results relating proportional economic loss risk with the different zones of tropical storm intensity, the mode of the gridcodes was lowest in central Florida. This may have resulted for several reasons. Perhaps central Florida does not have a lot of property that can be damaged, because there might be fewer people living there. Maybe this area is mainly covered by nature reserves or swamplands, relative to what is typical in the rest of the state. Clearly, these issues should be followed up on.

         Next, regarding proportional economic loss risk, it may be that the gridcode mode was higher in non-urban areas because of confounding variables. As an example, a study indicated that there was a correlation between southern Baptist ministers and liquor stores (Lecture). However, rather than both variables being linked to one another, both are instead correlated to a third variable which affect both, such as the size of the city, or the fact that both are in the south, and where its hot, people get thirsty. Thus, it may be that my three hypotheses are not necessarily incorrect, but rather that there are confounding variables which have yet to be explored.

         The effect of damage from storm surges was meant to be captured by only analyzing urban areas of Florida within a set distance from the shore, although this time constraints did not permit the completion of this aspect of the analysis. However, even under a category 5 hurricane, storm surges reach only reach 18 or more feet above normal sea level (Grotzinger et. al., 2007, pg. 483). Therefore, had separate histograms been completed comparing urban areas under 18 feet near the coast with urban areas of all elevation away from the coast (outside a buffer), one might be able to identify the impacts of damage as a result of winds and storm surge separately. Perhapse such operations can be included in future studies

         Measuring mortality risk is a challenging task for researchers. As discussed by Combs et. al., there are several important decisions which go into classifying mortality statistics which can effect the the mortality risk of an area. For instance, not all deaths which are used to calculate mortality risk may have been as a direct or immediate result of Hurricanes. Rather, accidental deaths, such as from falls from ladders while storm-proofing homes, may or may not be included, depending on the way the data was collected. Imagine the deaths which could occur from traffic accidents from people fleeing an area, or being killed by structurally damaged buildings after the fact. Furthermore, it was unknown to what extent the mortality risk data layer took into account the effect of evacuation of an urban area prior to hurricanes making landfall. One would expect that relative mortality risk would be higher in urban areas (as well as non-urban ones) in which there are large populations with low percentage of evacuees. Therefore, data on absolute population and the proportion of evacuees might be useful to examine alongside relative mortality risk.

         There is also the issue of error which must be discussed. Some error may have been unavoidably induced, such as when data projections had to be transformed so that all data was being displayed under a single projection. For example, some layers were originally meant to be displayed using a Mercator projection, but had to be transformed into WGS 1984.

         Also, some data for mortality risk, proportional economic loss risk and frequency risk was not present for certain areas of Florida. As a result of this missing data, some of the statistical analysis, such as measuring the amount of clustering and its significance may have been affected. Missing data means that the histograms, scatter plots and box plots do not represent a complete picture of all of Florida and it urban and non-urban areas.

         All of the operations performed in ArcGIS may have inadvertently induced error into the resultant layers. One operation which may have had the greatest impact on the results of this project was converting the raster files for mortality, proportional economic loss risk, and frequency risk into shape files. This was necessary in order to perform my analyses involving clustering, as well as for assessing the relationship between urban and non-urban areas and high and low elevations and mortality risk, proportional economic loss risk and frequency risk. However, despite these qualifications, I am concerned regarding the division of the gridcodes within Florida. At first, the shape files appeared fine, and my analyses proceeded unencumbered. Yet, I eventually noticed that some polygons were larger than other, and then realized that raster cells of the same gridcodes were aggregated, rather than retaining their own separate polygons.

         The implications, I fear, are that certain gridcodes are underrepresented, those which were grouped into large polygons. Thus, instead of my histograms and scatter plots returning many gridcodes of the same class, only a single entry is recorded, and over a larger geographical area. This also compounds the problems I have regarding my elevation distribution data. Where there might have been the same gridcode classes grouped sided by side, but covering a varying topography, instead, a large identically gridcoded polygon was produced, for which the average elevation was calculated. Thus, the histogram, scatter plot and box plot outputs may not depict the true distributions of risk.

         Other error was purposefully done in order to generalize the data. For instance, for the spatial join operation, I designated ArcGIS to calculate the average elevation in each risk polygon. Therefore, any differences in elevation within a single polygon were lost, and the surface flattened. Selecting a different option for calculating elevation may have produced different elevation results. Thus, the results involving the distribution of mortality risk and proportional economic loss risk gridcodes over elevation of urban areas are not definitive. Other generalizations included adjustments to legends for the maps involving elevation data, since it was not appropriate to include elevation data to the sixth decimal point. Thus, I rounded to the nearest tenth of a foot.