Sinkholes have generated significant damage to buildings,
roads, and other human-built structures, and should be
considered natural hazards in their own right. In
sinkhole-prone areas where market insurance against sinkhole
damage is available, economic theory suggests that homes
located there should be valued somewhat lower than homes
located in areas where sinkholes are rare or nonexistent, in
recognition of both the risk faced by the homeowner in a
sinkhole-prone area, and the cost of insuring one's property
against that risk. Using sinkhole and Census data from the
Tampa Bay, Florida region in 1990, this paper finds no
evidence of a statistically-significant relationship between
the presence of sinkholes in a neighborhood, and the value of
homes in that neighborhood.
72 Are Home Values Affected by Sinkhole Proximity? Results of a Hedonic Price Model E. Spencer Fleury Department of Geography University of South Florida Introduction With its limestone bedrock, wa rm climate and high precipitation levels, Florida provides near-i deal conditions for sinkhole development. Additional contributing factor s in certain areas include high rates of urbanization and overpumpi ng of groundwater to meet agricultural demand. While sinkholes can be found throughout Florida, distribution is not even across the state, with the highest concentrations occurring in the west-central region, to the nor th and east of Tampa Bay (USGS, 1999). Though they lack the high profil e and sheer destructive force of hurricanes, floods, and other na tural hazards, sinkholes have on occasion generated significant damage to buildings, roads, and other human-built structures, and should be considered natural hazards in their own right. In sinkhole-pron e areas where market insurance against sinkhole damage is availabl e, economic theory suggests that homes located there should be valued somewhat lower than homes located in areas where sinkholes are rare or nonexistent, in recognition of both the risk faced by th e homeowner in a sinkhole-prone area, and the cost of insuring oneÂ’s property against that risk. Working with sinkhole and Census data from the Tampa Bay, Florida region in 1990, this paper uses a hedonic price model to look for a statistically significant relationship between the presence of sinkholes (and, in a separate set of regression s, the density of sinkholes) in a neighborhood and the value of homes in that neighborhood. The model did not find evidence of either type relationship. Background The decision to use 1990 data for this analysis has its roots in a policy decision made in 1991. In 1 969, the state of Florida put into
The Florida Geographer 73 place a reinsurance facility that coul d cover the risk of property loss to sinkholes. At that point, Florid a homeowners in sinkhole-prone areas had two options: they could either purchase sinkhole insurance, or they could gamble that their property would not be damaged by sinkholes. Either way, the risk of living in a sinkhole-prone environment was borne entirely by the homeowner. However, very few people purchased this optional availa ble coverage (Maroney, et al, 2005), and in 1991 the Florida Statutes we re amended to automatically include sinkhole coverage in every homeownerÂ’s insurance policy, at no specified additional cost (Eas tman, et al, 1995). Even though a 1993 study found that the problem of sinkhole losses was largely confined to the Tampa Bay area, the amended statutes made no distinction between homeowner policies issued for sinkhole-prone areas, and those issued for parts of the state where sinkholes were all but unknown (Maroney, et al, 2005). Though it was probably not the inte nt of the Florida Legislature to do so, by passing this piece of legislation lawmakers actually enacted a mechanism to encourage people to engage in a higher-risk behaviorÂ—purchasing a home in a sinkhole-prone areaÂ—while the full costs of those behaviors are distributed among people who choose not to engage in that same higher-risk behavior. Because homebuyers will be forced to pay for sinkhole in surance (a cost which is undoubtedly built into the price of each policy), they have no incentive to minimize their own risk by moving to an area where sinkholes are less likely to damage their property. The distribution of sinkhole risk was therefore altered and is now shared by those homeowners who face little to no risk of sinkhole dama ge, but are still obligated to pay into the insuran ce pool. Thus, the cost of living in sinkhole-prone areas is artificially lowered, which means that homebuyers are theoretically more likely to relocate there than they otherwise may have been. Methodology and data In real estate economics, hedoni c regression models are often used when a researcher wishes to control for the value of amenities such as square footage, number of bedrooms, and location, among others. For this reason, hedonic mode ls are frequently used to exam-
