Intimate Partner Violence-Related Fatalities in the United States

Intimate partner violence (IPV)—here defined as psychological aggression, physical or sexual violence, stalking, and/or coercive control between sexual, dating, marital, or romantic partners—has numerous documented impacts, including the homicide or suicide of the IPV victim, perpetrator, or other individuals (i.e., corollary victims). In the United States (U.S.), nearly half of all female and one-tenth of male homicide victims are killed by intimate partners (Fridel & Fox, 2019; Jack et al., 2018). Past research indicates that in two-thirds to three-quarters of heterosexual intimate partner homicides (IPHs), the male had been previously abusive toward the female, regardless of which person was killed (Campbell et al., 2003; Harden et al., 2019). Data on transgender and non-binary homicide decedents indicate that between 2013 and 2020, 21% of the murders were committed by an intimate partner (Human Rights Foundation, 2021). While an emerging topic, recent studies estimate that for 6% of youth and 7% of adult suicides, IPV was a contributing factor, with the people who died by suicide being IPV perpetrators, IPV victims, or corollary victims (Graham et al., 2022; Kafka et al., 2021, 2022a). Critically, these findings suggest that there may be missed opportunities to intervene in IPV prior to escalation to fatality. It is plausible that existing intervention models are ineffective, or in some cases, there may be a complete lack of intervention prior to lethal violence. As such, it is imperative that research further investigate circumstances associated with increased risk or protection for both IPV-related suicides and homicides to inform the development of effective prevention strategies.

Researchers interested in examining IPV-related fatalities in the U.S. can use two publicly available national datasets: the Federal Bureau of Investigation’s (FBI’s) Uniform Crime Reporting Supplementary Homicide Report (UCR-SHR) and the Centers for Disease Control and Prevention’s (CDC’s) National Violent Death Reporting System (NVDRS). This paper will begin with an overview of each national dataset, followed by a discussion of IPV measurement and data missingness. Authors then critically discuss research implications for using these data when examining deaths among marginalized groups, the types of statistical analyses best suited for these data, and novel methodologies for engaging in deeper analyses. We conclude by discussing potential pathways for future research.

The U.S. National Datasets

NVDRS and UCR-SHR rely on investigative data collected by coroners/medical examiners (CME), law enforcement officers (LE), and death certificates. This information is compiled for UCR-SHR at the jurisdictional level and for NVDRS at the state level. Both datasets can be easily merged with other federal datasets (e.g., American Community Survey) to examine spatial and place-based trends.

The Uniform Crime Reporting-Supplementary Homicide Reports

The Summary Reporting System (SRS) is a voluntary data collection system managed by the FBI whereby LE jurisdictions submit information on crime counts along with limited demographic and circumstantial characteristics (FBI, 2013). The FBI began collecting national data on urban crime in 1930 and began collecting supplementary data on homicides in 1962 (FBI, 2013). Continuous, national data since 1976 are publicly available (Fox & Fridel, 2016). Today, over 18,000 departments report crime data to the FBI through the SRS or a secondary system designed in the late 1980s called the National Incident-Based Reporting System (NIBRS; FBI, 2013, 2022). Participating agencies are trained by the FBI and can participate in a data quality audit every three years (FBI, 2013). In 2020, 85.24% of agencies participating in the SRS/NIBRS also participated in the UCR-SHR (FBI, 2022). The UCR-SHR is an incident-based dataset, where each observation is a single incident (one or more fatalities that occur at a single location at the same time) and contains age, race, ethnicity, sex, circumstance, and weapon used for up to 11 decedents and 11 homicide suspects, and the relationship between each suspect and the first decedent listed.

In 2018, the FBI announced that it would transition all data collection from the SRS to the NIBRS by 2021 (FBI, 2018). As of December 2020, a majority of states (78.84%) were still submitting through the SRS for some or all jurisdictions (FBI, 2022). Heretofore, the FBI has combined homicide data from NIBRS and the SRS before releasing the UCR-SHR, creating a seamless dataset for users. Moving forward, individual researchers will be able to merge NIBRS data with the historical UCR-SHR data using department codes (called ORI codes) to continue trend analyses, although some unit missingness may occur due to differences in agency compliance with the NIBRS transition (K. Martin, personal communication, December 6, 2022).

