The Growth of Illicit Drug Use and Its Effects on Murder Rates

Abstract

After years of reductions in the rate of murder in the United States, the national murder rate has increased since 2015. The causes of this trend are generally unknown, though there is some evidence related to narcotic drugs. Arrests related to heroin and cocaine had been stable between 2010 and 2014 before a sudden increase in 2015. Likewise, the number of murders related to narcotic drugs has increased since 2013, with a jump in 2015. Increased rates of these crimes parallel recent dramatic growth in overdoses involving heroin. However, the causal relationship between the recent opioid crisis and the rise in murder rates is missing from the literature. I used OxyContin reformulation as an exogenous shock to illicit markets. OxyContin reformulation led some people who misused OxyContin to switch to illicit opioids. Previous work has shown that areas with higher rates of OxyContin misuse experienced faster growth in heroin overdoses post‐reformulation. I tested whether this growth in illicit drug use caused an increase in crime. After reformulation, I find significantly greater relative increases in murder rates in states with high pre‐reformulation rates of OxyContin misuse. The results support a causal link between the opioid epidemic and crime.

1. Introduction

After years of declining murder rates in the United States, the national murder rate began trending upward in 2015. Figure 1 illustrates this shift, showing a consistent decrease in homicides ¹ per 100,000 people until 2014, followed by a subsequent increase, based on data from the National Vital Statistics System (NVSS). The factors contributing to this recent rise in homicide rates remain largely unexplained.

FIGURE 1.

National Homicide Trends in the U.S. (1999–2017). Figure 1 shows the national homicide trends per 100,000 people from 1999 to 2017, based on ICD‐10 codes X85‐Y09 (Assault). Homicides related to the 9/11 terrorist attacks have been excluded from the data. Source: National Vital Statistics System (NVSS).

At the same time, the nation has witnessed a dramatic escalation in overdoses involving heroin and synthetic opioids. Between 2014 and 2015, deaths involving synthetic opioids—particularly illicitly manufactured fentanyl—surged by 72.2%, while fatal overdoses involving heroin increased by 19.5% (Seth et al. 2018). These trends signal a growing misuse of illicit opioids during this period, which may have far‐reaching societal consequences, including potential impacts on crime rates.

Several researchers have suggested that the opioid crisis may be driving the recent increases in homicide rates, although existing evidence remains largely correlational (Rosenfeld et al. 2021, 2023). While there is substantial evidence linking heroin use to property crimes (Anglin and Speckart 1988; Barton 1976; Smart and Reuter 2022; Uggen and Thompson 2003), Mallatt (2022) is among the few studies providing causal evidence linking the demand for illicit opioids to an increase in heroin possession crimes. However, the relationship between increased opioid use—particularly heroin—and violent crimes, such as murder, though suggested by some studies (Moore and Schnepel 2021; Sim 2023), remains underexplored, with much of the earlier research predating the current escalation of the opioid epidemic and the rise of synthetic opioids in the U.S. (Pacula et al. 2013; Szalavitz and Rigg 2017).

This study aims to address this critical gap by examining the impact of the 2010 reformulation of OxyContin—an exogenous shock to illicit drug markets—on murder rates. Leveraging data from both the NVSS and the Uniform Crime Reporting (UCR) Program, I find that in the period following the reformulation, states with higher pre‐reformulation rates of OxyContin misuse experienced either an increase or a slower decline relative to states with lower rates of Oxycontin misuse in murder rates. This trend is particularly evident among white male victims, a group that also showed a significant rise in heroin overdose deaths post‐reformulation. These findings suggest a potential link between the opioid epidemic and rising homicide rates, indicating that the surge in murders may be partly attributed to the heightened risks associated with the increase in use of illicit drugs.

Establishing causal evidence of the link between rising opioid misuse and the increase in murders is crucial for developing informed criminal justice and public health policies aimed at mitigating the additional harms caused by the opioid epidemic. This study highlights the unintended consequences of the OxyContin reformulation, showing that it may have inadvertently contributed to a rise in murder rates. The findings emphasize the need for comprehensive strategies that address opioid misuse as a key component of broader crime reduction efforts.

2. Review of Literature and Conceptual Framework

2.1. Opioid Epidemic and OxyContin Reformulation

The opioid crisis in the United States has evolved in three distinct waves, as identified by the Centers for Disease Control and Prevention (CDC). The first wave, driven by the increased use and misuse of prescription opioids, led to a significant rise in overdose deaths. Around 2010, the crisis entered its second wave, characterized by a shift from prescription opioids to heroin as the primary driver of opioid overdose fatalities. The third wave began in 2013, marked by a surge in overdose deaths related to synthetic opioids, particularly due to the growth of illicitly manufactured fentanyl.

Research indicates that approximately 80% of heroin users initially misused prescription opioids (Muhuri, Gfroerer, and Davies 2013). This transition from prescription opioids to illicit drugs has been partly driven by policies aimed at curbing access to prescription opioids, such as Prescription Drug Monitoring Programs (PDMPs), and the introduction of abuse‐deterrent formulations.

A key policy in this regard is the 2010 reformulation of OxyContin, one of the most frequently misused prescription opioids (Cicero, Inciardi, and Muñoz 2005). The reformulated version of OxyContin, approved by the FDA in 2010, included abuse‐deterrent features that made it harder to crush or dissolve the pill, thereby increasing the non‐monetary costs of misuse. Several studies have shown that this reformulation successfully reduced prescription opioid deaths (Cicero and Ellis 2015; Cicero, Ellis, and Surratt 2012; Coplan et al. 2013). But, research suggests that the reformulation may have pushed some users toward illicit drugs like heroin (Cicero and Ellis 2015).

A pivotal study by Alpert, Powell, and Pacula (2018) found that the reformulation of OxyContin led to an increase in heroin overdose deaths, as the difficulties associated with misusing the reformulated drug drove some individuals to turn to heroin instead. Similarly, W. N. Evans et al. (2019) concluded that the reformulation particularly increased heroin overdose death rates in regions with more established heroin markets. Furthermore, Powell and Pacula (2021) suggested that the reformulation might have led to increased use of other synthetic opioids and non‐opioid drugs over the long term.

Beyond overdose deaths, the increase in heroin use following the OxyContin reformulation has had broader societal impacts, including reductions in labor force participation, an increase in disability insurance claims (Park and Powell 2021), a rise in child maltreatment (M. F. Evans et al. 2022), and higher rates of Hepatitis C infections (Beheshti 2019).

