Using arrest and prescription data to examine the relationship between intimate partner violence and opioid prescriptions in the United States, 2006–2012

Abstract

Background:

Research has established a strong and positive correlation between substance use and intimate partner violence due to a complex interplay of individual, situational, and contextual factors. This study seeks to further explore this relationship in the context of the recent opioid crisis in which millions of Americans have been diagnosed with an opioid use disorder. Specifically, we analyze how opioid prescriptions relate to intimate partner violence within and between counties over time throughout the rise of the opioid crisis.

Methods:

This study employs an integrated dataset that merges crime data from the Federal Bureau of Investigation, demographic data from the American Community Survey, and prescription opioid pill counts from the Drug Enforcement Administration to study the relationship between opioid pills prescribed per person and levels of intimate partner violence arrests from 2006–2012. Fixed-effects and mixed-effects techniques are both used.

Results:

Increases in opioid pill distribution volume within-counties over time are related to increases in intimate partner violence arrest volume (p ≤ .001). Additionally, counties which have higher amounts of opioid pills in circulation tend to experience higher levels of arrests for intimate partner violence than counties with fewer pills (p ≤ .001).

Conclusions:

Policymakers who are dealing with the effects of the opioid crisis should consider the relationship between opioids and intimate partner violence when attempting to address either of these issues. Based on the results of this study, addressing opioid dependence and mitigating the extent of the crisis may also reduce intimate partner violence.

1. Background

Intimate partner violence (IPV) affects millions of Americans every year (Black et al. 2011; CDC 2018c). IPV includes physical violence, sexual violence, threats of physical or sexual violence, stalking, and psychological/emotional violence by a current or former partner or spouse (CDC 2018a; NIJ 2007). Approximately 36.4% of women and 33.6% of men experience sexual violence, physical violence, and/or stalking by an intimate partner in their lifetime, although not all of these instances are actually reported (CDC 2018c). To view the ubiquity of this public health issue from another perspective, IPV accounts for about 15% of all violent crimes (Truman and Morgan 2015) and 16% of all homicides (CDC 2018a). Further, data from 1980–2008 demonstrate that about 42% of female homicide victims were murdered by a current or past partner, compared to 7% of male homicide victims (Cooper and Smith 2011).

While there are many correlates of IPV including childhood abuse and mental health complications (Afifi et al. 2009), neighborhood characteristics (Copp et al. 2015; Jackson 2016), income (Kaukinen and Powers 2015; Okuda et al. 2015), and previous experience of IPV (Afifi et al. 2009) or victimization (Kuhl et al., 2015), research has consistently found a strong relationship between IPV and substance use. A staggering majority of IPV perpetrators engage in substance use (Brookoff et al. 1997; Easton et al., 20003; Moore and Stuart 2004) and studies demonstrate that substance use facilitates or exacerbates IPV (Stuart et al. 2013). Studies attempting to understand the mechanism behind the link between substance use and IPV suggest that an increased likelihood of IPV may be related to a complex interplay of factors, such as irritability due to withdrawal, cravings for drugs, and frustration around a partner’s inability or refusal to acquire money to purchase drugs, the pharmacological property of the substance, the context of the situation in which the drug was ingested, preexisting individual characteristics, and/or the gendered power and control dynamics of the relationship (Cafferky et al. 2018; Gadd et al. 2019; Gilchrist et al. 2019; Radcliffe et al. 2019).

Studies focusing on the relationship between using opioids and intimate partner violence demonstrate a positive association between use and experiencing IPV (Moore et al., 2011; Smith et al. 2012; Stone and Rothman 2019). A meta-analysis by Stone and Rothman (2019) concluded that 36–94% of women who have used opioids are a survivor of IPV and 58% of men who have used opioids perpetrate IPV. Additionally, Moore et al. (2011) conclude that opioid-dependent fathers report more physical, sexual, and psychological aggression toward the mother of their child over the course of their relationship compared to a control group of men with no history of drug or alcohol abuse that lived in the same community as the drug-dependent fathers. These men also report greater levels of aggression directed at them from the mother than the control group.

