Abstract:
85 percent of published economics and political science findings can be independently verified * One in four significant results disappear under different reasonable choices, and one in four studies contain undetected coding errors * Published results cluster just above significance thresholds, which is evidence of publication bias * Policymakers should rely on independently reproduced evidence and avoid scaling up programs based on a single study * Journals and research entities should expand open data and robustness disclosure mandates.
