Proceedings of a Workshop—in Brief

Crime Rates During the COVID-19 Pandemic

Proceedings of a Workshop—in Brief

The National Academies of Sciences, Engineering, and Medicine’s Committee on Law and Justice convened a workshop through its Planning Committee on Crime Rates during the SARS-CoV-2 or COVID-19 pandemic (hereinafter referred to collectively as “COVID-19” or “COVID-19 pandemic”) on November 10, 2022, to explore crime rate changes during the pandemic, potential explanations for those rates, and opportunities for future methods, data, and research. Specifically, it sought to (1) explore existing data on the trends in multiple criminal offenses during the pandemic; (2) explore existing explanations for the crime rate changes in multiple offense types during the pandemic for their scope, logical consistency, empirical support, and limitations, with special attention to explanations related to the pandemic and associated population restrictions (e.g., stay at home orders, social gathering restrictions, etc.), as well as the diffusion and availability of firearms; and (3) discuss methodological issues, data infrastructure needs, and research gaps to inform understanding of crime problems and rates. Additional details, including a recording of the workshop, can be found online.¹

EXISTING DATA ON THE TRENDS IN MULTIPLE CRIMINAL OFFENSES DURING THE PANDEMIC

Jeff Asher, AH Datalytics, began the workshop with a discussion of existing data on national homicide and gun violence trends in the United States. Asher drew on an article he wrote for the New York Times² utilizing data from a variety of sources including the National Incident-Based Reporting System (NIBRS) and the FBI Supplemental Homicide reports and outlined the methodological issues and uncertainty³ of the data sets he examined (e.g., margins of error, speed, scope, level of detail).

Asher reported that homicide increased nationally starting in late 2019, remaining elevated through most of 2021, and decreased slightly to date in 2022. There was a sharp rise in the months of May, June, and July of 2020, which corresponded with the relaxing of stay-at-home orders, the murder of George Floyd, and associated

___________________

¹ For additional information regarding the workshop, visit: https://www.nationalacademies.org/event/11-09-2022/crime-rates-during-the-covid-19-pandemic-a-workshop

² For additional information, see https://www.nytimes.com/2021/09/22/upshot/murder-rise-2020.html

³ Uncertainty may be defined as “…limited (incomplete or imperfect) knowledge and information about current or future environmental, social, economic, technological, political and institutional conditions, states and outcomes and the implications or consequences of these current or future conditions, states and outcomes” (Brouwer and De Blois, 2008).

protests.⁴ The data also show an acceleration in the share of murders in the United States that are being committed via firearms (from roughly 60% in the 1990s to 77% in 2020).

However, Asher stressed that the data are incomplete. National crime and policing data sources each have their limitations, he said. The Summary Reporting System (or Uniform Crime Report) was slow and vague (only includes the 7 major crime categories), but it was precise in that it provided consistent data year-to-year. In 2021, the FBI transitioned to NIBRS, which expands the categories of crime collected to over 50 and provides much more specificity on incidents. However, only 65 percent of the U.S. population lived in areas where law enforcement agencies were reporting this data in 2021 (as compared to 95 to 97% coverage under the Summary Reporting System). This means that only around 50 to 60 percent of murders were reported in the system in 2021, as only 52 percent of law enforcement agencies reported data that year. Such insufficient reporting requires estimating roughly 35 percent of crime (as opposed to previously estimating 3 to 5%), which results in wide confidence intervals, wide margins of error, and imprecision.

The FBI Quarterly Reports also have their limitations, Asher said. While the reports are much more timely, they are incomplete, as only 60 to 80 percent of law enforcement agencies report data, and similar to the Summary Reporting System only include the 7 major crime categories. Asher also commented on the availability of quarterly data, noting that NIBRS has failed to meet the 60 percent threshold of agencies reporting set by the FBI for the past six quarters, limiting the ability of researchers to understand national trends over the course of the pandemic.

Next, Jillian Carr, Purdue University, spoke about a study⁵ she co-authored on localized domestic violence (DV) trends in Chicago, Illinois. This study utilized cell phone block-level activity data, administrative 911 data, and crime data from the City of Chicago to estimate the effects of the city’s stay-at-home order (SAHO) on calls for police service, crimes recorded by police, and arrests made relating to DV. According to Carr, the SAHO announcement increased time spent at home, leading to a decrease in total calls for police service, and yet a 6.8 percent increase in DV-related calls. Carr also reported a drop in official reports by police officers and arrests for DV crimes by 7.4 percent and 26.4 percent, respectively.