However, not all studies report this link between opioid use and IPV. A meta-analysis by Cafferky et al. (2018) finds that heroin is the only drug in their comparative study that is not significantly linked to IPV. They suggest that this finding could reflect the sedative properties of heroin or the low number of heroin effect sizes in their study. Further, Gilchrist et al. (2019) point out that there are quite a few gaps in the literature that hinder our ability to better elucidate the link between substance use and IPV. They highlight that the characteristics of the situation in which the abuse happened are probably quite important (e.g., were perpetrators using or withdrawing from substances at the time, was the substance used socially or with a partner, and the time-sequence relationship between substance use and the abuse situation that occurred). Indeed, one of the theories researchers use to explain the link between substance use and IPV is the biopsychosocial model, which places emphasis on an interplay of mediating and moderating factors that, taken together, explain the link between substance use and IPV (Cafferky et al. 2018). However, it is difficult for individual studies to take all of these factors into account in their analyses which could in turn influence the results of these individual-level studies (Cafferky et al. 2018).

The deeply intertwined relationship between substance use and IPV is particularly troubling in light of the recent opioid crisis. Largely thought to stem from a misinformation campaign directed by pharmaceutical companies (Van Zee 2009), the number of opioid-related deaths has risen astronomically in the U.S. since 1999 (CDC 2018b). In 2017, about 68% of drug overdose deaths involved opioids (Scholl et al. 2018), leading the United States Department of Health and Human Services to declare the opioid crisis a public health emergency. Contemporaneous estimates highlight the severity of the problem: An estimated 130 Americans die each day from opioid-related overdoses (U.S. Department of Health and Human Services 2019).

Scholars have long established the individual-level relationship between substance use and IPV. However, studies have yet to consider how this relationship might change over time, particularly in the context of the ongoing and widespread opioid crisis in which the United States currently finds itself. Additionally, while there has been a large body of research focusing on the extent of opioid dependency and misuse, it is crucial to understand how this crisis has impacted other aspects of society and social ills like IPV. Furthermore, studying how the opioid epidemic relates to levels of IPV is crucial for informing policy and care aimed at mitigating the effects of this crisis in the United States for both those with an opioid use disorder and survivors of IPV.

1.1. Current study

Using longitudinal data spanning between 2006 and 2012 from the Drug Enforcement Administration (DEA), the Federal Bureau of Investigation’s (FBI) National Incident Based Reporting System (NIBRS) program, and the Census Bureau’s American Community Survey (ACS), the current study is broadly situated in investigating the relationship between the number of opioid pills prescribed in a county and the amount of arrests a county experiences due to IPV. In the United States, counties are subdivisions of states and often have some form of governmental authority (U.S. Census 2016). There are over 3,000 counties spread across the 50 American states.

Specifically, we have two research questions. First, how do changes in the number of opioid pills prescribed over time relate to changes in the number of arrests for IPV within counties? This question is answered using a fixed-effects approach which models change in the number of pills prescribed per person in conjunction with changes in IPV over time. We hypothesize (H1) that increases in the number of pills prescribed per person will relate to increased levels of arrest for IPV over time. Our second research question is concerned more with comparisons between counties than comparisons within counties. Controlling for a variety of important county-level characteristics which might relate to IPV (e.g., poverty, opiate arrests, race, unemployment, urbanicity, etc.), do counties which have higher pill prescription rates experience higher levels of IPV than counties with lower prescription rates? Due to numerous prior studies having established a covariance between opioid use and IPV (e.g., Moore, Easton, and McMahon 2011; Smith et al. 2012; Stone and Rothman 2019), we expect (H2) that counties with higher rates of pills prescribed per person will experience the highest amounts of IPV arrests.

2. Methods

2.1. Data and sample

To test the research questions described above, we use a county-level, hybrid dataset that contains data collected from the United States between 2006 to 2012. These data contain information from the NIBRS (National Archive of Criminal Justice Data 2017), the ACS (U.S. Census Bureau 2013), and the newly released opioid prescription pill counts from the DEA (Washington Post 2020a).

NIBRS is an FBI program which provides rich incident-based data on crimes that come to the attention of law enforcement. As part of the Uniform Crime Reporting (UCR) program, law enforcement agencies in the NIBRS database submit information about crime events to their state UCR officials, including the type of offense and demographic characteristics of victim(s) and offender(s). Monthly reports from law enforcement agencies are combined to reflect yearly estimates. For this project, arrestee-level files (often with millions of arrestees) were aggregated and reshaped at the county-level, thus providing us with county-level data on all arrests within a given county for each year between 2006 and 2012. Using the NIBRS data allows us to access detailed information about arrests for both opioids and IPV.

