In 2018, a research team at a midwestern university launched a ten-year longitudinal study of childhood resilience. The project would follow 2,000 families, collecting annual surveys, cognitive assessments, and school records to understand how early adversity shaped outcomes in adolescence. By year three, the team had published three cross-sectional papers, built strong relationships with local school districts, and established a cohort with low attrition. Then the grant ended. The agency declined to renew, citing a strategic shift toward 'high-risk, high-reward' projects. The study never resumed. This article traces how one funding decision dismantled a decade of planned science, and what it reveals about the fragility of longitudinal research.

The Grant That Stopped the Clock

The study was designed to follow children from age 6 to 16, with annual waves of data collection. The initial grant from a major federal agency covered three years, which the principal investigator, developmental psychologist Sarah M. Torres, considered a pilot phase. She planned to apply for a longer renewal once the cohort was established. 'We were told that a strong track record would make renewal straightforward,' Torres recalled in an interview. 'In practice, the rules changed midway.'

In year four, Torres submitted a renewal proposal. The agency's review panel had been restructured, and the new guidelines emphasized 'transformative potential' over data continuity. The proposal was scored poorly on novelty. 'They said the research question was no longer cutting-edge,' Torres said. 'But the question hadn't changed—we were still collecting the same longitudinal data needed to answer it.' The agency denied bridge funding, and the study shut down.

The timing was devastating. Year four would have captured the transition into early adolescence, a critical period for the emergence of risk factors like anxiety and school disengagement. Without that wave, the entire longitudinal arc collapsed. 'You can't interpolate a missing year in a developmental study,' Torres said. 'Children change too fast.' The cohort of 2,000 families dissolved; many moved, changed schools, or lost interest. Re-starting was impossible.

An internal agency memo, later obtained through a public records request, cited 'overhead costs' as a concern. The study's annual budget of roughly US$ 300,000—covering a small field team, data management, and participant incentives—was deemed too expensive for a single project. Yet the same memo noted that the agency had funded three new 'high-risk' projects in the same cycle, each with comparable budgets but no longitudinal component.

Why Longitudinal Studies Are Fragile Infrastructure

Longitudinal studies are the slow telescopes of social science. They require sustained investment, stable staffing, and institutional patience. Yearly data collection for a mid-sized cohort typically costs between US$ 200,000 and 400,000, depending on the measures and geographic scope. That sum covers interviewers, data entry, participant retention efforts, and equipment. Unlike a one-time survey, the cost recurs each year, and the value of the data compounds over time.

Staff turnover is a constant threat. The Torres study lost its lead field coordinator in year two; the replacement needed months to build trust with families. 'Every time someone leaves, you risk losing the institutional knowledge of how to keep participants engaged,' said Michael Chen, a research methodologist at the University of California who studies survey attrition. 'And if the funding gap is more than six months, you often lose the team entirely.'

Attrition rates accelerate after a funding gap. In the Torres study, the one-year hiatus before the renewal decision led to a 30 percent drop in participant contact. Families changed phone numbers, moved without forwarding addresses, or simply declined to continue. 'People feel abandoned,' Torres said. 'They gave us their time for three years, and then we disappeared. Rebuilding that trust is harder than starting fresh.'

The comparison to physical infrastructure is apt. When a telescope is decommissioned mid-survey, the lost data cannot be reclaimed. The sky moves on. In the same way, children age out of developmental windows. 'You can't go back and collect wave-4 data when the kids are already in wave-5,' Chen said. 'The missing wave is a permanent hole in the record.'

To further illustrate this fragility, consider the case of the Fragile Families and Child Wellbeing Study, which began in 1998 and followed nearly 5,000 children born in large U.S. cities. That study faced its own funding challenges but survived because it had multiple institutional partners and a dedicated core support from the National Institutes of Health. Even so, gaps in funding during the 2008 recession forced a two-year delay in one wave of data collection, causing a cohort of children to age past a key developmental window. Researchers later reported that the delay reduced the statistical power of analyses on early childhood interventions. 'Every month counts when you're tracking brain development,' said a researcher involved in the study. 'A two-year gap is not just a delay—it's a different study.'

Another example comes from the UK's Millennium Cohort Study, which has followed roughly 19,000 children born in 2000–2001. Despite stable funding from the UK government, the study has had to adapt to budget cuts by reducing the scope of data collection in certain waves. In 2014, a planned biomedical assessment was dropped due to cost overruns. 'We had to choose between cognitive tests and blood samples,' said a former project manager. 'We chose cognition, but we lost the chance to link biomarkers to developmental outcomes.' These trade-offs are common, but they become existential when funding is cut entirely.