When exploring possible explanations for these changes, Carr pointed to SAHO, stress and strain, and the reporting behavior of witnesses as likely causes (e.g., neighbors home to report disturbances during the day), along with the impact of changing policing practices and intake methodology in Chicago (e.g., given the pandemic, officers may have been spending less time in people’s homes to mitigate the spread of the virus). Carr pointed to similar results in Los Angeles,⁶ and in a follow-up, 18-department study. She concluded by suggesting that regardless of the crime studied, it is difficult to disentangle any effects when the reporting process changes and the outcome also changes.

Following Carr, Richard Rosenfeld, University of Missouri-St. Louis, reviewed data from two research articles⁷ he co-authored on changes in crime rates across 31 cities in the United States for ten different offenses including homicide, DV, burglaries, larcenies, and motor vehicle thefts.⁸ He explained this research was rooted in routine activities theory, hypothesizing that as activities tend to be concentrated to households, one could expect decreases in crimes and lower crime rates.

According to Rosenfeld, the study Crime, quarantine, and the U.S. coronavirus pandemic used Google cellphone

___________________

⁴ On May 25, 2020, Minneapolis police officers arrested George Floyd, a 46-year-old Black man, after a store clerk alleged that Floyd made a purchase using a counterfeit $20 bill. During the arrest, Derek Chauvin, a 44-year-old White man, knelt on Floyd’s neck for approximately nine and a half minutes, killing him. During the arrest, three other officers aided Chauvin while bystanders recorded the incident. The video of the homicide became the catalyst for global racial justice protests and calls for police reform. In 2021, a jury found Chauvin guilty of unintentional second-degree murder, third-degree murder, and second-degree manslaughter and was sentenced to 22.5 years in prison. The other officers on the scene were found guilty of federal civil rights violations.

⁵ For additional information, see https://www.journals.uchicago.edu/doi/10.1086/713787

⁶ For additional information, see https://www.nber.org/papers/w29429

⁷ For additional information, see https://onlinelibrary.wiley.com/doi/full/10.1111/1745-9133.12557; see also https://counciloncj.org/mid-year-2022-crime-trends/

⁸ Notably, not all offenses were available for each of the 31 cities in the sample presented.

location data, deaths attributed to COVID-19, and unemployment data to conduct interrupted time series analysis, along with incident-level crime data from open data portals of local police departments. He described that at the outset of the pandemic, robberies and property crimes (except for motor vehicle theft) dropped. Homicides and serious assaults rose immediately after the murder of George Floyd, well after the pandemic had begun. Following the height of the pandemic, robberies and property crime climbed substantially, and homicides dropped slightly. These findings are reflected in Asher’s data, which Rosenfeld referenced during his presentation. He also said that based on data from 15 cities, there was little change in DV rates, but that these results should be taken with caution due to the sample size. Rosenfeld suggested his findings support the routine activity hypothesis. He observed that decreases in these offenses were associated with increases in residential duration and later increases in robberies and property crime corresponded with the sharp increase in inflation. However, he noted that COVID-19-induced reduction in police presence and response, declining police legitimacy, and rising legal cynicism after George Floyd’s murder are important factors to consider when looking to address methodological issues, data infrastructure needs, and how to inform understanding of crime rates during the pandemic.

Finally, Rosenfeld outlined the advantages (e.g., timeliness) and disadvantages (e.g., change over time) of incident-level data accessed from open data portals of local police departments. He indicated that reasonably reliable and timely crime data across multiple jurisdictions currently must come from private sources and observed that, as long as the federal statistical system for police-based crime data relies on voluntary contributions from local law enforcement agencies data limitations will persist.

