Partial Results from a Contract Issued in Response to the New York City Police Reform and Reinvention Collaborative Plan
by John K. Roman
Gregory Haugan
Benjamin Schapiro and
Sofia Rodriguez
NORC at the University of Chicago
May 2024
In 2021, the New York City Mayor’s Office of Criminal Justice (MOCJ) contracted with several research centers at John Jay College of Criminal Justice to support research and/or technical assistance related to 6 of the 132 reform initiatives contained within the New York City Police Reform and Reinvention Collaborative Plan (“the Plan”). The Plan was the result of more than 85 listening sessions, roundtable discussions, town halls, and stakeholder engagement meetings conducted by NYPD and community partners. In March 2021, the NYC Council adopted the Plan via Resolution 1584 pursuant to State Executive Order Number 203. The City’s Reform Implementation Working Group was tasked with implementing and monitoring the progress of the 132 reform initiatives within the Plan. This report from NORC at the University of Chicago is aligned with reform initiative 84 of the Plan as directed by MOCJ and overseen by John Jay’s Research and Evaluation Center.
Summary
What are the effects of drug-related arrests on New York City neighborhoods? Effects of criminal justice actions (including incarceration, arrest, misdemeanor conviction, and/or community supervision) extend beyond individuals and families to include the well-being of local communities. Collateral, unintended consequences of drug arrests include negative education and socioemotional outcomes for children whose parents have been incarcerated, shocks to local crime dynamics that impact levels of violence, and even effects on property values. Local economies and drug arrests affect each other in ways that are difficult to disentangle.
Drug arrests tend to cluster in particular places and those clusters may persist over long time periods. Places where drug arrests cluster may be more amenable to drug trafficking and less amenable to legal economic activity, and illegal drug activity may lead places to become less viable economically. Interactions between the micro economy and drug trade are not easily understood. Statistical models built with administrative data may lead to biased estimates. One solution is to employ statistical techniques to separate the effects of system-level shocks associated with drug arrests (e.g., policy changes, supply shocks), from the effects of local factors that may be tied up with local property values or general levels of violence (e.g., migration or gentrification). Data from system shocks may provide an unbiased estimate of the effect of drug activity on the microeconomy.
Using data from New York City’s Open Data Portal and other administrative data sets, researchers from NORC at the University of Chicago measured the collateral consequences of arrests for drug law violations. The spatial patterns of drug arrests over time were examined to test for a disproportionate impact on communities of color and economically disadvantaged communities. Annual drug arrests in individual census tracts were predicted by examining drug arrests in the rest of the city that year and weighting them by the share of all homicides in the city that occurred in the census tract in the baseline (i.e., pre-analysis) period. This is a variation on an instrumental variables approach known as a Bartik Instrument and is referred to as a “Bartik Shock.”
Results suggest that arrests for drug-law violations in a neighborhood reduce property tax assessments and the effect takes up to three years to manifest. The delay may be due to the time it takes for changes in drug arrests to affect perceptions of safety. On the other hand, drug arrests could have more immediate effects on perceptions of safety but delays in the administrative process of assessing property values could account for a lag in values.
Researchers also tested whether the effects of drug arrests vary according to a community’s socioeconomic status and racial composition. The results suggest that negative effects of drug arrests in the third year after the tax assessment are greatest in areas in the bottom quartile for median income. Effects of arrests on property values were also greater in communities of color, particularly driven by census tracts in the upper quartile for their share of the city’s Hispanic population.
The general goal of the analysis was to determine whether more drug arrests in a neighborhood are associated with changes in a community’s economic well-being. The results indicate that, on average, a one percent increase in drug arrests is associated with a two percent decline in assessed property values, and the effect is lagged: drug arrests this year tend to affect property tax assessments three years from now.
Introduction
What are the economic effects of drug-related arrests on New York City neighborhoods? Can variations in drug-related criminal justice contacts from 2009-2020, driven primarily by changes in policing and enforcement policies, be used to estimate the consequences of drug-related arrests on New York City communities? With census tracts as the unit of analysis, researchers from NORC at the University of Chicago estimated the economic consequences of drug-related arrests on property values (as measured through property tax assessments) and gun-related homicides.
The economy of an area and the number of drug arrests likely affect each other in ways that are difficult to disentangle. Many reasons why drug arrests may cluster in particular places cannot be readily observed with administrative data, and this may bias estimates obtained from statistical models. The solution is to employ a statistical technique that allows an analysis to separate drug-related arrests due to system-level shocks (e.g., policy changes, supply shocks) from arrests due to local factors (e.g., gentrification) that may be tied up with local property values or levels of violence. This allows for the calculation of estimates for the effect of drug arrests on property values and gun-related homicides that can be considered causal under certain assumptions. The following analysis suggests that a one percentage point increase in drug arrests decreases assessed property values by 2.2 percentage points, and the effect is lagged — i.e., drug arrests this year will tend to affect property tax assessments three years from now. On the other hand, the analysis does not find strong support for an impact of drug enforcement arrests on gun-related homicides.
Drug arrests are highly concentrated geographically. Numerous drug arrests in a small area can have discernable community-level effects or negative externalities. Collateral, unintended consequences of arrests may include negative education and socioemotional outcomes for children whose parents have been incarcerated. Property markets may be affected in various ways, and sudden changes in local crime dynamics known as shocks (e.g., new policies or raised drug prices) could have other impacts on communities.
The following analysis uses data from the New York City Open Data Portal and other administrative data sets to measure the collateral consequences of drug arrests. Researchers examined spatial patterns of drug arrests over time to consider how the patterns change in response to enforcement, including the city’s use of Stop-and-Frisk policing. The research team used variations in these spatial patterns to explore within-neighborhood effects and how changes in property markets and levels of violence in communities are aligned with variations in drug arrests.
