Statistical abuse uses analysis to support a preferred conclusion while hiding weaknesses in the evidence. Confirmation bias, data torturing, and model torturing can create a veneer of statistical validation. section 1.4
Watch for pressures that can steer research toward a desired result:
- A researcher is urged to use a favorable dataset to impress investors. section 1.4
- A nonprofit seeks evidence supporting its cause rather than evidence that might marginalize it. section 1.4
- A political candidate wants research supporting campaign policies. section 1.4
- A researcher with industry funding makes a reassuring claim about sugar consumption. section 1.4
The favorable-dataset request
When someone asks for a dataset chosen to showcase results, treat the resulting validation cautiously: the selection itself may have steered the conclusion. section 1.4