Planning committee member, Jamein Cunningham, Cornell University, moderated a discussion with the panelists. He noted that a general theme from the presentations is the discretionary aspect of how crime data is collected. First, the reporting officer has the discretion to report a crime or not, and then the law enforcement agency or jurisdiction decides if they are going to report their data to a federal collection. COVID-19 has changed the discretionary aspect of what police officers are doing, asking how this may impact our understanding of the available data. Panelists Asher, Carr, and Rosenfeld noted that police officers reporting fewer crimes is a plausible hypothesis. Asher also described challenges to police staffing that have resulted in longer response times, which could impact crimes reported. On this point, audience member, Cynthia Lum George Mason University, elaborated that it appears that the number of police labor hours did not change during the pandemic, but the quality of what they are doing during that time may have changed, which matters immensely for crime control. Carr further emphasized the importance of citizens’ trust in police and willingness to report crimes as an important factor, especially after the murder of George Floyd. Rosenfeld concurred that additional research is needed on the role of protest activity and corresponding drops in police legitimacy in generating crime rate changes during the pandemic.

EXPLORING THEORY-DRIVEN EXPLANATIONS FOR CRIME RATE CHANGE

The second panel of the day focused on addressing existing explanations for the crime rate changes in multiple offense types during the pandemic for their scope, logical consistency, empirical support, and limitations. María Vélez, University of Maryland, was the panel’s first speaker.

Vélez referenced a piece she co-authored for the Annual Review of Criminology,⁹ which presented a conceptual framework for understanding crime rates. The elements of the proposed framework include (1) changes in social controls (e.g., surveillance), (2) changes in criminal propensities and motivations (e.g., economic opportunity and aging populations), and (3) changes in criminogenic settings and situations (e.g., stabilization of illicit and secondary markets, routine activities).

Next, she reviewed four key concepts to consider when studying crime trends, derived from her co-authored piece, including (1) importance of offense type, (2) crime

___________________

⁹ For additional information, see https://www.annualreviews.org/doi/abs/10.1146/annurev-criminol-032317-092339

patterns often depend on data source, (3) consideration should be given to demographic patterning of trends, and (4) subnational heterogeneity.

Vélez expanded on the concept of subnational heterogeneity, noting that due to the highly concentrated nature of crime (particularly violent crime) there is great variation across places within cities (i.e., at the neighborhood level). According to Vélez, because a high percentage of a city’s crimes occur in concentrated locations, the effects of those trends (increases or decreases in crime) are not felt the same way across neighborhood places. Drawing from the National Neighborhood Crime Study and census tract data from Chicago and Washington, D.C., she illustrated neighborhood crime patterns for homicide and robbery rates, highlighting the variation in socioeconomic conditions, racial composition, and housing instability across communities seeing increases and decreases in crime.

To advance research in this space, Vélez described the need for an enhanced, targeted, and coordinated data infrastructure that includes (1) ongoing survey data collection on offending that can be connected to neighborhood context to capture demographic, socioeconomic, and policing conditions; (2) annual data on criminal legal system responses for the above neighborhoods; and (3) qualitative data collection for the above neighborhoods to understand on the ground dynamics related to crime changes.

In concluding, Vélez emphasized the importance of understanding how crime changes in society is pressing, as it exacts a toll on individuals and their communities, especially those that are racially minoritized.

Following Vélez, Amy Nivette, Utrecht University, discussed her contributions to a cross-national, multicity study into the impact of the pandemic stay-at-home restrictions on urban crime.¹⁰ According to Nivette, the study began in an effort to understand how changes in routine activities may lead to changes in crime. Theoretically, Nivette posited that changes in routine activities and opportunities, along with changes in motivations or propensities (e.g., strain) can affect criminal behavior by either reducing or increasing opportunities or in some ways affecting the propensity through the impact of the strain of the pandemic on individuals.

Using public access COVID-19 government response tracker data for 27 cities, researchers conducted an interrupted time series regression and utilized meta-analysis techniques to examine overall changes in different crimes. Nivette explained that stay-at-home orders are associated with immediate declines in all types of crime, with the exception of homicide. There was notable heterogeneity between cities. Researchers then examined the extent to which differences in decline were associated with the strength of a location’s stay-at-home order, using meta-regression techniques. Nivette said that in cities where orders were stricter, crime fell more sharply.¹¹

In looking past the initial stages of the pandemic, there was a trend toward returning to the average crime rate. However, Nivette noted that this study lacked long-term data, research into strain, and research into policies other than stay-at-home orders. She stated that these are opportunities for future studies that would bolster the logical consistency and expand the scope of the concepts presented here (e.g., studies into other policies implemented during the pandemic). This study also noted a lack of high-quality comparable data meant there was not an opportunity for researchers to look into whether the sudden shift in routine activities caused crimes to shift online, a topic also discussed during Q&A sessions.

