Talk:Data dredging

From Wikipedia, the free encyclopedia

You have new messages (last change).

Support merge - Data dredging is probably the best title (comment by John Quiggin, forgot to sign).

Do not merge - Bias through incorrect data-snooping is essentially different from the problem created by testing a hypothesis with the same data-set. For example, data-snooping bias may occur when dealing with an highly fluctuating set of data where every removal of a datapoint results in a new extreme, and so on. (Pc100935 11:59, 18 December 2006 (UTC))

Splitting a data-set parts A and B and then using part B to test a hypothesis formulated using part A is not recommended since these datasets can be highly correlated. Best practice is to formulate a hypothesis before looking at the data and use the data to test the hypothesis. If a hypothesis is based on existing data it should only be tested by collecting new independent data. (Pc100935 11:59, 18 December 2006 (UTC))

Retrieved from "http://en.wikipedia.org../../../d/a/t/Talk%7EData_dredging_6a3c.html"