Gentrification and Violent Crime in New York City

Abstract

Research has frequently referenced the influence of gentrification on crime, but only a few studies empirically assessed this relationship. Recent research has utilized innovative measures of gentrification and advanced statistical techniques, but many questions remain unanswered. One such question is whether and to what extent gentrification influenced crime in New York City. The current study used a quantitative operationalization of gentrification that was grounded in qualitative information and hybrid fixed-effects regression to assess whether changes in violent crime rates in New York City were associated with gentrification. Results indicate that sub-boroughs that experienced greater rates of gentrification featured significantly larger declines in assault, homicide, and robbery and that this relationship did not vary significantly over time.

Keywords

gentrification New York City violent crime

Introduction

Crime rates in most major U.S. cities increased between the 1960s and 1980s but then declined dramatically during the 1990s. This decline coincided with the gentrification movement, which began to occur in earnest in U.S. cites during the 1980s and continued during the 1990s and 2000s. While much research has referenced the impact of gentrification on crime, the few studies that empirically assessed the association of gentrification and violent crime produced conflicting results. Some studies reported a positive association (Covington & Taylor, 1989; C. M. Smith, 2012; Taylor & Covington, 1988; Van Wilsem, Wittebrood, & De Graaf, 2006), some reported a negative association (McDonald, 1986; Papachristos, Smith, Scherer, & Fugiero, 2011; C. M. Smith, 2012), and others reported that the association varied over time (Kreager, Lyons, & Hays, 2011; Lee, 2010). The differences in findings resulted from limitations of the analysis strategy including the operationalization of gentrification, how omitted variable bias was addressed, and how the influence of time was controlled.

Much of the previous research analyzed variation in census measures of the socioeconomic and racial/ethnic composition of neighborhoods between decennial censuses to identify gentrified neighborhoods. In contrast, recent studies incorporated non-census measures such as coffee shops (Papachristos et al., 2011; C. M. Smith, 2012) or mortgage lending information (Kreager et al., 2011) to allow for refined measurement of gentrification. These innovative operationalizations were not without limitation, however, as the information sources used were difficult to access for other cities.

Omitted variable bias was more of a problem for early research on gentrification and crime, which utilized descriptive or ordinary least squares regression. Recent research addressed potential omitted variable bias by using instrumental variable (Lee, 2010) and fixed-effects regression (Kreager et al., 2011; Papachristos et al., 2011). Each of these strategies was limited, however, in that instrumental variables can be difficult to identify, while conventional fixed effects regression requires that all of the independent variables vary over time.

Related to potential omitted variable bias was that much of the previous research inadequately controlled for the influence of time. Previous research recognized that the gentrification process unfolded over time and therefore utilized longitudinal data but most did not taken this further than analyzing data across multiple time points or incorporating lagged measures of crime. The importance of this issue was highlighted by Kreager et al. (2011), who found that investment in gentrifying neighborhoods was positively associated with property crime until a “tipping point” of investment was reached, after which investment was negatively associated with property crime.

This article links the decline in violent crime with the spread of gentrification in the 55 New York City sub-boroughs between 1980 and 2009 and advances the study of gentrification and crime in three ways. First, the selection of a quantitative operationalization of gentrification was grounded in qualitative data collected from The New York Times, which was attuned to how residents of New York City perceived cultural variation in neighborhoods during the study period. Second, omitted variable bias was addressed by using a hybrid fixed-effects regression strategy that allowed for the incorporation of time-variant and time-invariant measures. Third, the analysis strategy controlled for the influence of time in two ways, thereby allowing for the determination of whether the influence of gentrification on violent crime rates varied over time. Results indicate that sub-boroughs that experienced greater rates of gentrification during a given decade were more likely to feature lower rates of aggravated assault, homicide, and robbery at the end of the decade. Further results indicate that the negative association of gentrification with each of these crimes was maintained after controlling for variation across time and within traditional predictors of neighborhood crime.

Gentrification Research

Many aspects of gentrification have been debated, but much of the early scholarship sought to explain the emergence of gentrification, which resulted in the development of two explanations. The production explanation argued that gentrification occurred because of the perception that neighborhoods were undervalued and that restoration or development efforts would increase property values to their maximum potential (N. Smith, 1987). In contrast, the consumption explanation argued that gentrification was the result of increased availability of service-oriented jobs and urban amenities that made cities more alluring to young, primarily White, middle-class professionals (Ley, 1996). Contemporary scholarship adjudicated that gentrification was a combination of production and consumption factors, as the process required that neighborhoods contained a combination of undervalued properties and access to amenities that were attractive to middle-class professionals.

An area where gentrification scholars have yet to come to a consensus, however, pertains to the identification of gentrified neighborhoods. Qualitative studies typically discussed how gentrification influenced a single or small group of neighborhoods commonly accepted as gentrified. Quantitative studies, however, took a more varied approach. These studies often utilized a threshold strategy where neighborhoods were classified as “gentrifiable” if they featured a particular characteristic or characteristics at the beginning of a decade and “gentrified” if they experienced a change in the identified characteristic or characteristics over a one-decade period. For example, Atkinson (2000) identified gentrified neighborhoods as those that experienced gains in the proportion of workers employed in professional occupations. An important limitation of quantitative studies was that most relied on official data sources, which often did not include measures of local culture, thereby limiting the measurement of an important aspect of the gentrification process.

Recognizing the limitations of relying solely on census measures to operationalize gentrification, recent research has supplemented census-based strategies with non-census-based measures such as coffee shops (Papachristos et al., 2011; C. M. Smith, 2012) or mortgage lending information (Kreager et al., 2011). The inclusion of these variables allowed for a more refined operationalization of gentrification as such information was usually available on an annual basis and distinguished gentrification from other forms of neighborhood revitalization such as incumbent upgrading. These measures were somewhat limited though. For example, Papachristos et al. (2011) recognized that the location of coffee shops was influenced by city planning efforts, individual tastes, and residential preferences, all of which were difficult to measure with available data. In addition, coffee shops were clustered in the central business district, which meant that a potentially large number of neighborhoods were excluded. Similarly, the home mortgage measure used by Kreager et al. (2011) captured investments made in neighborhoods that gentrified as well as those that did not. Recognizing the strengths and limitations of supplementing census-based operationalizations of gentrification, the current study utilized a quantitative operationalization that was grounded in data collected from The New York Times to incorporate a measure of cultural interpretations of neighborhood change.