The National Violent Death Reporting System

Initiated in 2002 with the participation of six states, NVDRS is an enhanced surveillance system dedicated to collecting information about violent deaths by triangulating data from LE, CME, and death certificates. NVDRS collects data on homicides, suicides, legal intervention deaths (fatal injuries inflicted by active-duty LE), terrorism-related deaths, unintentional firearm injury deaths, and deaths that might have been due to violence but the intent of which was unclear. The number of participating states has grown over time and as of 2018, includes all 50 states, the District of Columbia, and Puerto Rico; although, data have not yet been released for all participating entities at the time of writing.

NVDRS compiles extensive information about each fatality, including more than 600 unique variables, substantially more than UCR-SHR. NVDRS abstractors are guided by a detailed coding manual that provides guidance for recording known risk factors and circumstances for each death (e.g., IPV; CDC, 2021). In addition to the required core data fields, NVDRS offers states and districts an optional IPV module that can be used to supplement reporting (CDC, 2021). Only eight states and the District of Columbia have continuously used the optional IPV module (Rebecca Wilson, CDC, personal communication, August 20, 2019).

Along with quantitative data fields, NVDRS abstractors also compile case narratives summarizing the LE and CME reports. The level of detail in the case narratives varies, though they each are typically a paragraph or two in length (Mezuk et al., 2021). According to the coding manual, NVDRS case narratives briefly summarize the fatal incident, describe precipitating circumstances for the death, and may provide additional details that are not captured elsewhere in NVDRS quantitative data fields (CDC, 2021, p. 25).

NVDRS provides aggregate public data through the online WISQARS Module and a streamlined process for requesting the Restricted Access Data (RAD) file for detailed research aims (CDC, 2022a, b). The RAD file is organized at the decedent-level (one observation per decedent), although, multi-victim incidents share a common identification number, allowing researchers to link deaths from a single incident (e.g., homicide-suicides, mass shootings).

Identifying and Operationalizing IPV

UCR-SHR and NVDRS allow for the identification of IPV-related cases through variables that capture information about (a) the relationship between homicide decedents and suspects, and (b) circumstances related to the fatality (Table 1). The UCR-SHR has two variables: relationship and circumstance., and the NVDRS contains multiple variables (intimate partner violence, intimate partner problems, jealousy, and stalking) and narrative information that can flag an IPV-related fatality (CDC, 2021; Fox & Fridel 2016).

Table 1 Intimate partner and intimate partner violence definitions in the UCR-SHR and NVDRS
Full size table

Data Quality

Both datasets have been criticized for limitations in identifying IPV-related fatalities. Notably, the datasets rely on investigative data from CME/LE, which may or may not reveal IPV history. IPV is an under-reported crime, with about 56.52-59.16% of IPV survivors ever having contacted LE about IPV (Ogbonnaya et al., 2021). Furthermore, one study found that IPV was typically (>70%) only documented in NVDRS suicide case narratives if an IPV incident had occurred within the past two weeks, suggesting that more distal experiences of IPV were probably under-identified (Kafka et al., 2022a). Current reporting protocols may also miss intimacy in dual relationships (e.g., employer dating an employee), obscuring the total number of IPHs (Campbell et al., 2007).

IPV in UCR-SHR

UCR-SHR has been criticized for inaccurate labeling of decedent-suspect relationships, with one study showing that over a third of decedent-suspect relationship labels were incorrectly identified (Pizarro & Zeoli, 2013). Relationship is only specified for the first decedent, which can lead to incomplete and/or inaccurate data in multi-victim homicides. For example, in an IPH with a corollary child victim, the child could be listed as the first decedent, but the relationship may still be (incorrectly) labeled as an intimate partner. Even if entered correctly with the intimate partner as the primary decedent, one would not know the relationship of the child to the suspect (e.g., biological child, child of dating partner). UCR-SHR also lacks relationship categories for ex-non-married partners, who perpetrate approximately 7% of IPH and 5% of IPV-related homicides of corollary victims (Graham et al., 2021). Further, the relationship category of “homosexual relationship” is no longer culturally appropriate and does not differentiate between married and dating same-sex partners. The lover’s triangle variable is also outdated, and it’s phrasing suggests a shared responsibility for violence rather than suspect accountability and does not necessarily indicate IPV. Unlike NVDRS, UCR-SHR does not preserve the qualitative data related to circumstances in its public dataset, prohibiting researchers from identifying additional IPV-related cases themselves. While UCR-SHR has some utility in describing IPH trends, it is limited in its capacity to describe IPV-related fatalities.