74 ine questions related to the impact of various hazards on home prices (see Nourse, 1967; Palm, 1981; Brookshire, et al, 1985; Tobin and Montz, 1994; Kiel and McClain, 1995; Dale, et al, 1999; among others). Results obtained using hedonic regression models often contradict those of other studies using di fferent models, which suggests that the specific nature of the hazard may be a crucial factor. A search of the literature did not turn up hedonic studies of any potential relationship between sinkholes and home prices. Data sources used for this analysis were the Florida Geological SurveyÂ’s sinkhole database and the 1990 US Census. Blockgroup-level Census data (described be low) for the four counties of the Tampa Bay area (Hernando, Hillsbor ough, Pasco and Pinellas) were used. All sinkholes reported in the four-county area between 1964 and 1990 were included in this analysis; sinkholes reported after 1990 or observations lacking a reporti ng year were dropped. Sinkhole locations were entered into ArcGIS, an d were linked to the block group in which they were located. Ordinary least squares (OLS) and Probit regressions were then run in orde r to characterize any potential relationship between sinkhole location and home values. These regressions were run once using regionwid e data; the OLS regressions were also run once for each county. The OLS and Probit regressions actually examine different, yet still related, questions. The OLS regressions investigate the relationship of home values to sinkhole density within each block group; the Probit regression instead focuses on the mere presence of sinkholes in a block group, with no adjust ment for either the number of sinkholes or the geographic size of the block group. The median home value variable is included in both the OLS and Probit regressions, though as the dependent variab le in the former and as an explanatory variable in the latter. Figu re 1 illustrates the distribution of median home values across the region in 1990. In addition to median home value, the following variables are included in the model: Sinkhole density : The value for this variable is derived from dividing the number of sinkholes reported in each
The Florida Geographer 75 block group by the area of the block group. Block group sizes vary widely across the region; this variable was created as a method of mitigating this problem. It is also the Figure 1. Median housing value Tampa Bay, Florida (1990).
76 key explanatory variable in the OLS model. Sinkhole density across the Tampa Bay re gion is shown in figure 2. Figure 2. Sinkhole density, Tampa Bay, Florida (1990).
The Florida Geographer 77 Number of bedrooms : Because median s quare footage data were not available, aggregate counts of homes organized by number of bedrooms were includ ed to serve as proxies for home size. Nonwhite population : Neighborhoods with significant nonwhite populations can have lower property values than whitedominated neighborhoods; however, such a relationship is by no means inevitable (Palmore, 1966; Boston, et al, 1972). This variable is included as a mean s of separating out any racebased home value disparities that may occur. The variable is formatted as a percentage. Median household income : Neighborhoods in which residents are wealthier tend to have highe r property values. It is possible to make a causal argument in either direction (are the high property values the result of the wealth of the neighborhoodÂ’s residents, or are wealthier re sidents attracted by the higher property values); however, the ex act nature of the relationship between income and home pri ces is not relevant here. Vacancy rates : This variable is formatted as a percent of each block groupÂ’s housing stock th at was vacant in 1990. This variable is included as a mean s to identify block groups with large numbers of abandoned or empty houses. Intuitively, we would expect block groups with higher vacancy rates to have less demand for residential pr operty, which should have a negative impact on housing value. Median structural age : As homes age, their values generally decline relative to newer hom es. However, because new construction can lead directly to the formation of new sinkholes (White, 1988; Patton and DeHan, 1998; Soriano and Simon, 2002), it is difficult to predict beforehand how this variable will interact with th e rest of the data. Sinks_present : This is a binary indicat or variable used as the dependent variable in the Prob it regression. It is not included in the OLS regression. Its value is 1 for block groups where a sinkhole had been reported prior to 1991, and is 0 for block
78 groups with no reported sinkholes. Total housing units: This is included as an explanatory variable in the Probit regression. It is included as a means of controlling for the greater likeli hood of sinkhole reporting in block groups with higher populations, as well as the possibility of new sinkhole generation brought on by higher levels of new construction in growing areas. This analysis makes use of a level of aggregation that some readers might find troubling. Spec ifically, both sinkhole occurrence and median home values are measured at the Census block group level; some might ask why the actual sale prices of individual homes were not plotted and mapped in rela tion to the nearest reported sinkhole. While this almost certainl y would have been the preferred method of proceeding, data limitations forced this approach. For one thing, home sale data in Hillsboro ugh County is no longer available for