Our hybrid dataset also draws on information from the 2008–2012 five-year cycle of the ACS to provide county-level demographic information. The ACS is an ongoing, annual survey that gathers housing, economic, and demographic information about respondents and their households. During a five-year cycle, the ACS samples each county once, thus making these variables time invariant. The ACS uses a two-phase stratified random sampling process and boasts very high response rates (97% and higher for 2008–2012; American Community Survey Office 2013, 2014). We use the ACS to provide demographic variables that contextualize the numerous between-county differences that exist across the United States.

In addition to the NIBRS and the ACS, our dataset also draws on newly released prescription opioid pill count data from the DEA. Providing pivotal information to this study, the DEA data captures the total number of opioid pills prescribed by doctors and dispensed to patients by pharmacies in each county in the United States between the years 2006–2012. These data, which were released following a successful legal challenge by the Washington Post and HD Media (see Washington Post 2020b), are unique in that they offer the only information on the number of opioid prescriptions and total number of pills distributed within each county currently available to researchers.

The common identifier in each of these datasets is the FIPS (Federal Information Processing Standards) county code that uniquely identifies each county in the United States. Using this critical piece of information, each of the data sources were merged to create one large, county-level dataset containing all information used in this study. While the NIBRS and the ACS contain data that extend through 2017, the DEA data capturing opioid pill distributions are not currently available past the year 2012. As such, this restricts our analysis to the years 2006 to 2012 (seven years of data) and leaves us with a longitudinal panel of 1,493 counties captured at a total of 9,008 measurement time points. Counties are the units of analysis.

2.2. Measures

2.2.1. Dependent variable

Descriptive statistics for all measures can be viewed in Table 1. The dependent variable in this study is the number of IPV-related arrests within each county per year. This time-variant count variable comes from the NIBRS data. NIBRS does not automatically categorize an arrest as IPV-related. However, the key information needed to classify an arrest as IPV-related is present in the NIBRS data. As such, this variable includes arrests which meet all of the following three conditions. First, the arrest had to have one offender and one victim. Second, the crime had to be categorized as an against-person (as opposed to against-property) offense, which includes – among other things – assault, kidnapping, forced sex, and homicide. Third, the victim had to be somehow romantically linked to the offender. We include incidents where the victim and offender were spouses, common-law spouses, boyfriend/girlfriend, same sex partners, or an ex-spouse. On average, each county had about 111 IPV arrests each year (SD between counties = 250.822). As a time-variant measure, there is a substantial amount of change over time within counties on a year-to-year basis (SD within counties = 51.895). In 2006, the total number of arrests identified as IPV-related in this dataset is 125,381. By 2012, this number had increased to 152,033. As a count variable, this measure is heavily skewed right (p ≤ .001). To help correct for this, the forthcoming models use the natural log of IPV arrests as the outcome.

Table 1.

Descriptive characteristics of variables used in analyses; n = 9,008.

	Mean	Standard deviation (between)	Standard deviation (within)	Range
Time variant
IPV arrests	111.416	250.822	51.895	0–3,697
Opioid pills prescribed per person	39.684	28.951	9.841	0–541
Opioid arrests	12.244	47.654	19.746	0–1,131
Time invariant
Percent poverty	8.257	3.340	–	1.592–23.444
Percent insured	86.040	5.103	–	54.800–97.400
Percent unemployed	8.224	3.578	–	0.400–26.100
Percent White	87.110	14.106	–	17.405–99.896
Population size / 10,000	8.804	22.693	–	0.075–407.920

2.2.2. Focal independent variable

The focal independent variable in this study is the number of opioid pills prescribed within each county during each year between 2006 and 2012. This is a time-variant variable from the DEA data. In its unadjusted form, the descriptive characteristics of this variable are – to say the least – alarming. The number of opioid pills prescribed across the U. S. increased from 190,000,000 in 2006 to 225,000,000 in 2012. The total number of opioid pills prescribed between 2006 and 2012 in the United States is approximately 73,590,000,000 (or, for round numbers, 74 billion pills). The unadjusted mean number of pills prescribed each year in each county is 3.5 million [SD between counties ≈ 9.5 million, SD within counties ≈ 1.5 million; the upper limit of the range is 233 million pills distributed within a county in one year’s time]).

