Information Needs and Data Harmonization—Two Sides of the Same Coin?

Community firearm violence prevention network

The CFVP Network (https://endfirearmviolence.org/) supports research that develops and tests behavioral and medical firearm-related interventions. The network consists of four groups: (1) data and methods; (2) community engagement; (3) policy, implementation, and economics; and (4) communication and dissemination. As of November 2024, the network supports six projects located across the country with a range of intervention strategies (e.g., providing legal guidance to victims). The data and methods group’s tasks include aligning study measures with one another and creating research data management practices that ‘enable robust cross-project analyses that accelerate the science of preventing firearm injuries’ (Sokol et al., 2024, p. 1122). Since research in this area has only recently returned, CFVP began with the goal of harmonizing measures and study results across sponsored projects. Sokol et al. (2024) delve into the benefits of CFVP’s approach and observe that even rigorous studies may not be generalizable, fail to measure key phenomena, or lack adequate data to assess the effects of interventions. Meta-analyses, too, are sometimes hindered by incomparable measures, varied timelines, and limited data (p. 1123). To circumvent these issues, the CFVP requires all six NIH-funded projects to collect data using a harmonized codebook at two assessment points.

Data harmonization

Data harmonization refers to a research process that allows researchers to compare data across varied sources and formats. In their primer on data harmonization efforts for COVID-19, Cheng et al. (2024) argue that harmonization has three dimensions: syntax, structure, and semantics. Syntax refers to data formats (e.g., CSV and HTML files) that require processing before it is possible to harmonize datasets. Structure, or conceptual schema and the organization of data points, describe how ‘variables relate to each other within a dataset’ (p. 2). Semantics describe the ‘intended meaning of words’ and symbols that comprise a dataset.

A broader definition of data includes

Homogeneity across datasets means that harmonization may succeed in one context but fail in another. Cheng et al. (2024) treat syntax and format as synonymous, but information scientists distinguish between the two.

Despite challenges in harmonizing data on a grand scale, we know that harmonization is effective. The Helping to End Addiction Long-term (HEAL) Prevention Cooperative (HPC) has utilized data harmonization to examine a range of evidence-based interventions aimed at preventing opioid misuse (Ridenour et al., 2023). Each of the 10 HPC projects has distinct characteristics (e.g., intervention strategies, theoretical models, settings, and populations) related to opioid misuse. The U.S. Institute of Museum and Library Services (IMLS) sponsored a harmonization effort in 2022 to evaluate public library programming outcomes (Adkins et al., 2023). CFVP’s harmonization effort bridges and combines research, and the number of harmonization-based research applications is growing due to changing scientific norms and requirements for data sharing (Nelson, 2022), as well as the ‘deluge’ of data (Hey and Trefethen, 2003) available to researchers.

Firearm injury-prevention research and information needs of researchers

As suggested earlier, the opportunity for research teams to utilize federal funding for FIP research presents new avenues for identifying patterns of violence and testing interventions aimed at reducing violence and promoting safety. The passage of the Dickey Amendment and the subsequent moratorium on funding resulted in a decline in peer-reviewed FIP research articles, starting in 1996 (Galea et al., 2018). Recent work focusing on FIP has set agendas for an entirely new generation of research. Cooper et al.’s (2022) survey found that FIP researchers rate suicide (29.6%), gun violence (20.1%), and violence prevention (18.3%) as the most important topics to study. During the COVID-19 pandemic, FIP research emphasized topics such as domestic violence, firearm storage, stay-at-home orders, and suicide (e.g., Anestis et al., 2021; Duncan et al., 2020; Lyons et al., 2021). Many researchers equate FIP and automobile safety research, which contributed to a 90% reduction in deaths per vehicle mile driven in the U.S. Automobile safety was achieved through behavioral, engineering, policy, and cultural changes (Roche et al., 2023, p. 583) at all levels of society, and FIP researchers replicating the strategy employed by auto safety researchers hope to produce similar outcomes.

