The opioid epidemic, ongoing since the 1990s and increasingly deadly during the COVID-19 pandemic, is linked to another long-standing and hard-fought public health crisis: HIV. People who inject drugs (PWID) experience elevated rates of HIV acquisition and transmission, a tendency that has been exacerbated by the opioid epidemic. Because this group is a potent vector through which HIV remains active in our society, changes in policies that affect PWID can have broader public health ramifications.

In a new study published in the International Journal of Drug Policy, researchers found that without careful and deliberate mitigation efforts in place, criminal justice reforms that result in large, rapid decreases in incarceration rates can lead to an increased risk of HIV spread through communities.

To identify this unintended hazard, authors of the paper had to overcome substantial research challenges: collecting data on PWID is difficult, and there are multiple overlapping sources of influence through which policy might affect HIV transmission. The Brookings Institution’s Center on Social Dynamics and Policy worked with colleagues across several institutions to create a sophisticated computational simulation tool that can provide insight into potential policy effects. Recently published research includes a description of this tool along with findings from our application of it to multiple types of relevant policies.

In each of the five metropolitan areas considered, simulations indicate that criminal justice reform policies that result in extensive reductions in incarceration rates increase relationships involving PWID that have the potential for HIV transmission (figure 1). On average, there is an approximately 10% increase in incidences of such relationships compared to simulation runs in which incarceration rates remained at current levels. There are only small changes (mostly slight increases, although we estimate a slight decrease in Baltimore) in these relationships associated with criminal justice reform policies that result in moderate decreases in incarceration rates.

These overall metropolitan area impact estimates can mask some potentially important variation in who is affected. Figure 2 disaggregates the effects of extensive incarceration rate reduction on relationships with the potential for HIV transmission by race in three metropolitan areas. The patterns depicted differ across each of these contexts, with white individuals experiencing the greatest increase in San Francisco and Black individuals in Miami and New York City.

Criminal justice reform initiatives are rooted in efforts to advance social justice and meaningfully enhance racial and ethnic equity. In general, these goals are well-aligned with public health. Thus, the potential for increased spread of HIV resulting from reductions in incarceration rates is counterintuitive. However, this finding should be taken with two important caveats in mind.

The first is that the simulated policy effects are restricted to a relatively short time horizon of three years. Potential positive effects of incarceration reform including decreases in HIV risk behaviors, stabilization of sexual networks, increased access to drug treatment, and HIV care and treatment might eventually offset any initial uptick in community-level HIV prevalence that follows criminal justice reforms.

Second, these simulated effects are absent any concerted efforts to mitigate them. Prior research suggests that ensuring HIV prevention and treatment among justice-involved populations and the persons with whom they interact in the community requires a multipronged approach that accomplishes the following: (1) bridges gaps between HIV and behavioral health services in correctional settings and those in the community, (2) engages formerly incarcerated men and women into HIV testing and treatment, and behavioral health programs in the community, (3) builds trust in the medical system and autonomy among formerly incarcerated men and women, (4) facilitates community reintegration by reducing barriers to employment, housing, and social benefits, leveraging social capital, and (5) integrates case management and patient navigation in post-release and reentry programs. Many of these approaches are documented to have strengthened reintegration and to have increased HIV care engagement and retention. That is, policymakers can proactively and strategically engage in efforts that buffer against unintended, negative consequences of criminal justice reform.

When attempting to limit or prevent increases in HIV prevalence stemming from criminal justice reform, policymakers should keep in mind that “one size does not fit all.” In our simulations, effects of reductions in incarceration rates vary across geographic settings and, within them, across racial and ethnic groups. Therefore, consideration of underlying contextual factors should play a key role in crafting public health strategies that accompany criminal justice reform.

Beyond the immediate implications for criminal justice reform, this work is a good example of the importance of exploring interactions between policy domains rather than only looking at them in isolation. The relationships between areas like health and criminal justice, as well as education, labor, and others, are important and understudied—and often wholly overlooked – to the detriment of policies they govern and people they affect.

      

The obesity epidemic continues to be a serious concern for the health and welfare of Americans, with over 40% of adults in the United States struggling with obesity. Childhood obesity is of particular concern, not only because it is associated with an array of immediate health challenges, but also because it increases the likelihood of obesity in adulthood. According to the Centers for Disease Control and Prevention, 13.7 million children and adolescents struggle with obesity, including 18.4% of children aged six to 11 years and 13.9% of children aged two to five years.

Childhood obesity is potentially influenced by a large number of factors. One that has been understudied until recently is the prenatal environment. As researchers have begun exploring the role that this might play, compelling evidence has emerged indicating that there is a meaningful  correlation between maternal consumption of sugar-sweetened beverages (SSBs) (e.g. soda and sweetened fruit drinks) and childhood obesity and overweight (i.e. high BMI) status. In a new study, my colleagues and I attempted to disentangle the contributors to childhood obesity, including prenatal factors, in hopes of improving the quality of data that policymakers need to address this complex health crisis.

Whether and to what extent maternal consumption of SSBs during pregnancy itself drives child obesity and high BMI is an important, multi-faceted question. On the one hand, this consumption may just be a proxy for parental food preferences and their household’s overall consumption patterns, which can both affect children’s BMI. In other words, this might simply be a case of correlation but not causation. Alternatively, biological pathways might account for some of the observed relationship. For example, during the second trimester, the fetus begins to ingest amniotic fluid in addition to receiving glucose through direct blood transfer, and chronically high glucose levels driven by maternal consumption of SSBs may be matched by chronically high levels of fetal insulin production that can trigger increased lipogenesis (the metabolic formation of fat) as well as altered expression of genes and proteins related to metabolic functioning in the offspring. That is, maternal consumption of SSBs might have a direct, causal impact on children’s physiology in ways that affect their BMI.

These two explanations have different implications for future research as well as policy and practice. In our study exploring the impact of SSB consumption during pregnancy, my colleagues and I used a rich set of longitudinal data from a high-quality, long-duration cohort study of children, and a lab-validated microsimulation model of childhood consumption, growth, and physiology from infancy through adolescence to infer caloric intake throughout childhood; this process removes some known, major influences on BMI, such as those related to physical activity. We then estimated the association between maternal consumption of SSBs during pregnancy and later caloric intake by the child. Our results suggest that, as expected, much of the observed correlation between prenatal exposure to SSBs and later childhood predisposition to high BMI reflects other social and environmental influences (e.g. the food parents feed their children). However, the results leave open the possibility that there may be biological effects of prenatal SSB exposure that operate either independently of or in concert with these other factors.

This work should motivate additional research to more definitively characterize the biological effects of prenatal SSB exposure. Such research might employ data that are collected with this purpose in mind. Even as that research proceeds, given the overwhelming evidence of other negative health outcomes associated with SSB consumption, policymakers might still wish to consider actions that would reduce maternal consumption of them during pregnancy.

