SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

We examine neighborhood externalities that arise from the perceived risk associated with the proximity of a registered sex offender's residence. We find large negative externality effects on a property's price and liquidity, employing empirical techniques that include a fixed-effects OLS model, a correction for sample selection bias and censoring using a Heckman treatment, and a three-stage least-squares model to account for simultaneity bias in the joint determination of a home's sale price and liquidity. Additionally, we find amplified effects for homes with more bedrooms (a proxy for children) and if the nearby offender is designated by the state as “violent.”

A household's largest asset is typically its investment in real estate, rendering the decision to buy or sell a home one of the most important that a household makes. Homeowners usually consider an array of factors in accordance with their tastes and preferences, such as a property's physical characteristics and especially its location. As households carefully examine a neighborhood's (dis)amenities, ranging from school quality to distance from work to crime risk, the location can make or break such a calculated decision. For example, perceived crime risk associated with a registered sex offender living next door or down the street may very well be a “deal breaker,” or at least substantially lower the reservation price and perhaps lengthen the time a home spends on the market. The central purpose of this study is to estimate the extent to which the presence of registered sex offenders affect nearby home prices and liquidity (as measured by time on market).

Market participants and their agents have access to a tremendous amount of information, even the proximity of a convicted sex offender to a particular property. Anecdotally, a number of real estate professionals report that they or their clients check the online sexual offender registry when evaluating a property, assessing crime risk along at least one important margin. Since the 1994 Wetterling Crimes Against Children and Sexually Violent Offender Registration Act, states have been required to register individuals convicted of sex crimes against children. Amendments passed in 1996, popularly known as Megan's Law, added a requirement to inform the public if a convicted sex offender moved into a neighborhood, as most states now have registries fully accessible online.1 In fact, one Richmond, Virginia, based realtor noted that she has a smart phone app, which allows her to show a list of nearby offenders to prospective buyers (Hazard 2010).

Given the ease of access of online sex offender registries and anecdotal evidence about their usage by real estate market participants, is there empirical evidence that registered sex offenders actually affect real estate market outcomes in an economically meaningful way? We examine precisely this question, utilizing data from relatively suburban/rural areas of central Virginia. We utilize a Multiple Listing Service (MLS) dataset to extend the current literature in a number of ways to gain novel insights into this question. Because MLS data include all listed homes during a period of time and whether or not they sold, this study accounts for selection bias by incorporating unsold homes in the analysis. In addition to estimating the effect a nearby sex offender has on sale price, we examine the effect on a home's liquidity, measured by time on market, using a three-stage least-squares (3SLS) model. Additionally, our study accounts for the possibility that a property is located near more than one registered offender, an outcome that becomes more likely as sex offender registries grow over time. Indeed, we estimate the effect of multiple nearby sex offenders and also whether differentiated crime risk preferences are captured by market prices and liquidity. That is, does the market differentiate between risks posed by violent versus nonviolent offenders? And, do households who need more bedrooms (i.e., families with more children) place a substantially higher value on the risk associated with living near sex offenders?

We answer these questions within several sections. First, we survey existing literature to identify critical gaps and to introduce the original contributions of this study. Second, we provide an overview of our data, its sources and computation of key variables of interest. In the third section, we detail the methodological approach of baseline price models and the utilization of a Heckman (1979) model to address possible selection bias. In addition, we discuss identification strategies where reverse treatment effects (postmove-in and move-out of registered offenders) are estimated to address possible selection and endogeneity issues. Fourth, baseline OLS and Heckman sample-selection results are presented. In the fifth section, we explain a different methodological approach that addresses potential bias resulting from the joint determination of price and liquidity, including both its theoretical rationale and our identification strategy. Following that, we present 3SLS baseline results along with a number of extensions. Finally, we present conclusions based on all of the above.

Filling Important Gaps in the Literature

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

While sex offender registries are not new, their widespread dissemination over the Internet is a relatively recent phenomenon. All states now provide detailed data about the locations, physical descriptions and pictures of nearby sex offenders, even details about the charges for which they have been convicted. Consequently, examining the effects of such registries is also relatively new. Utilizing these data sources, a small number of recent studies have found significant housing price effects for those located within one-tenth of a mile of a registered sex offender. While these studies find generally consistent empirical effects, they leave a number of important gaps—notably the failure to model and test for (a) the joint determination of housing price and liquidity, (b) sample selection issues, (c) effects of multiple nearby sex offenders and (d) idiosyncratic risk with respect to violent offenders or the number of children/bedrooms in a home.

Two recent studies found that sex offenders residing nearby had a significant impact on home prices. Linden and Rockoff (2008) analyzed housing effects in Mecklenberg County, North Carolina, a large area which contains Charlotte. Data on the move-in date for the most recent address of each offender were matched to a dataset of home sales over a 10-year period. By studying price trends, they found significant reductions in home prices across radii of less than 0.1 miles and 0.1–0.3 miles when an offender moves in. Controlling for home and property characteristics as well as neighborhood and time fixed effects, regression results indicated that the presence of a sex offender within 0.1 miles reduces home prices by roughly 4.0% after the offender's move-in date. Living beyond one-third of a mile from an offender had no significant effect on price in their study. The other recent study, Pope (2008), analyzed the effect of proximity to a sex offender in Hillsborough County, Florida, which contains Tampa. His dataset included move-in and move-out dates for every house an offender has occupied since being placed on the registry. Controlling for housing characteristics and time and neighborhood fixed effects, regression results indicated that a location within 0.1 miles of a registered sex offender reduced housing values by 2.3%. He also found that prices rebound when an offender moves out, a treatment that we utilize in a later section of this article. Like the Linden and Rockoff (2008) study, the effect dissipated quickly beyond 0.1 mile, as there was no significant effect on price for homes located between 0.1 and 0.2 miles from an offender.2

A critical limitation of both aforementioned studies is their data, which do not include unsold homes or a measure of time on market for homes that did sell. Consequently, their results may suffer from simultaneity bias as they fail to model and test for joint determination of a home's selling price and its time on market (i.e., its liquidity) and sample selection bias as they cannot account for the effect of a nearby sex offender on homes that were marketed but failed to sell. The time it takes to sell a home is often crucially important for sellers who are buying elsewhere, affecting holding costs and the seller's reservation price. The proximity of a nearby sex offender may reduce the number and size of potential buyers’ offers and therefore change the entire path of the home's marketing period, including the length of time it stays on the market and its eventual selling price. Failure to control for endogeneity of price and liquidity, then, may result in incomplete or biased estimation of a sex offender's impact on real estate outcomes. Moreover, failure to model time on market omits a key variable of interest, critically important to real estate market participants in its own right. Unlike previous studies, we utilize realtor data from a MLS, which includes the date at which each property was listed for sale, allowing the time on market to be calculated. Therefore, in addition to estimating a standard spatial hedonic pricing model, we contribute to the literature by estimating a jointly determined price-liquidity model using a 3SLS specification. Unlike previous studies, we utilize a Heckman model with MLS data on sold and unsold homes to address sample selection issues.

Authors of both studies discussed above present a number of reasons why their results may not be representative. First, each study restricted the sex offender sample by dropping transient sex offenders from their analyses. Pope (2008) dropped observations of sex offenders who resided in their respective homes less than six months, while Linden and Rockoff (2008) dropped ones who had been in residence less than a year. As a result, they only captured the effect of an “established” sex offender, who has settled into a neighborhood for some time.3 However, a sex offender may have moved in during any part of the marketing period, impacting buyer and seller behavior during that time (e.g., influencing a seller's reservation price or potential buyers’ offers and ultimately the final selling price of the home). Utilizing a property's listing data from the MLS data enables this study to focus on the effect of any sex offender who lived nearby during a home's marketing period. Real estate market participants who inspect the sex offender registry may observe the presence of a sex offender in a binary sense (i.e., whether a sex offender lives nearby or not), seeing crime risk associated with both transient and established sex offenders. Therefore this study estimates a more representative effect of any sex offender, transient or not, who resides near a marketed property.