One purpose of our project was to ensure that the CFVP properly supports and meets the information needs of FIP researchers. By information need, we mean the ‘information objects’ (e.g., datasets and textual documents) required to conduct rational, empirically valid research (Buckland, 1991; Wilson, 2006). Case and Given (2016) note that online information has alleviated concerns about information access for many scientists, but social networking, gatekeeping, and interpersonal knowledge sharing—the ‘invisible college’—remain crucial. They found that social scientists frequently utilize administrative data. Healthcare providers need to be aware of research findings and their patients’ conditions (p. 253), so they consult textbooks, drug guides, and other resources in applied settings. Scholars who work ‘between disciplines,’ like our interview participants, whom we discuss later, operate in a ‘scattered’ information environment (p. 247).

Data infrastructure

Despite an increase in funding for FIP research, anecdotal reports from CFVP stakeholders suggest that data remains scattered, unavailable, or are only accessible in sites with limited access. Public health researchers aim to identify the underlying causes of population health issues to inform targeted behavioral and medical interventions. These researchers rely on accessible, accurate data.

FIP research has long been hindered by structural barriers, including a lack of funding and missing or low-quality data (Jamieson, 2013). FIP researchers also need access to data describing communities, as well as information about the social, psychological, and behavioral levels of phenomena. Magee (2023) finds that firearm injury and fatality data are rarely accessible at the neighborhood level. A series of reports by the National Opinion Research Center (NORC) describe the fragmented state of America’s firearm data infrastructure (Roman, 2020a, 2020b, 2020c; Roman and Cook, 2021). These reports find that data are scattered across sites that could function during the funding moratorium, including the Gun Violence Archive, the Centers for Disease Control and Prevention (CDC), the RAND Corporation, Boston University, the American College of Surgeons, and state administrative systems. Data about injuries, available from hospital trauma units, document psychological and social factors that drive firearm violence and injury (Richardson, St Vil and Cooper, 2016). However, it is not hospital workers’ responsibility to clean and disseminate these data.

FIP researchers also need data generated using consistent definitions and reporting procedures, but neither is always supported by existing systems. Gobaud et al. (2023) assessed Gun Violence Archive (GVA) data from four US cities and found it was useful for studying urban areas. However, systematic biases and missing data, including data on nonfatal shootings, limit its research utility. To identify areas of overlap between the GVA and official records, Magee (2023) studied ‘the Nonfatal Shooting Review database, which is generated using information from both police incident reports and internal documents’ (p. 303) in Indianapolis, Indiana. The CDC’s National Violent Death Reporting System includes fatality data missing from Indianapolis, but it provides limited geographic information (Barber, Cook and Parker, 2022). Durkin et al. (2020) note that states collect consistent data from ‘crime reports and public health systems’ (p. 33). However, because each state has its own needs and priorities, the quality of data varies.

Network stakeholders

To conduct this study, we interviewed CFVP stakeholders in three groups:

• Healthcare and public health researchers (e.g., sociologists and statisticians) affiliated with the CFVP,

• CFVP staff and administrators, and

• FIP criminologists who are not affiliated with the coordinating center, but whose research needs and experiences we sought to compare to those of public health researchers.

These groups possessed diverse perspectives about the needs of, and issues faced by, FIP researchers. Public health and medical workers are the primary contributors to CFVP intervention evaluation studies and work closely with the center’s staff. CFVP staff were positioned to discuss the relationships among research studies and the broader FIP research landscape. Criminologists mostly relied on crime data and did not face the same problems as public health researchers during the funding moratorium.