One approach that policymakers might consider, which would mirror many past policy efforts, would seek to change health behaviors at the individual level. Policy efforts in this vein might include such things as advocating that physicians strongly highlight SSB recommendations during prenatal consultations or public messaging campaigns.

However, based on recent work that I have contributed to, a more effective policy approach would take a broad perspective that considers how individual behaviors are both driven and constrained by larger-scale influences: environmental, social, economic, and governmental structures. For example, a large number of American families do not have access to running drinking water, and many others have encountered water that is unsafe or unpalatable. In settings in which it is easier or less costly to procure SSBs than bottled water, this may increase consumption substantially, with profoundly negative health impacts. Addressing this will likely require action that encompasses policy areas such as infrastructure, food systems, and environmental regulations.

In general, the research suggests policymakers should consider actions to reduce maternal SSB consumption during pregnancy. The best way to substantially and sustainably do so would be to identify the combinations of obstacles that individuals and communities face—which may differ widely across contexts—and to work across sectors and between levels of organizations and institutions to craft solutions.

      

Health disparities along racial and socioeconomic lines are a persistent challenge for policymakers and researchers looking to improve health outcomes. The COVID-19 pandemic has highlighted and at least temporarily exacerbated these patterns: one estimate indicates that despite comprising only 13% of the U.S. population, non-Hispanic Black people accounted for 34% of COVID-related deaths. The causes driving these vastly different outcomes are numerous, often generational in timescale, sometimes inter-related, and geographically dependent, all of which make effective research and policy responses difficult.

A new book from the National Institutes of Health (NIH) seeks to provide policymakers and researchers tools to close the gap with bold, novel approaches. One such approach is a rapidly advancing and interdisciplinary research framework broadly referred to as complex systems science. This framework is uniquely equipped to grapple with the inherent complexity of health disparities and, as our previous work has shown, other policy research areas concerned with progress toward social justice.

Moving past isolating individual factors

The ultimate goal of health disparities research should be to uncover key drivers and craft policies and practices that can effectively and sustainably address them. Until recently, a challenge in reaching this goal has come from reliance on methods focused on isolating and separately quantifying the contribution of individual factors.

The impact of such analyses is limited in two ways. First, the data necessary to convincingly estimate the effects of individual factors is difficult to find or collect (especially in small populations). Second, and more fundamentally, even when efforts to quantify the contribution of individual drivers of health disparities are successful, they are limited in their ability to contribute to meaningful, sustainable solutions because health disparities are caused by multiple, overlapping factors that often operate on large time scales (i.e. across generations) and differ substantially across individuals and settings. Taking a broad perspective that acknowledges and explicitly grapples with this complexity is the best way to effect positive change.

Embracing complexity in health disparities

Complex systems science provides a way forward in addressing health disparities research, policy, and practice. Researchers and policymakers can use complex systems approaches to explore interacting mechanisms and answer new kinds of questions concerning why observed levels of health disparities occur, which effect pathways or leverage points might matter most, why past or existing policies and interventions have observed effects in a given context, and how novel proposed policies or interventions might affect different communities (including heterogeneous effects and unintended negative consequences). In short, we can address questions about what works, for whom, and why.

For example, our research has shown that observed physical activity patterns are generated by a combination of individual, social, and environmental factors. People’s exercise behaviors are inextricably intertwined: they are more likely to engage in these activities when others do as well because of additional opportunities (e.g. the presence of casual ball games or running groups), social influence (e.g. following the example of friends and family), and collective effects on the built environment (e.g. demand for safe bike lanes in a community). Strategies that explicitly incorporate this insight can be much more effective than ones that concentrate solely on individual-level factors such as information and motivation.

To achieve this wide-angle perspective, complex systems science incorporates a variety of emerging research tools and techniques, including both qualitative and quantitative methodology (and synergistic combinations of both). In our chapter of the NIH publication, we provide an overview of when and how to use these approaches, highlighting topics such as the importance of selecting methods that are appropriate for specific research goals, the role of theory and data, assembling an effective research team, and communicating findings to policymakers and other stakeholders.

Beyond health, towards social justice

Beyond providing a point of entry for researchers and policymakers unfamiliar with complex systems science, this work contributes to a broader conversation that is occurring among economists and other social scientists around research methodology that promotes social justice. The economic and social upheavals caused by the COVID-19 pandemic, coupled with high-profile incidents of police violence inflicted upon Black people, have made it starkly clear that there is both need and appetite for systemic reform. In the best tradition of science, many economists have committed themselves to finding ways to help with this endeavor. Alongside commendable efforts to study the potential impact of large-scale changes in policy and practice, the field is also beginning to revisit  long-standing conventions about theories, assumptions, and methods that can affect our ability to understand and address the structures that perpetuate inequalities.

Over the past several years, we have demonstrated the promising role that complex systems science can play in advancing this goal. For example, we have been able to compare how different policy approaches to tobacco retail control might affect disparities in smoking  across different racial and socioeconomic groups across settings. We have considered the possible effects of different college tuition subsidy programs (sometimes called “free college”) on equitable access to higher education. We have provided rapid-response public health guidance on effectively and equitably responding to the COVID-19 pandemic. In forthcoming work on exposure to HIV transmission, we argue that working across policy areas—specifically, coupling judicial reform with public health interventions—is a promising way to reduce pervasive disparities. And finally, we explore ways in which complex systems science can productively be applied to tackle large-scale, deeply entrenched problems affecting billions of people worldwide by considering the dynamic intersection between obesity, undernutrition, and climate change.

The way forward for researchers and policymakers seeking to reduce health disparities and ultimately move toward a more just society has historically been confounded by complexity. Perspectives and tools derived from complex systems science can reveal new avenues for substantive progress.

      

The COVID-19 pandemic has already killed 500,000 Americans and infected more than 28 million. Although national cases have recently been declining, the next few months are a critical time period that is fraught with uncertainty about the possible introduction and spread of new highly contagious SARS-Cov-2 variants, the pace of the rollout of effective vaccines, and nationwide availability of resources for testing and other containment efforts.

New research from The Brookings Institution Center on Social Dynamics and Policy provides guidance on how policymakers might navigate this critical time and contain a potential fourth wave until vaccination ramps up to levels high enough to provide widespread protection. Working closely with the St. Louis Department of Health, we developed a highly realistic, fine-grained simulation framework that captures empirically specific geography and demography of this metropolitan area of 2.5 million people along with contact patterns driven by where citizens live, work, attend school, and mix socially. We also characterized local distributions of individual-level COVID-19 case properties based on the most current, high-quality empirical research and represented current control measures in place across the region. This new model builds on our previous national-scale simulation for COVID policy decision support.

The project’s goal was to assist the health department in assessing containment policies and practices to minimize cases of COVID-19 in the coming six months–taking into account local conditions, planned and potential vaccine rollout schedules, and multiple scenarios for spread of new virus variants. In addition to the insights for St. Louis itself, several results from the analysis are of broad interest.