Another limitation of the existing literature is that by estimating only the effect of a single nearby sex offender, it does not account for multiple nearby sex offenders or whether a home is near a cluster of sex offenders. For example, Pope (2008) specifically limited his data to neighborhoods that had not previously contained a sex offender, where one, and only one, sex offender moves in nearby. He states that, “Restricting the data in this way is not costless. Estimates of the average treatment effect for this selected portion of the data may not provide an accurate measure of the housing price effects of areas with multiple sex offenders” (Pope 2008, p. 608). Yet, sex offender registries in many states are constantly expanding due to the fact that it is extraordinarily difficult to be removed from most registries, short of dying,4 and newly released sex offenders have to live somewhere. Therefore, over time the probability that a given home is near a sex offender and the probability of living near multiple sex offenders both increase. Prior estimates will have decreasing applicability as time passes, which is precisely why we do not limit our sample in this way; rather, we control for multiple sex offenders and estimate the effect of the existence of additional sex offenders who reside in the vicinity of marketed homes. In this respect, this study's estimates are more generalizable than prior research.

An important open question in the literature is whether market participants differentiate risk based on their own characteristics or characteristics of a nearby sex offender. It is reasonable to assert that a key driver of crime risk preferences of buyers and sellers is whether these households have children. Yet, differentiated neighborhood effects by homeowner preferences have not been analyzed in the literature. Data on buyers’ or sellers’ number of children are not available. However, responses to the 2009 American Housing Survey, cited by Horn (2011), reveal that households with children occupy housing with 3.2 bedrooms on average, compared to 2.5 bedrooms for households without children. Therefore, we use number of bedrooms in a home as a proxy for households with more children and utilize interaction terms to estimate the effect of nearby sex offenders on homes with more bedrooms.

Finally, we ask whether sex offenders themselves pose different risks. The Commonwealth of Virginia differentiates sex offenders based on the crime of which they have been convicted. Not only are the criminal statutes listed on a sex offender's registry profile, but Virginia also designates whether the crimes of which the offender has been convicted were “violent.” When a home buyer or seller views the registry, she or he knows whether sex offenders live nearby in addition to the offenders’ characteristics and whether they were convicted of a violent crime. Other states take similar measures. For example, Florida classifies sex offenders as to whether they are a sexual “predator.”5 Thus, one additional factor we explore is whether the market differentiates between “violent” and “nonviolent” sex offenders in its assessment of crime risk.

Data

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

The data for this research are from several sources. Data on real estate transactions consist of observations of residential properties on the market between July 1999 and June 2009 and come from a central Virginia MLS, which includes Lynchburg and surrounding areas. The initial housing data contained 21,453 observations of both sold and unsold properties. As noted by Levitt and Syverson (2008), MLS data are entered by real estate agents and can be incorrect or incomplete. As a result, the data were carefully culled for incomplete, missing or illogical data. The final dataset consists of 12,426 sold and 7,295 unsold properties.6 The data collected from the MLS include typical property characteristics (square footage, bedrooms, baths, etc.), location and market and calendar information (list date, sale date, length of listing contract).

Information about sexual offenders is contained in the Virginia Sex Offender and Crimes Against Minors Registry and is available on a public website maintained by the Commonwealth of Virginia. However, because a number of registered sex offenders move often, on a given date each state's registry reflects only a snapshot of sex offenders’ current addresses. A buyer or seller who checks the registry around a particular location one year later may find a different pattern of sex offender residences due to move-ins or move-outs. To assist in our research, we obtained a unique archived dataset from the Virginia State Police, which contains each of the registered sex offender's past addresses, including all move-in and move-out dates. Each observation also contains a registered sex offender's current address, along with a number of other personal characteristics (e.g., age, sex, race, description of the perpetrated crime and whether that crime was a violent crime) that appear on the registry. Over approximately 10 years, there were 2,031 sex offenders who resided at 4,601 different addresses in our data. The archived data allow us to utilize historical real estate data effectively, essentially replicating the data that would have been accessible to homebuyers and sellers at any time during a property's marketing period, including at the time of the transaction. Note that datasets composed only of a snapshot of present addresses do not allow a researcher to know which offenders lived near homes that were sold in the past. Analysis without such archived data is at best incomplete and is potentially biased. Unlike other studies, we did not cull the dataset for transient sex offenders based on length of stay at a residence alone, although we did remove some whose obvious transient indicators did not seem useful for our study. For example, we removed offenders if they were listed as homeless or as staying at a hotel or some other institution.

After obtaining the longitude and latitude of each marketed property and registered sex offender address, we use the great-circle distance formula to calculate the distance from a given house on the market to each sex offender.7 For each home we calculated binary variables based on whether the nearest sex offender who was still in residence on the home's sale date was located within (or ≤) a 0.1 mile radius or the following concentric rings: 0.1 mile < nearest offender ≤ 0.25 mile, 0.25 mile < nearest offender ≤ 0.5 mile and 0.5 < nearest offender ≤ 1 mile. We also calculated a continuous variable that counts the number of sex offenders located within a one-mile radius, in addition to the nearest one designated by dummy variables. These are the primary variables of interest in subsequent sections.

Methodology—Baseline Hedonic Models

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

Our principal objective is to isolate the effect of a nearby registered sex offender on neighborhood real estate market outcomes. In this section we focus on a registered sex offender's effect on the sale price of a home, utilizing a cross-sectional hedonic pricing model as the baseline. While hedonic pricing models are commonly used to determine the value of specific property attributes and surrounding (dis)amenities by estimating marginal effects on the sale price of the property,8 both Linden and Rockoff (2008) and Pope (2008) show that cross-sectional analysis alone may produce biased estimates of a treatment effect (presence of a nearby sex offender) due to unobserved or omitted variable bias and spatial autocorrelation. It is possible that the presence of a nearby sex offender residence is correlated with unobserved property and location characteristics and with a trend in local housing prices. In the next section we describe spatial and temporal controls, which increase the probability that the estimated sex offender effect is exogenous. In later sections we describe additional identification strategies utilizing alternative treatment measures and a “reversal treatment,” where we not only analyze the effect a sex offender has when he or she moves in, but also the effect a sex offender has when he or she moves out.

Fixed Effects and Spatial Modeling

We begin with a cross-sectional approach to provide a baseline reduced form estimate of the effect of a nearby sex offender, employing a traditional hedonic model that accounts for heterogeneous characteristics of both homes and their locations. We use the following functional form:

  • display math(1)

where SPi is a vector for property selling price,9 Xi is a vector of property-specific characteristics10 and time-specific and macroeconomic control variables,11 LOCi is a vector for location control, SOi, the variable of interest, equals one if the nearest registered sex offender contemporaneously lived within one of the rings (described above) around marketed homei, zero otherwise, and ε is an error term.

Hedonic analysis of the housing market requires some control for spatial heterogeneity because location itself (LOCi above) is a key source of differences in housing prices. Following Pope (2008), we chose census-block groups to control for unobserved heterogeneity across these areas so that the explanatory variables’ effects are identified from variation within a given area (or even in a given year, as is the case for time fixed effects). According to the U.S. Census, census tracts are “small, relatively permanent statistical subdivisions of a county … designed to be homogenous with respect to population characteristics, economic status, and living conditions.”12 Yet, census-block groups are subsections of census tracts and the smallest spatial area for which the U.S. Census tabulates sample data. This study uses block groups from the 2000 census, which on average contain between 600 and 3,000 people, usually around 1,500. Our sample of houses falls within a total of 163 census block groups in central Virginia. Following Davis (2004) and Heintzelman (2010), we cluster standard errors at the census-block-group level, rendering the error term, ε, robust to spatial autocorrelation at the census-block-group level. Heintzelman (2010) explains that including such neighborhood location controls and clustering standard errors at the neighborhood level eliminates a critical source of omitted variables bias and spatial autocorrelation.13 In addition to clustering, we use robust, heteroskedastic-consistent standard errors.