Due to these overwhelmingly high numbers of opioid pills distributed within each county on an annual basis, we divided the pill count by the county’s raw population size (a variable from the ACS). This produces our focal independent variable, which represents the number of opioid pills prescribed per person within each county at each year between 2006 and 2012 (M = 39.684 pills per person per year; SD between counties = 28.951). Due to concerns that there may be a reciprocal relationship between substance use and IPV (e.g., Afifi et al. 2019; Rothman, Stone, and Bagley 2018), we use a lagged measure of opioid pills (at t – 1) to provide for a more parsimonious time order. Perhaps unsurprisingly, there is a substantial amount of variation within counties over time in pill counts per person (SD within counties = 9.841 pills per person).

2.2.3. Control variables

The control variables come from both the NIBRS and ACS data sources. Based on the NIBRS data, we control for a time-variant measure of the number of arrests in each county for opioids per year (M = 12.244; SD between counties = 47.654; SD within counties = 19.746). From the ACS, we include a number of time-invariant measures which control for the demographic characteristics of each county including the percent of the county’s population living in poverty (M = 8.257; SD = 3.340; range = 1.592 – 23.444), the percent with health insurance (M = 86.040, SD = 5.103), and the percent who are unemployed (M = 8.224; SD = 3.578; range =0.4 – 26.1). The percent of the county’s population which is White is also controlled (M = 87.110; SD = 14.106). Finally, since there will be more IPV-related arrests in counties with very large populations, we controlled for the county’s population size. Due to naturally high values, we linearly transformed this variable by dividing by 10,000 and include a measure of population size / 10,000 in our analyses (M = 8.804, SD = 22.693). Descriptive statistics for all measures can be viewed in Table 1.

2.3. Analytic strategy

The nature of our research questions necessitates analyzing panel data with the use of multilevel regression techniques. Our first research question asks how the number of opioid pills prescribed per person relates to levels of IPV within a county over time. An intraclass correlation of .90 demonstrates that a two-level model where time (level 1) is nested within counties (level 2) is most appropriate for this analysis. Additionally, a significant (p ≤ .001) Hausman test confirms that a fixed-effects model is needed for the purposes of answering our first research question. In this context, our fixed-effects regression models within-county change between the years 2006 and 2012. Therefore, using a fixed-effects model to investigate within-county differences in IPV arrests is warranted for both conceptual and statistical reasons.

The second research question in this study is concerned with comparing how the relationship between opioid pills prescribed per person and the number of IPV arrests varies between counties. This necessitates the use of a mixed-effects model, which models both within- and between-county differences. Again, an intraclass correlation of .87 indicates that a multilevel model with time at level 1 and county at level 2 is most appropriate.

Since our outcome is a skewed count, we explored for the possibility of using a Poisson or negative binomial link in our multilevel models with an untransformed dependent variable. Since the variance of the outcome is not close to its mean, the Poisson assumptions were violated, meaning the negative binomial link would have been preferential. However, multilevel fixed- and mixed-effects negative binomial models are quite sensitive and often have numerous convergence problems. Our two-level, negative binomial models reflected these sensitivities and would not converge. Therefore, the negative binomial link was removed and we reverted to a standard fixed-effects and mixed-effects modeling approach with a logged dependent variable. We rigorously tested all models for statistical instability and multicollinearity. Upon inspection of variance inflation factors (all values were under 2) and tolerance statistics (tolerances all above .4), no symptoms of multicollinearity were detected and any threat of multicollinearity impacting the results was ruled out. Due to minor amounts of missing data (less than 5% or less missing on all measures), we utilize listwise deletion. All analyses were estimated using Stata 15.1.

3. Results

The graph in Fig. 1 shows the average number of IPV arrests in a county based on the number of opioid pills prescribed per person from 2006 to 2012. The x-axis represents time in years and the y-axis represents the number of IPV arrests a county reports. The dark, solid line depicts the number of IPV arrests for a county with the top 25% of pills prescribed per person. In contrast, the light, dashed line represents the number of IPV arrests for a county with the bottom 25% of pills prescribed per person. The data illustrate a relatively stable trend in the number of IPV arrests for counties with the bottom 25% of pills prescribed. However, the average amount of IPV arrests for counties with the top 25% of pills almost doubles from approximately 24 to 45 arrests from 2006 to 2012. These descriptive results provide initial evidence of what appears to be a substantial difference in IPV arrests between counties that have experienced the opioid crisis more intensely compared to those that have not.

Fig. 1.

Average number of intimate partner violence arrests within counties based on the amount of opioid pills distributed within each county, 2006 – 2012.

