Medical Physics Seminar – Monday, March 5, 2018
Data Analysis Pitfalls: P-Hacking, Dredging, Harking and Other Barriers to Translation
Michelle Svatos, PhD (hosted by Dr. Larry DeWerd)
Medical Device Executive, Celestial Medical - USA
In an age of growing data abundance, the collection, manipulation, and filtration of data are be-coming increasingly important to counter high levels of noise. If not done carefully, intuition and plausibility may be superceded by hunting for correlations, and conclusions may be based on finding “significant†results even if they were outside the original experimental intent. Using pre-set P-values as a metric of statistical significance can inadvertently incentivize the practice of op-timizing the analysis for that value, or P-hacking. Failing that, there is the temptation to Hypothe-size After Results are Known (HARK). Be aware these practices may result in misleading conclu-sions, irreproducible work, and failure to translate to clinical use, among other problems. Best practices to avoid such quandaries will be discussed.