• First, we found that control measures which are working well against current variants are unlikely to be sufficient to contain the pandemic if new, more infectious strains of SARS-CoV-2 become widespread—even with vaccination rollout proceeding at the planned pace and assuming continued adherence to social distancing. While the details of both epidemiological conditions and current policies vary across the nation, this finding is a potentially cautionary one for other large cities and for states.
• Second, and echoing our previous national-level work, we find multiple distinct robust policies centered around testing and contact tracing that we project could contain even the more infectious variants of SARS-CoV-2 without requiring renewed large-scale shutdowns of business or schools, underscoring the potential importance of sustained investment in capacity for testing and tracing.
• And third, the model suggests that the spread of more contagious variants of the virus could be strongly curtailed by relatively modest increases in proper and consistent mask usage among the population. Our simulations underscore the importance of the recent CDC guidance on masking.

The new work shows how the national-scale framework we developed previously can be adapted and tailored to provide guidance to state and local policymakers and can be extended to help manage new uncertainties that continue to arise in the COVID-19 pandemic. Our analysis and close collaborative engagement with the St. Louis Department of Health also demonstrates the continued utility of the agent-based computational modeling approach as a decision support tool to assist policymakers at many levels of government in developing highly customized policies in the face of the substantial uncertainty that remains part of managing the COVID-19 pandemic.

Click here to learn more about TRACE-STL >>

The TRACE-STL team includes Ross A. Hammond, Matt Kasman, and Rob Purcell. Learn more about the Center on Social Dynamics and Policy here.

      

Home | About TRACE-STL | TRACE-STL interactive dashboard

TRACE (Testing Responses through Agent-based Computational Epidemiology) is an agent-based computational model developed by a team from Brookings and Washington University in St. Louis, with the specific goal of providing insights into how policies that use testing and contract tracing might help contain the COVID-19 pandemic. It draws on the extensive body of evidence about both the current and past epidemics, and is also designed to manage a high degree of remaining uncertainty about some of the parameters it uses. The version of the simulation model described here has been developed by the Brookings Institution Center on Social Dynamics and Policy in close consultation with the St Louis Department of Health in order to explore how the COVID-19 pandemic might be effectively contained in St. Louis through summer, 2021.

Setting and population

TRACE-STL has been adapted from the original TRACE model to represent the St Louis metro region (St. Louis city and seven surrounding counties, namely Franklin, Jefferson, Madison,  Monroe, St. Charles, St. Clair, and St. Louis counties). The simulated population of approximately 2.4 million individuals is based upon a well-validated “synthetic population” that draws on multiple high-quality data sources (developed by RTI as part of the National Institutes of Health Models of Infectious Disease program) and situated in a geographically detailed representation of the St Louis area. Simulated individuals (“agents”) in our model meaningfully represent residents in city of St. Louis and surrounding counties. Agents have geospatial associations (e.g. where they live and work), demographic attributes, and infection states. Together, these shape the set of other individuals with whom they have social contact in ways that might transmit the virus. This approach allows us to compare cumulative levels and timing of population infection rates across a number of different combinations of active policies and practices in a simulated but highly realistic setting. We summarize our model here, and describe it in complete detail (including mathematical equations, data sources, and computational implementation) in a forthcoming manuscript.

COVID-19 Infection

TRACE simulates transmission of the COVID-19 virus between individuals and the progression of the infection within individuals. Infection progression in our simulations uses a variant of the classic “Susceptible-Exposed-Infectious-Recovered” epidemiological model that is intended to specifically represent COVID-19 (Figure 1).

A flow chart of Covid-19 “states” and possible “state transmissions” in the model

Figure 1. COVID-19 “states” and possible “state transitions” in our model

Simulated individuals who have never experienced a COVID-19 infection (or vaccination) start as “susceptible.” If contact with an infectious individual transmits the virus, then they become “exposed.” After a set incubation period, they become “infectious.” Some individuals have shorter latent periods and are infectious before they display symptoms (i.e. are “pre-symptomatic”), while others never display symptoms (i.e. are asymptomatic). We allow infectivity—that is, one’s ability to transmit the virus to others with whom they interact—to differ across both individuals and infectious type. This allows us to represent the presence of “super-spreaders” who are highly contagious as well as a lower likelihood of non-symptomatic individuals transmitting the disease (e.g. by coughing). Finally, when the infection has run its course, an individual is “recovered.” For the purposes of this model, we assume that anyone in the “recovered” state cannot be re-infected with COVID-19 during the remainder of the simulation (e.g. over the following < 6 months). The degree and duration of immunity conferred by prior infection remain open questions in the scientific literature.

Our simulation incorporates a rollout of vaccinations that reflects current expectations for this endeavor (locally determined rate of rollout, vaccine eligibility criteria, etc) for the time period simulated. Individuals who receive vaccinations are effectively placed in the “recovered” state.

Contact between agents

On any given day, individuals interact with others in ways that might transmit the COVID-19 virus. We simulate multiple settings in which such interactions can occur: in the home, workplace, or school with which each individual is associated as well as contact that might take place outside such settings (e.g. shopping). Contact is based on both a realistic geographic depiction of settings in which each individual spends time as well as an empirically-driven assignment of whom they interact with in those settings (e.g. a school-age child will have a much larger proportion of their daily interactions outside of home and school with other school-age children than with senior citizens, and these interactions are most likely to occur between children who live nearby. Contact in our model also is intended to reflect current conditions: for example, there is county-level variation in the extent of remote learning taking place that influence contacts in school settings. Policy interventions (described below), can alter the contact structure of the population and thus the transmission dynamics.

Model conditions and operation

Each simulation starts with numbers of currently infected and recovered individuals based on recent, zip-code-specific data (adjusted for the possibility of undercounting in these data). The model is allowed to run for six simulated months, effectively representing the course of the pandemic in St. Louis and the surrounding region through early summer, 2021.

Figure 2. Illustrative simulation visualization, St Louis metro region. Agents are color coded by disease state (Blue=Susceptible, Red=Infected, Green=Recovered).

Across model runs, we systematically explore variation in policy and practice to represent different combinations of pandemic containment options. Specifically, we explore different settings for:

• Testing. How many daily PCR or antigen tests are available, who is given priority for testing, and the accuracy of the test technologies used.
• Contact tracing. How many contacts of symptomatic or Covid-positive individuals can be traced (and themselves tested and requested to quarantine) each day.
• Social distancing policies. Restrictions on in-person business activity in St. Louis city or the surrounding region, designed to reduce contact and transmission.
• Mask usage. How many people wear masks outside of the household, and the effectiveness of mask use (i.e. compliance with current CDC recommendations).

In addition, we explore variation in epidemiological conditions to account for potential variation in the extent of current undercounting of active COVID cases or for an increase in infectivity that might result from the widespread introduction of new COVID variants. The full set of scenarios that we explore over nearly 50,000 simulation runs are summarized in Table 1.

