What Predicts Where Sex Offenders Live? An Examination of Census Tract Data in Minnesota

Former Prisoners and Sexual Offenders in Communities

Existing research has shown that many aspects of crime are neither evenly nor randomly distributed across neighborhoods, including aggregate crime rates (Sampson & Groves, 1989; Shaw & McKay, 1942; Morenoff, Sampson, & Raudenbush, 2001), individual offending (Anderson, 2002; Oberwittler, 2004), individual victimization (Miethe & McDowall, 1993; Sampson & Lauritsen, 1990), and even recidivism (Kubrin & Stewart, 2006; Mears, Wang, Hay, & Bales, 2008; Stahler et al., 2013). Certain demographic and economic characteristics can explain the distribution of such phenomena. Overall, areas with increased levels of economic disadvantage, residential instability, and racial and ethnic heterogeneity, among other factors, tend to have higher rates of all types of criminal activity.

Just as criminal activity is not randomly distributed, neither are former prisoners and parolees. A growing body of research has focused on the communities that prisoners come from and return to (e.g., Clear, 2002; Clear, Rose, & Ryder, 2001; Kirk, 2009; Kubrin & Stewart, 2006; Massoglia, Firebaugh, & Warner, 2013; Mears et al., 2008; Rose & Clear, 1998; Stahler et al., 2013). The Urban Institute’s multistate longitudinal study of newly released prisoners, the Returning Home series, found that many prisoners are returning to a small number of neighborhoods marked by multiple indicators of concentrated disadvantage (Baer et al., 2006). More than half of the recently released inmates that Visher and Farrell (2005) followed in Chicago were concentrated in just 7 of the city’s 77 neighborhoods. These neighborhoods had above-average rates of poverty and disrupted households and below-average rates of high school graduation compared with the rest of the city’s neighborhoods. In Maryland, Visher, La Vigne, and Travis (2004) found that almost 60% of state prisoners return to the city of Baltimore. Of the majority of prisoners who returned to Baltimore, more than one third resided in 6 of the city’s 55 neighborhoods. These 6 neighborhoods had higher levels of unemployment, poverty, and disrupted households compared with the rest of the city.

An abundance of research has demonstrated that former prisoners often encounter barriers to stable housing (Metraux, Roman, & Cho, 2007; Roman & Travis, 2004) and tend to return to the same or similar disadvantaged neighborhoods they came from (Clear, 2007; La Vigne, Kachnowski, Travis, Naser, & Visher, 2003; Lynch & Sabol, 2001; Massoglia, Firebaugh, & Warner, 2013; Visher et al., 2004). Additional research has shown that community reentry can be even more difficult for sex offenders compared with all others types of offenders (Bonnar-Kidd, 2010; Farkas & Miller, 2007; Hipp et al., 2010; Levenson, 2008; Socia et al., 2014). That is, much like all convicted felons and former prisoners, sex offenders may be restricted from private or public housing, and they face the same economic limitations that prevent them from renting or purchasing properties. On top of these common reentry hurdles, sex offenders in some states and municipalities face residency restrictions and registration requirements, and are subject to community notification. Their sentences or supervision conditions may also restrict them from living with minors, which could, in effect, preclude them from living with many family members or friends as many returning prisoners often do (Baer et al., 2006).

Confirming that sex offenders may face more limited and disadvantaged living situations upon release from prison, Hipp and colleagues (2010) found that, compared with all other types of offenders included in their sample, sex offenders released from California state prisons moved into neighborhoods with higher levels of disadvantage and residential instability. Moreover, the sex offenders in Hipp et al.’s sample continued to move into more disadvantaged neighborhoods on subsequent moves, indicating a propensity for downward residential mobility among sex offenders. Socia (2014) found disproportionate concentrations of sex offenders in upstate New York communities with higher levels of concentrated disadvantage (i.e., poverty, unemployment, female-headed households, higher proportions of Black residents), available housing, and affordable housing.