We interviewed 15 people: five researchers working on CFVP-supported projects, four CFVP faculty and staff members, and six criminologists identified through a review of FIP research projects in Google Scholar.¹ The researchers were multidisciplinary scholars who used diverse data types and research methodologies. Six participants worked in public health, four in crime and criminal justice, two in biostatistics, and two in public administration. Biopsychology, community psychology, economics, education, nursing, organizational psychology, pediatrics, psychology, quantitative psychology, social work, sociology, and trauma surgery were each represented by one researcher. Seven used mixed methods, and five were quantitative researchers. Two reported that they used computational methods, and two were experimental researchers. Nine stated that they relied on secondary data for their research, while six used original data they collected.

We did not employ a formal sampling approach when selecting interview participants, because our project’s goal was not to produce generalizable scientific knowledge—it was a user study. We elaborate on this point in the section below. However, we did recruit what might be called a maximum variation sample of individuals ‘to document unique or diverse variations’ among FIP researchers (Palinkas et al., 2015; Patton, 2002). Central to our effort was the goal of developing a resource including archived and indexed FIP datasets. To develop this resource, we needed to understand FIP researchers’ information needs.

Semi-structured interviews

Because this was a quality improvement (QI) project, our goal was not to collect generalizable information but to support CFVP stakeholders. At least two interviewers conducted each interview, providing participants with opportunities to elaborate on their responses and address unanticipated topics that arose during interviews. Each interview began with an explanation of the study’s purpose and a clarification that the project did not constitute ‘human subjects research.’ For good measure, and to align with good research practice, we asked for participants’ consent to participate in, record, and then transcribe our interviews. Next, we asked each researcher about their discipline, the type of research they conduct related to FIP, the research methods they have used in this work, the data that would enhance their research, and data-related problems they have faced in the past (e.g., availability, cost, and documentation). Next, we asked participants to envision an ideal data resource for their work and the support they needed from CFVP to succeed (e.g., help accessing restricted or sensitive data). Our participants did not represent all FIP researchers, but given the small size of our research community, our findings were sufficient. See Appendix A for the interview questions that we asked.

We used an investigator triangulation strategy, adapted from Lincoln and Guba (1985; as cited in Korstjens and Moser, 2017), to increase the trustworthiness of our analysis. This required all three authors to participate in the coding, analysis, and interpretation of the interviews. Two authors conducted the initial thematic analysis by reading through each interview multiple times and identifying common themes to address our research questions. When reading interview transcripts, we created memos that noted participants’ experiences and identified relevant quotations. Next, we reviewed the interviews as a group, assigning thematic labels to highlighted passages. We followed this iterative process until no new information emerged. Finally, the third author reviewed the codes, conducted spot checks, and synthesized themes in an Excel spreadsheet to confirm our analysis. The result was an organized set of textual passages from domain experts that we used to answer our research questions.

Some FIP research problems are not unique

Many of the problems that our interview participants described are common. We found that (1) controlling for confounding variables is challenging, much like in other research domains. Our 15 interview participants also said (2) valuable data, including datasets with key measures (e.g., an individual’s history of mental health issues) are not always reported in official data; (3) laws and regulations, such as the Health Insurance Portability and Accountability Act (HIPPA) impedes data collection for research (albeit for good reasons); and (4) data collection is time consuming. For example, two criminologists discussed using U.S. Federal Bureau of Investigation (FBI) data. Closed cases are available through Freedom of Information Act (FOIA) requests; however, case records can be thousands of pages long. Obtaining the data can be time-consuming, and officials can only provide data as resources permit. Complicating matters, FOIA requests require researchers pay government agencies for coarsened datasets, so many researchers scale back their work, use low-quality data, or abandon projects altogether.

Many types and sources of data exist

Another finding relates to differences in data types and their sources. We asked researchers where they found data and whether they collected it themselves. Some mentioned the complicated nature of FIP research, explaining that taking a national picture of firearm violence requires data from many organizations and perspectives, including from different disciplines. For instance, hospitals collect the medical history of victims. Social scientists (e.g., sociologists and criminologists) need data about social risk factors, such as individual socioeconomic status, education, sexual orientation, race, and age. Social workers have a broader perspective on victims’ trauma and adversity, so their insight is valuable, but it is typically stored in unstandardized case notes. Police and FBI data include extensive information about offenders, including crime details, perpetrator motivations, and victim information. The CVFP network exemplifies how interdisciplinary collaboration can bridge fragmented data sources—medical, social, and criminal justice—to generate a more comprehensive understanding of firearm violence. However, it is worth noting the expansive scope of researchers’ information needs.