John Boman, Bowling Green State University, focused on strain theories as an explanation for crime changes during COVID-19, within the context of the opioid epidemic. Placing a key focus on strains and stressors, general strain theory incorporates both individual- and macro-level factors in the explanation of crime and drug use. Drawing from his background in longitudinal

___________________

¹⁰ For additional information, see https://www.nature.com/articles/s41562-021-01139-z

¹¹ Nivette presented that Lima, Peru, had one of the strictest lockdowns and declines in crime, while cities like Malmö and Stockholm, Sweden had weaker, voluntary lockdowns and much smaller declines in crime.

research on opioids using five different federal data sets coupled with original data, Boman highlighted that the United States experienced more than 100,000 overdose deaths in a 12-month time span for the first time between May 1, 2020, and April 30, 2021. Approximately three-quarters of these overdose deaths were opioid-related. Boman observed that this timeline corresponds with breaks in black market supply chains during COVID-19 lockdowns, which once re-established, began fueling the spike in overdose deaths. In summary, he said this implicates the stress/strain of the COVID-19 pandemic as a leading correlate of the dramatic increase in opioid overdose deaths the United States experienced in 2020 and beyond. Regarding the needs for future research in the opioids and substance abuse areas, Boman observed that, there is a distinct lack of data from rural communities and stressed the need for projects accounting for micro- and structural-level factors drawing from multifaceted theoretical approaches.

Caterina Roman, Temple University closed the second panel. She discussed her work studying COVID-19’s effect on drug markets and gun violence in Philadelphia, which focuses on within-city variation (i.e., across census tracts) based in social disorganization theory as an explanation of crime rates.

The study, conducted in collaboration with graduate student Nicole Johnson, examined spatial and temporal patterns of shooting victims between January 2017 and June 2021 and differences in neighborhood-level predictors before and after the onset of the pandemic. Roman reported that although the typical social disorganization factors were significant predictors of increasing rates of neighborhood shootings before the pandemic, they were not significant predictors during the peri-pandemic period of March 2020 to June 2021. Roman described significant variables associated with increasing shooting rates peri-pandemic including high-drug-market-activity tracts¹² and investigatory stops. Roman suggested that drug markets shifting towards highly profitable drugs (e.g., fentanyl and synthetic opioids) could be studied in an effort to understand increases in crime rates within highly disorganized neighborhoods where there are routine activities and strong economic motivations but weak, informal social control.

Planning committee member, Charis Kubrin, University of California, Irvine, moderated a discussion with the panelists. Panelists and audience members engaged in a discussion with moderator Kubrin of whether current criminological theories are sufficient to understand changes in crime occurring during the pandemic, especially considering displacement of crime online, with Boman and Vélez noting that no one theory is likely to be explanatorily sufficient.

DIFFUSION AND AVAILABILITY OF FIREARMS: EXISTING EXPLANATIONS AND METHODOLOGICAL LIMITATIONS

The final panel of the workshop focused on issues concerning the role of firearms in explaining recent crime trends and the methodological difficulties that make drawing conclusions challenging. The first speaker, Jens Ludwig, University of Chicago, discussed general explanations for gun violence as applied to the pandemic.

Ludwig began by describing an increase in murders and gun violence (80 to 90% of all murders are committed with firearms) in Chicago in 2020 as compared to 2019. Ludwig noted that for those murders that we know the motivations, the largest share occur because of an interpersonal altercation. He then hypothesized three factors that might be contributing to more shootings during this period: (1) anything leading to more social friction, (2) anything leading to less de-escalation, and (3) anything resulting in more lethal arguments (e.g., more guns in public areas). The first change observed during the pandemic related to gun violence, he said, is an increase in social friction. While the number of social interactions during the pandemic may have decreased, the likelihood that any given social interaction leads to social friction may have increased, as data suggests that there has been an increase in the prevalence of mental health problems during the pandemic especially among younger age groups who are most likely to be involved in gun violence as either victims or perpetrators. Second, disruptions to social services and the criminal justice

___________________

¹² According to Roman, neighborhoods without high rates of drug market activity did not experience comparable rates of increase in gun violence in the peri-pandemic period.

system meant there were fewer opportunities for de-escalation of situations involving guns. Finally, Ludwig explained that there is suggestive evidence that more guns are being carried in public areas. In August 2019, Chicago police recovered illegal guns roughly at a rate of 1 per every 250 traffic or street stops; in June 2020, that rate increased to about 1 per 50 traffic or street stops. In addition, gun sales in Illinois increased by 40 to 50 percent in 2020 and 2021 as compared to 2019.