Table 1. Policy variations simulated

To see full results from the simulations, see the TRACE-STL Dashboard.

For media or collaborative inquiries regarding TRACE-STL, please contact: Shannon Meraw SMeraw@brookings.edu

TRACE-STL was constructed in close collaboration and consultation with the St Louis City Department of Health.

The Brookings Center on Social Dynamics TRACE-STL team includes:

Ross A. Hammond, Ph.D.,  Director, Center on Social Dynamics & Policy, Senior Fellow, Economic Studies, The Brookings Institution; Betty Bofinger Brown Distinguished Associate Professor, Public Health and Social Policy, The Brown School, Washington University in St. Louis; External Professor, The Santa Fe Institute

Matt Kasman, Ph.D., Assistant Research Director, Center on Social Dynamics & Policy, The Brookings Institution

Rob Purcell, Research Programmer, Center on Social Dynamics & Policy, The Brookings Institution

A scientific manuscript with full documentation for the TRACE-STL model is in preparation and we will update this page with a link to this information shortly.

Home | About TRACE-STL | TRACE-STL interactive dashboard

      

Home | About TRACE-STL | TRACE-STL interactive dashboard

TRACE (Testing Responses through Agent-based Computational Epidemiology) is a collaborative effort by researchers from The Brookings Institution and Washington University in St. Louis to produce a sophisticated computational simulation model to inform policy responses to the COVID-19 pandemic. It draws on the extensive body of evidence about both the current and past epidemics, and is also designed to manage a high degree of remaining uncertainty about some of the parameters it uses.  Previous work focused on U.S. national-level analysis and policy. The version of the simulation model described here has been extended and adapted to represent the St Louis metro area geographically and demographically and is focused on how the COVID-19 pandemic might be effectively contained in St. Louis through summer, 2021. For more details on the model design, see “About TRACE-STL” above.

Overall, we conducted over approximately 50,000 simulations representing over 1,000 different combinations of containment policy options across a range of epidemiological parameters (see “About TRACE-STL” for details). Below, we summarize four key results from our initial analysis of these simulations.

[1] With current infectivity, existing control measures and the planned vaccination campaign are likely sufficient to contain spread

Our simulations show that current containment measures in place in the St. Louis metro region (testing, quarantine, mask protocols, and social distancing) can contain the spread of COVID-19 and continue suppression of the epidemic, if:

• Infectivity remains low (e.g. new, more contagious variants do not become widespread)
• The current rates of testing, test positivity, and adherence to control measures are maintained

Figure 1a. Proportion of residents with active COVID-19 infections in St. Louis city at each time point during the simulated time period. Each blue line represents an individual simulation run that reflects our best available estimates of current epidemiological and containment conditions, with the thickest line indicating the median one. The dashed line indicates the peak value, which might have implications for things such as strain on health care resources.

Figure 1b. Cumulative proportion of residents who experienced COVID-19 infections in St. Louis city during the simulated time period. Each blue line represents an individual simulation run that reflects our best available estimates of current epidemiological and containment conditions, with the thickest line indicating the median one with respect to total cumulative count. The dashed line indicates the total value reached during the median run, which might have implication for outcomes such as incidence of severe cases and mortality rates.

In the simulations, continued control and reduction in numbers of new cases can be attained in this scenario even without widespread closures of businesses or schools. For full details of the simulated assumptions, as well as parallel results for the entire St Louis metro region, see the TRACE-STL Dashboard.

[2] If infectivity increases, additional control measures will be needed to prevent a substantial increase in new cases

Our simulations show that significant increases in infectivity, such as would be expected if new COVID variants become widespread, are likely to undermine existing control measures in the short run, resulting in a large additional wave of new infections over the coming several months. Final cumulative infection rates in such scenarios are projected in the simulations to be 6-10 times higher than today’s levels.

Figure 2a. Proportion of residents with active COVID-19 infections in St. Louis city at each time point during the simulated time period. Each blue line represents an individual simulation run that reflects our best available estimates of current containment conditions and a greater prevalence of more infectious variants of COVID-19, with the thickest line indicating the median one. The dashed line indicates the peak value, which might have implications for things such as strain on health care resources.

Figure 2b. Cumulative proportion of residents who experienced COVID-19 infections in St. Louis city during the simulated time period. Each blue line represents an individual simulation run that reflects our best available estimates of current containment conditions and a greater prevalence of more infectious variants of COVID-19, with the thickest line indicating the median one. The dashed line indicates the value reached during the median run, which might have implication for outcomes such as incidence of severe cases and mortality rates.

For full details of the simulated assumptions, as well as parallel results for the entire St Louis metro region, see the TRACE-STL Dashboard.

[3] Multiple robust policy options exist for increased control of COVID spread, even with higher infectivity

Even with much higher infectivity (e.g. from new variants), our simulations find multiple distinct policy options to effectively control the spread of COVID and keep growth in new infections relatively low. Three are detailed here and a potential fourth is discussed separately below.

[a] Increased investment in regional contact tracing capacity

Our simulations find that increasing the region’s capacity to successfully and quickly trace contacts of confirmed or suspected COVID cases can significantly contain spread even with high-infectivity variants. The graphs below show reductions in cumulative infection rate by >60% by increasing contact tracing capacity ten-fold (policy B vs policy A in Figure 3a), while leaving all other policies unchanged.

Figure 3a.

[b] Increased investment in PCR testing

Our simulations show that increases in the rate of PCR testing in the region would also substantially cut spread of infection. This is true even with all other policies held constant, but is especially effective in combination with increased contact tracing [see policy B vs policy A in Figure 3b]

Figure 3b.

[c] Restrictions on in-person business activity 

Our simulations show that these policies can significantly curtail spread [gold lines below in Figure 3c], but only when undertaken region-wide across the St Louis metro [top panel of Figure 3c] instead of unilaterally in the City [bottom panel in Figure 3c]

Figure 3c.

Due to the high degree of intermixing of contact across city lines with nearby regions in the metro area, simulation show policies must be region wide to maximize impact.

[4] Effective and widespread mask usage continues to be an important means of control over the spread of COVID-19, especially if new variants become widespread

Our model allows for varying assumptions regarding both the level of adherence to mask-wearing (what fraction of the population regularly wears masks when in contact with other people outside their home) and the assumed “effectiveness” of masks in blocking transmission (relative to an “ideal” of well-fitted, properly worn masks in a clinical setting). Simulations show both of these factors can have significant impact on spread, especially when infectivity of the pathogen is high.

Table 1. Cumulative Percentage of St. Louis City Population Infected After 6 Simulated Months across varying combinations of masking assumptions, holding other control measures to our best estimates of current conditions and assuming a greater prevalence of more infectious variants of COVID-19.

This suggests that investments in maximizing proper use of masks across the population over the coming months may be an important policy goal.