Some researchers whom we interviewed discussed their need for specific information (e.g., ‘was an injury self-inflicted or related to domestic violence?’). Authorities do not always collect these data, so they are not uniformly available. There is also missing information about nonfatal injuries because some victims do not seek medical care, or their wounds might not be recorded as gun related.

Differences among geopolitical jurisdictions indicate a need for baseline, shared measurements to conduct national research. One participant said that they have been ‘trying for years’ to obtain access to gun violence data in their state, but Colorado, ‘very much a gun rights state,’ employs a data scientist. ‘They have all the linked data—everything,’ including SNAP eligibility, type of injury, prior criminal histories, and other factors. However, all state governments need the funding, expertise, and motivation to make that happen. Another participant pointed out that ‘there is a strong difference in the policies that might have an uptake in red states versus blue states.’ Clarifying this statement, they said:

These examples highlight the need for standardized yet adaptable data systems that account for the diverse policy environments and patterns of firearm violence across American political jurisdictions.

Participants discussed 43 sources of data. Table 2 lists these sources, categorized by subject (e.g., crime and criminal justice), which reflect the fragmented nature of the U.S. national firearm data infrastructure. Several researchers collected or used qualitative data, and quantitative researchers agreed that qualitative data is essential for understanding the root causes of violence, such as individuals’ motivations and treatment outcomes for interventions. However, these data are often not suitable to draw generalizable claims, which makes them less suitable for secondary use.

Creatively combining data can cause problems

Combining datasets can reveal new findings, and three of our interviewees described creatively ‘Frankensteining’ data. However, they also recognized the problems that occur when using data from outside their academic disciplines. It is not enough for researchers to use data to generate factual knowledge; researchers must also understand the data, including its limitations and idiosyncrasies, as well as the methods used for data collection, imputation strategies, and the underlying phenomena being measured. Without understanding data, individuals or larger teams of researchers can easily misrepresent statistical findings or draw erroneous conclusions.

One example of a researcher expressing concerns about the combination of data related to a multi-measure instrument, the Conflict Tactics Scale (CTS), which is provided by Straus et al. (1996). This instrument provides a way to describe violence involving intimate partners, and one interview participant adopted and expanded the CTS for their work. However, they also said

A lack of established multi-variable measures relating to firearm injuries prompted this researcher to acknowledge the field had much more to learn. Unfortunately, a lack of knowledge about ideal measures also made it difficult for researchers to collect the data they might need in the future.

Existing databases change over time. Researchers sometimes add new questions to study questionnaires in healthcare settings, and lawmakers may mandate research on pressing issues. Regardless of how researchers measure data, one of our participants made an important point: incorporating new questions in large-scale data collection instruments ‘in a way that doesn’t completely fracture the structure that was already created’ can be a challenge unto itself. One criminologist discussed the Bureau of Justice Statistics’ National Crime Victimization Survey and the National Incident-based Reporting System, noting that both have changed over time. The NCVS includes multiple questions pertaining to cargo theft and cybercrime because the U.S. Congress wanted to gather information about these crimes. However, by adding these questions, the government also confused people used to prior versions of the survey data.

Political constraints are often manageable

Several of our interviewees indicated that state and national politics can pose barriers to conducting research about the causes and correlates of firearm violence, but typically, these barriers were manageable. Elected officials interpret the purpose of the 2nd Amendment of the U.S. Constitution differently, and some researchers said that any research examining firearm violence has the potential to become political. However, others said their work was supported by everyone because they only studied violence prevention. A more significant political issue related to data from police departments and local governments. Five study participants emphasized the importance of establishing relationships with government agencies. Building trust to obtain data successfully was a political process because it required demonstrating an ongoing commitment to protect departments against unfair criticism, ensuring that data was used for agreed-upon purposes to promote the common good.

