This is more true than JiT knows. If someone asked me "Do you know any scientist who has committed fraud?" I would have to say yes. However, the fraud they committed had to do with accounting, not the actual scientific process--as far as the scientific process goes, this organization is actually quite good. So I'd be part of that 14%, but my results would enver add to the number of papers that contained fraud.
Secondly, a lot of work in science never gets published. There's no good venue for publishing experiments that fail, for example, so anyone who commits fraud and fails anyway would be part of the 14% but not part of the frauds in published literature.
Third, there's a question of "What IS fraud?" If they allowed the individual researchers to define it they'd get as many answers as there are researchers. If THEY define it, however, their definition is open for debate. For example, is removing obviously erronious data from the dataset fraud? Some people say so, others do not. How much manipulation can you do to the data before it's fraud? Some manipulation is obviously necessary; however, too much distorts the data. As a professor once told me, take the log enough times and you eventually get a straight line. Is inclusion of profesional opinion fraud? I've made arguments to multiple federal agencies that amount to "because I say so", despite the fact that I can't cite a single source for my argument. These were based on extensive field work done in the past in the area, but they remain unsupported via the literature. Any of that can be included as fraud, or not, depending on the researcher's mood.