The full set of results from our models can be explored using the interactive TRACE-STL Dashboard, a user-friendly tool that allows for adjustment of and comparison across epidemiological and control conditions as well as the selection of viewing city or region-wide outcomes. In the results presented above and in the Dashboard, we focus on overall infection rates (both over time and cumulative). We do this because they have important public health implications (e.g. for health care capacity and incidences of severe or fatal Covid cases). However, because the TRACE model places simulated individuals in a realistic geographic environment, we have the ability to disaggregate these “top level” numbers to identify where Covid prevalence is higher or lower geographically under different conditions.

These two maps show two different scenarios that illustrate the potential importance of considering the geographic location of infections. In the first, overall prevalence appears to be driven largely by a large number of cases in one part of St. Louis. In the second, there are multiple areas that experience large numbers of Covid infections.

Figure 6a and b. Map of St Louis City, with zip code boundaries and shading showing simulated cumulative infection rates

Disaggregated results and spatial analysis will allow for additional work with TRACE-STL to focus on disparities and equity, as well as potential policy options to address these. Additional results will be added to this page as they become available.

Home | About TRACE-STL | TRACE-STL interactive dashboard

      

The COVID-19 pandemic has killed over 300,000 people worldwide to date and has led to large reductions in economic activity as people take steps to protect themselves and as governments implement policies designed to control the virus’ spread. In the United States, these intense public and private social distancing efforts have indeed helped to control spread of infection, but have come with mounting economic costs and potential health risks of their own. Across the country, state and local governments are grappling with decisions about when and how to reopen workplaces, schools, and social venues and how to create an environment in which their citizens feel safe enough to resume these activities.

Decision-makers must balance the imperatives for reopening to restore economic activity, education, and social life against the epidemiological risks of renewed transmission. Reopening too soon, or without the right epidemic control measures in place, is likely to produce additional waves of infection. Experience from other epidemics suggests these waves could produce surges of infection as high or worse than what the country has experienced so far. The science of epidemics tells us clearly that until a large fraction of the U.S. population has immunity—whether via a widespread vaccine or recovery from previous infection—the risk of resurgent infection will not go away. Given that a successful vaccination effort is likely many months into the future, and that we are currently far from widespread immunity, can the risks of reopening be mitigated or managed? Is there a “middle path” between indefinite shutdown and a freely spreading virus with inevitable high tolls of disease? In this essay, I argue that testing and modeling can help us navigate the uncertain terrain ahead.

INVESTING IN TESTING

The answer may lie in testing, as investment in a strong capacity to test Americans (for presence of or immunity to COVID-19) and policies based on testing have the potential to substantially mitigate the spread of infection, facilitating reopening some or all of the shuttered parts of American life while managing risks. Testing could be used in at least three different ways. First, testing to detect active infections could be combined with contact tracing (identifying those who may have been exposed) and quarantine to help contain emerging clusters of disease. Given sufficient test-and-trace capacity, this type of policy could replace some or even potentially all of the mass social distancing measures currently being used to contain epidemic spread. Other countries (including New Zealand, South Korea, Singapore, and Germany) are using versions of this approach already.

Second, even in the absence of contact tracing, sufficiently widespread and accurate testing could be used to adjust social distancing adaptively— turning shutdown measures on or off to respond to resurgent epidemics, exempting those who might be immune from distancing, or emphasizing protective measures for those at highest risk. Some states in the U.S. are already working toward this goal.

Decision-makers must balance the imperatives for reopening to restore economic activity, education, and social life against the epidemiological risks of renewed transmission. Reopening too soon, or without the right epidemic control measures in place, is likely to produce additional waves of infection.

Third, testing can give scientists much more accurate (and much needed) data to inform our understanding of who is at highest risk from COVID-19, how much spread is occurring among people with no symptoms (especially in children), and how much immunity the U.S. population is developing. These data can, in turn, lead to better projections and better planning.

THE CHALLENGE OF TESTING

Testing comes with its own challenges, as to be successful in mitigating an epidemic, any policy involving testing must be carefully crafted. Policymakers designing a testing approach must consider a number of factors. First, at least two different types of tests are available that give different information: whether a person is currently infected or whether a person has antibodies indicating they have had the disease in the past. For either kind of test, a testing regime must consider how many tests can be administered per day, how accurate the tests are, and how quickly the results become available.

In addition, a policy must define who is given the test—for example, to anyone with symptoms, to those employed in essential parts of the economy, to a random sample of the population, or to highrisk groups.

Finally, a policy must define specific containment actions and how the information from the tests will be used as part of these. Actions that may be part of such a policy include quarantining those with active infections, tracing the contacts of those who are sick, releasing those with immunity from workplace closure precautions, and so on. Many of these come with their own considerations such as how much capacity is there to trace contacts quickly, and what fraction of people will adhere to social distancing or quarantine rules.

The success of any particular configuration of testing-based policy in containing disease outbreaks will depend in part on the choices above (features of the policy) and in part on factors outside policymakers’ control (features of how the disease spreads). These factors include what fraction of cases of COVID-19 are asymptomatic, how contagious those with active infections are and for how long, and the length and degree of any protection conferred by antibodies.

THE VALUE OF MODELING

Given the complexity of these choices, and the uncertainty about many of the factors above, decision-makers considering a testing-based policy will benefit from the use of modeling. Quantitative dynamic models have been used effectively as policy tools in many previous epidemics, both to forecast the potential course of spread and as “policy laboratory” to understand the potential consequences of interventions. When confronting such a complex challenge, this type of model offers important advantages for policy design over reliance on either “mental models” (intuition) or the use of data alone. They allow counterfactuals and projections across diverse settings to be considered, past experience and extensive theory from combatting past epidemics to be incorporated, and experiments with many different policy options to be conducted within the model in ways that would not be feasible or ethical in the real world. For some types of model, the kind of diversity (demographic, geographic, social, and medical) that characterizes countries such as the U.S. can be taken into account to yield insights that are not “one size fits all.”

The use of models is not without drawbacks, and it is essential that modeling be used effectively and responsibly, following best practices on the part of both model designers and model consumers. But such models offer enormous potential to help answer key questions in the current situation, such as: How much testing and tracing capacity would it take to relax some or all social distancing without creating a large second wave of infection? What is the best way to use testing across a wide range of scenarios and uncertainty? Quantitatively, how well might various testing policies do—how many new cases would still occur, how quickly, and for whom? Models can also help tune policies to be as efficient as possible—maximizing the degree to which key economic and educational activities can be resumed while minimizing epidemiological risks. An example of the insights that such models can offer for COVID-19 containment policies based on testing can be found in the TRACE project, which evaluates thousands of potential policy configurations to find options that are robust across the enormous uncertainty facing decision-makers.

In the search for a “middle path” forward—a path that avoids the harms to health and the economy of either indefinite lockdown or unmitigated spread of the virus until widespread immunity offers a more lasting solution—testing can play a starring role, and models can help policymakers design approaches that are as effective, as efficient, and as robust to uncertainty as possible.

