This project will examine the short-, medium-, and long-term health, social and economic effects of the COVID-19 pandemic on an underserved and vulnerable population of young adults who have been part of the Fragile Families and Childhood Wellbeing Study since birth along with their primary caregivers who are entering middle age. Moreover, we will incorporate planned linkages to multilevel COVID-19 related health, social, and economic measures to identify how temporal and geographic variations contribute to outcomes of the two-generation sample.

Behavioral scientists aim to understand human behavior at the level of the individual (psychology), the level of the group (sociology), and within specific political and economic contexts (economics, political science). Social media data provide a potentially important avenue for learning about human behavior in real time by tapping multiple aspects of an individual’s beliefs, behaviors, emotions, and social networks. Such “data in the wild” provide a lens into human behavior and attitudes not available through traditional survey research. However, it is often unclear how to reconcile the behavioral processes that generate the millions of data points available with the carefully designed measurement strategies that produce typical social scientific data or how these data can yield the clear metrics of validity and reliability necessary for behavioral research. Such data are also rarely protected by the types of ethical safeguards provided to survey respondents. Hence, although these data may offer important new avenues for understanding human beliefs and behavior, harnessing these data in ways useful to social scientists is a major scientific challenge.

Computer scientists know how to “tame” these data — clean them, mine them, and make sense of them. They are leaders in using social media to investigate current events, such as tracking behavior related to social movements and virus outbreaks across the globe. The way they approach the data, however, does not typically adhere to established social science research designs or incorporate rigorous checks of the data’s validity or reliability as it relates to the individuals providing the data or the constructs measured. They also do not focus on understanding the generalizability of the data to the population at large, but instead target specific descriptive and learning tasks, making some of their methods, algorithms and results less useful to the broader research community. In order to extract significant research value from social media data, computer scientists and social scientists must integrate their expertise — or converge — to create and adapt computer science algorithms and data mining methods in ways that adhere to the design structures, measurement rigor and ethical protections of social science.

Big data featuring neuroimaging information collected from large population-based samples have spurred the emergence of population neuroscience research. However, traditional methods for neuroscience research are based on nonrepresentative samples that deviate from the target population, such as convenience and volunteer samples. The lack of representativeness may distort association studies of brain-cognition mechanisms.

The research team’s collaborative work on the Adolescent Brain Cognitive Development Study motivated this project, which presents these common problems in empirical neuroimaging studies, to fill the gap in statistical methodology between survey and neuroscience research. The team develops new strategies to adjust for nonrepresentativeness in association studies with complex and nontraditional survey designs, and to quantify the potential impact of sampling features on statistical and substantive inferences.

The overall objectives are to identify population heterogeneity in the association studies between imaging and cognitive ability measures and generalize multilevel regression and poststratification as a robust framework for inferences based on nonprobability samples. The software delivery with computational scalability and step-by-step guidelines will provide practical recommendations and tools to map the relationships and adjust for selection bias when making population inference. This interdisciplinary project will strengthen the validity and generalizability of population neuroscience research, deepen new association understandings of brain and cognition, and facilitate policy intervention.

We are only beginning to clarify the ways the COVID-19 pandemic has resulted in substantial changes to American neighborhoods. There has been an excess of permanent business closures, particularly among small neighborhood businesses most vulnerable to social distancing, such as local barbershops and nail salons. COVID-19 outbreaks in late September 2021 caused 2,000 neighborhood schools to close for an average of six days in 39 states.

A burgeoning body of research has tried to understand the forces driving these trends, focusing on infectious disease transmission at the individual level or economic models at the business level. What is not considered is the context in which these changes are taking place. By context, we mean the neighborhood community environment that holds the opportunities, restrictions, risks, and flexibility for post-pandemic growth. The community environment includes:

1. Job opportunities in business sectors robust to social distancing;
2. Comprehensive broadband internet access to facilitate telemedicine, online schooling, remote work, and online grocery shopping;
3. Parks and walkable streets to facilitate socially distanced physical activity and social interaction to mitigate social isolation brought on by the pandemic; and
4. The provision of medical care through the availability of alternate health care providers and pharmacies.

Access to these neighborhood resources is not equally distributed across America, reinforcing risk for vulnerable populations, including older adults, children and adolescents, racial/ethnic minorities, and those in rural areas. However, a lack of national, standardized, longitudinal metrics of the local neighborhood environment has hindered the ability to identify which communities are most vulnerable to the immediate and longer-term consequences of the pandemic for a host of behavioral, psychological, social, and economic outcomes.