Turning to our multivariable analyses, Table 2 displays the results from a fixed-effects model regressing IPV-related arrests on predictors. The model is significant. The number of opioid pills prescribed per person is positive and significantly related to the number of IPV-related arrests. This suggests that as the number of opioid pills prescribed within a county increases, the number of IPV arrests that are reported in subsequent years will also increase. Additionally, the number of opioid-related arrests shares a positive and statistically significant relationship with the amount of IPV-related arrests in a county, signifying that counties which record more opioid-related arrests also report higher numbers of IPV-related arrests over time.

Table 2.

Results from a fixed-effects model regressing IPV-related arrests on predictors; n = 9,008.

	bz	SE
Time variant
Opioid pills per person (at t-1)	.138	.028***
Number of opioid-related arrests	.002	.000***
Intercept	2.823	.096***
Model statistics
F	47.82***
Between cluster SD	1.652
Within cluster SD	.525

^*p ≤ .05

^**p ≤ .01

^***p ≤ .001.

Table 3 presents results from a mixed-effects model that regresses IPV-related arrests on predictors. The model is also significant. Both of the time-variant measures share significant relationships with the number of IPV arrests in a county. The number of opioid pills per person and the number of opioid-related arrests are positive and significantly linked with the number of IPV arrests a county reports. This suggests that as the number of opioid pills prescribed per person and the number of opioid-related arrests increases, respectively, so does the number of IPV arrests. All of our time-invariant variables also demonstrate statistically significant relationships with the outcome. The percentage of the population who have health insurance, are unemployed, and the overall population size are positively related to the number of IPV-related arrests. This indicates that counties with higher insurance rates, higher unemployment rates, and a larger population, respectively, tend to experience more IPV-related arrests. Finally, the percent of a county’s population that is living in poverty and that is White are negatively and significantly related to the number of IPV-related arrests in a county. This demonstrates that as the percentage of people in poverty and White people in a county increases, the number of IPV arrests is expected to decline.

Table 3.

Results from a mixed-effects model regressing IPV-related arrests on predictors; n = 9,008.

	b	SE
Time variant
Opioid pills per person (at t-1)	.215	.024***
Opioid-related arrests	.002	.000***
Time invariant
Percent poverty	−.041	.015**
Percent insured	.069	.009***
Percent unemployed	.110	.013***
Percent White	−.026	.003***
Population size / 10,000	.024	.002***
Intercept	−2.081	.766**
Model statistics
Wald χ2	805.05***
Between cluster SD	1.374
Within cluster SD	.525

^*p ≤ .05

^**p ≤ .01

^***p ≤ .001.

3.1. Ancillary analysis

As briefly mentioned in our measures section (and further discussed in our limitations section), some literature has suggested a reciprocal relationship between opioid use and IPV. Some authors have hypothesized that IPV may result in physical or mental health issues for which opioids may become a coping mechanism (Stone and Rothman 2019). To account for this, we additionally tested for a reciprocal relationship by switching our main independent variable and dependent variable. Upon regressing opioid pills per person on the number of intimate partner arrests (as we did in the main analyses, we lagged the focal independent variable at t-1) and our control variables, we find a positive and statistically significant relationship between IPV arrests and the number of opioid pills prescribed per person. However, with a standard error of .00004 and a coefficient size of only .0002, the interpretability of this effect is suspect as a standard error that small would make a range of incredibly small effects appear to be statistically significant. We would like to thank an anonymous reviewer for their suggestion to investigate this relationship.

4. Discussion

Using county-level data from several different sources, this study has investigated how IPV rates changed in accordance with the number of pills distributed in a panel of counties between 2006 and 2012. Using fixed- and mixed-effects regression techniques, our findings generally demonstrate that the opioid pill distribution trends within American counties are strongly linked to IPV. Our first hypothesis (H1) proposed that increases in the number of pills distributed within each county would relate to an increased incidence of IPV arrests over time. Results relevant to this hypothesis demonstrated that changes in opioid pill distribution trends were indeed strongly related to increases in IPV arrests over time, thus offering strong support to H1. The second hypothesis (H2) proposed that counties with higher pill distribution numbers tend to experience the highest rates of IPV arrests. Like H1, H2 also received strong support based on a longitudinal mixed-effects model. Overall, this study demonstrates that there is a clear, robust, and positive relationship between the volume of opioid pills distributed within a county and the extent of IPV arrests occurring.