Contemporary scholars also debated the impact of gentrification on crime. After two decades of increase, crime rates in many U.S. cities declined during the 1990s. At the same time, many cities saw an increase in the number of affluent and educated residents (Jargowsky & Sawhill, 2006), expansion of owner-occupied housing (Herbert & Kaul, 2005), increased housing values, and the deconcentration of poverty (Jargowsky, 2003; Jargowsky & Yang, 2006), all of which were commonly associated with gentrification. These simultaneous trends suggest a negative correlation of gentrification and crime. Whether and how these trends were associated with neighborhood crime in New York City remains to be determined.

Gentrification and Crime

Research on gentrification and crime has typically drawn on the social disorganization/collective efficacy and routine activities theories. Research by O’Sullivan (2005) was an exception, as this study drew on economic theory to determine whether changes in crime predicted gentrification. Social disorganization/collective efficacy argued that neighborhoods characterized by concentrated disadvantage, racial and ethnic heterogeneity, and high residential mobility were more likely to feature higher crime rates because these characteristics decreased the potential for community development. Residents of such neighborhoods were less likely to use or respond to informal means of social control and therefore were more likely to use formal controls such as the police (Sampson, 2012; Sampson, Raudenbush, & Earls, 1997). Gentrification, however, presents a problem for this framework as it breaks up concentrations of poverty by bringing middle-class residents into disadvantaged areas, which should lead to reductions in neighborhood crime. Conversely, the gentrification process was also associated with increased racial heterogeneity and residential mobility, which have been found to be positively associated with crime.

The other theory frequently used to explain the gentrification and crime relationship was the routine activities theory. According to this perspective, crime was more likely to occur when a suitable target (high value) lacking in capable guardianship converged in space and time with a motivated offender (Cohen & Felson, 1979). Given these tenets, gentrification should be positively associated with crime as the process introduced young, middle-class professionals into disadvantaged neighborhoods populated by impoverished residents who were often resentful that gentrification was occurring in their neighborhood. Gentrifiers were more suitable targets than incumbent residents because they were more likely to possess high-value goods and were not necessarily familiar with techniques for protecting themselves in public (Anderson, 1990). The motivation to offend among incumbent residents might have been guided by wealth disparities or by resentment that incumbent residents felt toward gentrifiers (Freeman, 2006; Taylor & Covington, 1988). Furthermore, gentrification disrupted local guardianship as new residents were less familiar with the neighborhood and incumbent residents may have been unwilling to act as capable guardians due to resentment of the gentrification process.

While previous research has drawn on similar theoretical arguments, empirical analyses produced mixed results, especially with regard to specific forms of violent crime. Research on gentrification and crime in Baltimore during the 1970s found a positive association with assault and homicide (Taylor & Covington, 1988) and robbery (Covington & Taylor, 1989). Similarly, research on this relationship in Los Angeles during the 1990s found that gentrification was positively associated with assault and robbery, but not significantly associated with homicide or rape (Lee, 2010). In contrast, O’Sullivan (2005) found that gentrification in Portland, Oregon, during the 1990s was negatively associated with assault and robbery. Furthermore, analyses of gentrification and violent crime in Chicago during the 1990s and early 2000s identified a negative association with homicide and robbery (Papachristos et al., 2011) but also with gang-related homicide (C. M. Smith, 2012). The contradictory findings with regard to specific forms of violent crime were due to limitations pertaining to the operationalization of gentrification, length of time assessed, and the analysis strategy used.

The strategy used to operationalize gentrification had important implications because it influenced which areas were included in analyses. For example, Covington and Taylor (1989) operationalized gentrification through a single measure of property value, which treated all areas in Baltimore as gentrifiable. In contrast, the more restrictive measures of the number of coffee shops used by Papachristos et al. (2011) and C. M. Smith (2012) only allowed areas where coffee shops could be built to be classified as gentrifiable. While more limited, operationalizing gentrification through the number of coffee shops was an important contribution because it incorporated changes in local culture. The incorporation of culture was important because gentrification was a more subjectively defined urban phenomenon, compared with others like concentrated poverty or residential segregation and therefore was influenced by residents’ perceptions. The current study addressed this by grounding the selection of a census-based operationalization of gentrification in information collected from The New York Times, which was attuned perceptions of neighborhood change among New York City residents. This ensured not only that all of the neighborhoods in New York City had an equal probability of being considered gentrified but also that the operationalization of gentrification was not arbitrarily determined.

The inconsistent results of previous research were also the result of how omitted variable bias was addressed. Research by Lee (2010) addressed this by using instrumental variable regression, while Kreager et al. (2011) and Papachristos et al. (2011) used fixed-effects regression. As described in the analysis strategy section, the current study followed the lead of Kreager et al. (2011) and Papachristos et al. (2011) and controlled for unmeasured variable bias by using fixed-effects regression. Unlike conventional fixed-effects regression, which required all measures be time variant, the hybrid fixed-effects regression technique used in the current study allowed for the inclusion of time-invariant measures that may be equally or more important predictors of crime than the time-variant measures (Allison, 2005).