The key question investigated by the research team is whether drug-related arrests generate externalities, positive or negative. Researchers examine whether arrests have collateral consequences for property values (measured by property tax assessments) and levels of violence (measured by homicides). As a secondary research question, the study looks at whether these effects differ by median neighborhood income, neighborhood racial composition, and drug arrest type (e.g., sale or possession). Researchers measure the effects of collateral consequences of violations by modeling the effects of drug arrests at the census tract level.
ESTIMATING COLLATERAL CONSEQUENCES
Previous research literature finds that arrests are highly concentrated geographically (Johnson and Roman 2022). Studies also suggest that aggregating individual outcomes underestimates community-level policy effects (Cook and Ludwig 2000). In particular, research posits that labor supply and wage demands react to the presence of crime victimization risks, and the risk of victimization is an important component in the price of housing (Chalfin 2015; Clark and Cosgrove 1990). This study investigates these issues by estimating the economic consequences of clustered arrests for drug-law violations.
Key measures include:
- Changes in assessed property values.
- Changes in gun-related homicide levels.
The study relies on an instrumental variables approach. Arrests may impact property values (or homicides). Alternatively, property values (or homicides) may impact arrests. Of course, any number of confounding factors might affect both arrests and property values. Failure to control for these factors could bias the estimates. For example, a property developer’s choice for the location of a new luxury apartment building could affect the values of surrounding properties and may also impact drug-related arrests (e.g., by pressuring police to increase patrol time in the neighborhood). An instrumental variables strategy allows researchers to attempt to isolate variations in drug-related arrests due to system-wide shocks in the broader socio-economic environment. In other words, the strategy helps to separate variations due to system-wide shocks — which should not suffer from dual-causality or confounding variables problems — from variations due to neighborhood-level shocks, which likely do suffer from such problems.
Researchers analyzed drug arrests from 2006 through 2020, a time with large variations in arrests due to changes in laws and policies. For example, a 2013 New York court decision limited the once-routine use of Stop, Question and Frisk (SQF), a policing strategy that allowed officers to search and question people with little justification. The court ruling increased the evidentiary requirements to justify citizen stops (Rudovsky and Rosenthal 2013). More recently, changes in marijuana possession laws led to declines in the number of drug-related arrests in New York, although trends varied by demographic group (Patten et al. 2019). These changes provide a natural policy experiment for researchers to isolate the effects of changes in drug arrests due to policy differences or other factors associated with outcomes of interest.
Prior Research
Arrests, prosecutions, and incarceration for violations of drug laws increased dramatically in the United States over the past four decades. Researchers may disagree about the net effect of expanded criminal enforcement, but there is little disagreement about its cost (Pettit and Western 2004). Known as collateral consequences in the legal literature, or negative externalities and excess burden in the economic literature, the cost of increased justice intervention falls into three categories. First, individuals arrested for violating criminal laws may experience a cascading set of negative consequences. Second, communities that experience high arrest rates and that lose many residents to incarceration experience an array of consequences. Third, racial and class disparities in the application of the criminal law generate other social, cultural, economic, and political costs.
Research confirms the collateral consequences that individuals and communities may experience following contact with law enforcement for drug-related offenses, including barriers to employment and lower earnings, temporary or permanent ineligibility for social benefits, housing instability, voter disenfranchisement, risk of deportation for non-citizens, and poor educational and health outcomes for children of incarcerated parents (Kirk and Wakefield 2018). Studies have shown that enforcement actions in response to drug-related offenses may actually contribute to drug market violence and higher rates of homicide (Werb et al. 2011).
Organizations involved in drug distribution may respond to increases in enforcement with violence. Enforcement activities may disrupt otherwise non-violent dispute resolutions among groups and individuals involved in illegal drug markets, and competitors may move aggressively to fill gaps created when previously dominant suppliers are removed from lucrative drug markets (Miron 1999). Increased enforcement also comes with its own costs as racial and class disparities are apparent in the enforcement of drug-law violations and the politicized “War on Drugs” (Kirk and Wakefield 2018; Alexander 2010; Wakefield and Uggen 2010; Blumstein and Beck 1999; Mauer 1999; Tonry 1996).
One way to understand the collateral consequences of drug law violations would be to measure how violators are processed through the criminal justice system and examine whether the process leads to differential impacts on an individual’s trajectory of criminal justice contact. A key challenge with this approach is that drug arrests are the result of processes that are somewhat independent of drug use or criminal behavior. Crime clusters in small-area geographies. Evidence for clustering is clear enough that it has earned the name “law of crime concentration of places” (Weisburd, Groff and Yang 2012; Johnson et al. 2020; Weisburd 2015). The geographic concentration of arrests (and pre-trial detention) exists even before the observation of individual behaviors or infractions (Fogliato et al. 2021; Skeem, Montoya and Lowenkamp 2022). When places serve as the units of analysis for selecting higher and lower levels of enforcement, it is instructive to understand the consequence of higher and lower levels of enforcement (as measured by arrest) on places.
The impact of individuals’ involvement with the criminal justice system (including incarceration, arrest, misdemeanor conviction, and/or community supervision) extends to their families and communities. A body of scholarship argues that aggregating individual outcomes to measure community-level effects underestimates community-level policy impacts (Cook and Ludwig 2000). At the community-level, high rates of incarceration and the displacement of convicted individuals from the community can destabilize families and destroy the informal social networks that contribute to lower crime rates (Stewart et al. 2020). The presence of crime hotspots and a police response in a community (and residents’ fear of victimization) impact the local economy through lowered property values and harms to local businesses (Buck et al. 1991; Burnell 1988; Haurin and Brasington 1996; Lynch and Rasmussen 2001; Thaler 1978; Ceccato and Wilhelmsson 2020; Greenbaum and Tita 2004).