Sex Offender Clustering

Like previous hedonic studies on sex offenders, the baseline model above is limited in that the effect of a sex offender is only captured by a binary variable, not accounting for the effect of multiple sex offenders clustered around a particular home. As mentioned in the literature review, prior studies dealt with this by confining their sample to homes that were only near a single sex offender. Therefore, we build on the baseline model (1) by extending it in the following way:

  • display math(2)

where ADD_SOi is the number of additional sex offenders located within one mile of the ith property. For example, consider a home that is located near a small cluster of three registered sex offenders. The nearest sex offender may be located within 0.1 mile and would already be captured by the corresponding binary variable, SOi. The two additional sex offenders who are located within one mile would be counted in ADD_SOi.14 Hence, this allows us to not only control for the presence of additional sex offenders clustered around a particular home, but also estimate the impact of additional sex offenders. Prior studies have simply omitted properties with multiple nearby offenders from their analyses.

Heckman Sample Selection Model

Prior studies model the impact nearby sex offenders on sale price with data limited to sold homes only. Yet, a sample restricted to only sold homes may suffer from a sample selection bias, as the dependent variable is observed only for a restricted, nonrandom sample. A Heckman sample selection model corrects the selection bias with a two-stage approach, by estimating:

  • display math(3)

where λ is the inverse Mills ratio evaluated at , where was estimated at the initial stage by the following bivariate probit:

  • display math(4)

where Z represents the following variables as defined above: Xi, LOCi, SOi, ADD_SOi, TOMi. The Heckman model utilizes the inverse Mills ratio as a regressor in the price equation to correct for selection bias by adjusting the conditional error terms so they will have means of zero. Again, the estimation approach is only possible for samples that include both sold and unsold homes due to the binary dependent variable in the first stage probit, which is a methodology previously unexplored in this literature.

Sex Offender Externality as a Time- and Distance-Weighted Continuous Variable

While the prior literature uses dummy variables to capture the presence of a sex offender and its corresponding externality, we develop an alternative methodology by constructing a time and distance-weighted continuous variable. An alternative measure of the externality can be thought of as a robustness check of the relationship between key interest variables. We incorporate Turnbull and Dombrow's (2006) approach in which they capture a market competition externality associated with other nearby homes on the market at the same time. They measured competition from nearby homes by constructing a sum of overlap days on the market weighted by distance. Their competition variable increases with the number of competing properties, the number of days properties are on the market together and also their proximity. We adapt their measure, a detailed explanation of which is presented in the simultaneous equation section below, to construct a continuous measure of the external affect a sex offender has on a home's price. Equation (5) below represents “sex offender density” (SOD), which measures a distance-weighted overlap between the days a given home i was on the market and in the presence of a nearby sex offender j, summed across all offenders who lived within one mile:

  • math image(5)

where D(i, j) is the distance between a given home on the market, i, and a given nearby registered sex offender, j, provided that they live within a one-mile radius, s(i) is the sell date for a home i, l(i) is the listing date for home i, mo(j) is the move-out date for a given registered sex offender j, mi(j) is the move-in date for registered sex offender j. SOD is a positive function of (1) proximity to sex offenders, (2) overlap between the sex offender's time of residency and the marketing period and (3) multiple sex offenders. An additional feature of this measurement is that it accounts for the impact of sex offenders who had resided nearby for only part of the home's marketing period.

Relative to binary indicators, this approach has three primary benefits. First, the construction of this variable allows sex offenders to have a quadratically larger effect the closer they are to a given home on the market. The intuition here is that a sex offender located next to a property would have a larger external effect than one located down the street, and much larger than one located on the other side of the neighborhood. Second, we account for the overlap between the marketing period and the sex offender's stay. Clearly, a sex offender who resides nearby during the entirety of the marketing period will be noticed more by the market and will have a larger external effect than a sex offender who moves out halfway through the marketing period. Third, SOD sums the effect of multiple sex offenders by making the calculation for all nearby sex offenders within one mile, accounting for the fact that a home may be located near a cluster of sex offenders. Therefore, in addition to the above specifications, we will estimate the following:

  • display math(6)

Move-In Identification Strategy and the Endogeneity of Sex Offender Locational Choices

The cross-sectional approach estimates the effect a nearby sex offender has on a home after the sex offender has already moved in. Yet, sex offenders’ locational choices could be endogenous to unobserved heterogeneity of the particular home or locale they choose. For example, a sex offender may be lured by the lower prices at or near a home with some unspecified negative externality (e.g., an unusual colored home, overgrown landscaping and other eyesores that would affect nearby home prices). It is not unreasonable to assume that registered sex offenders, who likely have limited employment opportunities, may also be more likely to tolerate negative externalities and atypicalities if such externalities make home prices more affordable. If that assumption is valid, the estimated effect of the nearby sex offender could actually reflect the underlying eyesore or atypicality, not the crime risk associated with the nearby sex offender per se. Therefore, even after for controlling for neighborhoods (i.e., census block group), cross-section results could still identify a within-area spurious correlation between some unobserved heterogeneity and sex offenders’ locational choices.

To ensure that cross-sectional estimates are properly identified, we employ a strategy similar to Linden and Rockoff (2008). If an underlying negative externality already exists at or near properties that registered sex offender choose, then the pre-existing externality would reduce neighboring home prices in the period immediately prior to the sex offender moving in. Hence, we estimate the following model:

  • display math(7)

where Equation (7) is identical to Equation (2) with the addition of the identification treatment parameters SOi_MI, which are binary indicators of the move-in of the nearest sex offender just after (i.e., 90 days) a particular home i was sold. Like the original treatment dummy variables, these are binary concentric circles at the same distances (0.1 mile, 0.25 mile, 0.5 mile and 1 mile). Indeed, if the sex offender's locational choice is endogenous to an underlying unobserved negative externality, then the SOi_MI parameters reflect the fact that homes have recently sold for a discount as a result of that particular negative externality, casting serious doubt on the cross-section results’ identification.15 However, insignificant SOi_MI parameters would provide strong evidence that registered sex offender locational choices are not endogenous to price reducing unobserved heterogeneity and the externality we estimate after the sex offender moves in is properly identified.

Move-Out Reversal Treatment

Once a sex offender has taken residence in a particular home, that home appears on the online registry and is searchable to prospective real estate buyers and sellers. If a sex offender who moves into a neighborhood creates a negative externality, the effect should be reversed when the sex offender moves out. Failure of the reverse treatment to empirically cancel the effect would cast doubt on the identification of a causal effect in the original estimate. Following similar methodology to Pope (2008), we estimate the following function:

  • display math(8)

where Equation (8) is identical to Equation (2) with the addition of the reversal treatment parameters SOi_MO, binary indicators of the move-out by the nearest sex offender within a year prior to the date at which home i was listed. Like the dummy variables in the original treatment, these are also binary concentric circles at the same distances as before (0.1 mile, 0.25 mile, 0.5 mile and 1 mile). By controlling for the existence of contemporaneous sex offenders (SOi and ADD_SOi), the move-out treatment compares properties for which the previous nearest offender moved out before property i was listed for sale against a control group in which there was no “move-out.” In this treatment, insignificant SOi_MO coefficient estimates (i.e., not statistically different from the control group) provide evidence that the external effect dissipates relatively quickly after the treatment has exited. Note that a negative coefficient would indicate that the effect of a previous offender has not fully dissipated and a positive coefficient would indicate that prices have rebounded to higher levels than control groups. Insignificant coefficient estimates in this quasi-experimental framework would provide further evidence that the original sex offender treatment is properly identified.