      

There is increasing political and public interest in college subsidy policies that reduce or eliminate the cost of college attendance for students. Because these programs represent large expansions of the role the federal government plays in higher education, it is worth considering their potential benefits and costs. Proponents of college subsidy programs argue that they will increase access to college for individuals who otherwise would not attend college. It is possible to assess the extent to which this is true for proposed college subsidy programs as well as a number of alternatives. I present four recommendations for evaluating and comparing the possible benefits of different college subsidy programs.

• Decisions about design and implementation should be informed by rigorous, high-quality, and meaningful data analyses of how a given program will affect the composition of colleges and universities, and which students will benefit. I present an example of how this can be done with computational simulation modeling.
• Simulation results indicate that many college subsidy programs may have no substantive impact on enrollment in selective colleges, or might even have unintended, negative effects. For example, simulations of programs like the ones proposed by Senators Elizabeth Warren and Bernie Sanders show that they would slightly increase the proportion of students at subsidized, selective public colleges from the highest income quintile and decrease in students from the bottom three quintiles. Therefore, proponents of college subsidy programs should be careful about overestimating their impact on college access.
• It is important to understand how a program’s design details shape its impact on enrollment. The combination of which students are eligible to be subsidy recipients and at which colleges their costs would be subsidized affects how programs influence enrollment patterns.
• It is advisable to consider a wide range of alternative programs and potential outcomes. College subsidy programs act through changes in demand: who considers attending these colleges and where they apply and ultimately enroll. However, they do not explicitly affect supply: the number of spots available in selective institutions and how those institutions make admissions decisions. It is quite likely that policies that are designed to act on supply (either alone or in conjunction with college subsidy programs) might have a much larger impact on whether and where traditionally disadvantaged students attend college. Such policies might involve a substantial investment in expanding high-quality public university system options or incentivizing colleges to alter their admissions policies (e.g., by making eligibility for college subsidy programs contingent on admitting a certain percentage of subsidized students). In addition, policy researchers can find ways to rigorously evaluate the potential impact of programs on colleges that can respond rapidly (e.g., community colleges) and on how program effects may differ across the country.

College Enrollment Problems 

A number of Democratic Party leaders have advanced proposals for federal programs that address college affordability. Elizabeth Warren and Bernie Sanders have each proposed plans that would eliminate some or all existing student loan debt and ensure “free” college options by covering all tuition and fees (as well as some additional expenses) at public higher education institutions. Pete Buttigieg has proposed a means-tested approach that would provide fully subsidized tuition at public colleges for families with annual incomes up to $100,000 and partially subsidized tuition for families with incomes up to $150,000. Other plans, such as those supported by Vice President Joe Biden and Senator Amy Klobuchar, would cover tuition and fees only at community colleges. Because these proposals (which I refer to from here collectively as “college subsidy programs”) represent, to varying degrees, large expansions of the role the federal government plays in higher education, it is worth carefully considering their potential costs and benefits.

The direct costs associated with attending college are substantial and have grown rapidly over the past several decades. According to a recent College Board report, the average total tuition and fees charged at private, non-profit colleges for the 2019-20 school year is $36,880 (an increase from $23,890 in the 1999-2000 school year, adjusted for inflation) and $10,440 at in-state four-year public colleges (up from $3,510). These figures do not include supplemental expenses such as room and board, nor do they reflect student aid offered through existing grant programs. Generally speaking, nationwide college subsidy programs can be expected to shift much of the cost of college attendance from individual families to the federal government. Smaller scale (i.e., state and local) programs that subsidize college costs have resulted in ongoing, annual direct costs in the tens and hundreds of millions of dollars. Any similar federal college subsidy program can be expected to represent a substantial yearly expenditure. It is thus advisable to carefully estimate expected costs associated with proposed programs and determine ways that program design can deter sharp increases in tuition and fees charged by colleges.

Just as important, if not more so, is identifying who will benefit from college subsidy programs. There is a large and consistent body of research that demonstrates the positive effects of college attendance (especially through degree completion) for individuals on myriad important outcomes including employment, earnings, health, and family stability. These benefits accrue from a combination of credentials, skills gains, mentoring opportunities, and peer relationships that college attendees experience and receive. Proponents of college subsidy programs argue that they will increase access to college, with individuals who otherwise would not attend college being able to reap the benefits associated with a college education. I believe that it is possible to assess the extent to which this is true for proposed college subsidy programs as well as a number of alternatives. I present four recommendations for evaluating and comparing the possible benefits of different college subsidy programs here.

Recommendation 1: Identify potential effects of college subsidy programs on enrollment ahead of time.

Because these policies represent large governmental expenditures and have the potential to alter how Americans think about higher education, both policymakers and the public should have a sense of how a given program will affect access to higher education in general and across different sectors before it is enacted. Decisions about design and implementation should be informed by rigorous, high-quality, and meaningful analyses of how a given program will affect the composition of colleges and universities, and which students will benefit.

Estimating the potential effect of policies is generally difficult, but it is especially challenging here due to the complex nature of the processes that determine college enrollment. Enrollment is the end product of application, admissions, and enrollment decisions. Students and colleges are not independent: admissions and enrollment decisions are inherently zero-sum (i.e., one student’s admission to a selective college implies another’s rejection, and a student’s decision to attend one college precludes enrollment elsewhere). Students and colleges can affect one another’s outcomes and adapt their behavior over time: selective colleges adjust the number of students they admit based on enrollment in prior years, and students adjust their application behavior based on recent admissions outcomes. And students and colleges are not uniform in their attributes and strategies, with these differences having important implications for how a given policy change might affect different colleges and students.

Analyses that extrapolate from observed effects of existing college subsidy programs without explicitly incorporating the complexity inherent in college enrollment are likely to be misleading in two key respects. They may only capture the immediate impact of potential college subsidy programs; these effects might change substantially over time, however, as colleges and students adjust their behaviors to an altered landscape. Secondly, they may inaccurately predict the effects of programs that target different sets of students and colleges than the programs currently in existence. This may occur due to divergence in how different sets of students and colleges respond to subsidy eligibility and because the outcomes for those directly targeted by programs are influenced by the decisions and outcomes of those who are not.

A viable alternative is to turn to approaches such as “agent-based modeling” (ABM), a computational modeling technique that can explicitly simulate individual college and student behavior over time, thus capturing the complexity of college enrollment dynamics. ABMs represent the characteristics and actions of each simulated “agent” (in this case, college applicants and admissions departments) over time, with system-level patterns emerging from an accumulation of micro-level behaviors. These models are inherently dynamic and heterogeneous, allowing individuals with different attributes and behavioral traits to interact with one another and their environment, and to adapt their decision-making in response to these interactions or changes in environment. This approach is increasingly being used to guide policy and program design in areas such as tobacco retail control, childhood obesity prevention, infectious disease control strategies, school choice, and student assignment.