2.2. Crime and Opioid Use: Insights From Literature

Heroin, a highly addictive opioid, is distinct from prescription opioids as it is primarily obtained through black market channels and lacks legal protections (Resignato 2000). This market dynamic increases the risk of violence. Szalavitz and Rigg (2017) identified a correlation between the rise in heroin use during the 1970s and a concurrent increase in murders and violent property crimes.

Earlier studies, predating the current opioid epidemic, consistently link heroin use to property crimes (Anglin and Speckart 1988; Barton 1976; Uggen and Thompson 2003). However, the relationship between heroin uses and violent crime remains less clear. Some research suggests heroin users are more likely to commit robbery (Johnson et al. 1985), while others have found a negative association between heroin use and violent crime (Benson 2009). In summary, pre‐epidemic studies highlight a strong connection between opioid use and property crime but offer mixed evidence regarding violent crime (Pacula et al. 2013; Szalavitz and Rigg 2017).

Recent shifts in the demographics of heroin users may influence this relationship. Studies have documented an increase in heroin use among women, individuals aged 26–34, and non‐Hispanic whites from 2002 to 2010, while usage among older adults and Black individuals declined (Han et al. 2020). Seelye (2015) noted that nearly 90% of new heroin users in the past decade were white. These changes suggest that the ongoing opioid crisis differs from the heroin epidemic of the 1970s, which disproportionately affected inner‐city minority populations (Kolodny et al. 2015). These demographic shifts could potentially reshape the relationship between illicit opioid use and violent crime in the current crisis.

The opioid crisis has led to changes in crime patterns. Arrests related to heroin and cocaine, which had remained stable between 2010 and 2014, began rising in 2015 (Rosenfeld and Fox 2019; Rosenfeld et al. 2017). Similarly, murders related to narcotic drugs, including heroin, morphine, and codeine, started increasing in 2013 and peaked in 2015 (Rosenfeld et al. 2017). Additional studies have linked rising opioid overdose death rates to homicide rates (Rosenfeld et al. 2021, 2023).

Policy responses have also shaped crime trends during the opioid crisis. Studies show that must‐access Prescription Drug Monitoring Programs (PDMPs) increased the prevalence of heroin and opioid crimes, such as possession and dealing (Mallatt 2022), while simultaneously reducing violent crimes like murders and assaults (Dave, Deza, and Horn 2021). Moreover, states highly exposed to OxyContin misuse prior to the reformulation reported a rise in violent crimes following the drug's reformulation (Sim 2023). While must‐access PDMPs led to a 4%–8% reduction in retail opioid prescriptions (Kaestner and Ziedan 2019), the OxyContin reformulation resulted in a 40% decrease in OxyContin misuse. Although not directly comparable, these policies demonstrate the significant role of regulation in shaping both the prescription opioid supply and its potential downstream impact on crime.

2.3. Theoretical Framework and Research Approach

If there is a link between illicit opioid use and murder rates, it is essential to understand the underlying mechanisms driving this relationship. This connection could arise through four primary pathways: psychopharmacological violence, economic compulsive violence, systemic violence, and victimization. The first three mechanisms were proposed by P. J. Goldstein (1985). The psychopharmacological and economic compulsive hypotheses suggest that individuals may commit crimes either due to the direct effects of drug use or to finance their drug habits. The third mechanism, systemic violence, relates to gang activities and territorial conflicts within the drug trade. Additionally, this study introduces a fourth mechanism—victimization, as heroin users often become victims of violent crime (P. J. Goldstein 1985; Pearson and Hobbs 2003). Thus, the relationship between illicit opioid use, such as heroin, and murder may be driven either by an increase in offenders or an increase in vulnerable victims.

This study focuses on three primary ways that heroin use may be linked to higher murder rates: increased likelihood of users committing crimes, users becoming victims, and the expansion of illicit drug markets driving systemic violence.

The first hypothesis suggests that opioids and heroin use increase the likelihood of individuals committing crimes, including murder. P. J. Goldstein (1985) identified two possible reasons for this. First, drug use may induce violent behavior, though this is less applicable to opioids, which are known for their tranquilizing effects (Power 1994). Since OxyContin and heroin have similar psychopharmacological properties, the switch from one to the other would not significantly reshape their behavioral effects. Some studies, however, have found links between opioid use and psychopathology (Campbell and Stark 1990), as well as violence toward intimate partners (Moore and Schnepel 2021). Second, economic factors may drive users to engage in criminal activities to fund their drug use, which may escalate into violence, as shown by Uggen and Thompson (2003). Due to the overlap between psychopharmacological effects and economic motivations, I treat them jointly as factors contributing to an increased likelihood of crime.

The second hypothesis indicates that rising opioid use may elevate the risk of individuals becoming victims of violent crime. Heroin users, who often carry cash or drugs, are more likely to be targeted for robbery (P. J. Goldstein 1985; Pearson and Hobbs 2003). Evidence from the UK shows a strong link between self‐reported drug use and victimization (Bebbington et al. 2004). Moreover, opioid intoxication may increase vulnerability, making users easier targets (Darke 2010).

The third hypothesis focuses on the role of the illicit drug market. The expansion of this market can lead to increased violence among drug suppliers, driven by competition for territorial control. Research has consistently linked drug market violence with prohibition policies (Werb et al. 2011). Conflicts over control of drug distribution territories have been shown to lead to increased rates of assault and murder (W. N. Evans et al. 2022; P. Goldstein et al. 1997). However, the extent of this impact may vary depending on factors like drug prices and policy interventions (Gehring and Langlotz 2018; Mejia and Restrepo 2013; Millán‐Quijano 2020; Restrepo 2015; Sobrino 2019).

Given these potential mechanisms—psychopharmacological effects, economic compulsions, victimization, and systemic violence—this study empirically assesses the extent to which the opioid epidemic has contributed to rising murder rates in the U.S. By integrating insights from previous literature and leveraging the 2010 OxyContin reformulation as a unique policy intervention, this research tests these hypotheses using robust data and methodological approaches.

This study focuses on state‐level homicide rates, investigating the relationship between opioid misuse and murder through subset analyses that account for the characteristics of offenders, victims, and the circumstances surrounding murders. Specifically, it examines the homicide rates of distinct demographic groups, such as white male victims per 100,000 population, across different states and years. By exploring the link between opioid use and murder, this research aims to inform the development of more effective crime prevention and enforcement strategies. To the best of my knowledge, this is among the first studies to explore a potential causal relationship between the opioid epidemic and rising murder rates, using the reformulation of OxyContin as a key exogenous factor.