Important to this study, the law and economics literature describes complicated ways exposure to crime may affect labor and housing markets. Labor supply and wage demands increase in the presence of greater victimization risk, and risk of victimization is an important component in the price of housing (Chalfin 2015; Clark and Cosgrove 1990). Beyond the formal sanctions imposed by the criminal justice system, a range of informal consequences of drug law violations reach beyond the individual to impact families and whole communities.
Crime is not randomly distributed across places. Research finds that arrests are highly concentrated geographically (Johnson and Roman 2022; Weisburd, Groff and Yang 2012; Weisburd 2015; Brantingham and Brantingham 1995; Eck 1995; Spelman 1995; Weisburd et al. 2004). Even in neighborhoods with high crime rates, crime tends to cluster in a few specific locations while other areas are relatively crime free. Environmental criminology includes the concept of crime attractors, or places that have well known reputations for attracting illegal activity (Brantingham and Brantingham 1995). Drug markets may act as crime attractors, bringing in people wishing to buy and sell drugs and increasing the potential for violence in those places (Johnson and Roman 2022). A study of Miami neighborhoods found that drug activity impacts crime independently of other disorganization measures, suggesting that illicit drug activity itself could have violence-producing effects (Martinez, Rosenfeld and Mares 2008).
As drug behavior most often occurs between consenting parties, policing of drug-related offenses is proactive rather than reactive, making arrest data the most widely used official data available to analyze drug offenses (Beckett et al. 2005; Rosenfeld and Decker 1999). Arrests are intended to mitigate the harm caused by illicit drug activity, but enforcement is largely discretionary. Community relationships and fear of crime may actually worsen when police engage in overly aggressive enforcement tactics. The presence of law enforcement can also serve as a reminder of nearby crime and associated dangers (Gorr and Lee 2015; Rosenbaum 2006; Tonry 2011). Vigorous enforcement and mass incarceration in neighborhoods characterized by social and economic disadvantage may contribute to social disorganization, degrading community well-being and sparking a cycle of disparity-generating arrests (Neil and MacDonald 2023).
Data
The goal of this study is to quantify the economic effects of drug-related arrests on New York City neighborhoods from 2006 to 2020. The analysis relies on three components: measures of drug arrests, boundary measures of city neighborhoods, and an economic outcome that is comparable across different times and places. Researchers explored different levels of geography, from very small areas with a few thousand people, to very large areas and even the entire city, charting trends in drug-related arrests and quantifying their effects into something understandable and tangible. Data about drug-related arrests were obtained from the publicly available NYC Open Data portal. Arrests classified as drug crimes are grouped into five categories (Marijuana, Sales, Possession, Paraphernalia, and “Other”). Two small groups of arrests were eliminated: 1) arrests recorded at police precincts (as they do not represent arrests actually occurring in police buildings but are instead an artifact of the requirement to geolocate every arrest), and 2) arrests with errors in their geolocation suggesting they occurred outside of New York City. All remaining arrests were located in Census Tracts and Neighborhood Tabulation Areas (NTA).
A variety of neighborhood characteristics were examined across census tracts and Neighborhood Tabulation Areas. Measures were derived from the U.S. Census Bureau’s American Community Survey (ACS), utilizing 5-year rolling averages including estimates from past years of the ACS. The characteristics include variables measured at the level of census tract, such as total population, the proportion of residents receiving Supplemental Nutrition Assistance Program (SNAP) benefits, the percentage renting their housing, and residents that identify with different racial groups (e.g., White, Black, Hispanic, etc.). Tract-level details were used to omit areas with no housing units or inordinately low populations, indicating that they were primarily commercial plots or other areas inappropriate for the analysis, such as Rikers Island and Liberty Island.
Outcome measures were created using property tax assessments and reports of gun-related homicides. Gun-related homicide measures were obtained from police department data available from the NYC Open Data portal. Shooting incidents from 2009-2020 were filtered to retain all cases where a shooting victim’s death was attributed to murder rather than suicide or accident. The data included GPS coordinates and dates, and the study examined annual numbers of homicides in each census tract.
Tax assessment data in the Open Data portal provided annual, per-unit property values for specific addresses geocoded by census tract. The data contained both commercial and residential properties, as both may suffer negative consequences from increasing drug arrests. Like arrests, tax assessment data occasionally contained unlocatable properties (e.g., Henry Hudson Parkway, no address given), and these were eliminated from the analysis. Tax assessments sometimes involve extreme values, particularly in the most expensive commercial areas of Manhattan. To control for extremes, the highest one percent of all valuations in each census tract were eliminated before determining the median value of the tract, shielding the analysis from outliers or very large transactions which would inflate averages.
The analysis relies on property tax assessments as the most direct and reliable means of estimating the collateral consequences of arrests for small-area neighborhoods. Other place-based data were considered, including rates of business formation, length of business tenure, percentage of households below the poverty line, and percentage of households receiving SNAP benefits, but property tax measures were preferred for two reasons:
- Unlike many measures from the American Community Survey, property tax assessments were available at the level of census tract for each year between 2009 and 2020.
- Property tax assessments provide a measure of economic consequences without the complications presented by other measures (e.g., business income reports or commercial activity records).
Analytic Plan
The study presented researchers with key challenges. First, no matter when the timeframe for an analysis begins, some places will already have more drug law arrests and the local economy will likely reflect the disorder that accompanies high rates of illegal drug use and drug arrests. Second, the local economy and the number of drug arrests in an area affect each other in ways that are difficult to disentangle.