years prior to the late 1990s; for another, the FGS sinkhole database has always depended on volun tary reporting of sinkholes, and therefore suffers from a certain l ack of comprehensiveness (the 1990s in particular were lean years for the database, as funding for database maintenance dried up for much of th at decade). These two factors led to the development of the methodology used here, one that is not as precise as the ideal method but can still tell us something about the relationship between sinkhol e density and home values. Results Regionwide, both the OLS (table 1) and the Probit regressions (table 2) generated statistically si gnificant results; how ever, in no case were any of the variables of intere st significant. While median home value in a given block group shows a statistical relationship to every other explanatory variable in the mo del, there does not seem to be a connection between median home va lues and sinkhole densities. This suggests that the discounting predicted by economic theory did not occur here. (Some of the other expl anatory variablesÂ—in particular, the vacancy rateÂ—did not generate the kinds of results we might have expected before running the regr essions. And while these results merit examination, they go beyond the scope of this paper, and thus
The Florida Geographer 79 will not be discussed here.) Table 1. Regionwide OLS estimates. Dependent variable: median home value. Variable Coefficient t-statistic const 34459.4 5.9670 Sinkhole density -1004.23 -0.9534 Median household income 2.69503 28.8575* Vacancy rate 521.565 5.0075* Nonwhite population (pct. of total) -114.312 -2.9157* Homes with one or two bedrooms (pct. of total) -275.316 -4.5196* Homes with three or four bedrooms (pct. of total) -535.285 -8.8066* Homes with five or more bedrooms (pct. of total) 2139.38 5.9000* Median age of housing units -234.38 -3.4488* n = 1483 Adjusted R2 = 0.566116 *significant at p=0.05 Table 2. Probit estimates us ing binary dependent variable sink_present Variable Coefficient t-statistic const -0.186665 -1.2109 Total housing units 0.000214976 3.9162* Median home value -1.66425e-06 -1.5712 Median age of homes -0.0418238 -8.8272* n = 1491 Akaike information criterion (AIC) = 1176.9 Schwarz Bayesian criterion (BIC) = 1198.13 McFaddenÂ’s pseudo-R2 = 0.113904 *significant at p=0.05
80 A Probit regression, usin g the binary variable sinks_present as its dependent variable, was run in order to generate a second set of results to compare with the OLS results. The Probit regression offered Table 3. County-by-county OLS estimates. Dependent variable: median home value. Hernando Hillsborough Variable Coeff. t-statistic Coeff. t-statistic Constant 31735.2 1.1587 55360.2 5.9406 Sinkhole density -3099.19 -1.1186 741.351 0.4497 Median household income 2.29319 5.4061* 2.4366 18.1359* Vacancy rate 345.571 1.4545 186.415 0.8851 Nonwhite population (pct. of total) 36.8033 0.1801 -175.021 -2.8967* Homes with one or two bedrooms (pct. of total) -213.233 -0.8374 -373.12 -3.2903* Homes with three or four bedrooms (pct. of total) -7.96901 -0.0298 -720.688 -7.5710* Homes with five or more bedrooms (pct. of total) 1217.25 0.7610 1817.31 3.1781* Median age of housing units -1036.7 -3.2087* -25.8778 -0.2473 n = 79 n = 696 Adj. R2: 0.509 Adj. R2: 0.523 Pasco Pinellas Variable Coeff. t-statistic Coeff. t-statistic Constant 46384.1 2.6189 4241.16 0.5373 Sinkhole density -977.498 -0.8833 -2155.6 -0.6799 Median household income 1.1523 3.2194* 3.7057 20.9963* Vacancy rate 517.546 2.5159* 820.553 5.7169* Nonwhite population (pct. of total) -281.261 -2.0141* 8.95306 0.1567 Homes with one or two bedrooms (pct. of total) -69.4777 -0.4504 -218.262 -2.6124* Homes with three or four bedrooms (pct. of total) 213.011 1.1446 -636.588 -6.3491* Homes with five or more bedrooms (pct of total) 423.644 0.3693 1925.19 3.9087* Median age of housing units -1035.5 -4.7487* -363.374 -3.3427* n = 148 n = 568 Adj. R2: 0.705 Adj. R2: 0.704 *significant at p=0.05
The Florida Geographer 81 no evidence contradicting the result s of the OLS regression. Here, the median structure age and total housi ng units variables are both statistically significant and show the expected signs: positive for total housing units, negative for median st ructural age. These results lend some support for the hypotheses that sinkhole reporting is tied to population (sinkholes are more likely to be reported in areas where there are more people to find them ). Median home valueÂ—our variable of interest in this regressionÂ— is not statistically significant at p=0.05, thus confirming the resu lts of the OLS regression. County-by-county regressions (table 3) were included here to account for the localized nature of housing markets; it seemed possible that any sta tistically significant re lationship between sinkhole density and home values could po tentially be obscured by examining the results only at the wider regional level. However, the results of the county-by-county regressions mirrore d those of the regionwide analysis. The only explanatory variable that was significant for each countywide regression was median hou sehold income, which displayed a positive correlation with the dependent variable in each case. Conversely, the only