3. Data and Methods

Building on the theoretical framework outlined above, which highlights three key mechanisms—offender increase, victimization, and drug market expansion—this study empirically tests these hypotheses using a comprehensive dataset. By analyzing murder rates alongside opioid misuse and socio‐demographic data, the study aims to isolate the impact of the opioid epidemic on violent crime, particularly murder. The data sources and methodology employed ensure robustness and accuracy in assessing the relationship between drug use and crime.

3.1. Data

Rather than focusing on individual‐level characteristics, this study uses state‐level data from 2000 to 2017 to analyze the relationship between opioid and illicit opioid use, such as heroin, and murder rates. The analysis employs subsample analyses of specific demographic groups (e.g., white male victims per 100,000 population) across various states and years. The dataset includes four types of variables: crime‐related (murder) variables, OxyContin misuse rates, socio‐demographic controls, and policy variables.

Crime and murder data are sourced from two primary sources: the National Vital Statistics System (NVSS) and the Uniform Crime Reporting (UCR) Program. For robustness checks, I also use the National Incident‐Based Reporting System (NIBRS). While the NVSS provides the most accurate count of deaths (primarily homicides), the UCR offers more detailed information on the types of murders, though with some omissions in reported crime data. For example, in 2016, the UCR recorded only 91.3% of the murder and manslaughter victims that were reported by the NVSS. By combining both NVSS and UCR data, this study aims to present a more comprehensive picture of homicides in the U.S. ² NIBRS data, although available only for 10 states since 2005, is also incorporated for robustness checks due to its rich, detailed incident‐level information on the nature of violent crimes. ³

A key feature of my identification strategy is the use of the differential effect of OxyContin reformulation across states, based on pre‐existing rates of OxyContin misuse. The National Survey on Drug Use and Health (NSDUH) is used to measure the OxyContin misuse rate, which is strongly correlated with geographic prescription rates for oxycodone and OxyContin (Alpert, Powell, and Pacula 2018; Beheshti 2019; Park and Powell 2021). For the analysis, I aggregate state‐level misuse rates for the years 2004–2005, 2006–2007, and 2008–2009 to capture pre‐reformulation misuse patterns across states. This state‐level data is restricted and available upon request from the Substance Abuse and Mental Health Services Administration (SAMHSA). Aggregating these rates smooths out fluctuations in annual misuse rates and provides a more stable measure of pre‐reformulation exposure. ⁴

Additionally, I incorporate time‐varying state characteristics such as policy variables and socio‐demographic factors. Information on state policies—including Prescription Drug Monitoring Programs (PDMPs), pain clinic regulations, and marijuana laws (for both medical and recreational use)—is drawn from the RAND Marijuana Policy database (Powell, Pacula, and Jacobson 2018; Williams, Pacula, and Smart 2019). Demographic data, such as population size, age, and race, are obtained from Medicare SEER data, which provides improved imputation between census years. Variables like the unemployment rate and education levels are sourced from the Current Population Survey.

3.2. Analytic Approach

I employ a difference‐in‐differences (DID) empirical design centered on the implementation of the OxyContin reformulation policy. In a typical DID model, the treatment effect is estimated by comparing changes over time between treated and untreated groups. However, instead of adopting a traditional event study, which often divides the sample into binary treatment and control groups, I follow the approach suggested by Alpert, Powell, and Pacula (2018). In this approach, pre‐reformulation OxyContin exposure serves as a continuous treatment level. States with higher pre‐reformulation OxyContin misuse rates are likely to experience a greater impact from the reformulation, while states with lower misuse rates may be less affected. This continuous treatment design allows us to assess how varying levels of pre‐reformulation opioid misuse influence post‐reformulation crime outcomes.

The base year for the analysis is 2009, which serves as a clean baseline before users began switching to alternatives, ahead of the reformulation's rollout in late 2010 (Nolan et al. 2020). This approach avoids confounding effects that might have started in 2010 as the reformulation took effect.

To estimate the impact of the OxyContin reformulation on crime outcomes at the state level, I use the following model:

$$Y_{\mathit{\text{st}}}={\alpha }_s+{\gamma }_t+{\delta }_t\times {\text{OxyRate}}_s+{\theta }_t\times {\text{PainRate}}_s+X_{st}^{\mathit{\prime }}\beta +{ϵ}_{\mathit{\text{st}}},$$ (1)

where: $Y_{st}$ is the logarithm of the homicide rate (murders per 100,000) in state s and year t, ${\alpha }_s$ represents state‐fixed effects, ${\mathit{\gamma }}_{\mathit{t}}$ represents year‐fixed effects, ${\text{OxyRate}}_s$ denotes the pre‐reformulation OxyContin misuse rate in state s (measured for the years 2004–2009), interacted with year‐fixed effects to allow for a differential effect across time, denoted by ${\delta }_t$, ${\text{PainRate}}_s$ represents the non‐opioid painkiller usage rate for the years 2004–2009, also interacted with year‐fixed effects, with its impact denoted by ${\theta }_t$. $X_{st}^{\mathit{\prime }}$ is a vector of time‐varying state‐level controls, including socio‐demographic variables (education, age composition, and ethnicity), socioeconomic factors (unemployment rate), and policy variables (e.g., Prescription Drug Monitoring Program (PDMP) implementation, must‐access PDMP, pain management clinic policies, and medical/recreational marijuana policies). ${ϵ}_{\mathit{\text{st}}}$ is the error term.

The estimates for ${\delta }_t$ capture the annual relationship between pre‐reformulation OxyContin misuse rates and crime outcomes, using 2009 as the base year (i.e., ${\delta }_{2009}=0$), ⁵ allowing the analysis to isolate the true impact of the reformulation.

I include population‐weighted models to ensure that larger states, with more homicides, have a proportionate influence on the estimates, mitigating concerns that small states with very few homicides could disproportionately affect the results. Additionally, I cluster standard errors at the state level to account for within‐state serial correlation, which could otherwise bias the standard errors in panel data settings. The inclusion of state‐fixed effects controls for unobserved state‐specific characteristics that are constant over time, while year‐fixed effects account for national time‐varying factors, such as macroeconomic shocks, changes in national drug policies, or trends in physician prescribing practices for opioids, that might influence both opioid misuse and crime outcomes. This approach ensures a more accurate estimation of the causal effect of OxyContin reformulation on homicide rates.