Places with low economic activity and frequent disorder tend to attract drug distribution and, as a result, more drug arrests. At the same time, legitimate economic activity is likely to be deterred by drug market activity and extensive drug arrests. Commercial and business entities endeavor to locate their activities in neighborhoods with favorable security conditions. This can create a vicious cycle (each negative factor makes the other worse) and/or a virtuous cycle (each positive factor improves the other). This presents a dual-causality (or “chicken and egg”) problem. How can the effect of drug arrests on economic activity be separated from the effect of economic activity on drug arrests? To contain the effects of this problem, the following analysis relies on the longest available time-series of relevant data. Specifically, arrest data run from 2006 to 2020, beginning before the steep increase in SQF but not so far into the past as to confound the analysis with the long, secular decline in crime beginning in the early 1990s. Ending the examination of arrests in 2020 avoids most of the steep crime increases that emerged during the COVID-19 pandemic. Property tax assessment data were available beginning in 2009. Thus, the study focuses on the years between 2009 and 2020.
The analysis uses a technique that allows researchers to distinguish between system-wide and local-level shocks. Certain shocks are citywide and unrelated to the economic conditions in any given neighborhood. For instance, when courts ruled that SQF violated the constitution, this resulted in fewer SQF drug arrests everywhere in the city. By focusing on system-wide variations in drug arrests due to outside forces, the effect of changes in drug arrests on economic activity may be isolated.
Technically, the goal of the analysis was to identify the effect of drug arrests on two outcome variables at the level of census tracts: property tax assessments and gun-related homicides. Of course, unobservable variables will vary over time within a census tract and could be correlated with drug arrests. For example, when a property developer selects a site for a redevelopment project, police may face pressure to patrol that area more frequently or aggressively, which results in more drug arrests. The effect of drug arrests on property values would then be confounded by the effect of redevelopment project site selection decisions.
To solve this problem, the variation in drug arrests due to events unlikely to be associated with property tax assessments (e.g., a citywide policy ending SQF) were separated from variation due to local trends that could be (e.g., redevelopment projects). The analysis started with the observation that citywide drug arrests vary over time due to events that are arguably unrelated to property values or other outcomes (e.g., stop-and-frisk policies, fluctuations in wholesale drug prices, relaxation of marijuana laws, etc.). When these drug-arrest shocks happen, they are likely to affect some census tracts more than others.
Methods
The analysis first examines how the census tracts most impacted by citywide shocks are the census tracts with the most drug arrests at baseline (i.e., pre-analysis). This information was then used to separate variations in drug arrests in census tracts associated with citywide shocks from variations due to unobservable location-specific trends. The analysis measures the effects of drug arrests using only the variations due to citywide shocks using an approach known as instrumental variables, with a share-weighted measure of exposure to citywide drug arrest shocks as the instrumental variable. These are the volume of arrests outside of each tract in a given year, multiplied by the proportion of all drug arrests in the city that occurred in the census tract during the baseline period. More simply, the analysis is saying that drug arrests can be predicted in a census tract each year by looking at drug arrests in the rest of the city that year and weighting them by the share of all gun-related homicides in the city that occurred in that particular census tract in the pre-analysis period.
This is a variation of the instrumental variables approach known as a Bartik Instrument, often called “Bartik Shocks” (Goldsmith-Pinkham, Sorkin and Swift 2020). The approach relies on a key assumption: In the absence of these citywide shocks, arrests in the baseline period must be uncorrelated with the changes in the outcome. In other words, the analysis assumes that neighborhoods with different levels of drug arrests at baseline would have evolved similarly over time in terms of property tax assessments (or homicides), had it not been for the citywide shocks that affected arrests.
Several other potential sources for the instrumental variable were considered before the variable in this study was selected. Other options explored included 311 calls, which proved difficult to isolate to potential drug-related occurrences only; hospitalizations from overdoses, which appeared to result in only the location of hospitals and relative volumes and were thus insufficient to cover every tract in the city; and first-responder dispatch data, which were not able to be accessed. As a result, arrest data were used for the instrument in this analysis. (For a more detailed discussion of the instrumental variables approach used in this analysis, see the Appendix.)
RESULTS
DESCRIPTIVE STATISTICS
The analysis estimates possible influences on two dependent variables across New York City census tracts: median property tax assessments and gun-related homicides (Table 1). Census tracts comprised an average of just under 3,900 residents. About 30 percent of the residents on average lived in rental housing and 7 percent received food support from the federal SNAP program. The average median property tax assessment was $215,620, with a very high standard deviation ($673,000) due to a small number of tracts with very high assessment values. The mean (average) property tax assessment across all census tracts was $36,423. Census tracts averaged 0.13 gun-related homicides per year.
Key independent variables included several measures of drug arrests. Census tracts averaged slightly more than 29 drug arrests per year — approximately nine (8.59) arrests on average for marijuana and nearly 21 (20.53) non-marijuana arrests. Among non-marijuana arrests, about two-thirds (13.79) were for drug sales and one-third were for possession (6.01).
The study assumes census tracts with the most drug arrests at baseline (i.e., before 2006) will be the most affected by citywide policy changes that impact drug-related arrests. Researchers plotted each census tract’s percentage change in drug-related arrests between the baseline and final two years of observation (2006/2007 versus 2019/2020) against the number of drug-related arrests per census tract at baseline (Figure 1). Most census tracts fall below zero on the vertical or y-axis of the graph, reflecting the fact that over 95 percent of census tracts experienced a decline in drug-related arrests over the period of observation. The trendline slopes downward, indicating that census tracts with more baseline arrests experienced the greatest relative decline in arrests during the study period.