Results—Baseline Hedonic Pricing Models and the Reversal Treatment

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

Baseline—A Diminishing Effect across Farther Distances

Estimates of Equation (1) show that nearby registered sex offenders have a substantial negative impact on home prices, and this effect diminishes the farther a home is from a sex offender. Table 2 shows the effects of using different radii for measuring the effect a sex offender has on property values, controlling for a number of property, location and market characteristics.16 Pope (2008) uses a binary variable equal to 1 if a sex offender resides within 0.1 miles of a property and a second binary variable for values greater than 0.1 mile and less than 0.2 mile. Linden and Rockoff (2008) also use the 0.1 mile dummy, but they differ in their additional use of a 0.3 mile variable. In the proceeding analysis, we analyze the effect of the nearest sex offender location within the following mutually exclusive concentric rings: 0.1 mile, 0.25 mile, 0.5 mile and 1.0 mile.

Table 1. Summary statistics (sold homes)
       Mean        Std. Dev.     Min.    Max
Sale price ($)165,874.10       89,949.70 25,500.00749,000.00
Time on market (days)110.55   88.79    0.00963.00
Sold0.610.4801
Sex offender ≤ 0.10 Mile0.020.1201
Sex offender ≤ 0.25 Mile0.050.2101
Sex offender ≤ 0.50 Mile0.060.2501
Sex offender ≤ 1.00 Mile0.120.3201
Sex offender density (SOD)0.170.58010.03
Additional S.O. ≤ 1.00 Mile0.682.58034
Violent0.780.4101
Square feet1,924.02     782.09417.008,418.00
Age (years)26.4228.150267.00
Vacant0.330.4701
Bedrooms3.200.7818.00
Baths2.040.6916.00
Length of contract (days)186.83  102.980990.00
One story0.390.4901
New0.160.3701
Finished basement0.270.4401
Hardwood0.550.5001
Brick0.540.5001
Pool0.170.3701
Fenced yard0.170.3701
Walk-in closet0.210.4101
Acreage2.047.670248.66
Avg. fixed rate mortgage at sale date6.130.494.818.64
Virginia unemployment rate3.570.592.207.10
Consumer Sentiment Index86.1710.4555.30112.00
Leading Economic Indicators Index99.046.0484.20104.90
Fall0.190.3901
Winter0.260.4401
Spring0.300.4601
Summer0.250.4301
Table 2. The effect of a nearby registered sex offender on a home's sale price
 Baseline OLSBaseline OLS withHeckman SelectionSex Offender
 ModelAdditional S.O.ModelDensity
 [1][2][3][4]
Notes
  1. Columns (1) through (4) display sex offender variable coefficients in four specifications of a hedonic selling price model; robust t-statistics are in parentheses (errors clustered by census-block group); in the Heckman model, the selection probit coefficients are omitted from this table; ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.

Sex offender ≤ 0.10 mile−15,532.89***−12,272.64**−11,331.96*** 
 (−3.34)(−2.58)(−3.36) 
Sex offender ≤ 0.25 mile−10,109.87***−7,092.18**−6,912.36*** 
 (−3.71)(−2.20)(−3.03) 
Sex offender ≤ 0.50 mile−5,605.66**−4,285.34−4,027.20** 
 (−2.06)(−1.42)(−2.20) 
Sex offender ≤ 1.00 mile−3,796.23*−3,709.78*−3,488.17** 
 (−1.95)(−1.90)(−2.54) 
Additional S.O. ≤ 1.00 mile −694.55*−614.65*** 
  (−1.77)(−2.82) 
Lamda (inverse mills)  474.91 
   (0.22) 
Sex offender density   −4,508.44***
    (−3.25)
Property characteristics
Macroeconomic controls
Season fixed effects
Census-block groups
Year fixed effects
Observations12,42612,42619,78112,412
R20.71230.7126 0.7117

Table 2 shows that a registered sex offender living within 0.1 mile of one's home will reduce the value of surrounding properties sold by $15,533, representing a 9.2% discount for the average home in our sample. This is nearly three times the magnitude of similar estimates reported in Pope (2008), perhaps suggesting a much higher willingness to pay to avoid crime risk in predominantly suburban and rural central Virginia. To put this magnitude in relative terms, homeowners value avoiding this crime risk more than they value a pool, hardwood floors, a brick exterior, a walk-in closet or an additional bathroom. More precisely, homeowners value avoiding this risk as much as an additional 209 square feet in their homes, suggesting that homeowners might be willing to live in significantly smaller homes if they could avoid living so close to sex offenders.

As the radius to the nearest sex offender widens, Table 2 also shows that this effect is still present, albeit diminished. A nearby sex offender lowers property values $10,110, $5,606 and $3,796 when the nearest sex offender residence falls within widening radii. This result also differs from previous studies, which have not found an effect on property values beyond 0.3 miles, suggesting that there is a significant difference between the perception of “neighbor” in suburban/rural and urban areas. Previous studies have used data from more densely populated counties within urban areas. Suburban or rural residents may simply perceive homes within a larger radius as “neighbors,” resulting in a greater alertness or aversion to crime risk over larger distances. Indeed, this relationship remains robust across all model specifications. Moreover, declining magnitudes of coefficients across distance conforms to the reasonable expectation that risk falls as the physical distance from a perceived crime risk rises.

Additional Sex Offenders—A More General Baseline Estimate

Baseline estimates above and similar analyses in other studies are limited by the fact that they do not control for the effect of multiple sex offenders who may be clustered around particular homes. Controlling for additional sex offenders offers a more general baseline estimate of the effect of a sex offender, given that it becomes more common to live near multiple sex offenders as registries grow. Column (2) of Table 2 presents estimates that are qualitatively similar to the previous baseline. Homes located within 0.1 mile of a registered sex offender sell for about a $12,273 discount, or about 7.4% less on average, relative to the control group. The effect diminishes the greater a home's distance to the nearest sex offender: $7,092, $4,285 and $3,709 when the nearest sex offender falls within the 0.1–0.25, 0.25–0.5 and 0.5–1.0 mile radii, respectively.

The effect of the nearest single sex offender on a home's price is much larger than the additional sex offender. Table 2 shows that once a home is already located near a sex offender (i.e., within one mile), each additional sex offender within a one-mile radius will lower a property's price by about $695, or 0.4%. Of course, the effect of multiple sex offenders could be substantial if one lives within a large cluster of sex offenders. Still, this evidence suggests interesting aspects about the revealed risk preferences of the market. First, an additional nearby sex offender may signal increasing risk relative to an existing, single sex offender. Second, while people care most about the presence of a nearby sex offender in a binary sense, the fact that marginal effect of each additional offender is much smaller makes sense from a precautionary standpoint. In the short run, a family, for example, may take great precautions because they live near a single sex offender, yet they may not need to significantly alter their precautionary costs with the addition of another offender nearby. For example, if a homeowner builds a fence so that the children are less likely to wander over into the neighbors’ yards, she or he need not necessarily build two fences if another sex offender moves in. That is, the homeowner may take more precaution on the margin, but not proportionately. However, a more likely long-run outcome would be market sorting, particularly by households with children, rather than additional precautions.

Heckman Sample Selection Model—Results

After correcting for potential sold/unsold selection bias with a Heckman sample selection model, the effect of a nearby sex offender on a home's sale price is nearly identical to baseline OLS estimates, which include a multiple offender term. The Heckman results in Table 2 show that the presence of a sex offender living within 0.1 mile is associated with a 6.8% (or $11,332) drop in sale price, which is only nominally smaller than the OLS estimate in column (2). The other coefficients are nearly identical, which is unsurprising given that the coefficient estimate for λ is not statistically significant in the price equation. While the probit results are omitted from Table 2 for brevity, none of the sex offender variables of interest had statistically significant coefficients in the first stage probit, providing additional evidence that selection bias is not a significant concern.