To address the skewness in the homicide rate distribution, I apply a logarithmic transformation to normalize the outcome variable (homicide rate) and reduce potential bias in the estimates. Additionally, I conduct robustness checks using a Poisson regression model for count data, which estimates proportional effects across different baseline homicide rates while accounting for the discrete nature of the homicide data. This approach further validates the results, particularly in cases where homicide counts are small.

4. Results

4.1. Summary of Descriptive Statistics

Table 1 presents the descriptive characteristics categorized by pre‐reformulation OxyContin misuse status, distinguishing between High OxyContin Misuse States and Low OxyContin Misuse States. This binary classification, based on the OxyContin misuse rates from 2004 to 2009, uses the median OxyContin misuse rate (0.657%) as a cutoff for grouping states.

TABLE 1.

Descriptive statistics of U.S. states by OxyContin misuse status (2004–2009) prior to reformulation.

	All states	Low OxyContin misuse states	High OxyContin misuse states
Number of states	51	25	26
States in the west region	25	12	13
States with any PDMP in 2009	36	18	18
States with must‐access PDMP in 2009	0	0	0
States with legal medical marijuana in 2009	14	5	9
States with pain clinic policy in 2009	1	1	0
States with active medical marijuana dispensaries in 2009	2	2	0
Homicides per 100,000 population (2009)	5.12	6.13	4.15
OxyContin misuse rate (%) (2004–2009)	0.67	0.47	0.86
Painkiller misuse rate (%) (2004–2009)	6.48	5.79	7.13
Population with college or advanced degree (%) (2009)	27.63%	28.77%	26.53%
Non‐Hispanic white population (%) (2009)	81.15%	74.82%	87.23%
Share of population aged 25–44 (%) (2009)	26.31%	26.81%	25.84%
Unemployment rate (%) (2009)	9.81%	9.69%	9.92%
Mean population (2009)	6,015,128	8,530,623	3,596,383

Note: Table 1 presents the descriptive characteristics of the states included in the study, categorized into states with low and high OxyContin misuse rates based on the median misuse rate between 2004 and 2009. States with OxyContin misuse rates above the median are classified as “high misuse,” while those below the median are classified as “low misuse.” All statistics, except for the OxyContin and Painkiller misuse rates (which are averages from 2004 to 2009), are from 2009—the year immediately preceding the OxyContin reformulation. Data for the PDMP and marijuana policy variables are from the RAND Marijuana Policy database. Population, age, and race data are from the Medicare SEER database, and unemployment and education statistics are sourced from the Current Population Survey.

While this table uses a binary classification to provide an initial comparison, it is important to note that the primary analysis in this study treats OxyContin misuse as a continuous variable. The binary grouping in Table 1 is intended to highlight broad differences between states with higher and lower misuse rates but does not reflect the continuous nature of the treatment variable in the main empirical analysis.

The descriptive statistics show that states with higher OxyContin misuse rates tend to have a larger proportion of white residents, higher rates of overall painkiller use, and smaller average population sizes compared to low misuse states. These differences highlight the varying socio‐demographic and prescription opioid use patterns across states prior to the 2010 OxyContin reformulation.

4.2. Impact of OxyContin Reformulation on Homicide Rates

To estimate the impact of OxyContin reformulation on homicide rates, I analyze data from the National Vital Statistics System (NVSS). Figure 2 presents the event study, which examines the relationship between pre‐reformulation OxyContin misuse rates (2004–2009) and the logged homicide rates per 100,000 population over time.

FIGURE 2.

The Impact of OxyContin Reformulation on State‐Level Homicides (2000–2017). Notes: The outcome variable is the log of total homicides per 100,000 population at the state‐year level (mean: 1.504), calculated using data from the National Vital Statistics System (NVSS). The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The underlying regression model for this figure is specified in Equation (1) and covers the period between 2000 and 2017. The model clusters standard errors at the state level and includes state and year‐fixed effects. Control variables include several time‐varying state characteristics: Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. Additionally, the interaction term between the pain reliever misuse rate and year dummy variables is included. The year 2009 serves as the reference year (denoted by the dashed line), and the interaction term between the year 2009 and the pre‐reformulation OxyContin misuse rate is excluded to establish this reference. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2017) are equal to the 2009 reference level. The pre‐period F‐statistic for 2005–2008 is 0.16 (not statistically significant).

In the pre‐reformulation period (2005–2009), the relationship between OxyContin misuse and homicide rates appears relatively stable. The F‐statistic for the pre‐period (2000–2008) is 5.500, indicating some evidence of pre‐trends, particularly before 2005. However, further analysis shows that these differential pre‐trends are concentrated before 2005. The F‐statistic for 2005–2008 (0.16) is not statistically significant, supporting the assumption of parallel trends in the years immediately before the reformulation. This confirms the validity of using a difference‐in‐differences approach, with 2009 as the baseline year.

Post‐reformulation, I observe a statistically significant increase in homicide rates in states with higher pre‐reformulation OxyContin misuse. The post‐period F‐statistic (4.353) allows us to reject the hypothesis that post‐reformulation homicide rates (2011–2017) are equal to the baseline year (2009) at the 1% significance level. These findings suggest that the OxyContin reformulation, by driving users from prescription opioids (OxyContin) to illicit drugs such as heroin, contributed to a relative increase in homicide rates.

The coefficients for the interaction between the initial OxyContin misuse rate (2004–2009) and the post‐reformulation period (2010–2017) range from 0.20 to 0.45, implying that a 10% increase in the pre‐reformulation misuse rate is associated with a 2%–4.5% increase in homicide rates per 100,000 people. Nationally, this translates to 315–710 additional murders annually during the post‐reformulation period. However, these figures represent aggregate effects, and the impact varies significantly across states.

For instance, in states like West Virginia, where the pre‐reformulation OxyContin misuse rate was 1.13%—well above the national median—this corresponds to an increase in homicides of approximately 23.5%. With West Virginia's average pre‐reformulation homicide rate of around 5 per 100,000 people, this implies an increase of about 1.18 additional homicides per 100,000 people post‐reformulation. Such state‐level differences illustrate how the reformulation's impact was more pronounced in regions with higher misuse rates, while states with lower misuse rates saw proportionally smaller or negligible effects on homicide rates.

The time‐varying socio‐demographic controls and policy variables included in the model—such as education levels and the proportion of young people—do not show statistically significant effects, as shown in Appendix Tables A.2 and A.3. Therefore, the inclusion of these variables does not meaningfully impact the results. The nonmedical pain reliever misuse rates show negative coefficients, consistent with previous findings by Alpert, Powell, and Pacula (2018). This suggests that the increase in homicide rates is specifically linked to OxyContin misuse, rather than the misuse of other painkillers.