FIRST-STAGE MODELS: TESTS OF THE INSTRUMENTAL VARIABLE
Drug arrests in a community are likely affected by various citywide, statewide, or national changes, or “shocks” in the statistical models used in this study (described further in the Appendix). Shocks may have a different effect on arrests depending on the characteristics of a neighborhood. An increase in “stop and frisk” actions by police, for example, could have a different effect in communities depending on their underlying levels of crime and how aggressively they are already being policed.
A first-stage instrumental variables model tests whether specific shocks are strong predictors of arrests (Table 2). Three models portray separate statistical regressions, where the dependent variable is the actual arrests and the independent variable is the Bartik shock alone (Model 2.1), while Model 2.2 adds year and census tract fixed effects, and Model 2.3 adds additional census tract level control variables that vary over time. Specifically, Model 2.3 controls for annual census tract measures of population size, renters as a proportion of tract populations, and the percentage of the population receiving SNAP benefits. The results show that the Bartik shock is strongly predictive of actual arrests, and the results are stable across different specifications of the model. For each unit increase in the Bartik shock, drug arrests increase by approximately 0.9 in all three models.
The strategy used in this paper to identify the effects of drug-related arrests is known as instrumental variables or Two-Stage Least Squares (Pearl 2013). Each model presented below uses predicted drug arrests from the regression analyses in Table 2 as the independent variable. Because these values were predicted using the Bartik shock instrument, which arguably varies from systemwide shocks, it should remove variations in arrests due to local, neighborhood-level factors that would provide the most obvious sources of bias for the estimates.
SECOND-STAGE MODELS: EFFECT OF ARRESTS ON MEDIAN PROPERTY TAX ASSESSMENTS
A second-stage regression model estimates the effect of arrests on median property tax assessments (Table 3). Model 3.1 portrays results for the model with no fixed effects or controls. The results show a negative, statistically significant coefficient for the first-year lag of arrests (i.e., arrests in the year before the tax assessment), which is offset by a similarly sized, positive and statistically significant coefficient for the third-year lag (i.e., arrests in the third year before the tax assessment). Using lagged arrests is an effort to account for administrative delays in how property tax assessments are conducted, as well as potential inertia in property markets (i.e., markets likely take time to react to new neighborhood dynamics).
Model 3.1 does not include fixed effects for the census tract or for the year, which means it does not account for unobserved confounding variables that might affect all census tracts in a given year, or that affect a specific census tract across all years. For example, an economic downturn in New York City would impact all census tracts in the year it happens and could be correlated both with arrests and property values in a way that is not accounted for in the instrumental variables strategy.
Model 3.2 shows the effect of adding fixed effects for tract and year, while Model 3.3 includes additional control variables. The estimates in the second and third models are quite different from those in Model 3.1. This suggests that there may indeed be confounding variables, such as the previous example, that are uncontrolled for in Model 3.1 and could be biasing the estimates. The negative, statistically significant results for the third-year lag of arrests from Models 3.2 and 3.3 suggest that arrests have a negative effect on property tax assessments and the effect is realized in three years.
The delayed effect on tax assessments could be due to either (or both) of two sources. First, it may be that changes in drug arrests take time to have an effect on perceptions of safety and, ultimately, property values. The lag may be a result of such a process. On the other hand, it may also be that drug arrests have an immediate effect on perceptions of safety, but there may be delays in the administrative processes required to assess property values. The coefficient may be interpreted to mean that a one-percentage point increase in drug arrests decreases assessed property values by 2.2 percentage points. The median home in New York City would see a decrease of 2.2 percent in assessed value for each 1 percent increase in arrest volume, or approximately $800 of $36,423 (as of 2020).
Researchers plotted results from seven separate regressions (Figure 2). Each point represents a coefficient for the effects of drug arrests on median assessed property value (lagged between one and seven years prior to the tax assessment year). The results again indicate notable delays between when increases in drug arrests occur and when the impact on assessed property values may be realized. The first negative and statistically significant estimates for drug arrests occur three and four years before the tax assessment year. As additional time passes, the effect of arrests on property values appears to decline. Estimated effects are larger than those shown in Table 3. A one-percentage point increase in drug arrests leads to decreased property values three and four years before the tax assessment year of 3.2 percent and 3.0 percent, respectively.
DIFFERENTIAL ANALYSIS: EFFECT BY CENSUS TRACT DEMOGRAPHICS, MEDIAN INCOME, AND ARREST TYPE
Researchers tested other regressions (Table 4). Each analysis estimated whether the effects of drug arrests vary by community socioeconomic status as represented by a proxy measure: median income at baseline (2009). Specifications from Model 3.3 in Table 3 were re-run for different sub-samples of census tracts. Drug-related arrests might affect property tax assessments in a number of ways. Arrests might impact income and the ability of residents to afford housing. Arrests could increase residents’ perceptions of insecurity and reduce the attractiveness of the neighborhood, or increasing arrests could influence how real estate and commercial developers choose to locate their investments. Thus, it is difficult to predict which types of neighborhoods might be most affected by growing arrest rates.
Models 4.1, 4.2, and 4.3 run the analysis for census tracts in the lowest, middle two, and highest median income quartiles, respectively. The results suggest the negative impact of drug arrests emerging in the third year before tax assessment is experienced most acutely by census tracts in the bottom quartile for median income. The effect in Model 4.1 suggests a one-percentage point increase in drug arrests in the third year before the tax assessment decreases property values in census tracts with the lowest median incomes by 5.7 percentage points. The results in Model 4.3 suggest that arrests may actually increase property values in census tracts from the highest income quartile, an effect that takes less time as well, which is indicated by the positive and statistically significant estimate for the first lag of drug arrests.