A key area for future research, Ludwig said, is to understand whether these patterns are a temporary effect of the pandemic, or if this level of gun carrying will remain a steady state. Ludwig cautioned, “If these challenges stay in place and persist overtime, there is a chance that we might wind up returning to the sustained rates of violence that we saw towards the end of the 20th Century.”

Ludwig was followed by Michael Sierra-Arévalo, The University of Texas at Austin. He focused his presentation on what is known about firearm availability during the pandemic and how those changes are related to increases in firearm crime and violence during that same time period.

Sierra-Arévalo began with patterns in firearm availability. Using administrative data from the FBI’s National Instant Criminal Background Check System as a proxy for firearm sales, Sierra-Arévalo observed an almost 100 percent increase from 11 million firearm sales in 2019 to nearly 22 million in 2020. Sierra-Arévalo pointed to a spike in monthly sales in the first half of 2020 noting the period in late spring/early summer of 2020, which corresponds with the public protest and unrest following the police murder of George Floyd. National surveys of self-reported purchasing behavior show similar increases from 2019 to 2020: individual purchasers increased from approximately 13.8 million purchasers in 2019 to 16.6 million purchasers in 2020, and the proportion of first-time purchasers also increased from 2019 to 2020 (from 17.7 to 22.9%).¹³

To date, he noted, there is only one study that directly assessed whether firearm purchases explain changes in violence during the pandemic; Schleimer and colleagues (2021) found no relationship between state-level excess purchasing and non-domestic firearm violence, during the peak of the COVID-19 pandemic firearm purchasing spike.¹⁴

Sierra-Arévalo concluded by discussing limitations of available data. Trace data underestimates the total number of firearms recovered by police because not all recovered firearms are submitted for tracing, some submitted firearms have missing serial numbers, and some are, “ghost guns.” Moreover, currently available trace data does not include the type of crime in which the recovered firearm was used. Finally, Sierra-Arévalo noted that the effects of gun purchases on violence following the spike during the COVID-19 pandemic, and that due to the fact firearms are durable goods, the “time-to-crime” remains and important research question.

Morgan Williams, Jr., Columbia University, discussed the available literature on the role of law enforcement in effectively addressing gun violence, emphasizing the importance of factors such as officer staffing levels, gun proliferation and regulations, and incentives affecting enforcement decisions. Williams notes that the empirical evidence suggests that increased police expansion leads to decreases in crime, but can lead to increased racial disparities in certain “quality-of-life” arrests. He also noted that many law enforcement agencies are currently facing significant challenges in maintaining “optimal” staffing levels. Recent reforms in Missouri, which repealed the “permit-to-purchase”¹⁵ law, and resulting increases in gun violence, Williams said, coincided with a decrease in weapons arrests and an increase in officer exposure to gun violence. Williams outlined how

___________________

¹³ See Miller, M., Zhang, W., and Azrael, D. (2022). Firearm purchasing during the COVID-19 pandemic: Results from the 2021 National Firearms Survey. Annals of Internal Medicine, 175: 219–225.

¹⁴ See Schleimer, J.P., et al. (2021). Firearm purchasing and firearm violence during the coronavirus pandemic in the United States: A cross-sectional study. Injury Epidemiology, 8: 43.

¹⁵ “Licensing laws” or “Permit-to-Purchase (PTP)” laws require an individual who wants to purchase a firearm to apply for a license at a local law enforcement agency. The application may include a background check, fingerprinting, and potentially evidence of handgun safety training.” For more information, see https://www.jhsph.edu/research/centers-and-institutes/johns-hopkins-center-for-gun-violence-prevention-and-policy/research/licensing/

disentangling many of the factors driving both changes in policing and rising gun violence involves several challenges from the COVID-19 pandemic to coinciding reforms and unobservable factors (e.g., bail reform). He explained that moving forward, contemporary applied microeconometric techniques can yield insightful evidence on certain factors driving these trends, including policy experiments, other forms of quasi-random variation, and improvement in data collection and quality.