For robustness, I also estimate a Poisson regression model, which accounts for count data and offers advantages over log‐linear models for rare event outcomes like homicides (Silva and Tenreyro 2006). The results, presented in Appendix Figure A.3, align with the main findings, suggesting that the observed effects are robust to different model specifications.

4.3. Effects on Other Crimes

Further analysis investigates whether the OxyContin reformulation had any direct effects on other types of crime using data from the UCR Program. Prior studies have found a relationship between heroin use and property crimes, but the evidence for violent crimes remains ambiguous. My analysis shows no significant effects on theft, burglary, or assault (Appendix Figures A.4–A.7). However, there is some evidence of a direct effect on robbery, a crime that spans both property and violent categories. Figure 3 shows that the post‐period F‐statistic (2.007) is significant at the 10% level, while the pre‐period F‐statistic (0.641) is not, suggesting no pre‐reformulation trend. The interaction term for 2011 and the initial OxyContin misuse rate is 0.22, implying that a 10% increase in the pre‐reformulation OxyContin misuse rate correlates with a 2.2% increase in reported robberies for that year. Given the overall downward trend in robbery rates since 2006, reducing OxyContin misuse could have further accelerated this decline. This suggests that the reformulation did not just affect homicide rates, but also contributed to slower‐than‐expected reductions in robbery rates.

FIGURE 3.

The Impact of OxyContin Reformulation on State‐Level Robbery Rates (2000–2017). Notes: The outcome variable is the log of total robberies per 100,000 population at the state‐year level (mean: 4.395), calculated using data from the Uniform Crime Reporting (UCR) program. The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The regression model for this figure is specified in Equation (1) and covers the period from 2000 to 2017. Standard errors are clustered at the state level, and the model includes state and year‐fixed effects. Control variables include several time‐varying state characteristics, such as Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. Additionally, the interaction term between the pain reliever misuse rate and year dummy variables is included. The year 2009 is set as the reference year (represented by the red line), and the interaction term between 2009 and the pre‐reformulation OxyContin misuse rate is excluded to establish this reference. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2017) are equal to the 2009 reference level. The pre‐period F‐statistic for 2005–2008 is 0.489 (not statistically significant).

5. Mechanisms Behind Increased Homicide Rates

5.1. Victimization Hypothesis

To test the hypothesis that the OxyContin reformulation led to higher murder rates among specific victim groups, particularly in states with high OxyContin misuse rates prior to 2010, I stratify the data by race and gender using NVSS data. The underlying assumption is that the transition to heroin or other illicit drugs after the reformulation increased individuals' exposure to riskier environments, resulting in higher homicide rates. If this is true, populations with higher heroin overdose mortality should also exhibit higher murder rates as victims.

The results (Appendix Table A.4 and Figure A.8) show that in states with higher pre‐reformulation OxyContin misuse rates, homicide rates increased significantly for male victims, regardless of race. The effects are most consistent for white males (Figure 4), while patterns for Black and other males are more variable. Notably, the same subpopulations with higher homicide rates also experienced elevated heroin overdose mortality, particularly among males. In contrast, the impact of the reformulation on female homicide rates is negligible (Appendix Table A.4 and Figure A.9). While no substantial reformulation effects are found for women, heroin overdose deaths among females begin to rise later, around 2013, suggesting delayed effects on both overdose deaths and murders, though this may not be fully captured here.

FIGURE 4.

The Impact of OxyContin Reformulation on Homicides and Heroin Overdose Deaths Among White Male Victims (2000–2017). Notes: The outcome variables are the total number of white male homicide victims per 100,000 population at the state‐year level (Panel A, mean: 4.511) and the total number of white male heroin overdose deaths per 100,000 population at the state‐year level (Panel B, mean: 2.677). These outcomes are calculated using data from the National Vital Statistics System (NVSS). The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The regression model for this figure is specified in Equation (1) and covers the period from 2000 to 2017. Standard errors are clustered at the state level, and the model includes state and year‐fixed effects. Control variables include time‐varying state characteristics such as Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. The interaction term between the pain reliever misuse rate and year dummy variables is also included. The year 2009 is set as the reference year (denoted by the red line), with the interaction term between 2009 and the pre‐reformulation OxyContin misuse rate excluded. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2017) are equal to the 2009 reference level. The pre‐period F‐statistics for 2005–2008 are 1.769 for Panel A (not statistically significant) and 1.306 for Panel B (not statistically significant).

These findings suggest that the OxyContin reformulation had heterogeneous effects, with more pronounced impacts on male victims, particularly white males. This pattern aligns with previous research (Blunkett 2004), which shows that men are generally more likely to be victims of violent crime in public spaces, whereas women tend to experience violence in domestic settings. Since public spaces are often associated with obtaining illicit drugs, the reformulation's effects may have led to increased exposure to violence for male victims in particular.

5.2. Growth in Criminal Behavior

I test the hypothesis that, following the OxyContin reformulation, states with higher pre‐reformulation OxyContin misuse rates would experience a relative rise in murder rates where the offenders were suspected to be under the influence of illicit drugs. If the transition from prescription opioids to illicit opioids led to increased criminal behavior, individuals exposed to the transition would be more likely to commit homicides.

To examine this, I conduct subsample analyses based on the race and gender of the offenders, similar to the analysis focusing on victim characteristics. Since the NVSS does not provide information on offenders, I rely on data from the UCR, though this is limited to cases where the offenders were apprehended, potentially introducing bias due to underreporting or selective capture. The results (Appendix Table A.4, third row, and Appendix Figure A.10) do not provide strong support for the hypothesis. While there is some indication of an effect for white male offenders, the presence of pre‐existing trends raises concerns about confounding factors, which cast doubt on the robustness of these findings. Therefore, these results should be interpreted with caution, acknowledging that confounding variables and pre‐trends could influence the observed outcomes.

To further investigate the relationship between illicit drug use and violence, I use UCR supplementary homicide report data on murders committed during fights under the influence of narcotics. ⁶ While there is a small increase in narcotic‐related murders in states with high pre‐reformulation OxyContin misuse rates (Figure 5), the limited number of cases makes the results noisy and inconclusive. The small sample size, particularly in states with fewer homicides, presents additional challenges, suggesting these findings should be considered as suggestive rather than definitive.