Existing research shows that sex offender concentration does not lead to increased victimization of children, although it has been found to be associated with higher rates of sexual offending against adult victims (Socia, 2013b; Tewksbury, Mustaine, & Covington, 2010). Despite evidence that residency restrictions for high-risk sex offenders do little to prevent new sex offenses and may ultimately cause more harm than good (Durling, 2006; Duwe, Donnay, & Tewksbury, 2008; Levenson, 2008; Tewksbury, 2007; Walker, 2007; Zandbergen & Hart, 2006), an increasing number of communities are imposing such restrictions and developing innovative ways to limit sex offender residency (Lovett, 2012, 2013). In Minnesota, Level 3 sex offenders are the only category of sex offenders targeted by the few existing city ordinances that ban offenders from living in most places within city limits. Level 3 sex offenders are individuals convicted of sexual offenses that have passed through the Minnesota state prison system and have been deemed as the highest risk to reoffend. Of the nearly 300 Level 3 sex offenders who lived outside confinement in the summer of 2013, more than 40% resided in the city of Minneapolis. Of those Level 3 offenders who resided in Minneapolis, more than half lived in just 5 of the nearly 200 census tracts in Hennepin County.

The risk of victimization by a sex offender living in close proximity may be minimal (Duwe et al., 2008; Harris & Hanson, 2004), but the presence of sex offenders in neighborhoods has measurable effects on communities. Although broad community notifications informing residents about the presence of high-risk sex offenders may be useful for reducing recidivism (Duwe & Donnay, 2008), there is evidence that these community notifications may also increase fear among residents (Beck & Travis, 2004; Caputo & Brodsky, 2004; Kernsmith, Craun, & Foster, 2009; Pope, 2008; Zevitz, 2003, 2004). Furthermore, community awareness of resident sex offenders can also have an appreciable effect on home values. Studies of housing prices in Ohio (Larsen, Lowrey, & Coleman, 2003), North Carolina (Linden & Rockoff, 2008), and Florida (Pope, 2008) have provided evidence that the presence of registered sex offenders subject to community notification can reduce the values of nearby homes. In addition to the perceived risk of victimization, community residents have tangible incentives to keep sex offenders out of their neighborhoods.

Social Disorganization and Offender Housing Patterns

Social disorganization theory is a common theoretical backdrop to ecological studies of crime-related phenomena, including the present study. Elements of social disorganization theory have evolved since its inception at the Chicago School (Shaw & McKay, 1942), but many of the primary structural indicators of social disorganization have remained constant. Poverty, high residential turnover, and racial and ethnic heterogeneity can create disorder in neighborhoods, and in effect, increase criminal activity.

Since the 1980s, a large body of research has articulated the neighborhood mechanisms that work between ecological characteristics and neighborhood crime, including the breakdown of local peer networks, informal social control, and collective efficacy (e.g., Bellair, 1997; Morenoff et al., 2001; Sampson & Groves, 1989; Sampson, Raudenbush, & Earls, 1997; see Kubrin & Weitzer, 2003). Faced with poverty and given the short tenure of many households, residents in socially disorganized neighborhoods tend to interact less with one another, which erodes the collective ability of the neighborhood to provide informal surveillance and social control. Residents of socially disorganized areas have a limited investment in their neighborhoods and are either unwilling or unable to respond to local issues, including criminal activity.

Studies focusing on the residential patterns of all offenders and sex offenders in particular have also relied on social disorganization framework to guide research and explain findings (e.g., Clear et al., 2001; Hipp et al., 2010; Kubrin & Stewart, 2006; Mears et al., 2008; Socia, 2013a, 2014). Indeed, common measures of social disorganization have been salient factors in research similar to the present study (Socia, 2013a, 2014). Socia (2014) found that a neighborhood-level increase in concentrated disadvantage, housing availability, and housing affordability were all significantly associated with an increase in higher-than-expected rates of registered sex offenders in New York State. In addition to the lure of more affordable rental rates and housing availability, it appears that sex offenders are bound to more socially disorganized neighborhoods either by choice or by default (Hipp et al., 2010). Anecdotal evidence from around the country has demonstrated that organized local residents can push out and ward off sex offender residents (Lovett, 2012, 2013; Stahl & Rao, 2013).