The updated NIBRS, which will replace UCR-SHR, responds to some of these concerns, potentially increasing the utility of FBI data to describe IPV-related fatalities. “Lover’s triangle” has been removed as a level in the circumstance variable, and instead “domestic violence” is listed (FBI, 2021). Domestic violence is defined as physical violence, the use/threat of a weapon, coercion, intimidation, and property crime between current and former spouses or romantic/intimate partners; co-parents, parents, or guardians; or people cohabitating as spouses, parents, or guardians (FBI, 2021, p. 123). The relationship variable has additionally been updated to include ex-non-married partners and has removed “homosexual relationship” as a category (FBI, 2021). Earlier versions of NIBRS used UCR-SHR circumstance and relationship levels, however, so researchers will be unable to examine past trends (NACJD, 2021).

IPV in NVDRS

NVDRS records potential IPV through several circumstance variables: IPV, intimate partner problem (IPP), jealousy, and stalking. In NVDRS, IPV and jealousy are only recorded for homicide cases. Case narratives may also provide additional detail that can help substantiate whether a fatality was connected to IPV. In a recent study that examined IPV-related fatalities among youth, almost half (44.7%%) of all IPV-related fatalities (1.3% of IPV-related homicides, 93.6% of IPV-related suicides) were identified through study team coders reviewing case narratives rather than relying on the existing IPV variable (Graham et al., 2022). Another study focused on child decedents in IPV-related homicide found that nearly half (48.6%) had been missed by the IPV variable and were identified only through detailed case narrative review (Adhia et al., Fitzmaurice, & Hemenway, 2019).

While NVDRS records IPV circumstances for homicides, no such variable exists for IPV-related suicides. For NVDRS suicide cases, the IPP variable indicates that some problem, such as “divorce, break-up, argument, jealousy, conflict, or [romantic] discord …” contributed to the suicide (CDC, 2021, pp. 91-92). This variable is operationalized to include potential cases of IPV, but it may also capture non-violent relationship discord that occurred prior to a suicide. Additionally, the “stalking” variable in NVDRS records if stalking by an intimate partner, stranger, co-worker, or other acquaintance contributed to the death (CDC, 2021, pp. 113-114). As a result, identifying IPV-related suicides is challenging; available NVDRS proxy variables are broad and contain false positives. For homicide-suicides, however, it is possible to link related deaths by an incident identifier allowing researchers to find suicides connected to known IPV-related homicides (McNally et al., 2016).

Although the coding manual clearly delineates between suicide and homicide circumstance codes (i.e., IPV v. IPP), data suggests that a small percentage of NVDRS abstractors cross utilize these codes, with 0.38% of suicides having a positive indication in the IPV variable and 0.69% of homicides with a positive indication of IPP. Additionally, NVDRS data abstractors may not appropriately recognize cases of emotional abuse or coercive control as IPV and fail to select “present” for the IPV variable. There remains widespread variability in how IPV is conceptualized and formally defined in institutional, occupational, and political settings, with many entities focusing primarily on physical violence, while ignoring other forms of abuse (Breiding et al., 2015; Candela, 2016; Gill et al., 2021). Thus, it remains unclear how consistently IPV is documented when physical violence was not explicitly reported.

Rapid Review: IPV-Related Fatalities in Research

Measurement challenges in NVDRS and UCR-SHR are further complicated by inconsistent conceptualization and identification of IPV-related fatalities across research studies. To gauge how IPV-related fatalities are identified in the literature, we conducted a rapid review of studies published between January 1, 2019 and June 30, 2022 and identified 35 articles that used NVDRS or UCR-SHR to examine IPV-related homicide or suicide. Most manuscripts focused on documenting the role of IPV in homicide, while only a few manuscripts examined isolated suicide events beyond homicide-suicide. The majority of articles examined IPV-related fatalities alone (60%), while a subset of articles compared IPV-related fatalities to other violent deaths (40%; Table 2; see Supplementary materials for more details).