The contradictory findings were also the result of how time was incorporated, with regard to the study period as well as how changes over time were controlled. Study periods ranged from 5 years (Van Wilsem et al., 2006) to 20 years (Kreager et al., 2011) with an average of about 10 years. Within the overarching study periods, previous research analyzed annual variation (Lee, 2010; Van Wilsem et al., 2006), variation in 3 year averages (Papachristos et al., 2011; C. M. Smith, 2012), and variation between decennial censuses (Covington & Taylor, 1989; Kreager et al., 2011; McDonald, 1986). The use of smaller intervals makes intuitive sense but may not be necessary because neighborhood processes such as gentrification typically occur gradually and have little immediate influence on neighborhood crime (Kirk & Laub, 2010). Therefore, focusing on longer intervals and a broader study period, as is done in the current study, may highlight more of the impact of gentrification on crime rates.

Related to the issue of study period is how time was incorporated into previous analyses. All of the previous research analyzed variation over time, but Lee’s (2010) use of the 1994 Northridge earthquake as an instrumental variable was the first to incorporate time in a meaningful fashion. Lee (2010) found that gentrification that occurred shortly after the Northridge earthquake was positively associated with assault and robbery, which indicated that gentrification had short- and long-term implications for violent crime rates. Kreager et al.’s (2011) findings also indicated short- and long-term implications of gentrification for crime, but only for aggregate total and property crime rates. In contrast, Papachristos et al. (2011) and C. M. Smith (2012) incorporated a lagged measure of the dependent variable and found that gentrification was negatively associated with homicide, robbery, and gang-related homicide, respectively, even after controlling for counts of each crime during the preceding 3-year period. C. M. Smith (2012) also found that gentrification continued to be negatively associated with gang-related homicide after controlling for city-level period effects. While contradictory, the results of these studies indicated the importance of controlling for time, which is done in the current study by incorporating time-specific measures.

Current Study

The current study assessed the association of gentrification and three forms of violent crime in New York City for the period 1980 to 2009. This objective was achieved by working through two sub-goals. First, the selection of a quantitative operationalization of gentrification was grounded in data collected from The New York Times. Second, variation in gentrification and violent crime over a 29-year period was assessed to determine whether the association was varied over time, which was important given the dramatic decline in crime during the 1990s.

Why New York City?

The current study focused on New York City neighborhoods because the tremendous amount of research on gentrification in New York City has included anecdotal evidence that crime rates changed because of gentrification. The only study that empirically assessed this relationship in New York City was a descriptive analysis conducted by McDonald (1986) that included three neighborhoods in Manhattan and one in Brooklyn. As a point of reference, the New York City Department of City Planning (NYCDCP; 2011) recognized 188 populated neighborhood areas in 2011. Furthermore, the descriptive analysis strategy was unable to control for the influence of potentially omitted variables.

In many respects, New York City was similar to other large American cities in that it experienced broad changes to its population in recent decades. For example, crime increased in New York City between the 1960s and 1980s before dramatically declining during the 1990s (Greenberg, 2013; Zimring, 2011). In addition, the influence of rapid population loss and increased concentration of poverty and racial and ethnic minorities because of middle-class flight to suburban areas during the 1960s and 1970s have been well documented. Furthermore, similar to other large American cities, research has documented the role of immigration in shaping the population of New York City (Rosenbaum & Friedman, 2007). Finally, like other major American cities, efforts to revitalize New York City have been underway since the 1980s.

A few features of New York City distinguished it from other major U.S. cities. For example, research by Zimring (2011) indicated that the decline in crime between 1990 and 2009 was substantially larger in New York City than the declines in the 10 largest non-New York City cities. In addition, the size of the population of New York City was an important difference, as current population estimates put the population of New York City at slightly more than eight million residents, which was more than double that of Los Angeles and Chicago, the second and third largest cities, combined (U.S. Census Bureau, 2011). Furthermore, because expansion of New York City took place in a vertical rather than horizontal manner, it was also the most densely populated city in the United States with about 27,000 residents per square mile (U.S. Census Bureau, 2011). Finally, New York City was potentially unique due to the high proportion of renters.

Analysis Strategy

Omitted variable bias is an important consideration for longitudinal research as changes to the relationship in question may be due to changes in unmeasured variables. Recent studies of gentrification and crime utilized instrumental variable and fixed-effects regression strategies. Instrumental variable regression requires the identification of a variable that preceded the independent variable(s) theoretically and that was strongly correlated with the independent variable(s) while simultaneously being weakly correlated with the dependent variable. Finding such variables can be difficult. Lee (2010) utilized the earthquake in Northridge California in 1994, which met these qualifications as it produced a large gentrifiable housing stock and was not significantly associated with changes in crime. Lee (2010) has been the only assessment of gentrification and crime to use this approach.

In contrast to Lee (2010), Kreager et al. (2011) and Papachristos et al. (2011) utilized fixed-effects regression. Fixed-effects regression controls for unmeasured variable bias by allowing for a different intercept for each case so that variation within each case can be assessed independently. This strategy is not without limitations as all of the variables must be time variant and it is impossible to identify what unmeasured variables are excluded.

The current study explored instrumental variable and fixed-regression analysis strategies, but selected fixed-effects regression due to difficulties associated with the identification of an instrumental variable. Unlike prior research, the current study used a hybrid fixed-effects technique that allowed for the inclusion of time-invariant measures (Allison, 2005). The inclusion of time-invariant measures helped determine whether changes in the sampled neighborhoods were due to stationary characteristics such as access to transportation or if the changes were due to variation in population composition. The hybrid fixed-effects regression technique required the assessment of between-unit and within-unit variation for each of the time-variant independent variables, but only the coefficients for within-unit variation measures have a meaningful interpretation (Allison, 2005). Only the coefficients for the within-unit variations are presented in the results tables to facilitate interpretation of the results.¹