Sex Offender Externality as a Time- and Distance-Weighted Continuous Variable—Results

Here we explore expressing the sex offender externality as a time and distance-weighted continuous variable, finding qualitatively similar results to our baseline estimates from the previous section. Results in column (4) of Table 2 demonstrate that this alternative methodology also indicates a substantial negative impact on home prices of nearby sex offenders. SOD can be interpreted as a continuous multidimensional index of sex offender presence. An example could clarify the interpretation of a marginal change. Suppose registered sex offender j moves in next door to home i prior to its listing and resides there during the entirety of the home's time on the market. From Equation (5) above, the first term, (1 − D(i, j))2, is approximately one. Because the sex offender's residency and the time on market overlap completely, the second term also equals one. Therefore, our estimate suggests that such a sex offender (or this specific marginal change) would lower the home's price by about $4,508, or about 3%. In fact, three nearby sex offenders (situated as above) would produce an index value of three and lower a home's price by $13,524. Interestingly, the previous section's coefficient estimates tell a nearly identical story in this particular case. Those same three nearby sex offenders would lower a home's price by ($12,273 + 2 × $695 =) $13,663. While estimates will differ in other scenarios, it is striking how entirely different methodological approaches can yield such similar estimates of a particular externality, reinforcing the robustness of our estimates.

Move-In and Move-Out Treatments—Results

Cross-section evidence to this point has shown that registered sex offenders lower nearby home prices relative to a control group of homes not near sex offenders. Such results could be considered properly identified as causal if prices were not significantly different just prior to the sex offender moving in and prices were to rebound after the sex offender moves out. Following Linden and Rockoff's (2008) and Pope's (2008) quasi-experimental methodology, Table 3 offers such evidence. Column (1) provides evidence that sex offender locational choices are not endogenous to other unobserved negative externalities. Conditioned on other covariates, prices of homes that sold near the future sex offender residence (within 90 days of move-in) were not different from prices in other neighborhoods. Therefore, if prices are not significantly different just prior to a nearby sex offender moving in, and prices are lower after a sex offender moves in nearby, then the former serves as evidence that the latter effect is properly identified.

Table 3. The effect of a nearby registered sex offender on a home's sale price: Identification and move-out/in treatments
 S.O. Move-In After Sold (OLS)S.O. Move-Out Treatment (OLS)
 Dependent Variable: Sale PriceDependent Variable: Sale Price
 [1][2]
Notes
  1. This table presents results of sex offender move-in and move-out treatments; robust t-statistics are in parentheses (errors clustered by census-block group); ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.

Sex offender ≤ 0.10 mile−12,113.00**−11,468.59**
 (−2.59)(−2.45)
Sex offender ≤ 0.25 mile−6,776.27**−6,811.72**
 (−2.15)(−2.08)
Sex offender ≤ 0.50 mile−4,106.72−3,808.65
 (−1.43)(−1.24)
Sex offender ≤ 1.00 mile−3,475.91*−3,562.08*
 (−1.84)(−1.79)
Additional S.O. ≤ 1.00 mile−583.18−697.12*
 (−1.38)(−1.68)
S.O. move-in ≤ 0.10 mile after sold1,150.27 
 (0.54) 
S.O. move-in ≤ 0.25 mile after sold−4,936.85 
 (−1.36) 
S.O. move-in ≤ 0.50 mile after sold1,218.00 
 (0.45) 
S.O. move-in ≤ 1.00 mile after sold−1,853.53 
 (−1.08) 
Additional S.O. move-in after sold−204.69 
 (−0.30) 
S.O. move-out ≤ 0.10 Mile −5,281.20
  (−1.08)
S.O. move-out ≤ 0.25 mile −367.49
  (−0.12)
S.O. move-out ≤ 0.50 mile −2,693.90
  (−0.91)
S.O. move-out ≤ 1.00 mile −352.67
  (−0.16)
Additional S.O. move-out 97.75
  (0.29)
Property characteristics
Macroeconomic controls
Season fixed effects
Census-block groups
Year fixed effects
Observations12,42612,426
R20.71270.7126

Further, column (2) shows that prices rebound after nearby sex offenders move-out. Homes listed during the year after a nearby sex offender moved away have no statistically significant price effect relative to control group homes, for all radii, respectively. As in the cross-sectional approaches above, these regressions also control for contemporaneous sex offenders and obtain qualitatively similar estimates. As prices rebound after the sex offender moves out, the real estate market appears to respond predictably to new information regarding changes in externalities or potential crime risk. This offers further evidence that previous cross-sectional estimates are properly identified and builds a strong case that the estimates reflect a causal relationship.

Simultaneous Equations (3SLS) Methodology

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

Previous studies of sex offender externalities have not explored effects on a property's liquidity or the possibility that a property's selling price and liquidity are jointly determined. In this section we present a simultaneous equation model to address possible simultaneity bias in OLS estimates of price equations. Note that results presented in Table 2 show little evidence of selection bias in the OLS estimates, strengthening the case for a simultaneous price and liquidity (measured by time on market) estimation approach that does not correct for selection bias.17

Price, Liquidity and Identification

Given that housing is a heterogeneous good with attendant information costs, liquidity of residential real estate may be susceptible to shocks over time or across spatial characteristics. Essentially, housing prices do not instantaneously clear markets in response to shocks. Several papers, for example Stein (1995), Genosove and Mayer (1997) and Krainer (2001), have modeled real estate market characteristics that lead to price rigidity and illiquidity. Additionally, a recent paper by Turnbull, Zahirovic-Herbert and Mothorpe (2013) develops a search model in which sellers choose reservation price subject to market price and liquidity constraints. These models imply that housing prices and liquidity are jointly determined. Shifts in valuation over time or across neighborhoods can lead to changes in the prices at which properties sell and the average time they are on the market—an inverse measure of liquidity.

An empirical complication in this literature is the fact that a home's price and liquidity are simultaneously determined largely by identical factors. That is, the vector of factors that determine a home's price is usually identical to the vector of factors that determine how long it takes a home to sell, resulting in a system of equations which, by definition, are not identified. While a number of empirical studies acknowledge and model this simultaneity,18 the methods to identify price and liquidity equations have generally been ad hoc as authors make a case that some factors only affect price and not liquidity, or vice versa.

We follow Turnbull and Dombrow (2006) by constructing a joint determination model in which a property's sales price and time on market can be represented as separate functions with jointly distributed stochastic errors εsp and εTOM:

  • display math(9)
  • display math(10)

where TOM is time on market measured in days, X is a vector of house (and market) characteristics and C are neighborhood market conditions. Turnbull and Dombrow (2006) define C as a distance and marketing time overlap measure of competition from nearby houses for sale. An alternative measure L, divides C by the marketing period of property i to measure competition per day from nearby listings, where L(i) = ∑j (1 − D(i, j))2{min[s(i), s(j)] − max[l(i), l(j)]} / s(i) − l(i) + 1. Elements of this equation are as defined in the discussion of the SOD variable above.

To identify Equations (9) and (10), note that a change in competition while holding selling time constant is also the partial derivative with respect to listing density (i.e., ∂ϕp/∂C ≡ ∂ϕp/∂L). Therefore, the equation system can be rewritten as:

  • display math(11)
  • display math(12)

Therefore, using L and C vectors uniquely identify the simultaneous system without having to resort to an ad hoc justification for the addition or subtraction of covariates. See Turnbull, Zahirovic-Herbert and Mothorpe (2013) for a more complete discussion of this technique.