Table 2 Operationalization of intimate partner and identification of intimate partner violence (N=35)
Full size table

Researchers used several different strategies to document IPV-related fatalities. While most papers on homicide included a list of specific relationship criteria (e.g., spouse, non-married partner, 64.3%) to determine IPH, other articles were less detailed, either simply saying “intimate partner” (homicide: 28.6%, homicide-suicide: 66.7%, suicide: 100%) or lacking a clear operationalization (homicide: 7.1%). Most of the articles that we identified (homicide: 67.9%, homicide-suicides: 50%) relied on relationship as the primary identifier of an IPV-related fatality. The reliance on relationship type to identify IPV-related homicide limits analyses of the lethal cost of IPV by excluding incidents in which only corollary victims were killed (e.g., an incident in which a man only killed his ex-spouse’s new dating partner). Past research using the NVDRS has indicated that 28.5% of IPV-related homicides involve multiple decedents, indicating the importance of broadening one’s conceptualization of IPV-related fatalities when possible (Smith et al., 2014).

Several NVDRS articles used additional methods of identification including circumstance variables (IPV, jealousy, stalking) and by reviewing case narratives. Case narrative review is particularly critical for suicide studies, as indicator variables like IPV, decedent-suspect relationship, and jealousy are not available in suicide data. While the IPP variable may be used to reduce the cases one needs to review, the variable is not necessarily indicative of IPV, but could rather represent non-violent relationship discord. In two studies, there were no details about how IPV was identified or operationalized. Clear and consistent operationalization in the future will allow for more appropriate comparison across studies to support evidence-based practice and policy recommendations.

Missingness in Existing Datasets

Because UCR-SHR and NVDRS rely on existing administrative data and are voluntary programs, they can have problems with substantial missingness for key data points. Missingness can occur because an entire agency, county, or state declines participation in the program, referred to as unit missingness, or because individual variables (e.g., suspect age) are unknown, called item missingness.

Unit Missingness

Shifts in participation have led to unit missingness in both datasets. For UCR-SHR, jurisdictions opt-in to participate, so local data may be incomplete or entirely absent for certain years (Kaplan, 2021b). Florida has declined to participate in UCR-SHR since 1996, although the publicly available multiply imputed dataset compiled by Fox and Fridel (2016) has merged Florida’s state data with UCR-SHR to create a full national record. State-level unit missingness ranges from 0% (CT, DE, DC, HI, ME, NJ, ND, PR, RI, VT) to 54.98% (MS).

NVDRS programs have been funded using a staggered cohort model affecting unit missingness. NVDRS data were first collected in 2003 by six states. By 2005, 16 states were participating, 32 states by 2016, and all 50 states and the District of Columbia and Puerto Rico by 2018 (see CDC, 2021 for additional details). Staggered initiation affects researchers’ ability to track trends over time. Most NVDRS-funded states report comprehensive data for all agencies within that states’ jurisdiction, although exceptions can be made if a state chooses to report only on a subset of counties as long as those counties either (a) contribute 80% or more of violent deaths in the state or (b) account for more than 1,800 violent deaths, including ≥25% of homicides and ≤25% of suicides in the state (CDC, 2014).

Unit missingness makes it challenging to estimate the total number of IPV-related fatalities in either dataset. Studies comparing UCR-SHR to CDC’s original homicide tracking system (the National Vital Statistics System, based on death certificates) suggest that estimates between the two systems were consistent (Smith & Cooper, 2013). More recent comparisons of UCR-SHR to NVDRS, however, suggest that UCR-SHR underestimates homicides compared to NVDRS (Kaplan, 2021b).

Item Missingness

Item missingness may be an issue because the information was not disclosed during the death investigation process, CME/ LE omitted those details from their final reports, the details were not updated in the data system even though additional or more accurate information was discovered in ongoing investigations, or information was not comprehensively abstracted into UCR-SHR or NVDRS. Item missingness may occur in homicide data specifically because cases have not been cleared by local LE (i.e., an arrest has been made or the case has been closed due to exceptional circumstances like perpetrator suicide). The national clearance rate for homicide is 61.6%, ranging regionally as low as 46.2% (east north central Midwest) and as high as 68.6% (Mountain; FBI, 2017). The low homicide clearance rate affects item missingness particularly in variables that describe the homicide suspect.