Data and Measures

Units of Analysis

The current study analyzes variation in the 55 New York City sub-boroughs for 3 time points. Sub-boroughs were created through a collaborative effort of the New York City Department of Housing Preservation and Development and the U.S. Census Bureau for the purposes of conducting the New York City Housing and Vacancy Survey. Each of the 55 sub-boroughs contains about 40 census tracts and a population of about 100,000 residents. Sub-boroughs were larger than traditional proxies of neighborhoods but have been used to study gentrification and other urban-related processes in New York City (Newman & Wyly, 2006; Rosenbaum & Friedman, 2007). Research on crime in New York City has preferred to use the slightly smaller 76 NYPD (New York City Police Department) police precincts (Messner et al., 2007); the current study used sub-boroughs because their boundaries were coterminous with census tracts and police precincts.^2,3

Dependent Variables: Violent Index Crimes

With the exception of Van Wilsem et al. (2006), who analyzed victimization data, previous studies of gentrification and crime used official crime data to operationalize the dependent variable. While official crime statistics feature a number of known limitations, it makes more sense to analyze variation in official crime data than victimization data because research has noted a greater willingness and ability of gentrifiers to call the police (Freeman, 2006; Skogan, 1990). Therefore, changes in crime among the sampled areas were at least in part a function of more incidents recorded by the police. Given the contradictory findings of previous research, variation in rates of aggravated assault, homicide, and robbery reported to the NYPD were analyzed to determine whether gentrification was differentially associated with each type of crime. Rape was not included due to inconsistent reporting.

The dependent variables were the rates of aggravated assault, homicide, and robbery per 1,000 sub-borough residents. Counts of each crime for the years 1989 through 1991 and 1999 through 2001 at the Community District level were downloaded from Infoshare Online (Infoshare, 2010) and at the precinct level for the years 2005 through 2009 from New York City’s Citywide Performance Tool (New York City Mayor’s Office of Operations, 2011). Precinct crime counts were aggregated to the Community District level for the 2005 to 2009 period using a crosswalk file provided by Infoshare Online and then matched with sub-boroughs using a well-established technique that aggregated three pairs of Community Districts into three sub-boroughs. Once at the sub-borough level, 3-year averages around 1990 and 2000 were computed and then converted to the rate per 1,000 residents. Crime data for the final time point were created by averaging annual crime counts for the 5-year period 2005 to 2009 and then converting to the rate per 1,000 residents. Table 1 shows that the average sub-borough experienced a similar decline in violent crime between 1990 and 2009 that has been documented elsewhere (Greenberg, 2013; Zimring, 2011).

Table 1.

Crime Rate Per 1,000 Population Averages Across Sub-Borough Areas.

Year	1990	2000	2005/2009
Assault	6.08 (4.41)	3.10 (1.86)	2.13 (1.30)
Homicide	0.29 (0.26)	0.08 (0.07)	0.07 (0.05)
Robbery	12.84 (7.00)	3.95 (1.82)	2.82 (1.40)

Note. N = 55 sub-borough areas.

Operationalization of Gentrification

Empirical studies of gentrification overall and its association with crime specifically have frequently operationalized gentrification through changes in a single or small number of census attributes. For example, Covington and Taylor (1989) operationalized gentrification through increased home values, while Van Wilsem et al. (2006) operationalized gentrification through variation in neighborhood-level (socioeconomic change, residential mobility, percent minority) and individual-level (homeownership, number of cars, demographic characteristics) characteristics. This strategy limited the ability to incorporate cultural changes associated with gentrification, which was important given the subjective nature of the definition of gentrification. Papachristos et al. (2011) and C. M. Smith (2012) overcame this limitation by operationalizing gentrification through the number of high-end coffee shops, a measure of local culture. While innovative, this strategy was somewhat limited as the location of coffee shops was influenced by city planning efforts, individual tastes, and residential preferences. Furthermore, coffee shops were clustered in the central business district, which meant that a large number of neighborhoods that underwent similar changes were potentially excluded.

Recognizing the importance of cultural interpretations, the current study uses cultural information to inform the selection of a quantitative operationalization of gentrification. This allowed for the identification of gentrification throughout New York City. To do this, census-based operationalizations of gentrification developed by Bostic and Martin (2003) and Freeman (2005) were replicated with tract-level data from the 1980, 1990, and 2000 Censuses collected from the Neighborhood Change Database (NCDB) and the 2005 to 2009 Five-Year American Community Survey (ACS). All data were in Census 2000 boundaries.

Replication of the Bostic and Martin (2003) operationalization involved first identifying the number of tracts that featured a median family income below the average for New York City at the beginning of each decade and a median family income above the average for the city at the end of the decade. In the second step, each tract was assigned a ranking based on how much the proportion with college degrees, family income, homeownership rates, proportion aged 30 to 44, proportion White non-family households, proportion managerial and administrative workers, and the proportion with some college increased and how much the percentage poverty and Black decreased. The average of the ranked variables was computed and compared with the number of tracts identified in the first step. For example, if the median family income for 152 tracts changed from below to above the average for the city between 1980 and 1990, then the 152 tracts with the lowest average rank value were coded as gentrified during the 1980s.

To replicate the Freeman (2005) operationalization, tracts were classified as gentrifiable if they featured an average household income that was less than the median for the city and featured a proportion of housing built within the past 20 years lower than the proportion found at the median for the city at the start of the decade. Tracts that gentrified had to be gentrifiable at the start of the decade and feature an increase in educational attainment greater than the median for the city as well an increase in real housing prices during intercensal period.

The tracts identified as gentrified during each decade for each strategy were mapped using a shapefile that located tracts within commonly recognized neighborhoods (NYCDCP, 2011). Neighborhoods that contained at least one gentrified tract during a given decade were compared with a list of neighborhoods identified by The New York Times as gentrified between 1980 and 2009 to determine which census-based strategy better matched local cultural perceptions of gentrification. This list of neighborhoods was produced by conducting a simple content analysis of The New York Times articles published between January 1, 1980, and December 31, 2009, that contained the terms gentrification and New York City. A comparison of the neighborhoods identified by The New York Times with those identified by the census-based operationalizations identified the Bostic and Martin (2003) operationalization as a closer match and therefore was used to operationalize gentrification in the current study.⁴

Census-based operationalizations were given priority over The New York Times because media sources often report on “newsworthy” events while overlooking similar events in less noteworthy areas (Sampson, 2012, p. 189). Comparison of the quantitative operationalizations with The New York Times ensured that the richest possible measure of gentrification was used while incorporating the largest number of neighborhoods possible. The final measure of gentrification used in the current study was the percentage of tracts within a sub-borough identified as gentrified by the Bostic and Martin (2003) operationalization during each decade.