An additional benefit of estimating this system is obtaining an estimate for the effect of a nearby sex offender on time on market, an important dependent variable in its own right. Unlike price, the expected sign of nearby sex offender's effect on liquidity (time on market) is theoretically ambiguous, depending on the market and the relative preferences of its participants. For instance, a seller with intense preferences regarding the proximity of a sex offender is likely to have a low reservation price and may accept a steeply discounted offer in order to leave the neighborhood quickly once a sex offender moves in (provided that a buyer is willing to accept this tradeoff). In another instance, a seller's crime risk aversion may not be as high, and accordingly the price discount may not be as steep, potentially leading to a relatively longer time on market. There are a number of possible outcomes depending on the preferences of both buyers and sellers (extending beyond the simplicity of these examples); however, it is crucial to note that the expected sign is not obvious, making an empirical estimation of it one of the key contributions of this study.

Simultaneous Equation (3SLS) Model

Following Krainer (2001) and a number of empirical studies, including those cited above, we specify two market equations in which sales price and liquidity, measured by time on market, are jointly determined. Simultaneous equations take the form:

  • display math(13)
  • display math(14)

where the variables above are defined as in previous sections. We model simultaneity using a three-stage least-squares approach in which (13) and (14) form the system of equations between property price and time on market. In addition, 3SLS incorporates an additional step with seemingly unrelated regression (SUR) estimation to control for correlations between error terms.19 This functional form is similar to the baseline reduced form OLS equation, which also controls for additional registered sex offenders located within one mile. In addition, we will slightly modify the market equations above in the next section by incorporating interaction terms to help explain differences in risk preferences.20

Results—Simultaneous Model with Clustering

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

Consistent with our OLS model, the baseline 3SLS model reveals that nearby sex offender residences reduce nearby home prices and liquidity. Table 4 provides regression results from Equations (13) and (14). The price model reveals qualitatively similar estimates to the OLS estimates from Equation (2), although after accounting for the joint determination of liquidity, these coefficient estimates are slightly smaller. According to Table 4, a sex offender's residence located within 0.1 mile of a home reduces its price by $11,712, or about 7% for the average home. Also consistent with OLS models, the effect is smaller as a home's nearest registered sex offender resides further away. Table 4 shows that a nearby sex offender reduces home prices by $6,149, $3,875 and $3,339 when the nearest sex offender falls within the 0.1–0.25, 0.25–0.5 and 0.5–1.0 mile radii, respectively.

Table 4. The effect of a nearby registered sex offender on a home's sale price and days on market (3SLS baseline)
 3SLS Model3SLS Model
 Dependent Variable:Dependent Variable:
 Sale PriceDays on Market
 [1][2]
Notes
  1. This table presents results of simultaneous estimation of the effect of nearby sex offenders on a home's selling price and liquidity (time on market); t-statistics are in parentheses; ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.

Sex offender ≤ 0.10 mile−11,711.54***87.89***
 (−3.44)(3.38)
Sex offender ≤ 0.25 mile−6,148.71***45.49***
 (−2.67)(2.78)
Sex offender ≤ 0.50 mile−3,874.98**33.56***
 (−2.10)(2.65)
Sex offender ≤ 1.00 mile−3,338.71**23.80**
 (−2.41)(2.46)
Additional S.O. ≤ 1.00 mile−688.99***4.11**
 (−3.09)(2.58)
Property characteristics
Macroeconomic controls
Season fixed effects
Census-block groups
Year fixed effects
Observations12,36212,362
R20.7985−7.3355

Registered sex offenders not only affect a home's value, but also its liquidity or number of days it spends on the market. Table 4 displays a striking result: a home will take approximately 88 days longer to sell (or about 80% longer on average) if a sex offender is located within 0.1 mile. As one would expect, as the nearest sex offender is farther away, this effect diminishes in magnitude to 45, 34 and 24 days when the nearest sex offender falls within the 0.1–0.25, 0.25–0.5 and 0.5–1.0 mile radii, respectively. To put this in perspective, estimates indicate that a home selling during the “off season” (namely, the winter months, as compared to the summer) will take about 19 days longer to sell on average. However, selling a home near a sex offender has an effect approximately four times larger than selling a home in the winter.

Bedrooms and Children

As an additional identification strategy, we analyze observable margins across which buyer and seller preferences are likely to vary. If a homeowner or potential buyer has children, then perceived risk associated with living near a sex offender, and thus the negative externality, should be larger for a subsection of the market. Supported by evidence from the 2009 American Housing Survey, we can reasonably assume that homeowners with more children would be interested in homes with more bedrooms, all else equal, and would exhibit greater risk aversion to residing near a sex offender.21 Table 5 provides evidence to support this hypothesis. For example, a two-bedroom22 home located near a sex offender (<0.1 mile) would sell for little to no discount. However, a four-bedroom home near a registered sex offender sells for $24,947 less than a comparable four-bedroom home that is not near one. Indeed, each additional bedroom in a home within 0.1 mile of a sex offender amounts to an additional $12,476 discount on that home's sale price. Additionally, Table 5 shows a similar result for the effect of bedrooms on the liquidity of homes near a sex offender (<0.1 mile). A two-bedroom home near a sex offender takes 20 days longer to sell than a comparable home not near a sex offender, but a four-bedroom home takes 174 days longer to sell. In fact, each additional bedroom in a home near a sex offender will also stay on the market an average of 77 days more. These results support the hypothesis that families with more children (and thus more bedrooms) likely have a greater aversion to crime risk associated with living near registered sex offenders and are willing to pay more to avoid such risk.

Table 5. The effect of a nearby registered sex offender on a home's sale price and days on market: Bedrooms and violent interaction terms
 3SLS Model3SLS Model3SLS Model3SLS Model
 DependentDependentDependentDependent
 Variable:Variable:Variable:Variable:
 SaleDays onSaleDays on
 PriceMarketPriceMarket
 [1][2][3][4]
Notes
  1. This table adds bedroom (proxy for number of children) and “violent” offender interactions to the 3SLS estimation; t-statistics are in parentheses; interaction terms and their respective dummy variables are jointly significant at the 5% level in all equations; ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.

Sex offender ≤ 0.10 mile24,953.56**−134.51*902.8812.46
 (2.21)(−1.65)(0.14)(0.29)
Sex offender ≤ 0.25 mile−7,382.2919.88−6,661.5751.75
 (−1.13)(0.45)(−1.45)(1.59)
Sex offender ≤ 0.50 mile−1,366.4813.811,396.60−2.39
 (−0.22)(0.33)(0.36)(−0.09)
Sex offender ≤ 1.00 mile−18,581.72***124.97***−1,517.8814.35
 (−3.84)(3.00)(−0.62)(0.85)
Additional S.O. ≤ 1.00 mile−700.64***4.35**−589.68**3.83**
 (−3.13)(2.53)(−2.56)(2.21)
Sex offender ≤ 0.10 mile × bedrooms−12,475.99***77.22***  
 (−3.40)(2.62)  
Sex offender ≤ 0.25 mile × bedrooms353.629.22  
 (0.18)(0.71)  
Sex offender ≤ 0.50 mile × bedrooms−812.006.89  
 (−0.42)(0.54)  
Sex offender ≤ 1.00 mile × bedrooms4,828.57***−31.69***  
 (3.28)(−2.65)  
Sex offender ≤ 0.10 mile × violent  −17,432.10**111.90**
   (−2.45)(2.06)
Sex offender ≤ 0.25 mile × violent  −1,528.336.60
   (−0.32)(0.20)
Sex offender ≤ 0.50 mile × violent  −7,821.68*53.55*
   (−1.91)(1.79)
Sex offender ≤ 1.00 mile × violent  −3,437.9720.52
   (−1.29)(1.11)
Property characteristics
Macroeconomic controls
Season fixed effects
Census-block groups
Year fixed effects
Observations12,36212,36211,85311,853
R2 (overall)0.7990−8.22910.8011−8.4813

Violent versus Nonviolent Sex Offenders

When real estate market participants check the Sex Offender Registry in the state of Virginia, they observe that the Virginia State Police distinguishes between “violent” and “nonviolent” sex offenders on each sex offender's profile, referring to whether the perpetrated crime was determined to be a violent or nonviolent crime. It is reasonable to assume that violent sex offenders are perceived to represent heightened crime risk and thus a larger negative externality. Column (3) shows exactly this result, where a violent sex offender located within 0.1 miles reduces a home's sale price by $17,432, while a nearby nonviolent sex offender has no statistically significant effect. Moreover, column (4) shows that a nearby violent sex offender increases a home's marketing duration by 112 days, while a nearby nonviolent sex offender increases a home's time on market by about a two weeks. The greater impact of violent offenders on price and marketing supports the conclusion that homeowners’ perceived risk is driving these results, implying that the market differentiates between risks posed by registered sex offenders.