One factor that may have been overlooked in previous studies of sex offender housing patterns is the insulating effects of concentrated affluence and the deleterious effects of extreme poverty in the complete absence of wealth (Chiricos, Barrick, Bales, & Bontrager, 2007; Kubrin & Stewart, 2006; Massey, 1996, 2001; Morenoff et al., 2001; Sampson, Morenoff, & Gannon-Rowley, 2002). The harmful effects of coexisting indicators of disadvantage (e.g., unemployment, disrupted households) are intuitive and well-documented, but multiple studies have also shown that both extremes of income segregation (i.e., wealth relative to poverty and vice versa) can have unique effects on the same social phenomena (Morenoff et al., 2001). Brooks-Gunn, Duncan, Klebanov, and Sealand (1993) found that the presence of affluent neighbors, and not neighborhood-level poverty and joblessness, had a significant effect on individual-level childhood development. Sampson, Morenoff, and Earls’s (1999) multilevel analysis of Chicago neighborhoods revealed that concentrated affluence, more than poverty, consistently and significantly influenced outcomes such as informal surveillance of neighborhood children and willingness to engage in reciprocal relationships with other neighborhood residents (e.g., willingness to do favors for one another, frequency of get-togethers with neighbors).

Levels of economic inequality and income segregation have been increasing over the past few decades (Reardon & Bischoff, 2011). As affluent communities become more isolated from less affluent communities, resources also become more concentrated in certain areas, which can affect several community-level outcomes. Given the evidence provided by Sampson et al. (1999), that concentrations of income levels influence the willingness of neighborhood residents to intervene in the affairs of the neighborhood, it seems plausible that very wealthy neighborhoods could exert passive or active informal social control that prevents sex offenders from taking up residence in those neighborhoods. Conversely, the limited liability of residents in neighborhoods marked by extreme poverty in the complete or near absence of wealth may leave them unable or unwilling to deflect sex offender residents. Moreover, the relative anonymity that is provided by poor neighborhoods that lack collective efficacy and informal social control may be appealing to sex offenders, particularly the conspicuous high-risk sex offenders subjected to community-wide notification. Thus, in addition to common indicators of social disorganization and neighborhood disadvantage, the present study also uses a measure that captures the range of socioeconomic statuses that may exist in one community.

Sex Offenders in Minnesota

Although many sexual assaults are never reported to the police, recent data from 2008-2012 indicate there are, on average, a little more than 2,000 sex offenses reported to the police each year in Minnesota. In 2012, 829 of the 2,060 reported sexual assaults (40%) resulted in an arrest (Minnesota Bureau of Criminal Apprehension, 2013). Sentencing data from the same year show that 544 offenders were convicted and sentenced for a felony-level criminal sexual conduct (CSC) offense (Minnesota Sentencing Guidelines Commission, 2013).

Although a felony is the most serious offense level in both Minnesota and the United States, about 40% of those convicted of a CSC offense in Minnesota are sentenced directly to prison. The remaining 60% convicted of a CSC offense are sentenced to probation involving community supervision and/or are required to serve time (no greater than 365 days) in a local county jail. Of the sex offenders who are not sentenced directly to prison, the vast majority are sentenced to local confinement (Minnesota Sentencing Guidelines Commission, 2013). Offenders who violate their conditions of probation, however, are subject to having their supervision revoked and their prison sentence executed, resulting in admission to prison.

Since 1991, Minnesota has required convicted sex offenders to register their addresses with the Minnesota Bureau of Criminal Apprehension (BCA). Minnesota’s predatory offender registration (POR) law has been amended several times over the last few decades, primarily to widen its scope and increase the penalties for noncompliant offenders. Three years after the POR law was enacted in Minnesota, Congress passed the Jacob Wetterling Act, which required sex offenders in all states to register identifying information with law enforcement agents. Two years later, as part of Megan’s Law, the Wetterling Act was amended to allow for public dissemination of some registry information.

When Megan’s Law was passed in 1996, states were required to develop procedures to inform communities where sex offenders will be living. Yet, because states were given a lot of discretion in applying the law, community notification has varied widely across the United States. At one end of the spectrum are states that have different levels of community notification that vary according to sexual recidivism risk. That is, the higher the sex offender’s risk to reoffend sexually, the greater the degree of notification. At the other end of the spectrum are states that make information publicly available, often via the Internet, on all sex offenders regardless of risk.