Not all researchers want to share their data

While the Open Science movement has widespread support in academia and government, many researchers we interviewed expressed concerns about sharing the data they collect. Successful data reuse depends on both the quality of the data collected and the quality of accompanying documentation. While there was broad recognition that some data cannot be freely shared because of confidentiality and privacy issues, some of our participants discussed problems with sharing data because they wanted more time to analyze it. Interviewees were sometimes concerned about being ‘scooped’ by other researchers or having their findings misinterpreted or manipulated. Other participants mentioned being unsure of the best places to store sensitive data. As they pointed out, some (but not all) grants specify archiving study data in a repository. They felt it was essential to manage how data would be accessed, by whom, and to protect it from loss.

Data harmonization is contentious

Another challenge noted by some participants was the need for data harmonization across CFVP study sites. Fields such as psychology have freely available, standardized measurement scales, but those are generally lacking in FIP research. A CFVP staff member underscored the importance of standard measures and the need for ‘a compendium of measurement tools in a website […] that people could pull quickly.’ They said it is easy to do this with many mental health measures, but firearms data is more complicated, because the field is only now emerging.

Validating measures takes time and can be complicated by socioeconomic and other confounding factors. Furthermore, some research constructs are difficult to define in binary terms. For instance, one person raised concerns about survey questions regarding firearm storage. A gun owner might keep theirs in a locked case at home, but if it is carried, its status might change throughout the day:

The codebook for the CFVP network was created during a preliminary phase of research at individual study sites, well before data collection commenced. Some sites aspire to conduct research based on new topics that emerge during their research, that is, after the codebook was created. The result was problems with requirements issued by the coordinating center when balancing standardization and project autonomy.

Similarly, a data manager at the coordinating center described the data they are collecting, which includes demographics, social determinants of health, mental health, and substance use. Their data falls under multiple domains, including firearm violence, physical health, and protective factors. Within these domains are multiple constructs. Anxiety, depression, loneliness, and PTSD are measured using standardized tests. Standardized measures helped CFVP staff harmonize data, but not all researchers wanted to use those measures.

CFVP staff emphasized that project Principal Investigators (PIs) also need to understand how their data might be used in the future and how they want it curated for reuse. The PIs wanted clean data that was compatible with statistical analysis software. Collecting data on the same platform (e.g., in a uniform database) would streamline the data collection process, ensuring some degree of fidelity to a data codebook. However, researchers wanted to be able to shape their projects as desired. One individual said, ‘People underestimate the amount of work it takes to make a dataset publicly available and useful to other researchers’ because of the hidden, invisible labor associated with processing it.

The necessity of grants

Another prominent theme in our interviews, mentioned by eight participants, was the need for grant funding to support research, particularly in relation to the tenure and promotion process. Our participants were at various stages of their careers, and securing grant writing and external funding was critical for untenured researchers. During the period when funding for firearms violence research was scarce, researchers turned to other subjects (e.g., substance abuse and youth violence prevention). One participant was advised by a mentor to avoid FIP research because focusing on it before 2019 had been a ‘career killer.’ A younger researcher, beginning their career just as federal funding was restored, interpreted congressional guidance for firearm-injury research as an opportunity and chose to study the topic. Our study participants agreed that federal grants were central to FIP research because they are the only widespread funding source for researchers.