While this analysis falls short of being able to answer why the number of opioid pill prescriptions is so strongly related to IPV arrests, studies mentioned in the beginning of the manuscript may help to explain this phenomenon. Previous research suggests that the relationship between substance use and IPV is complex and involves multiple factors (Gadd et al. 2019; Gilchrist et al. 2019; Radcliffe et al. 2019). Among a host of other factors, these include the pharmacological properties of the drug used, history of use, preexisting psychological vulnerabilities and personal characteristics, the user’s mood, the context of the situation in which the substance was ingested, if the user was in withdrawal or under the influence at the time, and the broader power and control dynamics in the relationship (Cafferky et al. 2018; Gadd et al. 2019; Gilchrist et al. 2019; Radcliffe et al. 2019). Cafferky et al. (2018) assert that dependence with related withdrawal and cravings has a stronger association with IPV perpetration than substance intoxication alone. Additionally, a meta-ethnography by Gilchrist et al. (2019) determines that violence seems to occur when perpetrators are coming down from their high due to a heightened emotional state, especially when the survivor fails to acquire money for drugs. While these studies help shed light on the link between substance use and IPV, it is important to note that some feature samples from the United Kingdom, and none are specifically about prescription opioids, but heroin. However, based on findings in the studies described above, it seems possible that over-prescribing led to a mass dependency which prescription pills alone could not satisfy (due to their expirations and/or increases in tolerance), which in turn led to increases in IPV due to withdrawal- and craving-related frustrations.

Both series of models also reached a somewhat similar conclusion about opioid arrests. Results from our fixed-effects models demonstrated that increases in the number of opioid-related arrests were related to the number of IPV arrests. Extending this variable into a mixed-effects model, we also concluded that counties with higher amounts of opioid-related arrests tend to experience more IPV arrests than counties with lower levels of opioid-related arrests. This finding could be a signal that arresting people with substance use issues is not the solution to mitigating levels of IPV. As just one alternative, perhaps the solution lies in rehabilitation rather than arrest. In fact, some research finds that IPV interventions conducted in tandem with alcohol treatment show promise (Tarzia et al., 2017; Wilson et al., 2014), but more research is needed specifically regarding opioids.

Our mixed-effects models also identified a number of meaningful and statistically significant between-county relationships relevant to IPV. While it comes as no surprise that counties with larger overall populations and higher unemployed shares of the population tend to experience higher numbers of IPV arrests, there were several surprising findings. First, counties which have higher levels of IPV arrests were found to have significantly lower rates of poverty, but higher rates of health insurance coverage. Though unexpected, the poverty finding may indeed make sense when thought about in the context of arrest trends. Counties which have high poverty rates have a lower tax-revenue base and therefore carry a lower ability to finance their police agencies, meaning the overall number of officers may be lower in poorer counties. Having fewer officers generally means that there will be fewer arrests overall and, in a similar manner, fewer arrests for IPV-related incidents. Importantly, one of the implications for this finding may be that operationalizing IPV using arrests may lead to an underestimation of the relationship between opioid arrests and IPV occurrence and opioid pill prescriptions and IPV occurrence, particularly among counties with more poverty¹ . Regarding the insurance finding, prior studies have concluded that insurance companies have shouldered a huge portion of the financial burden of opioid prescription purchases (Katz et al. 2013). As such, future research could explore the possibility that insurance rates moderate opioid pill distribution patterns to impact a number of potential outcomes, including IPV.

Another notable finding lies in the racial composition of counties with more IPV-related arrests. Our results demonstrate that counties with a larger White population are associated with higher numbers of IPV-related arrests. However, numerous studies show women of color, and specifically Black women, experience IPV at a higher rate than their White counterparts (Black et al. 2011; Petrosky et al. 2017). Therefore, we hypothesize that this finding is a product of the fact that we are using official arrest data and points to the barriers women of color face in reporting IPV. For example, Black women may be less likely to report abuse due to numerous barriers: discrimination, considering Black men’s vulnerability to police brutality, and negative stereotypes, amongst other reasons (Decker et al. 2019; Nash 2005). Additionally, when a report is made, an arrest is more likely to happen if the survivor is White (McCormack and Hirschel 2018). Finally, our dataset involves aggregate, county-level data which precludes us from identifying the race of the victim and arrestee. Therefore, it is important to note that just because communities with high White populations tend to experience more IPV-related arrests, this does not mean that White people experience more IPV compared to their non-White counterparts. Each of these factors may be influencing our results and must be taken into consideration when analyzing this finding.