In Minnesota, which implemented the Community Notification Act on January 1, 1997, community notification applies only to offenders released from prison who are subject to POR. The degree to which the community gets notified depends on the risk level assigned to offenders prior to their release from prison by an End of Confinement Review Committee (ECRC), which is comprised of the prison warden or treatment facility head where the offender is confined, a law enforcement officer, a sex offender treatment professional, a prison caseworker, and a victim services professional. In assigning risk levels, the ECRC considers scores from actuarial risk assessment tools as well as additional factors that ostensibly may either increase or decrease the risk of re-offense (Duwe & Donnay, 2008).

For offenders receiving a Level 1 assignment (i.e., lowest public risk), notification includes victims, witnesses to the crime, law enforcement agencies, and anyone else identified by the prosecutor. For offenders given a Level 2 assignment (i.e., moderate public risk), notification includes those in the Level 1 information release plus schools and day care centers, and other organizations where individuals who may become victims of the offender are regularly found. For offenders assigned a Level 3 (i.e., highest public risk), broad public notification is required. More specifically, law enforcement is responsible for notifying the community where the Level 3 offender will be residing, generally by holding a public meeting in addition to distributing information through the media. Furthermore, following release from prison, the residential vicinities of Level 3 offenders are published on the Minnesota Department of Corrections’ (MnDOC) website (Duwe & Donnay, 2008). Historically, approximately 55% of predatory offenders released from prison receive a Level 1 assignment, 30% receive a Level 2 assignment, and 15% receive a Level 3 assignment.

In an effort to address inconsistencies across states regarding their registration and notification requirements, in 2006, Congress enacted the Sex Offender Registration and Notification Act (SORNA), which is Title I of the Adam Walsh Act (AWA). By introducing a comprehensive set of minimum registration and notification standards for states to follow, SORNA extended the provisions of the Wetterling Act by expanding the list of registerable offenses, requiring states to use a three-tiered classification system based on the convicted sex offense, requiring in-person appearances to verify registration information, and increasing the amount of registration information collected and disseminated. Moreover, SORNA required states to retroactively implement these standards on sex offenders within the criminal justice system by 2009. As of 2012, only 16 states (19 total jurisdictions) had substantially implemented the SORNA requirements, according to the Department of Justice’s Sex Offender Sentencing, Monitoring, Apprehending, Registering, and Tracking (SMART) office (U.S. Government Accountability Office, 2013). Results from the 2013 report prepared by the U.S. Government Accountability Office indicate that some of the more common challenges jurisdictions faced in implementing SORNA included conflicts between SORNA and state laws, the costs to implement SORNA, the retroactive application of the standards, the offense-based classification system, and the application of juvenile registration requirements. As one of the nonimplemented jurisdictions, Minnesota’s registration and notification practices have not been appreciably altered by the SORNA requirements.

Although a majority of states now have residency restriction laws for convicted sex offenders, Minnesota does not. Still, three cities within the state have passed ordinances limiting where certain sex offenders can live. In 2006, Taylors Falls and Wyoming each enacted ordinances that prohibit sex offenders who have committed a sex offense against a juvenile or been given a Level 3 designation from living within 2,000 feet of child congregation locations such as schools, playgrounds, churches, and bus stops. In May 2010, another city in Minnesota, Duluth, passed an ordinance that restricts Level 3 offenders from living within 2,000 feet of a church, playground, or day care.

By examining factors that predict where sex offenders live in Minnesota, this study makes three main contributions to the literature. First, previous studies have analyzed concentration for a city, county, or part of a state (Socia, 2013a, 2013b, 2014). We extend research in this area by examining concentration throughout the state of Minnesota. Second, by looking at different legislative categories of sex offenders, we not only account for residency restrictions laws, which previous studies have done, we also look at whether community notification has an impact on sex offender concentration patterns. Last, by adding a measure of concentrated affluence and relative inequality, this study also accounts for the protective mechanisms of concentrated neighborhood wealth, which previous similar studies have overlooked.