I have helped develop one such simulation model that is grounded in a strong body of evidence (i.e., rigorous empirical literature and our own analyses of nationally-representative data) about how students and colleges make application, admissions, and enrollment decisions. Previous iterations of this model have been used to explore how family resources affect whether and where their children attend college and to compare race based affirmative action policies with race neutral alternatives. My colleague and I were able to use the model to explore the potential impact of different prospective college subsidy programs on enrollment in selective colleges. We restricted our analyses in this way for two reasons. The first is that research suggests that attendance at these institutions has the strongest effect on later life outcomes (e.g., employment), especially for underrepresented minority and low-income students. And the second is that this set of colleges is unlikely to rapidly expand the numbers of students that they enroll because of their reliance on physical facilities and long-term personnel. Less selective institutions, including community colleges and for-profit institutions, generally face fewer of these constraints and may more rapidly expand their supply of available seats in response changes in demand for college, and do so in ways that we are unable to reliably predict without making a number of assumptions. The model is calibrated such that it represents college subsidy programs that would completely cover tuition and fees for eligible students attending eligible colleges.

I consider this model to be an example of an analytic tool that is rigorous, high-quality, and can meaningfully explore potential college subsidy program effects. The model represents three serial processes that occur in each simulated year: application, admission, and enrollment. Prospective students submit applications to a limited set of colleges, attempting to maximize their expected outcomes (i.e., secure admission at the most desirable colleges possible). Subsidized tuition at a college makes that option more attractive for potential recipients. College admissions departments admit a set of students who they consider to be the best candidates, selecting a sufficient number to fill out their freshman class given recent enrollment yield. Finally, students enroll in the most desirable college to which they have been admitted (with subsidized tuition again making a college more attractive for potential recipients). This model is outlined in more detail in a recent report.

Recommendation 2: Do not overestimate program impact prior to implementation, and be wary of the potential for unintended, negative consequences.

We used our model to conduct a series of “policy experiments.” That is, we compared simulated college enrollment after the implementation of 36 different hypothetical college subsidy programs to the current college enrollment landscape. The enrollment outcomes that we focused on were:

• Whether students enroll in any selective college
• Whether students enroll in a selective college where enrollment would be subsidized
• Whether students enroll in an “elite” college (i.e., the top 20% of selective colleges)

In the dynamic visualization linked here and featured below, we present changes in each of these outcomes for potential subsidy recipients overall as well as disaggregated by resource quintiles.[1]

The program design elements that we varied were:

1. Subsidy scale. The proportion of eligible students who are randomly selected to be offered college subsidies. The options are: 10%, 50%, and 100%.
2. Student eligibility. What is used to determine whether a student is eligible to be offered college subsidies. College eligibility criteria options are: Achievement, family income, both achievement and family income, and neither.
3. College type. Colleges where attendance would be subsidized for participating students. Options include: Least selective colleges, in-state public colleges, and all colleges.

Effects of programs on potential recipients’ enrollment

Our results indicate that a number of large-scale programs (i.e., ones that represent programs enacted at the federal level) would have little effect on who enrolls in selective colleges or even in which colleges students will enroll. In addition, policy specifications may result in unintended and undesirable effects, such as a decrease in recipients’ attendance at selective or elite schools (in turn affecting the composition of student bodies at prestigious institutions) or localization of program benefits among students who are already relatively advantaged. For example, simulations of programs like the ones proposed by Senators Warren and Sanders show that they would slightly increase the proportion of students at subsidized public colleges from the highest income quintile and a slight decrease in students from the bottom three quintiles. Therefore, proponents of college subsidy programs should be careful about overpromising their impact on college access. Doing so is likely to disillusion the public and prompt a backlash that could endanger future policies and programs that might be effective in reducing disparities in college enrollment.

Recommendation 3: Consider the role of program design details in shaping impact on enrollment.

The results displayed above represent a complete exploration of the following combinations of college eligibility, student eligibility, and program scale:

• College eligibility. Three college eligibility scenarios that reflect the colleges at which students would receive subsidized tuition. We simulate programs that represent ones where all colleges are eligible, so students receive subsidies even at the most elite private institutions; programs where subsidies are restricted to less selective public colleges, which may be thought of as public colleges that admit 75% to 90% of applicants (i.e., public colleges excluding most state flagships); and programs in which tuition would be subsidized at public institutions in a student’s state (similar to existing and proposed state-level college subsidy programs).
• Student eligibility. We explore four different eligibility criteria that determine whether students are restricted from receiving subsidies based on achievement (roughly equivalent to ones where the subsidy is available only to students with GPAs of at least 3.0), family resources (where the subsidy is available to low- and middle-income students), both achievement and resources, or neither. These choices are based on common differences in real-world subsidy criteria. For instance, West Virginia PROMISE selects entirely on merit; the federal Pell Grant program selects on need; California’s Cal Grant considers both merit and need; and many local Promise programs have no merit or need requirements.
• We vary the proportion of eligible students who actually receive offers of subsidized tuition. This is intended to represent the difference between programs similar to many of the smaller-scale subsidy programs that have been implemented to date (e.g., at the state level) and the universal elimination of tuition and fees for eligible students through a federal program. I focus the rest of my comments here on the latter case (i.e., one in which 100% of eligible students receive subsidies).

Several patterns emerge from my analysis:

• Effect sizes and directions. When subsidies have observable effects, these are experienced most strongly as increases in enrollment in subsidized colleges, then increases in selective colleges overall, and finally as decreases in enrollment in elite colleges. Subsidized college options attract students to apply and enroll in them, creating observable direct effects for eligible students (i.e., take-up of subsidies through attendance in subsidized schools).

Subsidies can also have indirect effects for these students: they change potential recipients’ application strategies (which colleges they consider and apply to) and enrollment decisions. This can result in increases in attendance in selective colleges overall in some cases (e.g., when enrollment in the least selective colleges of those considered is subsidized and eligibility is based on both resources and achievement) and decreases in others (e.g., when subsidies are offered to high-achieving students to attend any public college). It can also result in slight decreases in eligible students’ attendance at elite schools (e.g., when enrollment at public colleges is subsidized and eligibility is based on both resources and achievement).

Generally, direct subsidy effects are greatest for higher-resourced recipients, while indirect effects tend to be weakest for that same set of students. Students with greater resources are in a better position to take advantage of subsidies by gaining admission to subsidized schools but are also less likely to shift their application and enrollment strategies in ways that affect whether they attend any selective college.

• Student eligibility. Subsidy effects are greatest when restricted by both student achievement and resources. This is because higher achieving students are more likely to be admitted to selective colleges to which they apply and students from families with fewer resources are more responsive to college subsidies, and so the effect of subsidies on their enrollment is more pronounced.
• College eligibility. Subsidy effects are greatest when restricted to public colleges and weakest when all colleges are subsidized. The former condition induces potential recipients to substantially change their application and enrollment behaviors. Conversely, when all colleges are subsidized, students’ application behavior is only marginally affected (i.e., as subsidies only induce students to consider a somewhat wider set of schools) and enrollment not at all.