The findings from this study illuminate societal problems surrounding the opioid crisis that policymakers should be cognizant of while also pointing to shortcomings in the current response to IPV. Our results demonstrate that not only do we need to focus on curtailing opioid dependency and providing relief to those with an opioid use disorder, but policymakers in counties affected by the opioid crisis should allocate resources to efforts aimed at reducing IPV. Based on these results, a shortcoming in the United States’ current response to IPV may place the policy focus only on the IPV situation itself while failing to address the larger context in which the situation occurred. In addressing this shortcoming, one helpful framework to guide policymakers addressing the opioid crisis and IPV is the social-ecological model of violence prevention. This model considers the interplay between the individual, relationship, community, and society to understand and prevent violence (CDC 2020). According to this framework, policymakers cannot adequately address IPV without addressing the societal context in which it exists and thrives. Thus, legislators must account for the opioid crisis when addressing IPV.

Since this study is largely exploratory, future research should consider the role other covariates have within the relationship between opioids and IPV. Neighborhood research, for example, has shown that gender income inequality may be a predictor of IPV (Jackson 2016). Notably, missing from our analyses due to data limitations is gender. Future research should take it into account if data allow. Likewise, the small amount of macro-level IPV research suggests that laws and beliefs may also play an important role in IPV perpetration and victimization (Kovacs 2018). Additionally, being able to analyze this relationship within a larger time frame (i.e., past 2012) may help us to further understand the relationship between substance use and IPV over time.

Though this is among the first studies to assess how the opioid crisis has influenced levels of IPV, it is not without limitations. First, research documents a potential reciprocal relationship between substance use and IPV (Rothman et al., 2018). That is, while substance use has been thought to facilitate IPV, IPV could also influence substance use, particularly when viewed as a coping mechanism by survivors (Rothman et al., 2018). Though this analysis attempts to account for causal ordering by using a lagged independent variable capturing opioid pill distribution patterns, this reciprocity may nonetheless be pervasive across our analyses. Therefore, an important future direction of research is to investigate if IPV and opioids are reciprocally related. An answer to this question would be substantively impactful to policymakers, practitioners, and scientists across a broad range of fields. A longitudinal structural approach using directed acyclic graphs (DAGs; see Elwert 2013) would be ideal for such an investigation.

Additionally, operationalizing IPV using NIBRS arrest counts almost certainly underestimates the total number of instances of physical violence that is occurring between romantic partners. There are at least two reasons for this. First, there are plenty of situations in which survivors would not report IPV to law enforcement, including – but not limited to – physically being unable to access a phone, feelings of shame, being fearful of putting oneself in danger, attempting to avoid the attention of law enforcement, and so forth (Fleury et al. 1998; Wolf et al. 2003). Second, the NIBRS does not categorize an arrest as explicitly intimate partner violence-related. As a result, the criteria used to define an arrest as IPV-related are strict and imperfect, thus likely limiting the number of observed IPV-related arrests.

A final limitation concerns the time frame of this study. While the DEA’s data is currently the only source of information which provides the prevalence of opioid pill distribution patterns in the United States, the data is only available between 2006 and 2012. While this certainly captures an important time period spanning the rise of the opioid crisis, the unfortunate reality is that the crisis was far from being resolved at the end of our analytical time frame. Future data collection efforts similar to the DEA’s should be pursued that capture a more contemporaneous and modern time frame.

Despite having some limitations, this research is the first study of which we are aware to explicitly tie opioid pill distribution patterns to IPV arrests at the county-level. Not only do counties which have higher opioid pill distribution numbers tend to experience the highest amount of IPV arrests, fluctuations in pill distribution volume are also longitudinally linked to fluctuations in IPV arrests in a positive direction. As such, it would seem that opioid pill distribution trends carry implications for IPV-related incidents both between-counties as well as within-counties over time. Not only does this study contribute to research that has demonstrated an intricate and well-established link between IPV and substance use at the individual-level (e.g., Moore et al., 2011; Testa et al., 2003; Smith et al. 2012; Stone and Rothman 2019; Stuart et al. 2013), it speaks to a set of potentially impactful policies that could be implemented at the macro level (see Niolon et al. 2017 for examples): To sufficiently address IPV within a defined jurisdiction such as a county, local officials and policymakers may first have to find a way to mitigate the root causes of the opioid crisis occurring in the area.