Data and Method

Results

Descriptive Statistics and Correlations

Table 1 displays the means, standard deviations, and bivariate correlations for each of the measures included in the analyses. ² Census tracts in Minnesota had an average of 7.4 POR offenders and 2.4 Level 1 and Level 2 sex offenders. Nearly 9% of census tracts had at least one Level 3 sex offender. The measures of concentrated disadvantage, residential instability, and ethnic heterogeneity were all factor scores, so they averaged at 0 with standard deviations of 1. Because positive values are indicative of more wealth relative to poverty, the average ICE value of 0.27 suggests that neighborhoods in Minnesota skewed more toward affluence. The mean housing affordability value was −0.02, suggesting lower levels of housing affordability on average across the state. The average vacancy rate across Minnesota neighborhoods was about 8%, and neighborhoods had an average of 25 residents per square mile. An average of about 14% of neighborhood residents were a race other than White or Caucasian.

OPEN IN VIEWER

Table 1.

Correlations, Means, and Standard Deviations of Sex Offenders in Minnesota Census Tracts and Covariates (N = 1,329).

	1	2	3	4	5	6	7	8	9	10
1. POR count	1.00
2. Level 1 and Level 2 count	.81**	1.00
3. Level 3 dichotomous	.33**	.36**	1.00
4. Concentrated disadvantage	.46**	.51**	.26**	1.00
5. ICE	−.45**	−.45**	−.20**	−.76**	1.00
6. Residential instability	.28**	.33**	.16**	.59**	−.60**	1.00
7. Housing affordability	.22**	.21**	.06*	.31**	−.49**	.28**	1.00
8. Housing availability	.15**	.14**	.09**	.13**	−.12**	.06*	.06*	1.00
9. Population density	.16**	.27**	.17*	.46**	−.31*	.62**	.11**	.03	1.00
10. Ethnic heterogeneity	.20**	.25**	.10**	.58**	−.36**	.51**	.14**	.07*	.54**	1.00
M	7.40	2.38	8.9%	0.00	0.27	0.00	−0.02	7.9%	25.4	0.00
SD	6.32	2.97	0.28	1.00	0.24	1.00	0.29	8.42	33.1	1.00

Note. POR = predatory offender registration; ICE = Index of Concentrated Extremes.

*

p < .05. **p < .01.

Multivariate Analyses—POR Offenders

The Incident Rate Ratios (IRRs) from the four models predicting the number of POR offenders are displayed in Table 2. First, referencing Models 1 and 2, concentrated disadvantage had a positive and significant effect on the number of POR offenders. Before controlling for spatial autocorrelation, the incident rate of POR offenders in neighborhoods increases by 62% for every 1-unit increase in the level of concentrated disadvantage. Including the spatial lag control reduced the size of concentrated disadvantage’s effect by about 22%, but the effect remained positive and significant. Residential instability had a modest positive, yet significant, effect on the incident rate of POR offenders, and this effect increased in size after controlling for spatial lag.

OPEN IN VIEWER

Table 2.

Incident Rate Ratios From Poisson Regression Models Predicting POR Offender Counts in Census Tracts (N = 1,329).

	1	2	3	4
Concentrated disadvantage	1.622***	1.265**
	(0.056)	(0.076)
ICE			0.093***	0.193***
			(0.162)	(0.225)
Residential instability	1.064*	1.106**	0.973	1.012
	(0.029)	(0.030)	(0.045)	(0.051)
Housing affordability	1.878***	1.804***	1.269*	1.360***
	(0.100)	(0.093)	(0.100)	(0.014)
Housing availability	1.010***	1.007**	1.010*	1.006*
	(0.003)	(0.002)	(0.004)	(0.001)
Population density	0.998	0.996**	1.000	0.997*
	(0.001)	(0.001)	(0.001)	(0.001)
Ethnic heterogeneity	0.891***	0.953*	0.983	0.996
	(0.289)	(0.027)	(0.032)	(0.030)
Spatial lag		1.326***		1.233**
		(0.027)		(0.002)
Constant	0.002***	0.001***	0.003***	0.002***
	(0.066)	(0.068)	(0.056)	(0.192)

Note. Standard errors clustered by county. POR = predatory offender registration; ICE = Index of Concentrated Extremes.