To examine the effects of drug-related arrests by the racial composition of neighborhoods, researchers next tested census tracts with different resident demographics (Table 5). Regressions were run separately in Models 5.1, 5.2, and 5.3 for census tracts in the highest quartiles for residents reporting race/ethnicity of White, Black, and Hispanic, respectively. The results show that the effect of arrests on property values is most acute in communities of color and is particularly driven by census tracts in the upper quartile for the proportion of Hispanic residents in the population (Model 5.3). A one percentage point increase in drug-related arrests three years before property tax assessment reduces the median tax assessment by 4.6 percent in areas with the greatest share of Hispanic residents. The effect in census tracts with the highest proportion of Black residents (i.e., Model 5.2) is nearly identical, though not statistically significant. Estimated effects two and three years before tax assessments in areas with high proportions of Black residents are consistently negative and larger than the size of corresponding estimates for the overall population (Model 3.3).
Researchers also tested the effects of increasing arrests for different types of drug-related charges (Figure 3). The results revealed heterogeneity in the process by which drug arrests may reduce property values. While the effect on assessed property tax values was essentially the same for both marijuana and non-marijuana arrests in the top two panels, the effect of drug sales arrests and drug possession arrests had substantially different processes. The negative effect of non-marijuana drug sales on tax assessments appeared to occur immediately and only began to attenuate after the sixth year. In contrast, the effect of non-marijuana possession arrests generally declined until after the fifth year.
SECOND-STAGE MODEL: EFFECT OF ARRESTS ON GUN-RELATED HOMICIDES
In another test, researchers analyzed a second-stage regression model of the effects of increased drug arrests on gun-related homicides (Table 6). As before, Model 6.1 presents results for the model with no fixed effects or control variables. The result shows a negative, statistically significant coefficient for the first-year lag of arrests (i.e., arrests in the year before the observation year for homicides), which is offset by a much larger, positive, and statistically significant coefficient for the third-year lag (i.e., arrests in the third year before the observation year for homicides). Model 6.2 adds fixed effects for year and census tract, while Model 6.3 uses those factors as well as the additional control variables. Estimates in the second and third models are similar to the first, but notably smaller and not statistically significant. Given the results from the preferred model specification in Model 6.3, the null hypothesis that drug-related arrests have no effect on gun-related homicides cannot be rejected.
The research team examined seven separate regression analyses using a model similar to that of Model 6.3 (Figure 4). Instead of including all the lags in a single regression, the regression was run separately for each lag to address concerns with multicollinearity (i.e., lagged logs of arrest are correlated with one another).
Results were like those in Table 6 but show more statistical significance. The effects of arrest on gun-related homicide were positive, but near zero and not statistically significant for arrests one year prior to the year of homicide observation. The effect increased in magnitude and statistical significance two and three years before the observation year. After year three, the magnitude of the effect declined and lost statistical significance.
Together with the results of Model 6.3, these regression analyses suggest that a one percentage point increase in drug-related arrests may lead to a two-percentage increase in gun-related homicides two and three years later, but the results are not especially robust and may be sensitive to how the models are specified.
DIFFERENTIAL ANALYSIS: EFFECT BY CENSUS TRACT DEMOGRAPHICS, MEDIAN INCOME, AND ARREST TYPE
Another set of regressions tested whether the effects of drug arrests on homicide varied by community socioeconomic status, again as proxied by median income in the baseline year of 2009 (Table 7). Regression analyses estimated the effects of increasing drug arrests on gun-related homicide in different sub-samples of census tracts. Researchers hypothesized that drug-related arrests affect gun-related homicides primarily through their effects on street gang dynamics and the creation of power vacuums in the elicit economy. As such, effects were expected to be more acute in poor and marginalized census tracts where organized street crews are more likely to be present.
The results failed to support the hypothesis. Estimates generated by the analysis lacked statistical significance for census tracts in every income group. For the lowest income group specifically (i.e., Model 7.1), estimates were generally similar to those seen in Model 6.3 that represented the effect of drug-related arrests on homicide in all census tracts and failed to find effects of drug arrests that were statistically different from zero.
As with the previous estimation for the effects of drug arrests on median tax assessments, researchers tested the effects of increased drug arrests on run-related homicides controlling for the racial and ethnic compositions of New York City neighborhoods (Table 8). Census tracts in the upper quartile for their shares of the White, Black, and Hispanic populations at baseline were examined in separate regression models. The effects were generally near zero and consistently failed to reach statistical significance.
Finally, the research team compared the effects of different types of drug-related arrests (Figure 5). Again, the guiding hypothesis was that drug-related arrests would affect gun-related homicides mainly by disrupting the elicit economy. Any effects, therefore, should be more apparent from drug sales than from drug possession.
The analysis provides some support for the hypothesis. The effects of all drug arrests on gun-related homicides are similar (marijuana and non-marijuana). Positive effects are observed between one year and three years before the year of homicide measures. The effects can be seen to decline before that, with estimates near zero and below zero four to seven years prior to the measurement of gun-related homicides. The results show positive and statistically significant estimates for the effect of marijuana and non-marijuana drug-related arrests two or three years before the year of homicide measures. Moreover, the effect from increasing arrests for drug sales three years prior to the homicide measures is positive and statistically significant while the effect for possession-related arrests hovers near zero and is not statistically significant.
Conclusion
This study relied on patterns of drug-related arrests to examine whether increases in such arrests generate collateral consequences (positive or negative) for New York City neighborhoods. Researchers focused on two main outcomes: property values and gun-related homicides. To resolve the confounding effects between arrests and other outcomes of interests, the study used an instrumental variables approach and exploited likely variations in drug-related arrests due to known changes in policy and practice (i.e., Stop-and-Frisk).