The workshop’s final speaker was George Mohler, Boston College, who presented his work on the spatial concentration of gun violence during the pandemic, and the contagion of gun violence. His remarks largely drew from work using open-source shooting incident data from several U.S. cities.¹⁶ Mohler referred to the concentration of crime at hotspots and noted the data show stability of hotspot locations for shootings in the four cities studied pre- and post-pandemic. He said that victims of gun violence have similar demographics pre- and post-COVID and are disproportionately Black and Hispanic victims. Mohler reported increases in the contagion of gun violence in most cities, but reproduction numbers vary and there were several limitations including the small subset of cities studied.

Planning committee member, Ojmarrh Mitchell, Arizona State University, moderated a discussion with the panelists. Panelists addressed a number of questions including one from moderator Mitchell on the role of gun policy in explaining the increase in firearm-related offenses, with Sierra-Arévalo noting increasingly good evidence that policy changes (e.g., right to carry) are related to increases in violent crime. Williams suggested that a question for future research is whether these policy effects were exacerbated by the COVID-19 pandemic.

CONCLUDING THOUGHTS

In the workshop’s final session, John MacDonald, workshop planning committee chair, highlighted some key takeaways from the day’s discussion:

(1) improvements in data infrastructure are needed to better understand crime rate changes; (2) communicating about crime requires nuance and precision and it is important to discuss victimization and not just police action when describing crime; (3) the effects of the pandemic differ by place and country; (4) in terms of serious crime and the pandemic, the United States is an outlier in the availability of firearms; and (5) it is important to pay attention to the disparate impact of crime, noting when there is a crime rise it is going to disproportionately be the burden of low-income and racially and ethnically minoritized individuals. Other members of the planning committee (Jamein Cunningham, Charis Kubrin, Ojmarrh Mitchell, and Richard Rosenfeld) highlighted additional key areas for future research, including: understanding crime rate changes in rural areas, differences in victimization and offending by juveniles versus adults (particularly considering changes in opportunity for crime during the pandemic), potential changes in police-citizen interaction and use of force incidents, and other criminal justice system responses during the pandemic (e.g., courts, jails) and their impact on crime trends. Robert Crutchfield, Committee on Law and Justice chair, remarked in closing on the “importance of moving away from simple narratives and one variable explanations” for crime rate changes, given that crime and crime rates are complex phenomena and require complex explanations and interconnected solutions that account for the multi-faceted and systemic nature of crime.

___________________

¹⁶ For additional information, see https://pubmed.ncbi.nlm.nih.gov/36084751/; see also https://www.sciencedirect.com/science/article/abs/pii/S0047235221000684

DISCLAIMER This Proceedings of a Workshop—in Brief was prepared by Abigail Allen and Emily Backes, rapporteurs, as a factual summary of what occurred at the workshop. The statements made are those of the rapporteur or individual workshop participants and do not necessarily represent the views of all workshop participants; the planning committee; the Committee on Law and Justice; or the National Academies of Sciences, Engineering, and Medicine.

COMMITTEE JOHN M. MACDONALD (Chair), the University of Pennsylvania; JAMEIN CUNNINGHAM, Cornell University; CHARIS KUBRIN, University of California Irvine; OJMARRH MITCHELL, Arizona State University; and RICHARD ROSENFELD, the University of Missouri- St. Louis.

STAFF EMILY BACKES, Deputy Board Director, Committee on Law and Justice (CLAJ); ABIGAIL ALLEN, Associate Program Officer, CLAJ; and STACEY SMIT, Program Coordinator, CLAJ.

REVIEWERS To ensure that it meets institutional standards for quality and objectivity, this Proceedings of a Workshop—in Brief was reviewed by JAMEIN CUNNINGHAM, Cornell University. We also thank staff member RUTH COOPER for reading and providing helpful comments on this manuscript. KIRSTEN SAMPSON SNYDER, National Academies of Sciences, Engineering, and Medicine, served as the review coordinator.

SPONSOR This Workshop was supported by the National Science Foundation (# 2213343).

Suggested citation: Suggested Citation: National Academies of Sciences, Engineering, and Medicine. 2023. Crime Rates During the COVID-19 Pandemic: Proceedings of a Workshop—in Brief. Washington, DC: The National Academies Press. https://doi.org/10.17226/26920.

Division of Behavioral and Social Sciences and Education Copyright 2023 by the National Academy of Sciences. All rights reserved.

__________________