In addition to these general patterns, policymakers can also make use of simulated effects under specific combinations of program conditions to make decisions about whether and how to engage in program design for desired outcomes. For example, if the primary goal of a large-scale federal program is to maximize access to subsidized tuition at selective colleges for low-income students (i.e., those from families in the lowest income quintile), our simulation results suggest that they may wish to explore a targeted program that provides subsidized tuition at any public college, and restricts eligibility by both family resources and achievement.

Recommendation 4: Explore a wide range of policies, combinations of policies, and policy effects

Although the results above represent a moderately large number of simulated policy experiments, this exploration of policy options was far from exhaustive. The tool discussed here (or something similar) has the potential to provide additional insight into these policies as well as a much broader set of potential policies and combinations of policies. I want to highlight three areas that I believe can be productively explored further when making decisions about programs to propose and implement:

• Explore policies that act directly on colleges. It is perhaps unsurprising that our simulations showed a limited impact of large-scale college subsidy programs on enrollment in selective colleges. These programs exclusively act through changes in demand: who considers attending these colleges and where they apply and ultimately enroll. However, they do not explicitly affect supply: the number of spots available in selective institutions and how those institutions make admissions decisions. It is quite likely that policies that are designed to act on supply (either alone or in conjunction with college subsidy programs) might have a much larger impact on whether and where traditionally disadvantaged students attend college. Such policies might involve a substantial investment in expanding high-quality public university system options or incentivizing colleges to alter their admissions policies (e.g., by making eligibility for college subsidy programs contingent on admitting a certain percentage of subsidized students).
• Estimate potential policy effects on enrollment in less selective institutions (e.g., community colleges). The simulation tool noted here is only designed to reflect enrollment in selective colleges. As discussed above, this is both because of evidence that attending this set of schools has the largest positive impact on student outcomes (e.g., future employment) and because of uncertainty about how less selective institutions might rapidly respond to the introduction of subsidy programs. However, there are proposed policies that are specifically targeted at less selective colleges. In addition, it is likely that other college subsidy programs will substantially affect enrollment in less selective colleges, with important implications for student behavior and program costs. Therefore, policy researchers may want to find ways to expand upon this simulation tool (or one like it) to estimate these effects, even if they must acknowledge greater uncertainty when they do so.
• Understand how effects may vary geographically. There is meaningful variation in how colleges are geographically distributed relative to prospective students. In addition, there a number of existing state and, to a lesser extent, local-level policies in place that affect the college enrollment process. These can both have important implications for the impact of large-scale college subsidy programs. Benefits may be localized or experienced at different levels across different groups of students. Policymakers should consider these possibilities during design and implementation.

Summary

 There is increasing political and public interest in college subsidy policies that reduce or eliminate the cost of college attendance experienced by students. Because these programs represent large expansions of the role the federal government plays in higher education, it is worth considering their potential benefits and costs. Proponents of college subsidy programs argue that they will increase access to college for individuals who otherwise would not attend college. It is possible to assess the extent to which this is true for proposed college subsidy programs as well as a number of alternatives.

To do so, I recommend turning to rigorous, computational simulation models that embrace the heterogeneity and interdependence inherent in the college enrollment process. I highlight one such model here. Models like this one can provide important guidance prior to program implementation. They can help policymakers explore a wide array of potential program designs and identify whether programs might result in substantial positive impact, where there is the potential for unintended negative consequences, and how program effects can vary across students and colleges. In addition, these models are highly extensible: they can be productively built upon and used as the basis for additional analyses of program design options and potential outcomes.

      

States across the U.S. are considering paths to re-opening following months of stay-at-home orders and a widespread shuttering of the economy in response to the threat of COVID-19. Policymakers now face the task of crafting strategies that will allow resumption of activity without producing additional waves of infection that could do even more damage to health and economies. Looking to successful examples from around the world, many of these strategies involve widespread testing, coupled with contact tracing and selective quarantine. Yet many questions that are important in designing such policies remain difficult to answer. Getting the answers “right” may be the difference between successful containment or a damaging resurgence of infection. Some of these questions include:

• How effective can a test-and-trace policy really be in containing future waves of infection? Is such an approach feasible in the United States?
• How much testing capacity is needed for effective containment? How much capacity to trace contacts will be needed?
• How accurate must tests be?
• What is the most efficient way to use limited testing capacity?
• How might success depend on still-uncertain assumptions about the spread of the disease itself?
• What social distancing measures might still be needed to enhance containment?

To help answer these questions and provide specific guidance to decision-makers, we offer new analysis based on a model entitled TRACE (Testing Responses through Agent-based Computational Epidemiology), developed collaboratively by researchers at Brookings and Washington University in St Louis. Unlike many other COVID-19 models, TRACE is not a forecasting model. It is intended instead as a policy laboratory to assist in the design of effective containment policies using testing and contact tracing. By considering a very wide array of possible policy variations, capturing scenarios that encompass the extensive uncertainty still surrounding COVID-19, and providing specific quantitative inputs and outcomes, TRACE aims to be a practical tool to help decisionmakers manage many of the implementation decisions they face in crafting a re-opening strategy. TRACE is an agent-based computational model, allowing it to include variations in age, activity pattern, infectivity, and contact networks—all features that evidence so far suggests are important determinants of how COVID-19 spreads.

Our analysis based on the TRACE model identifies promising intervention strategies to successfully suppress the spread of COVID-19 while allowing relaxation of many or all of the mass social-distancing measures that have been in place across the country. Suppression means not just “flattening the curve” (by spreading out infections over time) but ongoing containment that curtails sustained spread and prevents large numbers of new cases. A primary goal of TRACE is to identify policies that yield true suppression of the epidemic while gradually relaxing social distancing.

Our results also suggest that while the ability of policies based on testing and contract tracing to effectively suppress epidemic spread can depend on the setting and timing—a single policy does not necessarily suit all circumstances—some policies are highly robust to the uncertainty facing policymakers about the underlying biology of COVID-19. All of the policies we simulated also underscore the importance of sufficient adherence by individual citizens to quarantine, self-isolation, or limited social distancing measures. This indicates that an important goal for policy may be to encourage adherence through consistent, widespread messaging, and to make self-isolation financially and logistically feasible.

The goal of re-opening large parts of the country while suppressing COVID-19 and preventing a large second wave of infection may well be possible, and not that far out of reach from our current capabilities. But to do this, we will likely need to refocus and re-orient our current approach to make best use of limited resources, and we will need to tailor features of any specific policy implementation to local conditions to maximize chances of success. TRACE was designed to help facilitate this process, and we hope it will prove a valuable resource for decision-makers working to bring our country through this crisis.

The TRACE team includes Ross A. Hammond, Matt Kasman, Joseph T. Ornstein, and Rob Purcell. You can find more information about the model here.