FIGURE 5.

The Impact of OxyContin Reformulation on Murders During Fights Under the Influence of Narcotics (2000–2016). Notes: The outcome variable is the total number of murders occurring during brawls under the influence of narcotics per 100,000 population at the state‐year level (mean: 0.028). These outcomes are calculated using data from the Uniform Crime Reporting (UCR) program. The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The regression model for this figure is specified in Equation (1) and covers the period from 2000 to 2016. Standard errors are clustered at the state level, and the model includes state and year‐fixed effects. Control variables include several time‐varying state characteristics: Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. The interaction term between the pain reliever misuse rate and year dummy variables is also included. The year 2009 is set as the reference year (denoted by the red line), with the interaction term between 2009 and the pre‐reformulation OxyContin misuse rate excluded to establish this reference. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2016) are equal to the 2009 reference level. The pre‐period F‐statistic for 2005–2008 is 1.354 (not statistically significant).

5.3. Gang Activity and Drug Trafficking Hypothesis

To test the hypothesis that OxyContin reformulation led to an increase in gang‐related or drug trafficking‐related murders in states with high OxyContin misuse rates, I use UCR supplementary homicide report data, focusing on gang activity and drug trafficking as the circumstances of murder. The dependent variable remains the logged murder rate per 100,000 population, restricted to gang‐related murders. ⁷

The analysis (Figure 6, Panel A) does not show significant evidence of an increase in gang‐related murders following the reformulation. Similarly, when limiting the sample to murders during drug trafficking (Figure 6, Panel B), I find no significant effects. These results suggest that the reformulation's impact on homicide rates is not primarily driven by gang or drug trafficking‐related violence. This is consistent with the idea that the reformulation disproportionately affected users transitioning to heroin and other illicit drugs, increasing their vulnerability as victims rather than driving systemic violence within organized crime networks.

FIGURE 6.

The Impact of OxyContin Reformulation on Murders Related to Gang Activity and Drug Trafficking (2000–2016). Notes: The outcome variables are the total number of murders related to gang activity per 100,000 population at the state‐year level (Panel A, mean: 1.073) and the total number of murders related to narcotic felonies per 100,000 population at the state‐year level (Panel B, mean: 0.148). These outcomes are calculated using data from the Uniform Crime Reporting (UCR) program. The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The regression model for this figure is specified in Equation (1) and covers the period from 2000 to 2016. Standard errors are clustered at the state level, and the model includes state and year‐fixed effects. Control variables include time‐varying state characteristics such as Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. The interaction term between the pain reliever misuse rate and year dummy variables is also included. The year 2009 is set as the reference year (denoted by the red line), with the interaction term between 2009 and the pre‐reformulation OxyContin misuse rate excluded to establish this reference. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2016) are equal to the 2009 reference level. The pre‐period F‐statistics for 2005–2008 are 1.766 for Panel A (not statistically significant) and 2.815 for Panel B (statistically significant at 5%).

6. Sensitivity Analysis and Robustness Checks

6.1. Using Complementary Datasets

To check the robustness of the primary analysis, two additional datasets are analyzed: UCR data and NIBRS data. These complementary datasets help confirm that the results are not driven by limitations in the primary data sources.

First, the UCR data, which provide the number of known offenses (murders), show trends that are consistent with those observed in the NVSS dataset. Figure 7 illustrates that states with higher pre‐reformulation OxyContin misuse rates experienced relatively larger increases in murder rates, similar to the NVSS results. The effect size is nearly identical, with a coefficient of 0.28 in 2011, further confirming the robustness of the main findings.

FIGURE 7.

The Impact of OxyContin Reformulation on Murders Using UCR Data (2000–2017). Notes: The outcome variable is the log of total murders per 100,000 population at the state‐year level (mean: 1.417), calculated using data from the Uniform Crime Reporting (UCR) program. The y‐axis represents the coefficients on the pre‐reformulation OxyContin misuse rate, interacted with year dummy variables. The regression model for this figure is specified in Equation (1) and covers the period from 2000 to 2017. Standard errors are clustered at the state level, and the model includes state and year‐fixed effects. Control variables include several time‐varying state characteristics: Prescription Drug Monitoring Program (PDMP) adoption, marijuana policy adoption year, population, age, race, unemployment rate, and education. The interaction term between the pain reliever misuse rate and year dummy variables is also included. The year 2009 is set as the reference year (denoted by the red line), with the interaction term between 2009 and the pre‐reformulation OxyContin misuse rate excluded to establish this reference. The pre‐period F‐statistic tests whether the pre‐reformulation estimates (2000–2008) are equal to the baseline estimate in 2009. The post‐period F‐statistic tests whether the post‐reformulation estimates (2011–2017) are equal to the 2009 reference level. The pre‐period F‐statistic for 2005–2008 is 0.222 (not statistically significant).

Second, the NIBRS dataset is used, though its coverage is limited to 10 states with data available since 2005. To maintain comparability, I limit the NVSS sample to the same 10 states. While the results from both datasets are generally similar, no significant jump in murder rates is observed post‐reformulation in either dataset (Appendix Figure A.11). It is important to note that the 10 states used in this analysis exhibit characteristics similar to high OxyContin misuse states prior to reformulation, including lower average homicide rates (4.60 compared to the national average of 5.12), lower population counts (around 4 million compared to the national average of 6 million), and higher painkiller and OxyContin misuse rates (6.83 and 0.69, respectively, compared to national averages of 6.48 and 0.67). Given the skewed sample toward high OxyContin misuse states, there may not be enough variation in independent variables to detect policy impacts in these states.

6.2. The Role of PDMP Adoption

To investigate whether the adoption of Prescription Drug Monitoring Programs (PDMPs) affects the analysis, I conduct a robustness check, as previous studies have found that PDMPs decrease violent crime (Dave, Deza, and Horn 2021) but increase heroin‐related offenses (Mallatt 2022). Appendix Table A.5 shows that PDMP adoption is not significantly associated with changes in homicide rates.

In column (1), where neither pre‐OxyContin misuse nor pre‐painkiller misuse rates are included, PDMP adoption shows no significant effect on homicide rates. Similarly, in column (2), where only pre‐painkiller misuse rates are included, the PDMP adoption coefficient remains insignificant. Columns (3) and (4) exclude the indicator variable for any PDMP implementation, based on Mallatt's (2022) findings that must‐access PDMPs increase the prevalence of heroin‐related crimes. The coefficient for must‐access PDMP is positive, implying a possible increase in homicide rates post‐PDMP adoption; however, none of the coefficients are statistically significant. This differs from previous studies, suggesting that PDMPs may have limited effects on homicide rates. Interestingly, Evans et al. (2022a, 2022b) found that the OxyContin reformulation had larger impacts on child neglect rates than PDMP adoption did, suggesting that reformulation policies may have more significant consequences than PDMPs in certain areas.