Control Variables

Information on the census-based control variables was collected from the NCDB for 1980, 1990, and 2000 and from American FactFinder for 2005 to 2009 (U.S. Census Bureau, 2011). Similar to gentrification, concentrated disadvantage has been operationalized in a number of ways. The current study, however, operationalized concentrated disadvantage with a slight variant of the index identified by Sampson and colleagues (Sampson, 2012; Sampson et al., 1997). The original index included the percentage of residents receiving public assistance, percentage living below poverty, unemployment rate, percentage with less than a high school education, percentage female-headed households, and percentage Black, but confirmatory factor analyses conducted with tract-level data revealed that the percentage with less than high school education did not consistently load highly on a single factor with the other variables and therefore was excluded. Residential stability is operationalized through change in the population who lived in the same home for at least 5 years. Finally, changes in immigration were controlled for by including a measure of the change in the percent foreign born during each decade.

Given the importance of public transportation to New York City residents, the current study incorporated a measure of subway access. Information on the location of subway entrances was downloaded from Spatiality, which was maintained by the Director of Mapping Services at the Center for Urban Research at The City University of New York (CUNY) Graduate Center, Steven Romalewski (2010). The locations of subway entrances were geocoded to census tracts, which was then used to determine the percentage of tracts within a sub-borough that contained at least one subway entrance.

Results

Table 2 presents descriptive statistics for the independent variables. On average, about 7% of tracts within each sub-borough gentrified during each decade. Similar to national trends identified by Ellen and O’Regan (2008) and Jargowsky (2003), the results for the measure of disadvantage show it became less concentrated. The statistics on residential stability show an increase during the 1980s, a decrease during the 1990s, and a dramatic increase during the 2000s. The increase in residential stability during the 2000s may have been the result of the housing market crash, which discouraged or prevented many residents from moving (Frey, 2012). An alternative explanation for the dramatic increase is that the use of the ACS for the final time point captured a greater number of individuals living in the same home for at least 5 years. Statistics on the percentage foreign born show the average sub-borough experienced an increase in this population during each of the decades covered by the study period. The statistics for subway entrances indicate that about 20% of tracts in the average sub-borough contained at least one subway entrance but that there was great variation.

Table 2.

Summary of Census Variable Change Scores at the Sub-Borough Level.

Decade	1980s (1980-1990)	1990s (1990-2000)	2000s (2000-2005/2009)
Percentage gentrified	7.02 (10.01)	7.56 (9.59)	6.46 (8.67)
Concentrated disadvantaged index	−0.04 (0.63)	−0.08 (0.71)	−0.11 (0.90)
Residential stability	3.11% (5.02)	−1.54% (3.28)	27.73% (3.52)
Percentage foreign born	4.48 (3.77)	6.78 (4.38)	0.46 (3.02)
Percentage of tracts containing at least one subway entrance	20.63 (12.39)

Note. N = 55 sub-borough areas.

Table 3 presents results for the hybrid fixed-effects regression analyses, which analyzed variation in the 55 sub-boroughs across three time points. Descriptive analysis confirmed that violent crime rates declined greatly during the 1990s and 2000s, so Models 1, 5, and 9 regressed each violent crime on dichotomous measures of time. The significant and negative coefficients for the 1990s and 2000s for all three types of crime confirm that the declines were statistically significant. Given the significance of these declines, the remaining models control for time to determine whether changes in crime rates were a reflection of changes in the gentrification measure and control variables rather than simply the result of the overall decline in violent crime.

Table 3.

Fixed-Effects Analyses of Violent Crime Rates.

	Assault				Homicide				Robbery
	1	2	3	4	5	6	7	8	9	10	11	12
Decade 1990s (1990-2000)	−2.988*** (0.342)	−2.841*** (0.400)	−2.922*** (0.293)	−2.743*** (0.353)	−0.207*** (0.023)	−0.206*** (0.026)	−0.203*** (0.020)	−0.200*** (0.024)	−8.881*** (0.629)	−8.612*** (0.747)	−8.804*** (0.597)	−8.498*** (0.717)
Decade 2000s (2000-2005/2009)	−3.955*** (0.342)	−4.524*** (1.046)	−4.023*** (0.293)	−4.793*** (0.922)	−0.224*** (0.023)	−0.238*** (0.069)	−0.228*** (0.020)	−0.254*** (0.062)	−10.019*** (0.629)	−12.007*** (1.952)	−10.098*** (0.597)	−12.319*** (1.874)
Concentrated disadvantage	—	0.799** (0.275)	—	0.349 (0.255)	—	0.064*** (0.018)	—	0.037* (0.017)	—	0.830 (0.514)	—	0.307 (0.519)
Residential stability	—	0.026 (0.039)	—	0.031 (0.034)	—	0.001 (0.003)	—	0.002 (0.002)	—	0.100 (0.073)	—	0.105 (0.070)
Foreign born	—	0.004 (0.054)	—	−0.011 (0.048)	—	0.003 (0.004)	—	0.002 (0.003)	—	0.100 (0.101)	—	0.082 (0.097)
Percentage of tracts with subway entrances	—	−1.488 (2.798)	—	−1.763 (2.867)	—	−0.175 (0.146)	—	−0.172 (0.150)	—	0.964 (3.420)	—	0.104 (3.451)
Percentage gentrified	—	—	−0.122*** (0.019)	−0.114*** (0.020)	—	—	−0.008*** (0.001)	−0.007*** (0.001)	—	—	−0.142*** (0.039)	−0.132** (0.041)
Constant	6.084*** (0.389)	3.287* (1.622)	5.229*** (0.481)	2.667 (2.079)	−0.207*** (0.023)	−0.203*** (0.020)	−0.206*** (0.026)	−0.200*** (0.024)	−8.881*** (0.629)	−8.881*** (0.629)	−8.612*** (0.747)	−8.498*** (0.717)

Note. N = 165 (55 sub-boroughs with three observations each). Standard errors in parentheses.

