Are Difficult-To-Study Populations too Difficult to Study in a Reliable Way?
Abstract. Replication studies, pre-registration, and increases in statistical power will likely improve the reliability of scientific evidence. However, these measures face critical limitations in populations that are inherently difficult to study. Members of difficult-to-study populations (e.g., patients, children, non-human animals) are less accessible to researchers, which typically results in small-sample studies that are infeasible to replicate. Nevertheless, meta-analyses on clinical neuropsychological data suggest that difficult-to-study populations can be studied in a reliable way. These analyses often produce unbiased effect-size estimates despite aggregating across severely underpowered original studies. This finding can be attributed to a neuropsychological research culture involving the non-selective reporting of results from standardized and validated test procedures. Consensus guidelines, test manuals, and psychometric evidence constrain the methodological choices made by neuropsychologists, who regularly report the results from neuropsychological test batteries irrespective of their statistical significance or novelty. Comparable shifts toward more standardization and validation, complete result reports, and between-lab collaborations can allow for a meaningful and reliable study of psychological phenomena in other difficult-to-study populations.