Books on Amazon
|Beck-Bornholdt/Dubben. Der Hund, der Eier legt Reinbek 1997
Def Significance/Statistics/Beck-Bornholt: a result is considered to be significant if the probability that it is an incidental finding is not greater than 5% - arbitrary (but internationally recognized). - The threshold should be lowered -> four-field test: Determines how likely it is that we suspect differences where none exist. - II 48 The chase after significance leads to new errors. - Significant results may be inconsequential. - Error: testing until a desired result comes out. - II 116 those who do not find a significant difference do not prove that there is no difference.
Statistics/Clinical Trials/Number/Significance: in a study with few patients major differences can only be confirmed or refuted. - II 114 - Problem: with a high caseload in the end everything is significant, but not necessarily interesting.
Significance/Statistics/Beck-Bornholt: the fixed level of significance should depend on what consequences a possible error has - e.g. an umbrella with 95% reliability is okay, a parachute is not.
|Link to abbreviations/authors|