dependent variable to show no statistical significance in any of the countywide regressions was sinkhole density. Based on these results, it seems likely that there is no significant relationship between home valu es and sinkhole location or density at anything greater than an extremely localized scale. Possible explanations, and potentia l directions for future research The most obvious possible expl anation for th e results described here is that homeowners and homebuyers may not generally consider sinkholes to be a significa nt threat to their property. This possibility presents an obvious an d straightforward avenue for additional research, which could be addressed via surveys and focus groups of homeowners and potential homebuyers in sinkhole-prone areas. It is also possible that homebuye rs may not generally be aware of the locations of sinkholes. The most accurate information on sinkhole locations is often proprietary information held by insurance com-
82 panies, who have historically been reluctant to share it. The existence of publicly-available informati onÂ—like the FGS sinkhole database, for exampleÂ—may not be widely known among the general public. Of course, regardless of its re latively low public profile, the FGS sinkhole database is still not a complete lis t of sinkholes within the state. The database relies on information provided by individuals who find a sinkhole and report it. In order for this to happen, a person with information on the location of a sinkhole must know where to report it, or to report it at all. A dditionally, database maintenance often depends on the provision of adeq uate funding by the state government, which is volatile from year to year. The inherent shortcomings of this database may have resulted in an inaccurate picture of sinkhole location and density across the region. Finally, some sinkholes may be us ed as Â“water featuresÂ” in new residential developments, as a means of adding value to nearby properties. Most homebuyers are un likely to distinguish between a man-made lake or pond, or a previously-existing sinkhole that has been intentionally converted to that purpose. Because of that, and because these water features are often seen as desirable amenities among homebuyers, it is at least co nceivable that any negative impacts of sinkhole proximity on home prices in other areas (particularly, in areas without ne wer, high-end developments making use of water features) have been obscured. This paper has demonstrated a lack of statistical evidence pointing to any relationship between home values and the presence of sinkholes in the Tampa Bay area. Th ese results could be due to homebuyer preferences, accessibi lity of relevant information, or a lack of available data for analysis. Any future research into the question of how sinkholes influence real estate markets should attempt to shed some light onto the underlying cause of the results presented here. References Boston, John; Rigsby, Leo C.; Zald, Mayer N. 1972. The impact of race on housing markets: A critical review. Social Problems, vol. 19, no. 3., pp. 382-393.
The Florida Geographer 83 Brookshire, David S.; Thayer, Mark A.; Tschirhart, John A.; Schulze, William D., 1985. A test of the expect ed utility model: evidence from earthquake risks. The Journal of Political Economy: vol. 93, no. 2, pp. 369-389. Dale, Larry; Murdoch, James C.; Thay er, Mark A.; Waddell, Paul A, 1999. Do property values rebound from environmental stigmas? Evidence from Dallas. Land Economics, vol. 75, no. 2, pp. 311-326. Eastman, Kevin L; Butler, Ann M.; Lilly, Claude C. 1995. The effect of mandating sinkhole coverage in Florida homeowners insurance policies. CPCU Journal, September 1995, pp. 165-176. Kiel, Katherine; McClain, Kather ine, 1995. House prices during siting decision stages: The case of an incinerator from rumor through operation. Journal of Environmental Economics and Management: Vol. 28, pp. 241-255. Maroney, Patrick F.; Cole, Cassandra R.; Corbett, Richard B.; Dumm, Randy E.; Eastman, Kevin L.; McCullough, Kathleen A. 2005. Preliminary Report: Insurance Study of Sinkholes. Tallahassee, FL. Nourse, Hugh O., 1967. The effect of air pollution on home values. Land Economics: vol. 43, no. 2, pp. 181-189. Palm, Risa I. 1981. Public response to earthquake hazard information. Annals of the Association of American Geographers, v. 71, no .3, pp. 389-399. Palmore, Erdman. 1966. Integration and property values in Washington, DC. Phylon, v. 27, no. 1, pp. 15-19. Patton, D.; DeHan, B., 1998. Water Issu es: Global, National, State, Ecosystem. In Fernald, Edward A.; Purdum, Elizabeth D., eds., Water Resources Atlas of Florida. Florida State University. Soriano, M.A.; Simon, J.J., 2002. Subsidence rates and urban damages in alluvial dolines of the Ce ntral Ebro basin (NE Spain). Environmental Geology, vol. 42, no. 5, pp. 476-484.
84 Tobin, Graham A.; Montz, Burrell E., 1994. The flood hazard and dynamics of the urban residential land market. Water Resources Bulletin, v. 30, no. 4, pp. 673-685. United States Census Bureau, 1990. United States Geologic Survey, 1999. Sinkholes of West-Central Florida. White, William B., 1988. Geomorphology and Hydrology of Karst Terrains. Oxford University Press. Oxford, UK.