Commonalities: Needs shared among research environments

Our final finding was a set of agreed-upon points that interviewees believed were necessary to support firearm injury-prevention research. Researchers indicated that they needed rich, clean, and comprehensive data with correct and uniform variables. Data collected and shared by hospitals, police, social workers, and other parties must be ‘open’ and provided by, or at least indexed within, easily accessible look-up resources. Databases, datasets, and other information resources must include social risk factors (e.g., prior convictions related to gun violence and sociocultural factors). Participants noted the need for different variables, depending on their academic discipline and research requirements. Some participants proposed a comprehensive state data hub or dashboard of public data that was well-maintained and monitored by a public agency (e.g., the CDC), which could be used across various jurisdictions to demonstrate relationships between variables and outcomes. That type of dashboard would have established baselines, allowing longitudinal studies to demonstrate the effects of medical and behavioral interventions.

Over half of our participants talked about the need to harmonize and standardize measures across different CFVP projects. Social risk factors (e.g., food insecurity, housing, and underemployment) were important constructs for researchers to measure. One interviewee said

This quote highlights the importance of using standardized measures across CFVP projects to facilitate cross-study comparisons.

Geographic granularity was also a necessary data attribute for many of our interviewees. While some individuals said that they simply wanted to have data available at the national level (e.g., the number of firearms in each county), their ideal research environment and ‘stack’ of supplemental resources included information that described states, localities, and even ZIP codes. Regarding geography, interview participants discussed the need for uniform reporting across political jurisdictions. A lack of uniformity across the country made it difficult to draw comparisons or reach conclusions about the effectiveness of interventions. Combining temporal, geographic, and socioeconomic data and isolating relevant variables could also enable individuals to develop targeted interventions in high-risk areas.

Our findings about geographic data also extended to questions about the timeliness and currency of data. Some research projects that our participants spoke about (e.g., behavioral interventions) were only possible when researchers recently collected data. Several interviewees expressed frustration that publicly available data was released 2 years after a given period ended. While older data can be used to answer theoretical questions, write background sections for grants and papers, and take a long view when describing social and behavioral phenomena, timely datasets are needed to address contemporary issues.

What are the information needs of CFVP stakeholders?

Contrary to our expectations, although some CFVP stakeholder information needs were unique to firearm violence prevention, most were not. Because the information needs of CFVP stakeholders were transferable to most other domains, we interpret our results to mean the answer to RQ1 is that researcher information needs are comparable to those of others, particularly in the social sciences.

The data needed by firearms violence researchers is complex, varied, and interdisciplinary. As framed by Case and Given (2016), FIP research is between disciplines. Furthermore, firearm violence, like many other public health issues, is not contained within political borders, yet local jurisdictions have very different policies, motivations, and abilities to collect, process, and store both hospital and police administrative data. To perform their work effectively, FIP researchers require data that pertains to social services, law enforcement, and health factors. It should also be curated (or at least indexed) in public repositories.

Although the data-related issues we identified were not unique, we did identify unique contextual challenges that are worth mentioning. For example, our findings corroborate Roche et al. (2023), who present five issues in studying gun violence:

Relatedly, some of the data problems our interviewees discussed included a lack of research funding to support scholarship and political constraints. Both contributed to a dearth of data and a lack of influence on data-sharing practices. A historic and lingering lack of political will to collect data (by hospitals and police) and fear of political retribution or withdrawal of funding made data both expensive and difficult for researchers to obtain. They also discussed the lack of coordination between states, which suggests that a consortium or network led by data managers, encompassing multiple states or regions, would be a significant improvement. However, to promote cross-organizational efforts, hospitals, coroners, urgent care centers, and law enforcement agencies all require training, political will, and funding to collect and report data for this cause.

What data-related problems were expressed by the people we interviewed?

Although most firearm-injury prevention research problems are not unique, our answer to RQ2 is somewhat more nuanced. As shown in Table 2, a variety of types and sources of relevant data are already available to researchers. Fragmentation in these resources presents a significant challenge that is unique to FIP research. The state and composition of these resources (given the political reality associated with decades of limited funding and research prioritization) limit their utility, and we believe this stems from the necessity of grant-funded research, even though, beyond this point, the problems expressed by CFVP stakeholders were still like those in other fields.