To address this limitation in the nation’s data infrastructure, we will augment, curate and disseminate data from our National Neighborhood Data Archive (NaNDA). This dataset includes a wealth of physical, social and economic characteristics of the local neighborhood across the United States (e.g., racial segregation, business density, environmental hazards, broadband internet access, and healthcare availability), in the years both before and since the pandemic. We will participate with the Consortium on Social, Behavioral, and Economic Research on COVID-19 to integrate, share, and analyze spatially referenced neighborhood data that can be readily linked to existing survey data, cohort studies, or electronic health records at various levels of geography. We will work with the COVID-19 Consortium Coordination Center to identify and create key neighborhood metrics that are priorities for research teams in the Consortium, including a set of common data elements (CDEs) on the social, behavioral and economic indicators of the COVID-19 pandemic at the neighborhood level. We will also develop new metrics of longitudinal neighborhood change in the decades preceding the pandemic, which can inform community risk and resilience since the pandemic.

Alzheimer’s disease and related dementias (ADRD), a leading cause of disability among older adults, has become a critical public health concern. The clock-drawing test, which measures multiple aspects of cognitive function including comprehension, visual spatial abilities, executive function and memory, has been widely used as a screening tool to detect dementia in clinical research, epidemiologic studies, and panel surveys. The test asks subjects to draw a clock, typically with hands showing ten after 11, and then assigns either a binary (e.g. normal vs. abnormal) or ordinal (e.g. 0 to 5) score. An important limitation in large-scale studies is that the test requires manual coding, which could result in biases if coders interpret and implement coding rules in different ways.

Several small-scale studies have explored the use of machine learning methods to automate clock-drawing test coding. Such studies, which have had limited success with ordinal coding, have used methods that are not designed specifically for complex image classification and are less effective than deep learning neural networks, a new and promising area of machine learning. More recently, machine-learning methods have been applied to digital clock-drawing testing, a form of the clock-drawing test that uses a digital pen and tablet. Despite some promising results on small-scale data, thus far these studies have only attempted to code binary categories.

This project develops advanced deep learning neural network models to create and evaluate an intelligent clock-drawing test Clock Scoring system – CloSco – that will automatically code test images. We will use a large, publicly available repository of clock-drawing test images from the 2011-2019 National Health and Aging Trends Study (NHATS), a panel study of Medicare beneficiaries ages 65 and older funded by the National Institute on Aging. Specifically, we will:

1. Develop an automated clock-drawing test coding system for both ordinal and continuous scores;
2. Evaluate the performance of the CloSco system and investigate the value of continuous scoring for dementia classification and longitudinal test models; and
3. Prepare and disseminate NHATS public-use files and documentation with ordinal and continuous clock-drawing test codes assigned using CloSco along with the CloSco deep learning neural network program.

If successful, the DLNN programs may offer a model for automating coding of other widely available drawing tests used to evaluate a variety of cognitive functions.

What do the teams that produce science and the networks in which they are included look like? How is credit allocated within them? Previously, analyses performed using co-authorship data have largely shaped our knowledge of these topics. . However, not all people who make important contributions to research projects appear as coauthors on all articles and some author positions are more prestigious than other positions. Evidence suggests that women and members of underrepresented racial and ethnic groups are disadvantaged in terms of author position. The same is likely true for research staff.

This project combines UMETRICS data covering over 23 million payroll transactions to over 175 thousand people paid on NIH projects at 31 universities which accounts for over one-third of federally funded, academic R&D with Torvik and Smalheiser’s updated Author-ity database of publications by biomedical researchers. The combined data allow us to analyze scientific collaboration networks, as distinct from coauthorships, and shed new light on relatively marginal populations in biomedicine. The interdisciplinary team combines emerging and established scholars and has successfully developed, analyzed, and distributed both datasets.

We will describe the composition of the teams and networks supported on research projects, answering questions such as: How large are the teams that actually conduct research? What factors relate to the size of teams? What types of people work on them in terms of gender, race, ethnicity and age, and in what capacities? How are each of them positioned in collaboration networks? We further study the association of researchers’ characteristics such as gender, race, ethnicity, age, and job title to the credit that they receive for their work. Lastly, we will study how these relationships depend on the gender, race, ethnicity, and age of the Principal Investigator (PI) and on funding mechanisms — for example, whether women and younger PIs are more inclusive and egalitarian than men or if some mechanisms are better in terms of inclusion. These questions are increasingly urgent as scientific teams expand but require information on who actually works on research projects beyond data on authorship alone.

The Longitudinal Study of American Youth (LSAY) is a three-cohort two-generation longitudinal study of national samples of public school students in the United States. The two original cohorts consisted of national probability samples of 7th and 10th grade students selected in 1987. These young adults are now 37 to 40 years of age and reside in all 50 states of the U.S. With continued support from the National Science Foundation, the LSAY will launch a new cohort of 7th grade students in the fall of 2015. The new cohort (called Cohort 3) will be exactly one generation younger than the students in the 1987 cohorts, allowing a generational comparison of changes in American family life, schooling, and society. The original cohorts were designed to study the factors related to student interest in science and mathematics, the development of skills in those disciplines, the selection of careers, and the development of sufficient scientific literacy to perform citizenship responsibilities in a democratic society. Cohort 3 will explore the same questions over the next two decades. The first 20 years of LSAY data are available through the Inter-university Consortium for Political and Social Research (ICPSR).