Cambridge Research Proposal Example: Journal editor to research integrity policy (Score 93)

Research integrity policy has expanded considerably at the institutional and funder level over the past decade, yet the mechanisms by which editorial practice at individual journals translates—or fails to translate—into measurable reductions in post-publication error remain poorly understood. Retraction databases record outcomes; they do not record the editorial decisions, workflow changes, or policy memos that preceded them. This proposal addresses that gap directly. The central research question is: To what extent do documented journal-level integrity interventions—defined here as formal editorial policy changes, author declaration requirements, or data-availability mandates introduced at identifiable points in time—predict subsequent changes in retraction rates, correction rates, or expressions of concern within the same journal, controlling for field, impact factor, and publisher type? Two subsidiary questions follow. First, is there a measurable lag between policy introduction and detectable outcome change, and does that lag differ systematically across disciplines or publisher categories? Second, do journals that introduce multiple simultaneous interventions show different outcome trajectories from those introducing single-policy changes? The practical motivation for this question arises from direct editorial experience. Working as a journal editor, I drafted and implemented an internal integrity policy memo that standardised how the editorial team handled authorship disputes, image-manipulation queries, and data-sharing requests. That exercise made visible a structural problem: editorial teams make policy decisions with almost no empirical basis for predicting whether a given intervention will reduce the rate of post-publication problems. The academic literature on research integrity has grown substantially, but it has largely studied retraction as an outcome variable without tracing the upstream editorial decisions that may have caused or prevented it. This proposal attempts to build that causal bridge, at least at the level of association, using publicly available data. Two bodies of scholarship bear on this question without yet speaking to each other adequately. The first is the retraction literature. Fang, Steen, and Casadevall's foundational analysis of retraction patterns across biomedical journals established that misconduct, rather than honest error, accounts for the majority of retractions in high-impact venues. Subsequent work by Brainard and You at Science, and the systematic analyses published through the Retraction Watch database, has refined our understanding of field-level variation and temporal trends. What this literature does not do is link retraction rates to specific editorial policy changes at the journal level. Retractions are treated as outcomes of author behaviour, not as outcomes that editorial governance might modulate. The second body of literature concerns research integrity policy design. Work by Bouter, Tijdink, and colleagues on questionable research practices, alongside the COPE guidelines and the San Francisco Declaration on Research Assessment, has produced normative frameworks for what journals should do. Empirical evaluation of whether those frameworks change behaviour is sparse. The few studies that exist—including work on registered reports reducing outcome-reporting bias—focus on specific methodological interventions rather than on the broader class of editorial integrity policies. No study, to my knowledge, has used a quasi-experimental design to compare retraction and correction trajectories before and after documented policy introductions across a cross-journal sample. The gap, stated precisely, is this: we have descriptive epidemiology of retractions and normative guidance for editorial policy, but no systematic empirical account of whether editorial policy changes predict post-publication outcome rates. This proposal is designed to produce that account at a scale feasible within an MPhil research year. The study uses a retrospective interrupted time-series design applied to a cross-journal panel dataset. This design is appropriate because the key independent variable—policy introduction date—is a discrete event that can be treated as an intervention point, and because retraction and correction counts are available as monthly or annual series through public databases. Data sources. The Retraction Watch database provides retraction and correction records with journal identifiers and dates. Journal-level policy information will be assembled from two sources: COPE membership records, which log when journals joined and which guidelines they adopted; and a systematic review of journal Instructions to Authors archived via the Wayback Machine, which allows approximate dating of when data-availability statements, authorship criteria, and image-integrity requirements were introduced. Publisher-level policy announcements (for example, mandatory data sharing introduced by PLOS journals in 2014, or the Nature portfolio's image-integrity requirements) provide additional intervention dates for journals within those portfolios. Sample. The target sample is approximately 150–200 journals drawn from the biomedical and life sciences fields, selected to include variation in impact factor quartile, publisher type (commercial, society, open-access), and policy-introduction timing. This range is large enough to support multivariate analysis while remaining manageable for the manual coding of policy-introduction dates that the Wayback Machine approach requires. Analysis. The primary analysis is a panel interrupted time-series regression with journal fixed effects, estimating the change in retraction and correction rates in the periods following documented policy introductions. Secondary analyses will test for heterogeneity by discipline and publisher type, and will estimate lag structures using distributed-lag models. The analysis will be conducted in R, using established packages for panel time-series estimation. The choice of a quantitative, observational design reflects a deliberate scope decision. A fully causal claim would require random assignment of policies to journals, which is not feasible. The study aims instead to produce a well-identified associational estimate that can inform future experimental or quasi-experimental work, and to demonstrate whether the signal is large enough to warrant that further investment. Data access presents the main feasibility constraint. The Retraction Watch database is publicly accessible for research use; its terms permit academic analysis of aggregate patterns. Wayback Machine archives are publicly available. COPE membership lists are published. No individual-level human data are involved, and no institutional ethics approval is required for analysis of publicly available bibliometric records. If the Wayback Machine archive is incomplete for a given journal, the fallback is to use publisher-level policy announcements as a proxy intervention date, which reduces precision but preserves the sample. Scope is the second constraint. Manual coding of policy-introduction dates for 150–200 journals is time-intensive. I estimate approximately six weeks of systematic coding at a rate of five to seven journals per day, which is feasible within the first term of the MPhil year. A pilot coding exercise on twenty journals, completed before the programme begins, will test inter-rater reliability using a second coder and refine the coding protocol. Provisional timeline. Months one to two: finalise sample, complete pilot coding, and establish the panel dataset. Months three to four: complete full coding and merge with Retraction Watch data. Months five to seven: run primary and secondary analyses, produce draft results. Months eight to nine: write up, incorporate supervisor feedback, and prepare for submission. The principal risk is that the Wayback Machine archive coverage is too sparse before 2010 to support pre-intervention baselines for a sufficient number of journals. If this materialises, the sample will be restricted to journals with adequate archive coverage, and the analysis will be reframed as a post-2010 study with explicit acknowledgement of the truncation. The MPhil in Research Methods at Cambridge provides the methodological training—particularly in quantitative social science methods, causal inference, and research design—that this project requires. The interrupted time-series design sits at the intersection of applied statistics and science policy research, and the programme's emphasis on rigorous method selection over disciplinary convention is directly relevant to a project that borrows from bibliometrics, health policy evaluation, and science studies simultaneously. Within Cambridge, the Meta-Research Innovation Center at Cambridge (METRIC) works on exactly the questions this proposal addresses: the empirical evaluation of research practices and the effectiveness of integrity interventions. The project would benefit from engagement with researchers in that group, as well as from the broader network of scholars working on open science and publication ethics across the Faculty of Human, Social, and Political Science and the Department of Public Health and Primary Care. Access to the University Library's bibliometric resources and to Cambridge's institutional subscription to journal metadata services would support the data assembly phase. The project does not require laboratory access, fieldwork travel, or proprietary data. Its resource requirements are modest: computational access for panel regression, library access for journal archive retrieval, and supervisory expertise in quantitative research design and science policy. These are resources the MPhil programme is well placed to provide. The expected contribution of this study is limited but defensible: a systematic empirical estimate of the association between editorial policy change and post-publication outcome rates, using a design that is more transparent about identification assumptions than the descriptive literature that currently dominates this field. If the association is weak or null, that finding is itself informative for editors and funders who are investing in policy interventions without an evidence base.