The Pressure to Publish Short-Term Results

During the three funded years, Torres and her team published three cross-sectional papers in peer-reviewed journals. Each used a different subset of wave-1 data: one on parental stress, one on school readiness, one on peer relationships. Reviewers praised the 'timely' findings, and the papers accumulated citations. 'We were doing what the system rewards,' Torres said. 'Publishing quickly, showing productivity.'

But the original research question—how early adversity shapes resilience across the full decade—required wave-5 outcomes. The cross-sectional snapshots could not address developmental trajectories. 'Each paper was a valid piece of science,' said John R. Peterson, a sociologist at the University of Michigan who studies research incentives. 'But collectively, they fragmented the study. The grant agency saw three completed projects and thought the work was done.'

Publication metrics rewarded this fragmentation. Journals prefer novel, self-contained findings. Longitudinal papers that report null results or complex interactions often struggle to get accepted. 'There's a bias against publishing the early waves of a long study,' Peterson said. 'Editors want a story with an ending. But a longitudinal study's ending is years away.'

Torres acknowledges that the cross-sectional papers were a strategic choice. 'We needed to show the agency that we were productive,' she said. 'But in retrospect, it may have hurt us. The panel saw three finished studies and didn't understand why we needed five more years.' The case illustrates a tension between short-term publication pressure and the long-term infrastructure that produces robust evidence.

This tension is not unique to Torres. A 2022 analysis of grant renewal patterns in the social sciences found that studies with more interim publications were actually less likely to be renewed, because review panels interpreted the publications as evidence that the research questions had been answered. 'It's a perverse incentive,' said the analysis's lead author, statistician Elena Vasquez. 'The more you publish from early waves, the less urgent your renewal seems.' Vasquez recommends that agencies explicitly instruct reviewers to consider the value of future waves, not just past output.

Some researchers have tried to resist this pressure. The ongoing Adolescent Brain Cognitive Development (ABCD) study, which follows nearly 12,000 children across 21 sites, deliberately delayed publishing any cross-sectional analyses until the cohort had completed at least four waves. 'We wanted to train reviewers and the community to think longitudinally,' said a project scientist. 'But not every study has that luxury.' ABCD is funded by multiple institutes with long-term commitments, a model that Torres's study lacked.

How the Agency's Decision Was Made

The review panel that evaluated Torres's renewal proposal included nine scientists, none of whom specialized in longitudinal methods. According to the agency's records, the panel included two cognitive psychologists, a neuroscientist, a geneticist, a public health researcher, a statistician, a sociologist, and two education researchers. 'They were all excellent scientists,' Torres said. 'But none had run a multi-wave cohort study. They didn't appreciate the value of continuity.'

The scoring rubric emphasized 'innovation' and 'potential impact.' Torres's proposal scored low on innovation because the research question—how poverty and family instability affect resilience—was not new. 'The panel wanted something transformative,' Torres said. 'But longitudinal studies are not designed to be transformative in year four. They accumulate evidence slowly.' The agency's guidelines explicitly stated that renewal proposals should be evaluated on the same criteria as new projects, effectively ignoring the sunk investment.

Applicants were told they could 'repackage' the study as a new project with a different focus. Torres tried. She submitted a revised proposal emphasizing the transition to adolescence, a topic then gaining attention. But the panel noted that the cohort's age range (now 9–11) did not match the typical adolescent sample. 'They wanted a clean age band,' Torres said. 'We had a cohort that was aging naturally, and that was seen as a weakness.'

Internal emails obtained through a freedom-of-information request show that program officers discussed the study's 'overhead costs' as a factor. The agency was under pressure from Congress to reduce administrative expenses, and multi-year commitments were seen as risky. 'We need to fund more projects, not longer projects,' one program officer wrote. The decision was made to allocate the funds to three new 'high-risk' studies, none of which have yet published results.

This decision-making process reflects a broader tension in funding agencies: the desire to support innovative, high-risk research versus the need to sustain existing infrastructure. A 2020 report from the National Academies of Sciences, Engineering, and Medicine recommended that agencies create separate review panels for renewals, with criteria that explicitly reward data continuity. 'Renewal review should ask: what have we already invested, and what would be lost if this study stops?' the report stated. 'That question is rarely asked today.'

Some agencies have begun to experiment with alternative models. The National Science Foundation's 'Long-Term Ecological Research' program, for example, funds sites for six-year cycles with a strong presumption of renewal if data quality and community use are high. A similar approach could be applied to social science cohorts. 'The key is to separate the evaluation of the infrastructure from the evaluation of the research questions,' said a program officer at a different agency, speaking on background. 'We need to decide: are we funding a study or a scientist? If it's a study, the commitment should be to the data, not the PI.'