*

p < .05. **p < .01. ***p < .001.

Consistent with expectations and similar research (Socia, 2014), more affordable housing and increased levels of housing availability were both significantly associated with increased rates of POR offenders (Table 2, Models 1 and 2). A 1-unit increase in housing affordability was associated with an 88% increase in the incident rate of POR offenders, and an increase in housing availability increased the incident rate of POR offenders by 1%. These relationships were reduced only slightly in size but remained significant after controlling for spatial autocorrelation. Areas with cheaper rents and more available units also have higher rates of POR offenders.

Contrary to expectations, population density and ethnic heterogeneity were both negatively associated with rates of POR offenders. Prior to controlling for spatial autocorrelation, population density was very slightly, negatively, and not significantly associated with the rate of POR offenders, but this relationship increased in size and became significant once spatial autocorrelation was controlled for. Similar to what Socia (2014) found predicting distributions of registered sex offenders, an increase in ethnic heterogeneity resulted in a decrease in the incident rate of POR offenders (−11%, p < .001), but the magnitude of this relationship was reduced in size after controlling for spatial lag.

Referencing Models 3 and 4 in Table 2, we observe that a 1-unit increase in ICE is associated with a nearly 91% decrease in the incident rate of POR offenders. This effect is reduced to an 81% decrease when the spatial autocorrelation control is added (Model 4), but remains statistically significant.

Including ICE in the models rather than concentrated disadvantage appears to change the relationship between residential instability and the rate of POR offenders. This effect is not statistically significant and very close to zero in Models 3 and 4. Housing affordability is positive and significant in Models 3 and 4 as it was in Models 1 and 2, but it is smaller in size in models that include ICE compared with models that include concentrated disadvantage. Also different in Models 3 and 4 compared with Models 1 and 2, the effect of ethnic heterogeneity on the incident rate of POR offenders living in neighborhoods is very close to zero and is not significant. The other relationships between neighborhood characteristics and the incident rate of POR offenders are the same or very similar in Models 3 and 4 compared with Models 1 and 2.

Multivariate Analyses—Level 1 and Level 2 Sex Offenders

Table 3 displays the results of analyses predicting the rate of Level 1 and Level 2 sex offenders both with and without the control for spatial autocorrelation and with measures of concentrated disadvantage and ICE. Comparing the results in Table 2 with the results in Table 3, we observe that concentrated disadvantage and ICE maintain similar relationships with the outcomes. Two notable differences are that housing availability and ethnic heterogeneity are only significant in one model (Model 1), and are both very close to zero and nonsignificant in the other three models. Overall, comparing the results in Tables 2 and 3 (particularly the models that control for spatial lag) shows that the more conspicuous sex offenders who have passed through prison, but are not subject to wide community notification (i.e., community meetings, media announcements; Table 3), have similar housing patterns to all of the sex offenders combined (Table 2).

OPEN IN VIEWER

Table 3.

Incidence Rate Ratios From Poisson Regression Models Predicting Level 1 Level 2 Sex Offender Counts in Census Tracts (N = 1,329).

	1	2	3	4
Concentrated disadvantage	1.802***	1.269**
	(0.102)	(0.077)
ICE			0.059***	0.170***
			(0.244)	(0.273)
Residential instability	1.101	1.128**	1.023	1.022
	(0.152)	(0.039)	(0.064)	(0.051)
Housing affordability	2.406***	2.525***	1.493	1.818***
	(0.239)	(0.180)	(0.216)	(0.175)
Housing availability	1.012*	1.001	1.009	1.003
	(0.001)	(0.004)	(0.005)	(0.006)
Population density	1.000	0.995***	1.002	0.996
	(0.003)	(0.001)	(0.003)	(0.002)
Ethnic heterogeneity	0.876***	0.977	0.994	1.022
	(0.039)	(0.052)	(0.047)	(0.053)
Spatial lag		2.394***		2.246***
		(0.209)		(0.203)
Constant	0.001***	<0.001***	0.001***	0.001***
	(0.103)	(0.104)	(0.107)	(0.137)

Note. Standard errors clustered by county. Census tract population totals included as exposure terms in all models. ICE = Index of Concentrated Extremes.