Items in both UCR-SHR and NVDRS may also be system missing, meaning that no value was entered into the data system, or that it was directly labeled as “unknown.” Uniquely, NVDRS also captures circumstantial item missingness with a single global variable, recording whether any circumstantial details about the death were available from CME/LE reports. For any substantive analyses of precursors for violent deaths, CDC recommends that investigators only include cases that include known circumstances information (CDC, 2019, p. 9). Thus, a considerable percentage of all NVDRS cases must be dropped prior to analysis of circumstances.

Specific NVDRS circumstance variables (e.g., mental health problems) are coded in a binary manner as “present” or “not present/not available/unknown” (CDC, 2021, p. 77-150; NIST, 2021). In effect, this latter category lumps together “no,” “not available,” and “unknown,” making it impossible for researchers to disentangle system missingness from unknowns from true negatives. In both datasets, any type of item missingness functionally creates complications for data analysis.

Current Item Missingness in UCR-SHR and NVDRS

The missingness in the datasets limits researchers’ abilities to examine complete and accurate information about IPV-related fatalities. For the present commentary, we analyzed data from UCR-SHR 1976-2018 (Kaplan, 2021a) and NVDRS 2003-2018 (CDC, 2020) to describe the percent missing and percent unknown among the IPV variables discussed above (Table 3).

Table 3 Percent item missingness and unknown in the UCR-SHR and the NVDRS
Full size table

Information missing on decedent demographics is relatively low in both UCR-SHR and NVDRS, which would be expected if decedents could be identified by their remains. Homicide suspect demographics are functionally missing in about 30-60% of cases across both datasets. The high-level of functional missingness (i.e., both unknown and system missing) in the suspect variables is likely due to the low national homicide clearance rate. Circumstances are unknown for 22.63% of homicides in UCR-SHR, 25.47% of homicides in NVDRS, and 11.76% of suicides in NVDRS. Missingness in the circumstance variables limits researchers’ abilities to identify IPV-related fatalities beyond decedent-suspect relationship.

Examining Marginalized Populations

UCR-SHR and NVDRS allow researchers to examine IPV-related fatalities among marginalized populations, although measurement challenges pose serious limitations. UCR-SHR and NVDRS both rely on investigative reports to identify decedent and suspect identities, however, marginalized populations are often misidentified or overlooked in official data sources.

UCR-SHR and NVDRS both use U.S. Census definitions of race and ethnicity (CDC, 2021; FBI, 2013). Unlike UCR-SHR, NVDRS treats race as a multiple-choice variable, allowing for individuals who identify as multiple races to be properly identified, thus decreasing monoracial bias in the data (Townsend et al., 2009). Multiracial individuals experience the highest rates of IPV of any U.S. racial/ethnic group (women: 56.6%, men: 42.3%), suggesting a need for further understanding IPV-related fatalities among this population (Smith et al., 2017). Further, racial and ethnic identities may be missed or inaccurately captured by CME/LE, obscuring real trends. For example, Native American/Alaskan Native decedents are thought to frequently be misidentified in state data, resulting in an underestimate of the burden of fatal violence experienced by this community (Echo-Hawk et al., 2019). Like multiracial individuals, Indigenous Americans have higher rates of IPV than other groups (women: 47.5%, men: 40.5%), and experience a disproportionate burden of homicide and suicide compared to other racial/ethnic groups; suggesting that IPV-related fatality may be of critical importance to study among Indigenous Americans (Heron, 2019; Smith et al., 2017). Recent investigations into COVID-19 related deaths indicate that people of Hispanic/Latino/a/x ethnicity were misidentified in medical records and death certificates leading to data inaccuracies, suggesting other official sources of data may have similar errors (Arana et al., 2022). UCR-SHR particularly seems to be plagued by missingness with the ethnicity variable, making the dataset poorly suited for studies focused on the experience of Hispanic and Latino/a/x Americans.

Transgender, non-binary, and gender diverse decedents of fatal violence face similar erasure in these datasets. UCR-SHR has decedent and suspect gender variables, however, it uses a binary definition of gender, and thus is not able to describe trends in homicides against gender diverse communities (Fox & Fridel, 2016). NVDRS records decedent and suspect sex, and since 2013, has also identified gender diverse decedents through a transgender indicator variable. Unfortunately, it is hard to know how complete the data are for transgender decedents. For 2013-2018, 99.66% for all fatalities has the transgender variable recorded as “no, not available, or unknown.” About 30% of transgender individuals report lifetime IPV, one-fifth of all transgender homicide victims are killed by an intimate partner, and exploratory research suggests that transgender individuals face unique risks for IPH, such as stigma, suggesting a need for better measurement of gender identity in violent death data (Human Rights Foundation, 2021; Sherman et al., 2021; Valentine et al., 2017).