6.3. Excluding Small States and Large State‐Specific Analysis

Additionally, I perform a robustness check excluding small states, and the results remain consistent, confirming that the relationship between OxyContin misuse and homicide rates is not driven by these states (Appendix Figure A.12). Recognizing concerns that large states with more reliable OxyContin misuse estimates might exhibit different trends, I conduct a separate analysis focusing exclusively on these larger states. This analysis confirms that the relationship between OxyContin misuse and homicide rates is observable even in these states with more stable data, further supporting the robustness of the findings. The results of this analysis are presented in Appendix Figure A.13.

6.4. Excluding Cities With High Murder Increases

Szalavitz and Rigg (2017) argue that there is a disconnect between the opioid epidemic and crime trends, citing that murder rates surged in large metropolitan areas such as Chicago, which are not the primary regions affected by the opioid epidemic, which tends to affect suburban and rural areas (Keyes et al. 2014). To assess whether the surge in murders in major metropolitan areas affects the results, I rerun the analysis excluding the five cities with the largest increases in murder rates: St. Louis, Baltimore, Detroit, New Orleans, and Chicago.

Appendix Figure A.14 shows that the results without these five cities remain consistent with the main findings. I also perform a similar analysis focusing on white male victims (the group most affected by the reformulation), and the results (Appendix Figure A.15) similarly show no substantive change when these five cities are excluded.

6.5. The Heroin Market Before the Reformulation

The primary analysis does not account for the pre‐reformulation levels of the heroin market at the state level. However, W. N. Evans et al. (2019) demonstrated that the maturity of the heroin market before the reformulation could influence the reformulation's effect on homicide rates.

To address this, I stratify states based on heroin overdose death rates from 2004 to 2009 to account for pre‐existing heroin market conditions. The results show that the impact of the pre‐reformulation OxyContin misuse rate is larger in states with higher heroin overdose death rates (Appendix Figure A.16). This suggests that states with more mature heroin markets before the reformulation were more affected by the policy change, further highlighting the role of pre‐existing market conditions in shaping the outcomes.

6.6. Sensitivity to Age of Victims

Given that younger populations have higher OxyContin misuse rates (Hadland et al. 2017), I conduct an additional analysis focusing on murders where the victims are aged 15 to 24. As expected, the results are consistent with the main analysis (Appendix Figure A.17), confirming that the general patterns hold even when limiting the analysis to younger victims.

6.7. Sensitivity to Reference Year (2010 vs. 2009)

A key analysis assumption is the choice of 2009 as the reference year. Given the concern that the reformulation's effects might not be immediate—due to existing stocks of the original OxyContin and the reformulation occurring in August 2010—I rerun the analysis using 2010 as the reference year to account for the possibility of delayed effects. However, using 2010 as the reference year introduces complications. While the FDA approved the reformulation in April, the newly formulated OxyContin did not start shipping until August. As a result, 2010 captures a transitional period with reformulation effects beginning to take hold. This transitional nature makes 2010 a less suitable reference year for assessing the full impact of the reformulation. In contrast, 2009 offers a cleaner baseline, unaffected by any reformulation changes, providing a more reliable point of reference for evaluating the reformulation's complete effects.

Figure A2 shows that using 2010 as the reference year essentially nullifies the observed effects of the reformulation on homicide rates, likely due to this partial exposure. In contrast, using 2009 as the baseline year provides clear evidence of the reformulation's effects. This is because 2009 fully precedes the reformulation, offering a clean and unaffected reference period. Additionally, several studies (e.g. Alpert, Powell, and Pacula 2018; Beheshti 2019; Park and Powell 2021) also advocate for using 2009 as the reference year for similar reasons—capturing a pre‐reformulation period before the population began shifting behavior in anticipation of the reformulation.

Thus, the choice of 2009 is not only supported by prior research but is also essential to avoid the ambiguity caused by the partial rollout of the reformulation in 2010. This selection ensures a clearer estimation of the reformulation's true impact, as relying on 2010 would obscure important dynamics that began earlier in the year, further weakening the policy's observable effects on homicide rates.

6.8. Addressing Continuous Treatment in Difference‐In‐Differences Models

Recent criticism of difference‐in‐differences models with two‐way fixed effects and continuous treatment variables (Callaway, Goodman‐Bacon, and Sant'Anna 2024) suggests that such models may not provide accurate estimates. To address this, I estimate the aggregated Average Treatment Effect on the Treated (ATT), as recommended by Callaway, Goodman‐Bacon, and Sant'Anna (2024). While there are fluctuations in the pre‐treatment period, the ATT stabilizes in the later years (Appendix Figure A.18), confirming that the main findings are robust to this methodological concern.

7. Discussion

This study provides evidence that the reformulation of OxyContin may have contributed to a relative increase in homicide rates while having little to no impact on property crimes. A plausible explanation for this discrepancy may lie in the nature of crime reporting. Unlike property crimes, which often depend on victims' willingness to report, murders are typically reported without exception due to the necessity of medical or legal intervention. Violent crimes, particularly murders and robberies, tend to be documented at higher rates because of external reporting from medical professionals or law enforcement. This consistent reporting may explain why the reformulation disproportionately affected homicide rates, as murder cases are more likely to be accurately captured. In contrast, the analysis found no significant effect on property crimes, possibly because property crimes are less likely to be consistently reported or may not be as directly impacted by the changes in the opioid market.

The discrepancy between the national trends in Figure 1 and the state‐level results in Figure 2 can be explained by differences in aggregation and how quickly the impact of the reformulation was felt across different states. Figure 2 captures the immediate effect of the reformulation starting around 2010 in high‐misuse states, where the shift to illicit opioids like heroin occurred more quickly and sharply, driving an earlier rise in homicides. Figure 1, however, presents national homicide rates, which aggregate data from all states, including those with lower OxyContin misuse rates, where the reformulation's effects on crime were delayed or less pronounced. This delay in lower‐misuse states resulted in the national increase in homicide rates becoming noticeable only around 2015. Therefore, the national trends appear later because they reflect a cumulative effect, which masks the early increases seen in states most affected by the reformulation. This highlights the need for state‐level analysis, as national averages may overlook important regional differences in the timing and intensity of the reformulation's impact.