*

p < .05. **p < .01. ***p < .001.

Multivariate Analyses—Level 3 Sex Offenders

The results of the probit models ³ predicting the presence of any Level 3 sex offenders in neighborhoods both with and without spatial lag controls and with concentrated disadvantage and ICE are displayed in Table 4. Models 1 and 2 show that concentrated disadvantage significantly increases the predicted probability of whether or not a Level 3 sex offender resides in a neighborhood. However, once we control for the presence of Level 3 sex offenders in contiguous neighborhoods, the size of this effect is more than cut in half, and it is no longer significant. Residential instability and housing affordability both positively increase the predicted probability of neighborhoods having at least one Level 3 sex offender resident in Model 2, but neither of these effects is significant. Housing availability significantly increases the predictive probability of the presence of Level 3 sex offenders in Model 1, but this effect is halved and reduced to nonsignificance once the spatial lag measure is added. An increase in population density significantly increases the predicted probability of Level 3 sex offender residents, but this effect reverses in direction and turns nonsignificant once spatial autocorrelation is controlled for.

OPEN IN VIEWER

Table 4.

Probit Regression Models Predicting Binary Presence of Level 3 Sex Offenders in Census Tracts (N = 1,329).

	1	2	3	4
Concentrated disadvantage	0.325***	0.148
	(0.030)	(0.115)
ICE			−1.386***	−0.991*
			(0.206)	(0.465)
Residential instability	0.018	0.024	−0.027	−0.013
	(0.047)	(0.055)	(0.055)	(0.061)
Housing affordability	−0.032	0.036	−0.204	−0.116
	(0.194)	(0.184)	(0.222)	(0.242)
Housing availability	0.011**	0.005	0.013*	0.008
	(0.004)	(0.005)	(0.006)	(0.007)
Population density	0.004**	−0.004	0.005**	−0.002
	(0.001)	(0.003)	(0.001)	(0.004)
Ethnic heterogeneity	−0.154***	−0.059	−0.052	−0.020
	(0.037)	(0.040)	(0.028)	(0.039)
Spatial lag		1.810		1.346
		(1.020)		(1.029)
Constant	−1.634***	−1.515***	−1.308***	−1.347***
	(0.078)	(0.106)	(0.081)	(0.084)

Note. Standard errors clustered by county. ICE = Index of Concentrated Extremes.

*

p < .05. **p < .01. ***p < .001.

Increased levels of ethnic heterogeneity decrease the predicted probability of the presence of Level 3 sex offenders (Model 1), but this effect is cut in half and turns nonsignificant once the spatial lag control is added (Model 2). This pattern is similar to the results displayed in the same models in Tables 2 and 3.

Turning to Models 3 and 4 in Table 4, ICE appears to be a more consistent predictor of whether or not a Level 3 sex offender resides in a neighborhood compared with concentrated disadvantage. An increase in ICE significantly lowers the predicted probability of having a Level 3 sex offender resident, and this effect decreases only slightly but remains significant when controlling for spatial lag (coefficient = −0.930, p < .05 in Model 4). Unlike Models 1 and 2 in Table 4, an increase in the percentage of non-White residents is positively and significantly associated with an increase in the predicted probability of Level 3 sex offender residents, even after controlling for spatial lag. Besides the effect of neighborhood race, Models 3 and 4 are very similar to Models 1 and 2 in Table 4 in terms of the size, direction, and significance of coefficients. ⁴

Discussion and Conclusion

The purpose of this research was to examine the factors that are associated with the concentration of sex offenders in a large geographic area with few residency restrictions. By examining multiple categories of sex offenders subject to varying levels of public notification, we were also able to assess what, if any, effect community notification has on the housing patterns of sex offenders. Moreover, we extended the literature on the housing patterns of offenders and sex offenders in particular by accounting for extreme poverty and affluence.