Conclusion

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

This study finds that registered sex offenders residing nearby substantially reduce a home's price and liquidity. These results are consistent with the notion that residents of central Virginia are aware of and utilize information available in online sex offender registries, made possible by Megan's Law, to evaluate perceived crime risk when making decisions regarding real estate. We estimate that a sex offender residence located within 0.1 mile lowers a nearby home's price by approximately 7% and substantially lengthens its time on market by as much as 80%. In addition, we estimate reverse treatments, finding that home prices are not significantly different just prior to a sex offender's arrival and prices rebound after sex offenders move-out, providing evidence that the cross-sectional approach produces properly identified causal estimates.

Our study contributes a number of unique elements to the literature by filling important gaps left unexplored by other authors. First, our data allow us to make methodological contributions by exploring potential estimation bias due to sample selection and simultaneity. Second, we emphasize the value of estimating price and liquidity jointly, and the importance of estimating the impact on liquidity in its own right. We consistently find large and statistically significant effects on time on market in 3SLS estimations. Third, by controlling for the presence of potentially multiple nearby sex offenders, we take the most general and representative approach to generating estimates of a sex offender's external effect on nearby real estate. Prior studies are more limited in scope and/or lack the appropriate data to generate such estimates. Fourth, we explore alternative methods for capturing the sex offender externality by estimating a time and distance-weighted continuous variable, finding qualitatively similar results to its binary counterpart. Fifth, we find evidence of differentiated market preferences, captured by the number of bedrooms in a home as a proxy for number of children in a household. Finally, we model unique effects of clustering and also estimate the effect of a “violent” versus “nonviolent” offender, finding that the market does in fact differentiate between the perceived riskiness of nearby offenders. Moreover, our results also show statistically significant effects out to one mile, which is consistent with the notion that residents in suburban and rural areas consider larger areas when defining who constitutes a “neighbor” while assessing subsequent risks.

Unlike other studies, we do not limit the sex offender sample to “established” sex offenders because we include homes near both transient and nontransient sex offenders in our dataset. One reason why our estimates are larger could be that previous studies only estimate the effect of an established offender who has lived in the same residence for an extended period of time. Even though they are in residence for fewer days, the presence of nearby transient offenders can be detected by market participants, potentially affecting reservation and offer prices and therefore all outcomes over the home's entire marketing period. Alternatively, neighbors may have less urgency to sell a home if they have more time to know (or know of) a nearby established sex offender. Or, perhaps buyers may be less reluctant to purchase a home near an established offender that has a longer track record of nonrecidivism. In any case, we did not restrict our sample because our aim at the outset was to take a more generalizable approach. However, differentiated effects of established and transient sex offenders could be an avenue of future research.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References

The authors thank participants at the Applied Research Seminar at the Federal Reserve Bank of Richmond, as well as the participants of the annual conferences of the American Real Estate Society and the American Real Estate and Urban Economics Association for helpful comments. We thank Steven Levitt, Alex Tabarrok, Paul Anglin and Jaren Pope for valuable comments on a prior draft. We are indebted to Dr. Ed Kinman for invaluable assistance in the use of GIS software and to Lt. William Reed, Jr of the Criminal Justice Information Services Division of the Virginia State Police for providing the historical record of sex offender residences. We would also like to give a special thank you to Velma Zahirovic-Herbert for invaluable help with programming in Stata. Any and all errors, of course, are our own.

  1. 1

    States were initially given leeway to determine how to make information available and which crimes were severe enough to warrant public notification. Additional legislation took effect in 2007, creating a national registry and imposing some standardization of state registration and notification requirements. In Virginia, the Sex Offender and Crimes Against Minors Registry is maintained by the State Police and is available at the following URL: http://sex-offender.vsp.virginia.gov/sor/.

  2. 2

    A slightly older third study by Larsen, Lowrey and Coleman (2003) found similar results in Montgomery County (Dayton area), Ohio. They matched data on sex offender addresses with home sales during one year, 2000. Controlling for a vector of house characteristics and neighborhood fixed effects, regression results indicate that living within 0.1 miles of a registered sex offender reduces property values by 17.4%. Significant but diminishing price effects were estimated out to a distance of 0.3 miles.

  3. 3

    In a footnote, Linden and Rockoff (2008, p. 1107) state, “it is possible that our estimates might not be representative of the effects of the average sex offender moving into a neighborhood if offenders who move frequently would cause different changes in property values than offenders who choose to live in a single place for an extended period of time. Unfortunately, measuring the impact of itinerant offenders is not possible, given their short durations and our reliance on sales data.”

  4. 4

    Note that a given state's registry could also shrink if it has a net outflow of sex offenders.

  5. 5

    Pope (2008) estimated the effect a “predator” had on home prices and did not find a significantly larger impact than “nonpredator” sex offenders. Our study, however, has a different finding with respect to “violent” sex offender.

  6. 6

    Consistent with other real estate studies, mobile homes and other outliers were culled, confining the data to a “typical” range of homes from $25,000 to $750,000. Empirical results in this study, however, are not sensitive to dropping these observations. As an additional quality check, a sample of the MLS data was compared to county government records, which contain data on price and housing characteristics. The MLS data were 100% accurate.

  7. 7

    This distance calculation (see below) approximates a straightline (“crow flies”) distance, rather than a driving distance. To illustrate why this might be preferable for this study, consider an example where a sex offender lives in an adjacent property behind a home. That sex offender might actually be relatively far away in terms of driving time. Yet, the sex offender in that situation would likely be perceived just as risky as a sex offender next door (on the same street), as they may pose equal risk to, for example, children playing in the yard.

    inline image

  8. 8

    Recent examples include neighborhood foreclosure effects (Agarwal et al. 2012, Harding, Rosenblatt and Yao 2009, Lin, Rosenblatt and Yao 2009) and urban green space programs (Voicu and Been 2008, Conway et al. 2010).

  9. 9

    Kuminoff, Parmeter and Pope (2010) survey 69 hedonic studies and report that 80% rely on linear, semi-log or log-log functional form. We have explored a number of nonlinear functional forms and our results remain robust. Rather than repeat all of estimated models with various nonlinear explanatory variables, the authors will produce results of alternative specifications upon request.

  10. 10

    We use the following property-specific variables: square footage, age, acreage, number of bedrooms, bathrooms, length of the listing contract, whether the home is a one-story, new, vacant and whether it has a brick exterior, hardwood floors, a pool, a fenced yard and a walk-in closet.

  11. 11

    We use the following time and macroeconomic controls: year the home sold, season the home sold, Consumer Sentiment Index, fixed-rate mortgage interest rate at the sale date, Virginia unemployment rate and the Leading Economic Indicator Index. The macro controls are monthly aggregates, which correspond to the month the home was sold.