Among the proposed mechanisms driving the rise in homicides, the growth in the number of potential victims stands out as the most likely explanation. States with higher pre‐reformulation OxyContin misuse rates experienced a more pronounced rise in homicide rates post‐reformulation, particularly among white male victims. This aligns with studies showing that males are more likely to be victims of violent crime in public spaces, where drug transactions and related violence occur. However, no substantial increases in the number of offenders by demographic category were observed, suggesting that mechanisms such as psychopharmacological or economic compulsive violence—typically linked to drug‐related crimes—are not major contributors to the rise in homicides. This challenges the assumption that opioid misuse leads directly to violent behavior from users themselves, suggesting instead that increased victimization may play a larger role.

Additionally, the data provide little support for an increase in homicides related to gang activity or drug trafficking. This null finding aligns with studies showing stable heroin prices post‐reformulation, limiting the economic incentives that typically drive gang violence. As previous research (Gehring and Langlotz 2018) suggests, supply‐side policies like the OxyContin reformulation may have limited impact on gang‐related murders if drug prices remain stable. In the U.S., heroin prices did not substantially increase following the reformulation, and the new “pizza delivery” model for drug distribution (Quinones 2015) likely reduced the visibility and territorial conflicts traditionally associated with gang activity.

TABLE 2.

Heroin prices in the United States (2007–2017).

	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017
Heroin retail prices (street prices) in the U.S. (US$ per gram)
Average (US$)	230	310	275	279	260	271	263	266	267	307	307
Average, inflation‐adjusted (2017 US$)	271	353	314	314	284	290	277	275	276	314	307
Average, adjusted for purity	634	822	871	1126	867	773	732	724	768	929	929
Average, adjusted for purity and inflation (2017 US$)	750	935	995	1266	944	825	771	750	794	949	929
Heroin wholesale prices in the U.S. (US$ per kilogram)
Average (US$)	71,200	57,500	65,053	65,750	55,583	61,000	61,000	68,333	59,500	53,333	53,333
Average, inflation‐adjusted (2017 US$)	84,173	65,463	74,327	73,911	60,570	65,125	64,185	70,753	61,534	54,470	53,333

Source: United Nations Office on Drugs and Crime (UNODC), Heroin and Cocaine Prices in the EU and USA (2017). Available at:https://dataunodc.un.org/drugs/heroin_and_cocaine_prices_in_eu_and_usa‐2017.

This study has several limitations that may influence the interpretation of the findings. One key limitation is the reliance on reported crime data—particularly from the UCR, NVSS, and NIBRS—which may introduce biases due to inconsistent reporting across states. Additionally, the limited availability of NIBRS data restricts the scope for subgroup analyses, and incomplete data on offender characteristics may obscure the full impact of the reformulation across different demographic groups. Furthermore, limited data on crime circumstances pose challenges for detecting significant effects related to gang activities. With nearly half of all murders (45%) in the dataset lacking specific circumstantial information, the data may be too limited to draw definitive conclusions. Given that gang‐related murders represent only a small portion of total homicides (5.76%), missing information on crime circumstances could disproportionately skew the analysis.

Another limitation concerns the variability in NSDUH OxyContin misuse rates. The low correlations between NSDUH misuse rates across different periods (e.g., 2004–05 and 2008–09) may raise concerns about data precision. Averaging these rates over multiple years has helped smooth fluctuations, providing a more stable estimate for assessing long‐term trends. Cross‐referencing NSDUH data with ARCOS MME data has bolstered reliability, despite a moderate correlation. However, while NSDUH remains the best available dataset for tracking state‐level trends, its limitations warrant cautious interpretation.

Additionally, the results show sensitivity to the choice of reference year, with notable variability in outcomes depending on whether 2009 versus 2010 is used as the baseline. Although 2009 serves as the more appropriate reference year due to the partial exposure to reformulation in 2010, the differences observed when using 2010 underscore the challenges of precisely isolating the reformulation's effects. This inconsistency may lead some readers to question the robustness of the findings, as the choice of reference year can influence the magnitude and direction of the observed effects.

8. Conclusion

This paper investigates whether the introduction of the abuse‐deterrent version of OxyContin had unintended consequences on violent crime, particularly homicide rates. By testing three potential mechanisms—an increase in vulnerable victims, offenders, or gang activity—the analysis reveals that the reformulation is associated with a significant relative increase in murder rates in states with higher pre‐reformulation OxyContin misuse rates. This rise is primarily driven by an increase in vulnerable victims rather than by an increase in offenders or gang‐related activity.

The findings show that the reformulation disproportionately impacted certain victim groups, especially males, as the shift from prescription opioids to more dangerous illicit drugs like heroin likely increased their exposure to violence. There is limited evidence to suggest a corresponding rise in offenders or organized crime, indicating that the reformulation's primary effect on homicides stems from increased victimization.

While this study provides important insights, it is subject to several limitations, including reliance on reported crime data, variability in NSDUH misuse data, and sensitivity to the choice of reference year. Despite these limitations, the findings contribute valuable insights into understanding the relationship between opioid reformulation and violent crime, an area that remains largely under‐explored.

Future research should address these limitations by incorporating more comprehensive datasets that capture a broader spectrum of opioid‐related criminal behavior. Expanding the scope of analysis to include other regions and longer study periods as data availability improves will enhance the generalizability of these findings. Additionally, examining how opioid policies intersect with socio‐economic factors—such as healthcare access, poverty, and regional drug markets—will be essential for designing more effective, targeted interventions to mitigate both opioid misuse and its associated social harms.

Ultimately, this study highlights the need for a more nuanced approach to drug policy, one that considers potential unintended consequences on public safety. By shedding light on the complex relationship between opioid misuse and violent crime, these findings underscore the importance of balancing supply‐side interventions with broader, integrated public health strategies that address the underlying factors driving opioid dependence and misuse.

Conflicts of Interest

This research was partially funded by the Pardee RAND Graduate School Dissertation Award and the CDC grant R01CE02999. The funding organizations did not request to review this paper prior to submission.

Supporting information

Supporting Information S1

Data Availability Statement

The data supporting the findings of this study are available from the National Center for Health Statistics (NCHS) and the Substance Abuse and Mental Health Services Administration (SAMHSA). Restrictions apply to the availability of these data, which were used under license for this study. The data are available from the corresponding author with permission from NCHS and SAMHSA.