Field-based Deviant Case Sampling — Outlier Cases in Naturalistic Field Research

The core logic — deliberately selecting cases that deviate from the norm — is the same. The field-based modifier specifies the data collection mode: data are gathered through sustained on-site fieldwork (observation, in-situ interviews, ethnographic engagement) rather than through remote surveys, administrative data extraction, or documentary analysis alone. This distinction matters because fieldwork access, relationship-building, and observational skill become additional methodological requirements.

There is no fixed rule, but one to four cases is common. The criterion is theoretical saturation — the point at which additional deviant cases no longer reveal new mechanisms or contextual factors. Given the resource demands of sustained fieldwork, most studies work with one or two intensively studied deviant cases, sometimes supplemented by briefer comparison visits to typical cases.

Define the norm using the best available evidence before entering the field: published benchmarks, administrative datasets, survey data from the broader population, or systematic literature review. Specify a clear quantitative threshold (e.g., outcomes two standard deviations above the mean) or a qualitative criterion (e.g., the only program in a region that retained all staff during a funding crisis). The criterion must be defensible and documented in the methods section.

Yes. It is common to combine deviant case selection with maximum variation sampling — selecting one outlier at each extreme — or with typical case sampling, where one deviant and one typical case are studied in parallel. Such combinations strengthen comparative analysis and are fully consistent with purposive sampling logic as described by Patton.

The case-selection logic can inform mixed-methods designs — for example, identifying statistical outliers in a large dataset and then conducting fieldwork at those sites. However, field-based deviant case sampling in its standard form produces qualitative data and serves interpretive or explanatory purposes rather than statistical estimation.