***

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

Models 2, 6, and 10 regressed each of the dependent variables on the control variables. As expected, concentrated disadvantage was positively associated with assault and homicide, indicating that sub-boroughs tended to feature higher rates of assault and robbery if they experienced a deconcentration of disadvantage. The measure of concentrated disadvantage was not significantly associated with robbery. This does not necessarily contradict prior research as these results only identify the association of each crime type with the within-unit variation in concentrated disadvantage. Similarly, the non-significant associations with the other independent variables may be indicative of a similar pattern.

Analyses of the association of gentrification and each of the crime types were supportive of the main argument in this study that the decline in crime in New York City was associated with gentrification (Models 3, 7, and 11). Specifically, the results indicate that sub-boroughs were characterized by lower rates of all three crimes at the end of a given decade when a greater percentage of the sub-borough gentrified. These findings are similar to those reported by Papachristos et al. (2011), who found a negative association of gentrification with homicide and robbery in Chicago during the 1990s and early 2000s. Furthermore, results show that gentrification was negatively associated with all three violent crimes even after controlling for traditional predictors of neighborhood crime (Models 4, 8, and 12).

Given the significance of the declines in each crime type and Kreager et al.’s (2011) finding of a curvilinear association of gentrification and total crime, additional analyses were conducted to highlight the importance of change in the independent variables over time (Table 4). The significant coefficients for the main association of gentrification and each of the violent crime types continued to indicate that sub-boroughs featured significantly lower rates of crime at the end of each decade when they experienced greater rates of gentrification. The interactions of the dichotomous time measures with gentrification, however, do not indicate a curvilinear relationship with any of the violent crimes. Instead, the non-significant associations of the time interactions indicate that the relationship of gentrification and each crime type was stable.

Table 4.

Fixed-Effects Analysis With Time Interactions.

	Assault	Homicide	Robbery
Decade 1990s (1990-2000)	−2.012* (0.810)	−0.190** (0.063)	−4.749** (1.525)
Decade 2000s (2000-2005/2009)	−1.921 (1.904)	−0.148 (0.142)	−2.082 (3.459)
Percentage gentrified	−0.104*** (0.028)	−0.009*** (0.002)	−0.105* (0.051)
Concentrated disadvantage	−1.290** (0.425)	−0.040 (0.032)	−2.588** (0.789)
Residential stability	0.069 (0.048)	0.004 (0.004)	0.256** (0.090)
Foreign born	0.003 (0.068)	0.001 (0.005)	0.109 (0.127)
Percentage of tracts with subway entrances	0.907 (3.409)	−0.050 (0.185)	10.516* (4.502)
Concentrated disadvantage*1990s	1.594** (0.537)	0.092* (0.040)	3.254** (0.987)
Concentrated disadvantage*2000s	2.420*** (0.524)	0.107** (0.039)	3.561*** (0.962)
Residential stability*1990s	−0.089 (0.089)	−0.006 (0.007)	−0.255 (0.162)
Residential stability*2000s	−0.121 (0.079)	−0.007 (0.006)	−0.394** (0.146)
Foreign born*1990s	−0.006 (0.077)	0.000 (0.006)	−0.092 (0.144)
Foreign born*2000s	−0.111 (0.093)	−0.002 (0.007)	−0.138 (0.172)
Subway*1990s	−3.030 (2.357)	−0.128 (0.184)	−12.384** (4.490)
Subway*2000s	−0.785 (2.454)	−0.064 (0.191)	−12.163** (4.660)
Gentrification*1990s	−0.002 (0.044)	0.004 (0.003)	−0.034 (0.079)
Gentrification*2000s	−0.002 (0.046)	0.005 (0.003)	−0.031 (0.083)
Constant	2.5204 (2.292)	0.189 (0.115)	4.273 (2.784)

Note. N = 165 (55 sub-boroughs with three observations each). Standard errors in parentheses.

***

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

In contrast to the associations with gentrification, the association of changes in concentrated disadvantage with each violent crime varied slightly. Results for assault and robbery indicate that sub-boroughs where disadvantage became more concentrated during the 1980s tended to feature lower rates of violent crime at 1990, which is counter to much of the previous research and appears to be isolated to this period. This pattern was unable to be assessed with currently available data but could be the result of decreased willingness of residents of disadvantaged neighborhoods to report crimes to the police (Freeman, 2006; Skogan, 1990).

Results for the 1990s and 2000s indicate that the association of concentrated disadvantage with assault and robbery was significantly different from that of the 1980s. To determine the magnitude of the association for the 1990s and 2000s, the coefficients for the time interactions are added to the coefficient for the main effect. For example, the actual coefficient for the association of concentrated disadvantage and assault for the 1990s was 0.304 (−1.290 + 1.594) and 1.13 (−1.290 + 2.420) for the 2000s. This indicates that sub-boroughs where disadvantage became more concentrated during the 1990s and 2000s were more likely to feature higher rates of assault at 2000 and 2005/2009, respectively. Results for robbery were similar.

The results for homicide indicate that changes in concentrated disadvantage during the 1980s were not significantly associated with homicide at 1990. The results for the 1990s and 2000s indicate that sub-boroughs that experienced gains in concentrated disadvantage during each of these decades were more likely to feature higher homicide rates at the end of the decade. Exploratory spatial data analysis indicated that this contradictory finding was a function of a relatively even spread of homicide throughout New York City during the 1990 time point.

