August 2005 | Volume 2 | Issue 8 | e124 | John P. A. Ioannidis
The article "Why Most Published Research Findings Are False" by John P. A. Ioannidis discusses the increasing concern that many published research findings are likely to be false. The probability of a research claim being true depends on factors such as study power, bias, the number of studies on the same question, and the ratio of true to false relationships in a scientific field. Ioannidis models the framework for false positive findings, showing that for most study designs and settings, it is more likely for a research claim to be false than true. He highlights several corollaries, including that smaller studies, smaller effect sizes, and greater flexibility in designs and definitions reduce the likelihood of true findings. Additionally, greater financial and other interests, as well as more teams involved in a scientific field, can further decrease the probability of true findings. Ioannidis argues that claimed research findings may often simply be accurate measures of prevailing bias. He suggests improving the situation through better-powered evidence, reducing bias, and enhancing research standards. Despite these efforts, most new discoveries will continue to stem from hypothesis-generating research with low pre-study odds, and statistical significance testing alone provides only a partial picture.The article "Why Most Published Research Findings Are False" by John P. A. Ioannidis discusses the increasing concern that many published research findings are likely to be false. The probability of a research claim being true depends on factors such as study power, bias, the number of studies on the same question, and the ratio of true to false relationships in a scientific field. Ioannidis models the framework for false positive findings, showing that for most study designs and settings, it is more likely for a research claim to be false than true. He highlights several corollaries, including that smaller studies, smaller effect sizes, and greater flexibility in designs and definitions reduce the likelihood of true findings. Additionally, greater financial and other interests, as well as more teams involved in a scientific field, can further decrease the probability of true findings. Ioannidis argues that claimed research findings may often simply be accurate measures of prevailing bias. He suggests improving the situation through better-powered evidence, reducing bias, and enhancing research standards. Despite these efforts, most new discoveries will continue to stem from hypothesis-generating research with low pre-study odds, and statistical significance testing alone provides only a partial picture.