The study relied on the instrumental variables approach to estimate the effect of drug arrests on property tax assessments and gun-related homicides. While instrumental variables are a common approach in social science, their application in this study is somewhat novel and not without limits. The analysis cannot separate the effect of drug-related arrests from underlying drug-related crime (although it could be argued that the most important source of variation in arrests captured in the analysis is citywide changes in drug enforcement policy).
The research team explored other instrumental variables—e.g., drug overdose deaths. Previous research suggests that users of illegal drugs tend to use newly purchased drugs at sites near the place of purchase. Thus, overdose deaths would tend to be near points of sale and could serve as a good indicator for the location of drug markets and perhaps even be a more suitable instrument than lagged drug arrest patterns. The authors’ attempt to collect overdose data from New York City Vital Statistics, however, was not successful. Other data (e.g., national measures of opioid overdoses) proved to be aggregated at too broad a level of geography.
The analysis did not find strong support to suggest an impact of drug enforcement arrests on gun-related homicides. The results showed some evidence that certain types of drug arrests (i.e., sales and distribution) may be seen in increased homicides two or three years later, the results were sensitive to how the statistical model was specified and not especially robust. On the other hand, the analysis found that, on average, a one percent increase in drug arrests may be expected to decrease assessed property values by 2.2 percent within three years.
The association between increased drug enforcement and lower property values appears to be driven disproportionately by dynamics in poor neighborhoods and those with large proportions of Black residents. Increased drug enforcement may actually increase property values in wealthier neighborhoods. The study did not specifically examine the channels through which negative effects operate in impoverished neighborhoods, but the results are at least consistent with speculation that growth in drug-related arrests deters economic investments and either directly or indirectly affects wages in ways that disproportionately hinder lower-income residents’ ability to make housing payments, resulting in further damages to social conditions and increased perceptions of insecurity. Residents may see benefits from increased drug arrests in the short term, but if property values depreciate due to greater enforcement, neighborhoods would begin to suffer from declining prosperity and the long-term erosion of the community’s financial viability.
RECOMMENDED CITATION
Roman, John K., Gregory Haugan, Benjamin Schapiro and Sofia Rodriguez (2024). Collateral Consequences: The Effects of Justice Processing for Violations of Drug Laws in New York City. A report from NORC at the University of Chicago. New York, NY: Research and Evaluation Center, John Jay College of Criminal Justice, City University of New York.
ACKNOWLEDGMENTS
The authors appreciate the leadership and staff of the Mayor’s Office of Criminal Justice that helped to shape this analysis. The authors also express their appreciation for the collaboration and support of colleagues at the John Jay College Research and Evaluation Center, Jeffrey Butts, Sheyla Delgado, Kathy Tomberg, Gina Moreno and Rebecca Balletto, who provided comments and suggestions as the report developed. All conclusions, however, are those of the authors. Funders and partners of NORC at the University of Chicago are not responsible for any findings presented in its publications.
REFERENCES
Alexander, Michelle (2010). The New Jim Crow: Mass Incarceration in the Age of Color Blindness. New York: The New Press.
Beckett, Katherine, Kris Nyrop, Lori Pfingst and Melissa Bowen (2005). Drug use, drug possession arrests, and the question of race: Lessons from Seattle. Social Problems 52(3): 419-441.
Blumstein, Alfred and Allen J. Beck (1999). Population growth in US prisons, 1980–1996. In Prisons, Tonry, Michael and Joan Petersilia (Editors), Vol. 26: Crime and Justice: A Review of Research, pp. 17–61. Chicago: University of Chicago Press.
Borusyak, Kirill, Peter Hull and Xavier Jaravel (2022). Quasi-experimental shift-share research designs. The Review of Economic Studies 89(1): 181-213.
Brantingham, Patricia and Paul Brantingham (1995). Criminality of place: Crime generators and crime attractors. European Journal on Criminal Policy and Research 3: 5-26.
Buck, Andrew J., Joseph Deutsch, Simon Hakim, Uriel Spiegel and J. Weinblatt (1991). A Von Thünen model of crime, casinos and property values in New Jersey. Urban Studies 28(5): 673–686.
Burnell, James D. (1988). Crime and racial composition in contiguous communities as negative externalities: Prejudiced households’ evaluation of crime rate and segregation nearby reduces housing values and tax revenues. American Journal of Economics and Sociology 47(2): 177–193.
Ceccato, Vania and Mats Wilhelmsson (2020). Do crime hot spots affect housing prices? Nordic Journal of Criminology 21(1): 84–102.
Chalfin, Aaron (2015). Economic costs of crime. The Encyclopedia of Crime and Punishment, First Edition. John Wiley & Sons.
Clark, David E. and James C. Cosgrove (1990). Hedonic prices, identification, and the demand for public safety. Journal of Regional Science 30(1): 105-121.
Cook, Philip J. and Jens Ludwig. (2002). Gun Violence: The Real Costs. New York: Oxford University Press.
Eck, John E. (1995). A general model of the geography of illicit retail marketplaces. Crime and Place, John E. Eck and David Weisburd, Editors. Crime Prevention Studies, Vol. 4: 67–93. Boulder, CO: Lynne Rienner Publishers.
Fogliato, Riccardo, Alice Xiang, Zachary Lipton, Daniel Nagin and Alexandra Chouldechova (2021). On the validity of arrest as a proxy for offense: Race and the likelihood of arrest for violent crimes. Proceedings of the 2021 AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society, 100–111.
Goldsmith-Pinkham, Paul, Isaac Sorkin and Henry Swift (2020). Bartik instruments: What, when, why, and how. American Economic Review 110(8): 2586-2624.
Gorr, Wilpen L. and YongJei Lee (2015). Early warning system for temporary crime hot spots. Journal of Quantitative Criminology 31(1): 25–47.