  12. 12
  13. 13

    For further discussion, see Heintzelman (2010), explaining: “Essentially, the spatial weighting matrices are implicitly specified to allow for spatial dependence within census-block groups and spatial autocorrelation within census blocks. That is, the spatial weighting matrices (one for the spatial dependence and the other for spatial autocorrelation) are implicitly discrete, where entries are 1 if two observations are in the same census-block group (spatial dependence) or census block (spatial autocorrelation) and zero otherwise. As shown below, my results are robust to changes in the spatial scale of both fixed effects and error clustering. Intuitively, these controls are using the time-series nature of my data to control for the fact that populations sort between neighborhoods based on characteristics that, to me, are unobservable. Left uncontrolled this sorting could lead to both omitted variables bias and spatial autocorrelation.” (Heintzelman 2010, p. 28).

  14. 14

    Some homes in our sample are located near a large cluster of sex offenders. We explored nonlinear terms, which yielded no statistically significant effect. In addition, our results are not sensitive to trimming outliers of ADD_SOi.

  15. 15

    Note that in order to isolate this treatment effect, this specification includes the covariates from the previous regressions to control for the property characteristics, location, time and the presence of sex offenders prior to the sale date. Also note after holding constant the presence of current sex offenders, the sale price of a home should be unaffected by any “sex offender effect” of the SOi_MI parameters unless the market is able to successfully forecast the locational choices of a sex offender before she or he actually moves in. If these parameters were to affect price, it would be far more likely that the catalyst is unobserved heterogeneity or some existing underlying disamenity, rather than the market's ability to see into the future.

  16. 16

    Control variables are discussed in the methodology section, and summary statistics are presented in Table 1. Most of these control variables are commonly used within the real estate literature and so for brevity discussion of results is limited primarily to the variable of interest: proximity to a registered sexual offender.

  17. 17

    Potential bias due to censoring is outside the scope of this article and we leave that for future research.

  18. 18

    For examples of early simultaneous estimations, see Yavas and Yang (1995), Forgey, Rutherford and Springer (1996), Huang and Palmquist (2001), Rutherford, Springer and Yavas (2001) and Knight (2002).

  19. 19

    According to Belsley (1988), 3SLS is used instead of 2SLS in estimating systems of equations because it is more efficient, particularly when there are strong interrelations among error terms.

  20. 20

    While the simultaneous equation model employed here models the joint determination of price and liquidity, like other studies of real estate markets, it omits an important aspect of simultaneity in these transactions: the seller's decision to list the property. Due to data constraints, this cannot be adequately modeled empirically, but it may be an important avenue for future research.

  21. 21

    Our estimates hold attributes such as square footage constant. So, additional bedrooms, holding square footage constant, essentially means that homebuyers are getting more, albeit smaller bedrooms. There are a number of reasons why they might want additional bedrooms holding square feet constant, but the most straightforward explanation is additional children.

  22. 22

    A two-bedroom home is used as the comparison because there are few one-bedroom homes in our sample. The results suggest there may be a premium for one bedroom homes near sex offenders, but the implication is likely a spurious one, given lack of observations with only one bedroom in our sample.

References

  1. Top of page
  2. Abstract
  3. Filling Important Gaps in the Literature
  4. Data
  5. Methodology—Baseline Hedonic Models
  6. Results—Baseline Hedonic Pricing Models and the Reversal Treatment
  7. Simultaneous Equations (3SLS) Methodology
  8. Results—Simultaneous Model with Clustering
  9. Conclusion
  10. Acknowledgments
  11. References
  • Agarwal, S., B. Ambrose, S. Chomsisengphet and A. Sanders. 2012. Thy Neighbor's Mortgage: Does Living in a Subprime Neighborhood Impact Your Probability of Default? Real Estate Economics 40(1): 122.
  • Belsley, D. 1988. Two- or Three-Stage Least Squares? Computational Economics 1: 2130.
  • Conway, D., C. Li, J. Wolch, C. Kahle and M. Jerrett. 2010. A Spatial Autocorrelation Approach for Examining the Effects of Urban Greenspace on Residential Property Values. Journal of Real Estate Finance and Economics 41(2): 150169.
  • Davis, L. 2004. The Effect of Health Risk on House Values: Evidence from a Cancer Cluster. American Economic Review 94: 16931704.
  • Forgey, F., R. Rutherford and T. Springer. 1996. Search and Liquidity in Single-Family Housing. Real Estate Economics 24: 273292.
  • Genosove, D. and C. Mayer. 1997. Equity and Time-to-Sale in the Real Estate Market. American Economic Review 87: 255269.
  • Harding, J.P., E. Rosenblatt and V.W. Yao. 2009. The Contagion Effect of Foreclosed Properties. Journal of Urban Economics 66(3): 164178.
  • Hazard, C. 2010. Sex Offenders as Neighbors a Realty Factor. Richmond Times-Dispatch August 17, D3.
  • Heckman, J. 1979. Sample Selection Bias as a Specification Error. Econometrica 47: 153161.
  • Heintzelman, M.D. 2010. Measuring the Property-Value Effects of Local Land Use and Preservation Referenda. Land Economics 68: 2247.
  • Horn, K.M. 2011.Does School Quality Influence Property Owner Capital Investment Decisions? Working Paper. Presented at the January 2011 ASSA Meetings, Denver, CO.
  • Huang, J. and R. Palmquist. 2001. Environmental Conditions, Reservation Prices, and Time on the Market for Housing. Journal of Real Estate Finance and Economics 22: 203219.
  • Knight, J. 2002. Listing Price, Time on Market and Ultimate Selling Price: Causes and Effects of Listing Price Changes. Real Estate Economics 30(2): 213237.
  • Krainer, J. 2001. A Theory of Liquidity in Residential Real Estate Markets. Journal of Urban Economics 49(1): 3253.
  • Kuminoff, N., C.F. Parmeter and J.C. Pope. 2010. Which Hedonic Models Can We Trust to Recover the Marginal Willingness to Pay for Environmental Amenities? Journal of Environmental Economics and Management 60(3): 145160.
  • Larsen, J., K. Lowrey and J. Coleman. 2003. The Effect of Proximity to a Registered Sex Offender's Residence on Single Family House Selling Price. The Appraisal Journal 71(3): 253265.
  • Levitt, S.D. and C. Syverson. 2008. Market Distortions When Agents Are Better Informed: The Value Of Information In Real Estate Transactions. The Review of Economics and Statistics 90(4): 599611.
  • Lin, Z., E. Rosenblatt and V.W. Yao. 2009. Spillover Effects of Foreclosures on Neighborhood Property Values. Journal of Real Estate Finance and Economics 38(3): 387407.
  • Linden, L. and J.E. Rockoff. 2008. Estimates of the Impact of Crime Risk on Property Values from Megan's Laws. American Economic Review 98(3): 11031127.
  • Pope, J.C. 2008. Fear of Crime and Housing Prices: Household Reactions to Sex Offender Registries. Journal of Urban Economics 64: 601614.
  • Rutherford, R.C., T.M. Springer and A. Yavas. 2001. The Impacts of Contract Type on Broker Performance. Real Estate Economics 29: 389409.
  • Stein, J. 1995. Prices and Trading Volume in the Housing Market: A Model with Downpayment Effects. Quarterly Journal of Economics 110: 379406.
  • Turnbull, G. and J. Dombrow. 2006. Spatial Competition and Shopping Externalities: Evidence from the Housing Market. Journal of Real Estate Finance and Economics 32(4): 391408.
  • Turnbull, G., V. Zahirovic-Herbert and C. Mothorpe. 2013. Flooding and Liquidity on the Bayou: The Capitalization of Flood Risk into House Value and Ease-of-Sale. Real Estate Economics 41(1): 103129.
  • Voicu, I. and V. Been. 2008. The Effect of Community Gardens on Neighboring Property Values. Real Estate Economics 36: 241283.
  • Yavas, A. and S. Yang. 1995. The Strategic Role of Listing Price in Marketing Real Estate: Theory and Evidence. Real Estate Economics 23(3): 347368.