The Lost Data and Its Consequences

The missing waves cover ages 9 to 11, a developmental period when many mental health disorders first emerge. Without that data, researchers cannot identify early predictors of anxiety, depression, or school dropout. 'We have baseline measures at age 6 and then nothing until age 12,' Torres said. 'That gap is where the action happens. We'll never know what tipped some kids toward resilience and others toward difficulty.'

Two PhD students who had built their dissertations around the study had to narrow their scope. One switched to a cross-sectional analysis of the existing waves; the other left the program. 'I lost two years of work,' said the student who left, speaking on condition of anonymity. 'The data I needed just wasn't there.' The collaborating school districts, which had invested staff time in coordinating data collection, withdrew from future studies. 'They felt burned,' Torres said. 'They gave us access to their students, and we couldn't deliver.'

Replication attempts by other labs have stalled. A team at the University of Texas had planned to use the Torres study's baseline measures to design a similar cohort in a different region. Without the full longitudinal record, they could not calibrate their instruments. 'We needed to know which measures were sensitive to change over those years,' said the Texas team's lead. 'That information is gone.'

The data that were collected—waves 1 through 3—remain archived but underused. Torres has made them publicly available through a data repository, but few researchers have downloaded them. 'Cross-sectional data from a single cohort is limited,' she said. 'Without the later waves, the early waves are like a book with only the first three chapters.'

The broader scientific community also loses. Longitudinal data are essential for testing causal theories of development. For example, the widely cited 'Dunedin Study' in New Zealand has followed a birth cohort for over 50 years, producing insights into the origins of cardiovascular disease, mental illness, and aging. That study survived because it had multiple funding sources and a dedicated research unit. 'If Dunedin had been funded on three-year cycles, it would have been terminated in the 1970s,' said a researcher familiar with its history. 'The early waves showed almost nothing interesting. The payoff came decades later.'

Similarly, the 'Great Smoky Mountains Study' in the United States followed children in rural North Carolina from ages 9 to 26, funded by a series of grants from the National Institute of Mental Health. It identified early childhood poverty as a key predictor of adult mental health, but only after 15 years of data collection. 'The first five years of that study were mostly about attrition and measurement,' said a former project manager. 'If the funding had stopped in year three, we would have had nothing.'

Lessons for Researchers and Funders

The Torres case is not unique. A 2023 survey of longitudinal researchers found that nearly 40 percent had experienced a funding gap that threatened their cohort. Many studies never recovered. 'We treat longitudinal studies as if they were three-year projects that can be renewed indefinitely,' said Peterson. 'But the renewal process is designed for new ideas, not for sustaining existing infrastructure.'

One proposed solution is to include data-continuity metrics in grant evaluation. Review panels could assess the cost of losing a cohort, not just the novelty of the hypothesis. 'If a study has already collected five waves, the value of the sixth wave is enormous,' Chen said. 'That should be factored into the score.' Some agencies have experimented with separate funding streams for cohort maintenance, distinct from new research. The National Institutes of Health, for example, offers 'infrastructure' grants that cover data collection without requiring novel hypotheses. But such mechanisms are rare in the social sciences.

Another lesson is the need for a 'minimum viable duration' for longitudinal studies. Researchers argue that funders should commit to at least five to seven years for any cohort study, with a clear plan for data collection across developmental transitions. 'Three years is barely enough to establish a cohort,' Torres said. 'You need at least five to get meaningful longitudinal data.' Longer commitments would require agencies to change their budgeting cycles, which are often tied to annual appropriations.

Institutional repositories for orphaned datasets could also help. If a study is terminated, the data should be preserved in a form that allows future researchers to combine it with other cohorts. 'We need a national archive for longitudinal data that outlasts individual grants,' Chen said. 'That way, even if a study stops, the data can still contribute.' But such repositories require funding and governance structures that do not yet exist.

Some funders are beginning to listen. The Wellcome Trust in the UK has introduced 'long-term cohort grants' that provide up to ten years of support with a simplified renewal process. The European Research Council offers 'Advanced Grants' that can cover up to five years, with the possibility of extension. But these are exceptions. Most social science funding still operates on short cycles that undermine the very science they aim to support.

Torres has since moved to a smaller university where she runs a cross-sectional survey on a related topic. She no longer applies for longitudinal grants. 'The system is not set up for the kind of science I want to do,' she said. 'It's set up for quick results. And that's a loss for everyone who wants to understand how people actually develop over time.'