Greenbaum, Robert T. and George E. Tita (2004). The impact of violence surges on neighbourhood business activity. Urban Studies 41(13): 2495–2514.
Haurin, Donald R. and David Brasington (1996). School quality and real house prices: Inter- and intrametropolitan effects. Journal of Housing Economics 5(4): 351–368.
Johnson, Nicole J. and Caterina G. Roman (2022). Community correlates of change: A mixed-effects assessment of shooting dynamics during COVID-19. PLoS ONE 17(2): e0263777.
Johnson, Nicole J., Caterina G. Roman, Alyssa K. Mendlein, Courtney Harding, Melissa Francis and Laura Hendrick (2020). Exploring the influence of drug trafficking gangs on overdose deaths in the largest narcotics market in the eastern United States. Social Sciences 9(11): 202.
Kirk, David S. and Sara Wakefield (2018). Collateral consequences of punishment: A critical review and path forward. Annual Review of Criminology 1(2018): 171–194.
Lynch, Allen K. and David W. Rasmussen (2001). Measuring the impact of crime on house prices. Applied Economics 33(15): 1981–1989.
Martínez, Ramiro Jr., Richard Rosenfeld and Dennis Mares (2008). Social disorganization, drug market activity, and neighborhood violent crime. Urban Affairs Review 43(6): 846–874.
Mauer, Marc (1999). Race to Incarcerate. New York: New Press.
Miron, Jeffrey A. (1999). Violence and the U.S. prohibitions of drugs and alcohol. American Law and Economics Review 1(1/2): 78-114.
Neil, Roland and John M. MacDonald (2023). Where racial and ethnic disparities in policing come from: The spatial concentration of arrests across six cities. Criminology & Public Policy 22(1): 7–34.
Patten, Meredith, Erica Bond, Cecilia Low-Weiner, Quinn O. Hood, Olive Lu, Shannon Tomascak, Darren Agboh and Preeti Chauhan (2019). Trends in Marijuana Enforcement in New York State, 1990 to 2017. New York: Data Collaborative for Justice, John Jay College of Criminal Justice, City University of New York.
Pearl, Judea (2013). Causality: Models, Reasoning, and Inference (Second Edition). Cambridge: Cambridge University Press.
Pratt Travis C. and Francis T. Cullen (2005). Assessing macro-level predictors and theories of crime: A meta-analysis. Crime and Justice 32(2005): 373–450.
Rosenbaum, Dennis P. (2006). The limits of hot spots policing. Police Innovation: Contrasting Perspectives, David Weisburd and Anthony A. Braga (Editors), pp 245–263. New York: Cambridge University Press.
Rosenfeld, Richard and Scott H. Decker (1999). Are arrest statistics a valid measure of illicit drug use? The relationship between criminal justice and public health indicators of cocaine, heroin, and marijuana use. Justice Quarterly 16(3): 685-699.
Rudovsky, David and Lawrence Rosenthal (2013). Debate: The constitutionality of Stop-and-Frisk in New York City. University of Pennsylvania Law Review Online, Vol. 162; 117-150.
Sampson, Robert J., Stephen W. Raudenbush and Felton Earls (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science 277(5328): 918–924.
Shaw, Clifford Robe and Henry Donald McKay (1942). Juvenile Delinquency and Urban Areas: A Study of Rates of Delinquents in Relation to Differential Characteristics of Local Communities in American Cities. Chicago, IL: University of Chicago Press.
Sherman, Lawrence W. (1995). Hot spots of crime and criminal careers of places. Crime and Place, John E. Eck and David Weisburd, Editors. Crime Prevention Studies, Vol. 4: 35–52. Boulder, CO: Lynne Rienner Publishers.
Skeem, Jennifer, Lina Montoya and Christopher Lowenkamp (2022). Place matters: Racial disparities in pretrial detention recommendations across the U.S. — Federal Probation: A Journal of Correctional Philosophy and Practice 86(3): 5-10.
Spelman, William (1995). Criminal careers of public places. Crime and Place, John E. Eck and David Weisburd, Editors. Crime Prevention Studies, Vol. 4: 115–144. Boulder, CO: Lynne Rienner Publishers.
Stewart, Eric. A., Patricia Y. Warren, Cresean Hughes and Rod K. Brunson (2017). Race, ethnicity, and criminal justice contact: Reflections for future research. Race and Justice 10(2): 119–149.
Thaler, Richard (1978). A note on the value of crime control: evidence from the property market. Journal of Urban Economics, 5(1): 137–145.
Tonry, Michael (2011). Less imprisonment is no doubt a good thing. Criminology & Public Policy 10(1): 137–152.
Tonry, Michael (1996). Malign Neglect: Race, Crime, and Punishment in America. New York: Oxford University Press.
Wakefield, Sara and Christopher Uggen (2010). Incarceration and stratification. Annual Review of Sociology 36(1): 387–406.
Weisburd, David (2015). The law of crime concentration and the criminology of place. Criminology 53(2): 133-157.
Weisburd, David, Shawn Bushway, Cynthia Lum and Sue-Ming Yang (2004). Trajectories of crime at places: A longitudinal study of street segments in the city of Seattle. Criminology 42(2): 283–321.
Weisburd, David, Elizabeth R. Groff and Sue-Ming Yang (2012). The Criminology of Place: Street Segments and Our Understanding of the Crime Problem. New York: Oxford University Press.
Werb, Dan., Greg Rowell, Gordon Guyatt, Thomas Kerr, Julia Montaner and Evan Wood (2011). Effect of drug law enforcement on drug market violence: A systematic review. International Journal of Drug Policy, 22(2): 87–94.