Information on sexual orientation is also often underreported or mischaracterized in these data, which limits the ability to identify disparities in IPV-related violent death rates for LGBQ+ populations (Haas et al., 2019). UCR-SHR has a relationship category to indicate a same-sex relationship, although it is unclear if abstractors endorse the variable if the partners were married. NVDRS added a sexual orientation indicator in 2013, whose value is missing in 4.52% of decedent observations and recorded as “no, not available, or unknown” in 86.73% of observations. LGBTQ+ individuals are at an elevated risk for suicide and experience higher rates of lifetime IPV, making it critical to examine IPV-related fatalities among this group (Chen et al., 2020; Johns et al., 2020).

Analytical Challenges

Mortality data are important for descriptive and associative studies; however, the data do not easily lend themselves to researching causal associations. These data are usually collected at a single time point, post-mortem, making it difficult to examine the temporal ordering of events prior to the death. Because mortality data only capture cases where people died, one cannot estimate the relative risk of death due to antecedent risk or protective factors because there is usually no comparison group of living people who survived in the data. This limitation represents an issue of selection bias, known as collider stratification bias which will lead observed correlations among circumstances in mortality data to present differently than they would in a general population of living individuals (Cole et al., 2010; Munafò et al., 2018). Thus, while researchers may want to identify which factors increase or decrease the risk for IPV-related fatalities, complimentary data sources would likely be needed to conduct causal research (e.g., by sampling controls).

UCR-SHR and NVDRS data do lend themselves, however, to descriptive research questions examining patterns, trends, and rates. For example, are there disparities in the rates of IPV-related violent deaths across subpopulations? Which circumstantial, relational, and individual characteristics coincide with IPV-related fatalities? How are IPV-related fatalities different from other types of violent deaths? Are there potential missed opportunities to intervene and prevent IPV-related fatalities? These are just a few examples of research questions that could be addressed using these data sources. UCR-SHR and NVDRS are nationally representative, only falling slightly short of a census of all violent deaths in the U.S. As a result, they provide an extensive breadth of information, offering researchers robust external validity compared to other datasets drawn from smaller geographic regions.

When analyzing data from UCR-SHR and NVDRS, researchers must also consider variability in data quality across space and time. For example, in NVDRS, some states use coroner systems while others rely on medical examiners to complete death investigations. Coroners are often elected and are not required to have medical training, while medical examiners are appointed and often require more training (Hanzlick, 2006; Ruiz et al., 2018). This has led to demonstrable variation in data quality based on the type of state system, as well as whether the state system is centralized or not (Posey & Neuilly, 2017; Warner et al., 2013). There are also bound to be local patterns in how deaths are investigated based on local agency characteristics (NIST, 2021). Finally, differences in abstractor training (for UCR-SHR at the jurisdiction-level and NVDRS at the state-level) could further lead to geographic and time-related variations in data quality. Researchers must take care to address clustering based on space and time in their analytical models, for example using fixed or random effects. While UCR-SHR has maintained the same variables since 1972, the CDC has continued to add new variables to NVDRS. For example, the stalking variable was added to NVDRS in 2013, precluding any investigations of IPV stalking-related fatalities for earlier years. Compounded with concerns about data quality, these factors can make it challenging for researchers to choose which cohort of states to analyze over what time periods for their analyses.

Novel Methodological Approaches

As premised in our Rapid Review of the literature, researchers are increasingly using UCR-SHR and NVDRS to examine the role of IPV in fatal outcomes. Methodologically, there are numerous approaches that can be taken to engage in deeper and more complete analyses of IPV-related fatalities using either dataset. First, deterministic or probabilistic linkages of fatality data at the individual-level to other data sources (e.g., health records) can provide more comprehensive information about decedent histories and lived experiences (Caves Sivaraman et al., 2022). A data use agreement can sometimes be arranged with states or local entities that can then provide identifiable information about decedents (e.g., decedent name), although to our knowledge identifiable data is unavailable directly from NVDRS or UCR-SHR. When using identifiable data fields, data security measures must be used to protect the identities of decedents and their loved ones. Taking appropriate data security measures and engaging in fully transparent negotiations with local data providers is an important step prior to data linkage.