While the results of the main analyses presented in Table 3 did not indicate an association of residential stability with any of the dependent variables, analysis of the interactions with time identified some variation with robbery. Specifically, results show that sub-boroughs that experienced gains in residential stability during the 1980s or the 2000s tended to feature higher rates of robbery at the end of each period. This is contrary to what was expected from the social disorganization perspective but makes sense from the routine activities perspective as residents of stable neighborhoods were potentially higher value targets.

The measure of subway access was included because of the importance of public transit to New York City residents and because previous research has found that public transit nodes attract criminal activity. While results of the main analyses did not identify a significant association of subway access with assault, homicide, or robbery rates, analysis of the variation over time show that subway access was positively associated with rates of robbery at 1990, but lower rates of robbery at 2000 and 2005 to 2009. Kelling and Coles (1996) offered a possible explanation in their discussion of efforts by the NYPD to remove graffiti and homeless people from subway stations during the 1990s. Furthermore, it makes sense that subway entrances would be more likely to feature an association with robbery, as it tends to be a crime of opportunity and would benefit from the easy escape routes offered by subway trains.

Discussion and Conclusion

This study contributed to research on the association of gentrification and violent crime by grounding the selection of a quantitative operationalization of gentrification in a culturally informed data source, a hybrid fixed-effects regression strategy to control for omitted variable bias, and the incorporation of time interactions. Results indicated that sub-boroughs that experienced greater rates of gentrification tended to feature lower rates of all three crimes at the end of each decade. Furthermore, results showed that this association was maintained after controlling for variation across time and within traditional predictors of crime.

The negative association of gentrification with violent crime is consistent with research by Papachristos et al. (2011) who also used fixed-effects regression. It remains unclear why gentrification was negatively associated with violent crime as no study to date has been able to assess the causal mechanism. Social disorganization theory predicted that gentrification could result in increased or decreased crime, while routine activities theory predicted that gentrification would result in increased crime. The negative association of gentrification with all three violent crimes offers stronger support for the social disorganization framework as scholars generally agree that the process results in the deconcentration of disadvantage, which has been found to be associated with reductions in crime.

Currently available data do not allow for the determination of the specific reason(s) for the identified negative associations, but it is possible the decline in violent crime was the result of an increased representation of middle-class culture in lower-class neighborhoods as middle-class culture was less likely to promote violent behavior (Anderson, 1990). Empirical studies of gentrification and crime only recently began to incorporate measures of changes to local culture, but this appears to be the most promising avenue for determining the causal mechanism of the association of gentrification and crime given the emphasis on changes to local culture in the broader research on gentrification.

The consistent negative association with gentrification across time was also interesting given Kreager et al.’s (2011) finding that the association of gentrification with total and property crime rates varied over time. Kreager et al. (2011), however, did not find that gentrification was significantly associated with violent crime. This inconsistency suggests two things. First, the operationalization of gentrification in empirical research may influence whether and to what extent a relationship with crime will be identified. For example, the home investment measure used by Kreager et al. (2011) was directly related to property value and therefore more likely to result in changes in property crime. In contrast, the coffee shop measure used by Papachristos et al. (2011) and C. M. Smith (2012) was more likely to be associated with changes in neighborhood culture such as how the use of violence was viewed. C. M. Smith’s (2012) assessment of the association of different operationalizations of gentrification with gang-related homicide explored this issue, but it remains unclear whether these findings are generalizable to more common forms of crime.

Overall, the results presented in this study indicate that gentrification in New York City was associated with declines in violent crime rates. This supports the logic of recent initiatives by city governments to promote revitalization strategies such as gentrification to reduce the prevalence of social problems such as poverty and crime. It is important to recognize that gentrification may not be the savior of cities, however, as there was widespread concern that gentrification disproportionately affected the poor and racial minorities by shifting them from disadvantaged areas with high potential for profit to disadvantaged areas with limited resources (Anderson, 1990; Freeman, 2006; Newman & Wyly, 2006). This was not explored in the current study because of the difficulties associated with measuring displacement and because research has yet to definitively determine that the process resulted in widespread displacement (Freeman & Braconi, 2004; Newman & Wyly, 2006).

Whether widespread displacement actually occurred may not matter as qualitative research has repeatedly described incumbent residents as being fearful of displacement (Anderson, 1990; Freeman, 2006). This fear of displacement may decrease the probability that incumbent residents will build relationships with gentrifiers, which means that redevelopment and crime reduction efforts might be slowed due to resistance among incumbent residents. Therefore, initiatives that encourage gentrification as a means of reducing crime should do so in a way that is least disruptive to local communities as multiple assessments of collective efficacy found that the influence of changes to the social structure of neighborhoods on crime rates were mediated by collective efficacy (Sampson, 2012; Sampson et al., 1997). The incorporation of local community members in the gentrification process may result in larger declines in violent crime due to the increased willingness of incumbent residents to work with gentrifiers on crime reduction practices.

Footnotes

Acknowledgements

Thank you to Samantha Friedman, Steven Messner, Glenn Deane, Joseph Gibbons, Ed Shihadeh, Rick Weil, and the anonymous reviewers at Crime & Delinquency for their constructive criticisms.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

Michael S. Barton is an assistant professor in the Department of Sociology at Louisiana State University. He received his PhD from the University at Albany. His research interests join the criminological and urban sociological literatures to explore the geographic contexts in which crime occur. His research has appeared in Journal of Criminal Justice, Youth & Society, and Law & Society Review.

References

Allison

P. D.

(2005). Fixed effects regression methods for longitudinal data using SAS. Cary, NC: SAS Institute.

Anderson

(1990). Streetwise: Race, class, and change in an urban community. Chicago, IL: The University of Chicago Press.

Atkinson

(2000) Measuring Gentrification and Displacement in Greater London. Urban Studies, 37, 149-165.

Bostic

R. W.

Martin

R. W.

(2003). Black home-owners as a gentrifying force? Neighborhood dynamics in the context of minority home-ownership. Urban Studies, 40, 2427-2449.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588-608.

