The Art of Statistics: Learning from Data

by David Spiegelhalter

Johannes Bohannon of the German Institute of Diet and Health conducted a study in which people were randomized into three groups and given either a standard, low-carb or low-carb with extra chocolate diet. They had a battery of measurements taken over three weeks, and the study concluded that the weight loss in the chocolate group exceeded that of the low-carb group by 10% (P = 0.04). This ‘significant’ result was submitted to a journal that deemed it an ‘outstanding manuscript’, and suggested that for €600 it ‘could be accepted directly in our premier journal’. On publication the press release from the Institute of Diet and Health led to stories in numerous news outlets, with headlines such as ‘Chocolate Accelerates Weight Loss’. But this was then all revealed to be a deliberate fraud. ‘Johannes Bohannon’ was really John Bohannon, a journalist; the Institute of Diet and Health did not exist; and the only real element was the data, which was not fabricated. But there were only five subjects per group, a large number of tests were made, and only the significant differences reported. The authors of this spurious article immediately owned up to their deception, but not all statistical frauds are done in order to show up weaknesses in the peer-review process.

Tilastollista huijausta parhainmillaan