Perhaps the most concise description of the data-related problems we identified can be addressed by the FAIR Principles. The FAIR (Findable, Accessible, Interoperable, and Reusable) Principles offer a set of guidelines and have been widely adopted, including by biomedical researchers (Wilkinson et al., 2016). Biomedical research is expensive and often funded by federal agencies and organizations that increasingly require researchers to develop data management plans. Some researchers have criticized the FAIR Principles because certain data should not be shared, particularly when governance, confidentiality, and ownership issues are involved. Our interviewees believed that making their data completely ‘open’ would violate the trust they had built with individuals who provided it. However, this view reflects a misconception of the FAIR Principles, which do not suggest that all data be public. Rather, what matters is that datasets are discoverable and described with rich metadata, including clear indications of local access restrictions (Landi et al., 2020).

There are many different reasons why individual researchers and organizations do not share data according to the FAIR Principles. For instance, (1) criminologists sometimes pay large sums to obtain data from police departments or the FBI using FOIA requests, so they do not want to share it with others for free. Furthermore, (2) health researchers cannot viably remove identifiable data from patient records, and organizations have a legal responsibility to protect subject confidentiality and privacy. Third, (3) many researchers do not want to be ‘scooped.’ They might prevent this by imposing an embargo on releasing data. Collectively addressing many of these issues cannot be done by individuals.

Where were the points of agreement about firearm violence and injury-prevention data among the people we interviewed?

The points of agreement we identified among interview participants (RQ3) primarily stemmed from efforts to harmonize data collected with support from the CFVP, and these may provide lessons for attempts to redevelop and streamline America’s firearm data infrastructure. The CVFP project aims to harmonize data from six studies and has identified considerable overlaps between projects by establishing a shared codebook and a set of measures for researchers to use. However, these teams still depend on supplemental resources to succeed. In our work, we found that the shared resources FIP researchers need are complex.

Beyond the practical problem of finding data, some participants described issues with others’ interpretation of data from outside their field; they had found studies that inaccurately reported data, having been creatively (but improperly) combined from multiple sources due to researchers’ misunderstandings. These authors may also not have read the ‘fine print’ about how secondary data was processed. A shared set of resources, including a harmonized dataset in an archive, such as ICPSR, resolves this problem. However, researchers still needed access to data from sources that will never be collected, curated, and stored by the same entity or organization. Data harmonization and leveraging information from multiple sources are not mutually exclusive activities but doing both at the same time is sometimes impractical for cost-related reasons.

Project-specific data harmonization efforts, such as the Helping to End Addiction Long-term (HEAL) Prevention Cooperative (HPC) and the data module for the Public Library Association’s Project Outcome initiative (Adkins et al., 2023), provide roadmaps to combine datasets when FIP researchers choose to harmonize datasets. The advantage of including multiple data types and resources is obvious: researchers can cross-reference and identify phenomena more easily and accurately. However, there are obstacles that all researchers face. Data are not uniformly collected across jurisdictions, and different methods of imputation and statistical maneuvering affect how data can and should be used.

Data harmonization was the primary strategy used by the CFVP network to help researchers work together, and it featured prominently in our interviews. However, even the small group of researchers affiliated with CVFP described challenges (i.e., establishing points of agreement regarding the creation of objective data sources) when trying to meet everyone’s needs. There was no guarantee that our interviewees could or would always reach consensus when trying to agree on shared data structures, syntax, and semantics (Cheng et al., 2024). Fortunately, harmonization is just one of many strategies that can be used to help FIP researchers meet their needs. This is particularly important because we also found there is no agreement about what data needed to be harmonized in the first place. Our interviewees discussed when harmonization does not work: (1) when political, organizational, or institutional systems prevent harmonization from taking place, (2) when economic issues prevent data harmonization, (3) when the syntax or type of datasets is incompatible, (4) when the structure of data are difficult to identify and transform to allow interoperability, and (5) when the semantics of data are sufficiently different that they render comparisons meaningless or of limited value.