While NVDRS and UCR-SHR data primarily catalogue information about people who died, there are some innovative approaches to design a case-control study using these data. For example, multi-death incidents (cases) could be compared to a counterfactual where only one person died, as plausibly other lives were spared (controls) (Logan et al., 2019; Lyons et al., 2020, 2021). Controls could also be sampled from complimentary data sources outside of NVDRS or UCR-SHR, allowing researchers to examine causal questions about factors that increase or decrease risk for IPV-related fatalities (Hernandez-Meier et al., 2019). Similar to data linkage work, data permissions and safeguards must be carefully established if sampling controls from external data sources, and care must be taken to select appropriate living controls so that it is possible to generalize research findings to the stated target population (Westreich, 2020). For example, sampling controls from IPV survivors based on criminal justice data would limit inference about IPV-related death risk to only survivors who have had criminal justice system contact.

NVDRS and UCR-SHR data can also be used to examine how certain risk factors cluster or coincide, suggesting overlapping or independent pathways to IPV-related fatalities. For example, latent class analyses (LCA) or other clustering techniques can reveal whether there are notable subgroups of decedents and how IPV history may be documented across subgroups (Wertz et al., 2020). LCA is a person-centered approach; rather than controlling for different variables, LCA examines how variables co-occur to form unique risk profiles (Scotto Rosato & Baer, 2012). Machine learning has also emerged as an important approach for data-driven modeling in large datasets like UCR-SHR or NVDRS. Machine learning studies usually examine correlation and use that information to predict outcomes or classify cases in a dataset (Westreich, 2020). While manual review is necessary in many cases to summarize details from the text, natural language processing can help extract meaningful data if the textual information is clearly operationalized and linguistically consistent (Boggs & Kafka, 2022). Recently, one study used NVDRS circumstance variables and case narratives to build a supervised machine learning model to identify IPV-related suicides (Kafka et al., 2022b), allowing for a comprehensive assessment of all types of IPV-related fatalities. This supervised machine learning tool is now publicly available to use and can help facilitate future research on the role of IPV in suicide. While machine learning is not used for causal research, it can be helpful for cataloguing novel variables in NVDRS and/or revealing strong relationships in the data. New approaches to summarize topics in death narrative text in NVDRS using machine learning are also under development and could hold promise for future research on IPV-related fatalities (Arseniev-Koehler et al., 2022).

The NVDRS case narratives are an untapped data source to examine additional details about IPV-related fatalities, including novel risk factors or circumstances, that are not available in the UCR-SHR dataset. While case narratives can provide unique details that are not captured elsewhere, the text can be brief, and details about IPV may be omitted or described inconsistently, limiting the depth of possible qualitative analyses. For example, Lyons and colleagues (2021) found that 5.4% of IPH cases in NVDRS described the decedent having sought a protective order prior to the death, but none of the case narratives explicitly mentioned whether firearm removal was included as a provision of the protective order. In another study, Kafka and her team (2022a) examined whether suicide occurred more often among IPV perpetrators or IPV victims, but the authors found that 9% of IPV-related suicide cases did not clearly describe the decedent’s role in the IPV. Case narratives can offer rich insights into the specific context for violent deaths, although limitations in detail and context must be considered.

Finally, both datasets can be used at the ecological level to examine the impact of policies and community-level characteristics on IPV-related death rates. For example, Siavaraman et al., (2019) used NVDRS data to examine whether a higher number of restrictive state gun laws was associated with lower state-level IPH rates. Other research has linked UCR-SHR data to U.S. Census data to estimate the relationship between community-level resource deprivation and gender equity characteristics to IPH among urban and rural populations (AbiNader, 2020; Reckdenwald & Parker 2012). While ecological studies cannot establish causality, they provide critical insight into the potential role of existing policy tools (e.g., firearm and social welfare policies) to shift the prevalence of IPV-related fatalities at the population-level.

