Why Some Methods Jump Fields and Most Die at the Border

Every few decades a method escapes the field that invented it and rewrites the rules somewhere else. Polymerase chain reaction started in molecular biology and ended up in forensics, ecology, and clinical diagnostics. Maximum likelihood estimation traveled from statistics into genetics, then into linguistics, then into epidemiology. Bootstrap resampling moved from computing into almost everything. These crossings get treated as happy accidents, but the pattern underneath them is not random.

The single biggest predictor of whether a method travels is whether it solves a problem that is structurally identical across domains, even when the surface content looks nothing alike. PCR amplifies a signal from a low-abundance target. That is not a biology problem. That is a detection problem, and detection problems are everywhere. When a method is encoding a general solution to a general problem, domain labels become permeable.

The second predictor is abstraction level. Methods that live close to the data, tied to specific instrument readouts or organism-specific quirks, tend not to migrate because porting them requires rebuilding infrastructure that the receiving field does not have. Methods that live one level up, at the level of inference or estimation or model comparison, travel more easily because the receiving field usually already has data in some form. This is part of why Bayesian hierarchical modeling spread so widely after computational costs dropped: the conceptual machinery was domain-agnostic even when the application had always looked domain-specific.

There is also a sociological layer that the citation literature undercounts. Methods move when people move. Postdoctoral researchers trained in one tradition who take faculty positions in adjacent fields carry techniques in their heads that do not always make it into explicit methods papers. The receiving field often reinvents a version of the tool independently, publishes it under a different name, and the cross-citation that would reveal the genealogy never appears. Historians of science who study priority disputes find this pattern repeatedly. The actual diffusion network is sparser on paper than it was in practice.

The failure modes are symmetric. A method can fail to travel because the receiving field never learns it exists, which is a communication problem. It can fail because the assumptions baked into the method do not hold in the new domain, which is a portability problem. And it can fail because the receiving field's gatekeepers, reviewers and editors and grant panels, treat the unfamiliar method as a defect rather than a contribution, which is an incentive problem. All three failure modes are common. The third one is probably the most correctable and gets the least attention.

The implications for science communication are direct. When a methods paper publishes in a specialty journal, the readers most likely to benefit from it in a different domain will never see it. Preprint servers have improved cross-domain visibility at the cost of signal-to-noise, because no one has solved the indexing problem at the level of structural problem type rather than keyword. You can search for 'sparse signal recovery' and miss a decade of relevant work filed under 'compressed sensing' in engineering and 'lasso regression' in statistics because the vocabularies did not converge.

Paying attention to methods sections, not just results, is the corrective. A result tells you what a team found. A method tells you what class of problem they solved and, implicitly, which other fields are sitting on the same unsolved version of it.