The effect of proximity to a registered sex offender's residence on single-family house selling price.

A selling price results from the negotiations between the seller and buyer. The presence of an offender may motivate owners to accept a low offer to consummate a sale, and the model employed here will capture that effect. However, in order to estimate the actual selling price effect of proximity to an offender, both price and marketing time should be investigated because, from the seller's perspective, extra time on the market lowers the present value of the selling price. Unfortunately, the transaction data set used in this study does not include reliable time on the market information. (11) Because marketing time is not included in the model used in this study, the selling price effect discovered may understate the effective selling price effect. In fact, if a sufficient number of owners wait for an undiscounted offer, failure to include time on the market would make it impossible to detect any selling price effect. While knowledgeable buyers may refuse to bid on a house located in close proximity to an offender's residence, or may lower their bid, the price offered by uninformed buyers will be unaffected. Therefore, without time on the market in the model, if almost all owners are willing to search until an uninformed buyer is located, no selling price effect should be detected.

The existence of knowledgeable buyers is a critical element in order for selling price discounts to occur. This study could not determine the percentage of knowledgeable buyers in the data sample, but information obtained from the sheriff's office indicates that they are present. Although the sheriff's office does not track web site hits, they reported receiving approximately 3,000 personal requests for information during the study year (about 10% of which came from real

Data

Two data sets are used in this study. The first contains information about registered sex offenders, including their addresses and classifications. This information was obtained from the Montgomery County Sheriff's Office. Their database contained information on 22 sexual predators where limited notification applied, 4 habitual sex offenders where limited disclosure applied, 2 habitual offenders where passive notification applied, and 221 sexually oriented sex offenders where passive notification applied. (12) Therefore, the offender data used in this study contains 26 offenders where limited disclosure applied and 223 offenders where passive notification applied.

The second data set consists of 3,208 single-family houses located in Montgomery County, Ohio, that sold during 2000. (13) Montgomery County contains 461.7 square miles and has a population of 558,427. There are about 226,182 occupied household units in the county and 142,371 of these are owner occupied. Dayton (population 167,475) is the county seat. Another 216,100 people live in the next nine largest cities in the county. Approximately 5% of the population reside in rural areas. Transaction data and property characteristics for sold houses were obtained from public records offices in Montgomery County and the Dayton Area Board of REALTORS[R]. Descriptive statistics of the transactions in this study are presented in Table 1.

Geocoding

An important variable in the study is the distance between each sold house and the residence of the nearest offender. To determine this distance, both the sold house and the offender data sets were geocoded over the Internet using software provided by Tele Atlas. (14) This produced the latitudes and longitudes for observations in both data sets. In addition, Tele Atlas rated each geocoded location for accuracy based upon how well the property description submitted matched the information in their database, and indicated the level of accuracy for each observation by attaching a quality rating ranging from 1 to 6. All geocodes that did not meet the highest quality rating of 1 were eliminated from the study. (15)

Next, using the geocoded latitudes and longitudes, ArcView[R] software was used to calculate the distance from each address in the sold house data set to the nearest address in the offender data set. (16) This distance was calculated twice: once from each sold house to the nearest offender who was subject to limited disclosure, and again from each sold house to the nearest offender subject to passive notification. In determining the closest offender, it was required that the offender be in residence at least one week before the purchase contract was signed.

In addition, the geocoded points were plotted on maps using the ArcView[R] software. A map of the location of sold houses and the residences of offenders subject to limited disclosure is presented in Figure 1, and a map of the location of sold houses and the residences of offenders subject to passive notification is shown in Figure 2. These maps show that the clustering of sold houses and offenders' residences are similar, The source of the shape file used to display Montgomery County in both exhibits was the 1990 Topologically Integrated Geographic Encoding and Referencing System (TIGER[R]) files from the U. S. Census. (17)

[FIGURES 1-2 OMITTED]

Methodology

There are three basic steps in the methodology. The first two steps are used to test the following two null hypotheses:

[H.sub.O]: Ceteris paribus, the selling price of a single-family house is unaffected by its proximity to the residence of a registered sex offender where limited disclosure applies.

[H.sub.O]: Ceteris paribus, the selling price of a single-family house is unaffected by its proximity to the residence of a registered sex offender where passive notification applies.

Stepwise Regression

The objective of the first step of the methodology is to identify variables to be included in a hedonic regression. To accomplish this task, a battery of property characteristic variables is subjected to stepwise regression. (18) Each variable must maintain a p-value of.05 or less in order to enter and remain in the model.

In the second step of the methodology, a binary variable, PROX, is added to the model, and a two-level analysis of covariance (ANCOVA) procedure is performed on the expanded model. (19) PROX is assigned a value of 1 if the subject house was located within a specified area relative to the nearest offender, or 0 if the subject house was located outside the specified area. A series of non-overlapping concentric rings (rings) is used to define the area where PROX equals 1. In the first iteration, PROX equals 1 if the subject house is located 0.1 mile or less from the nearest offender. In subsequent iterations, the maximum radius is expanded by increments of 0.1 mile and observations from previously tested rings are eliminated from the sample. For example, when the maximum radius is set at 0.2 mile (the second ring), houses located no more than 0.2 mile but more than 0.1 mile from an offender are compared to houses located farther away than 0.2 mile from the offender. (20) The iterative process is conducted twice, first for offenders subject to limited disclosure and again for offenders subject to passive notification, to determine the point at which PROX becomes and remains insignificant at the 95% confidence level. When estimating the model for offenders subject to passive notification, any observations located within the distance found to be significant for an offender subject to limited disclosure are eliminated from the sample in order to eliminate the effect of the more dangerous offender. (21) A significant negative estimated coefficient for PROX indicates that the average selling price of houses located within the specified ring is less than the average selling price of houses located farther away.

In the final step of the methodology, the percentage selling price effect for each ring is calculated by dividing the dollar price effect by the average selling price of houses located within the ring. The stepwise regression indicates that the basic hedonic model contains thirty-four variables, as shown in Equation (1). (22)

(1) SP = [alpha] + [[beta].sub.1]SQFT + [[beta].sub.2]AGE + [[beta].sub.3]LOT + [[beta].sub.4]FIRE + [[beta].sub.5]BATH3 + [[beta].sub.6]OWN + [[beta].sub.7]FULL + [[beta].sub.8]WINTER + [SIGMA][[beta].sup.34.sub.n=9]LOC + [xi]

where:

SP = the selling price of the subject property

[alpha] = the intercept

[[beta].sub.n] = the estimated coefficients

SQFT= the amount of living space in square feet

AGE = the age of the house in years

LOT = the size of the lot in acres

FIRE = the number of fireplaces in the house

BATH3 = a binary variable equal to 1 if the house has three or more bathrooms, equal to 0 otherwise (23)

OWN = a binary variable equal to 1 if the house is owner occupied, equal to 0 otherwise (24)

FULL = a binary variable equal to 1 if the house has a full basement, equal to otherwise

WINTER = a binary variable equal to 1 if the house sold during December, January, or February, equal to 0 Otherwise (25)

LOC = a vector of binary variables equal to 1 if the house is located in a particular tax district of 48 tax districts, equal to 0 otherwise (26)

[xi] = the error term