Discussion

This paper reviewed how NVDRS and UCR-SHR capture information about fatal violence, enabling researchers to examine circumstances for IPV-related fatalities, evaluate the impact of policy across states, and monitor trends and disparities in fatal outcomes across subpopulations. These research activities can inform key recommendations for potential interventions to prevent IPV-related fatalities. Still, several challenges to thoroughly examining violent deaths exist. Future researchers must consider which dataset and analytic approaches best fit their research aims and their conceptualization of IPV-related fatalities, while keeping data limitations in mind.

In our rapid review, we found that many studies exclude corollary victims and suicide decedents from their conceptualization of IPV-related fatalities. However, according to NVDRS data, more than one in four IPV-related homicides involve multiple victims (Smith et al., 2014) and many suicides may be connected to IPV, even if the suicide itself was not connected to a homicide (Kafka et al., 2022a). In fact, after a detailed review of death narratives in NVDRS, Kafka et al. (2021), and Graham et al. (2022), found that single suicides (i.e., isolated suicide deaths, not homicide-suicides) were the most frequently reported type of IPV-related fatal violence in the NVDRS narratives. Accordingly, broadening the operationalization of IPV-related fatalities to include corollary victims and suicide decedents is an important direction for future research. While this expanded operationalization may not be possible in UCR-SHR due to data limitations, researchers using NVDRS have the opportunity to use multiple strategies to identify IPV-related cases.

UCR-SHR and NVDRS collect data from secondary sources to record information about IPV and violent deaths. To improve the quality of information about IPV, identity, and lived experiences of decedents, more training and resources may be needed for CME and LE to help them thoroughly document this information during the initial death investigation (Blosnich et al., 2022). For example, CME or LE agencies could introduce systematic probes to ask about IPV during death investigation interviews if the death occurred in connection to a known domestic conflict. Similarly, it would be useful for LE to routinely disclose information about prior confirmed or suspected IPV incidents when they report data to the FBI and CDC.

Conducting supplemental inquiries about IPV, however, could be burdensome for local CME and LE agencies (NIST, 2021). In the meantime, training FBI and CDC abstractors to clearly record information about IPV if it was mentioned in existing death investigation records is a critical direction moving forwards. Providing a clear and standardized definition for IPV is key, and it is also important to consider emphasizing warning signs and indicators for coercive control, an underrecognized and underreported form of IPV (Candela, 2016). Relatedly, past research has indicated that individuals of marginalized identities are often misidentified, or IPV in their relationships is missed. Thus, both dataset abstractors and CME/LE agencies may benefit from standardized protocols to assess racial, ethnic, gender, and sexual identities, as well as routinized training and adequate resources to support this process (Haas et al., 2015).

As discussed previously, there are a number of methodological techniques that can support novel research, including linking data to other sources, sampling controls, machine learning and natural language processing, latent class analysis, and ecological models. The use of controls and ecological models may be of particular import to identifying effective interventions at the micro- and macro-levels, respectively. Case-control studies would allow researchers to identify if risk factors of IPV and those of IPV-related fatalities differ and to identify protective factors against fatality. Homicide and suicide research has historically operated from a deprivation-perspective focused on risk, rather than a strength-based perspective focused on protective factors that we can cultivate to increase safety. Linking fatality data to community-level data can assist comparisons of geographic areas that have certain policies (e.g., red flag laws), characteristics (e.g., high gun ownership), or programs (e.g., domestic violence high risk teams) to those that do not, evaluating the efficacy of policies and programs and suggesting potential avenues for community-based interventions. Notably, future studies must investigate antecedents to and interventions in IPV-related suicides, about which we know very little.

The current NVDRS expansion to collect data from all 50 states and UCR-SHR transition to NIBRS offers an opportunity for researchers to advocate for potential improvements. The inclusion of ex-non-married partners in the updated NIBRS suggests a responsiveness to long-held critiques of UCR-SHR and that FBI data reporting mechanisms may be amenable to change. As described above, there are several barriers to identifying IPV-related fatalities in both datasets, and researchers should advocate for better qualitative data that can be coded by the researcher (i.e., case narratives), better IPV-related training of abstractors and CME and LE agencies, and improved flags for IPV-related cases, particularly in suicide data.

Conclusion

In conclusion, future researchers should (1) continue to use these datasets to provide critical information about violent death outcomes; (2) clearly consider limitations of these data sources when conducting analyses and discussing results; (3) anchor research questions on actionable recommendations that can inform real-world practice and policy solutions; and (4) advocate for continued funding and rigor in data collection activities for the CDC and FBI.