When comparing the concentrations of POR and Level 1 and Level 2 sex offenders alone, we found few differences in the factors associated with these outcomes. Concentrated disadvantage, ICE, and housing affordability were all consistently significantly associated with the outcomes across different models with and without controls for spatial lag. Both concentrations of POR offenders and Level 1 and Level 2 sex offenders increased in the presence of concentrated disadvantage and more affordable housing, whereas these incident rates of sex offender residents decreased as levels of concentrated affluence increased. Most of these findings are consistent with similar research (Socia, 2014).

Comparing the rate of Level 3 sex offenders versus POR and Level 1 and Level 2 sex offenders combined is more difficult given the low number of Level 3 sex offenders distributed across the state, as well as the different analytical technique used to examine this outcome. Comparing only the direction and significance of the coefficients, ICE is the only measure consistently associated with all three outcomes both with and without the spatial lag controls. Taken together, regardless of how conspicuous these offenders were in the community, they are all less likely to be found in more affluent areas.

Our results indicate that concentrated disadvantage helps explain why certain neighborhoods are more likely to have a higher population of convicted sex offenders. However, it does not tell the whole story because concentrated disadvantage may not fully account for the potential protective effects of affluent neighborhoods (Morenoff et al., 2001). ICE was a strong and consistent predictor of sex offender concentration or presence in all of the models, which highlights the importance of including indicators of extreme wealth as well as indicators of extreme poverty in community-level research (Kubrin & Stewart, 2006; Morenoff et al., 2001). Just as sex offenders may be pushed into communities marked by poverty and other indicators of social disorganization, the other tail end of the economic distribution is insulated from the presence of all levels of sex offenders. Even after controlling for the affordability and availability of housing, an increase in the level of affluence relative to poverty is associated with a decrease in all categories of sex offenders, regardless of the presence of community notification.

Why do extreme levels of wealth and poverty exert such a strong influence on where convicted sex offenders live? When neighborhoods are more affluent, they typically have more social capital and greater access to social and institutional resources. Moreover, neighborhood affluence is positively associated with collective efficacy (Maimon & Browning, 2012), which increases the capacity for informal social control. The presence of a convicted sex offender in the community is generally perceived as a threat to public order. Sex offender concentration not only reduces home values, but it also inflames concern for the safety of children. Indeed, protecting children has ostensibly been one of the main reasons for the proliferation of sex offender residency restriction laws across the United States over the last decade. Because affluent neighborhoods are more likely to have social capital and collective efficacy, they possess the resources and means to prevent sex offenders from living in their communities.

Impoverished neighborhoods, however, seldom have adequate access to powerful social, economic, and political institutional resources. Nor do they tend to have the shared expectations for collective action that might result in an organized effort to successfully drive out sex offenders from their communities. Therefore, considering that sex offenders are widely regarded as social pariahs, they gravitate toward the areas that are least able to effectively resist their presence in the community.

Although this research has some strengths, it does come with three major limitations. First, this study is cross-sectional. Thus, we are only able to measure the factors that are associated with the concentration of sex offenders, and not factors that increase the concentration of sex offenders. As Hipp et al. (2010) notes, many neighborhoods across the United States are locked in a “cycle of disadvantage” (p. 559), confounded by poverty, crime, disorder, and the exit and reentry of offenders. It is possible that structural characteristics can increase the concentration of sex offender residents, whereas the presence of sex offenders can further erode certain structural characteristics. Data spanning multiple time points are needed to examine the potential causal relationships between structural characteristics and the concentration of sex offenders. Second, we did not specify the mechanisms that explain the relationship between ICE and the concentration of sex offenders. It appears that sex offenders are deterred from more affluent neighborhoods, and the price and availability of these markets do not explain this relationship. Future research should attempt to identify the community-level processes that may mediate the link between structural characteristics and the housing patterns of this conspicuous group of offenders, outside residency restriction laws and ordinances. Finally, there are likely many individual-level variables that explain the living patterns of all sex offenders who are not included in this research. For example, the prior criminal histories of these sex offenders or the severity of their offenses could either make it easier or more difficult for them to find housing in less disadvantaged areas.