Fig. 2

Three areas for reform to relieve research strain, outstanding questions for meta-research, and possible answers. Error detection: researchers need to be able to distinguish between reliable and unreliable research. A better system of quality control (both prior to and post-publication) might discourage research practices that inflate the rate of false-positive findings in the research literature (e.g. selective reporting; p-hacking; HARKing). At the same time, there should be incentives for researchers to remedy mistakes in their previous work, for example, through ‘living’ papers that can be easily updated. A more drastic change would be to require self-contained studies to be replicated, and for published results from long-term field studies to be revisited in subsequent years (e.g. before funding is renewed). Theory development: research in ecology and evolutionary biology sometimes fails to traverse the space between speculation and theory. In addition to hypothesis testing, answering big questions requires space for descriptive and exploratory research [10]. Detailed descriptions of natural history help calibrate theoretical models, and predictions of models should be tested in natural settings. To specify conditions under which findings are expected to replicate, authors can include ‘constraints on generality’ statements alongside inferences. When un-expected results are attributed to ‘context dependence’, specific contexts can be tested with new data. For cumulative research, foundational studies can be validated with close replications, and their generality assessed in different settings. Human resources: education programmes could increase the ability of researchers to work transparently and reproducibly, but honing these skills and conducting rigorous research is too often unrewarded. Any change to evaluation metrics requires careful consideration and measurement of unintended consequences (e.g. how to ensure costs are not disproportionately borne by less well-resourced research groups and universities). Much published research represents independent projects conducted by trainees, but reliability might be increased by coordinating multiple trainees on the same projects (including replication projects) and providing secure employment to people with specialised expertise (who can be professionally indifferent to the outcome of a particular study)