Covington

Taylor

R. B.

(1989). Gentrification and crime: Robbery and larceny changes in appreciating Baltimore neighborhoods during the 1970s. Urban Affairs Quarterly, 25, 142-172.

Ellen

I. G.

O’Regan

(2008). Reversal of fortunes? Lower-income urban neighbourhoods in the US in the 1990s. Urban Studies, 45, 845-869.

Freeman

(2005). Displacement or succession? Residential mobility in gentrifying neighborhoods. Urban Affairs Review, 40, 463-491.

Freeman

(2006). There goes the ’hood: Views of gentrification from the ground up. Philadelphia, PA: Temple University Press.

10.

Freeman

Braconi

(2004). Gentrification and displacement: New York City in the 1990s. Journal of the American Planning Association, 70(1), 39-52.

11.

Frey

W. H.

(2012). Population growth in Metro America since 1980: Putting the volatile 2000s in perspective. Washington, DC: Metropolitan Policy Program at Brookings.

12.

Greenberg

D. F.

(2013). Studying New York City’s crime decline: Methodological issues. Justice Quarterly, 31, 154-188.

13.

Herbert

C. E.

Kaul

(2005). The distribution of homeownership gains during the 1990s across neighborhoods. Washington, DC: Office of Policy Development and Research, U.S. Department of Housing and Urban Development.

14.

Infoshare. (2010). Infoshare online. New York: Community Studies of New York. Available from http://infoshare.org/

15.

Jargowsky

P. A.

(2003, May). Stunning progress, hidden problems: The dramatic decline of concentrated poverty in the 1990s (Living Cities Census Series, pp. 1-23). Washington, DC: Brookings Institution.

16.

Jargowsky

P. A.

Sawhill

(2006). The decline of the underclass center on children and families. Washington, DC: Brookings Institution.

17.

Jargowsky

P. A.

Yang

(2006). The “underclass” revisited: A social problem in decline. Journal of Urban Affairs, 28, 55-70.

18.

Kelling

G. L.

Coles

C. M.

(1996). Fixing broken windows: Restoring order and reducing crime in our communities. New York, NY: Touchstone.

19.

Kirk

D. S.

Laub

J. H.

(2010). Neighborhood change and crime in the modern metropolis. Crime and Justice, 39, 441-502.

20.

Kreager

D. A.

Lyons

C. J.

Hays

Z. R.

(2011). Urban revitalization and Seattle crime, 1982-2000. Social Problems, 58, 615-639.

21.

Lee

Y. Y.

(2010). Gentrification and crime: Identification using the 1994 Northridge earthquake in Los Angeles. Journal of Urban Affairs, 32, 549-577.

22.

Ley

(1996). The new middle class and the remaking of the central city. New York, NY: Oxford University Press.

23.

McDonald

S. C.

(1986). Does gentrification affect crime rates? In Reiss

A. J. J.

Tonry

(Eds.), Communities and crime (Vol. 8, pp. 163-203). Chicago, IL: The University of Chicago Press.

24.

Messner

S. F.

Galea

Tardiff

K. J.

Tracy

Bucciarelli

Markham Piper

Vlahov

(2007). Policing, drugs, and the homicide decline in New York City in the 1990s. Criminology, 45, 385-413.

25.

Newman

Wyly

E. K.

(2006) The Right to Stay Put, Revisited: Gentrification and Resistance to Displacement in New York City. Urban Studies, 43. 23-57.

26.

New York City Department of City Planning. (2011). Bytes of the Big Apple. New York: Author. Retrieved from http://www.nyc.gov/html/dcp/html/bytes/applbyte.shtml

27.

New York City Mayor’s Office of Operations. (2011). Citywide performance report. New York City Mayor’s Office of Operations. Retrieved from http://www.nyc.gov/html/ops/cpr/html/home/home.shtml

28.

O’Sullivan

(2005). Gentrification and crime. Journal of Urban Economics, 57, 3-85.

29.

Papachristos

A. V.

Smith

C. M.

Scherer

M. L.

Fugiero

M. A.

(2011). More coffee, less crime: The relationship between gentrification and neighborhood crime rates in Chicago, 1991 to 2005. City & Community, 10, 215-240.

30.

Romalewski

(2010). MTA Subway Data in GIS Format. [Spatiality Blog]. Retrieved from http://spatialityblog.com/2010/07/08/mta-gis-data-update/

31.

Rosenbaum

Friedman

(2007). The housing divide: How generations of immigrants fare in New York’s housing market. New York: New York University Press.

32.

Sampson

R. J.

(2012). Great American City: Chicago and the enduring neighborhood effect. Chicago, IL: The University of Chicago Press.

33.

Sampson

R. J.

Raudenbush

S. W.

Earls

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918-924.

34.

Skogan

W. G.

(1990). Disorder and decline: Crime and the spiral of decay in American neighborhoods. New York, NY: The Free Press.

35.

Smith

C. M.

(2012). The influence of gentrification on gang homicides in Chicago neighborhoods, 1994-2005. Crime & Delinquency, 60, 569-591.

36.

Smith

(1987). Gentrification and the rent gap. Annals of the Association of American Geographers, 77, 462-465.

37.

Taylor

R. B.

Covington

(1988). Neighborhood changes in ecology and violence. Criminology, 26, 553-589.

38.

U.S. Census Bureau. (2011). American Community Survey, 2005-2009. Retrieved from http://factfinder2.census.gov/faces/nav/jsf/pages/index.xhtml

39.

Van Wilsem

Wittebrood

De Graaf

N. D.

(2006). Socioeconomic dynamics of neighborhoods and the risk of crime victimization: A multilevel study of improving, declining, and stable areas in the Netherlands. Social Problems, 53, 226-247.

40.

Zimring

F. E.

(2011). The city that became safe: New York’s lessons for urban crime and its control. New York City, NY: Oxford University Press.