CONFIRMATION BIAS 1 Confirmation Bias A Ubiquitous Phenomenon
CONFIRMATION BIAS 1
• Confirmation Bias: A Ubiquitous Phenomenon in Many Guises – – – Raymond Nicholson, TUFTS Review of General Psychology 50+ pages non-technical 7 full pages of references • Francis Bacon (1620) – “…great and pernicious predetermination…” • Poyla (1954) – “…scientific thought as distinguished from everyday thought. ” 2
BIAS Treatment and Evaluation of Evidence Building a Case **no expectation of impartiality Motivated Bias Impartial Evaluation Unmotivated Bias Philosophy Logic Statistics Psychology Hempel (1945) Observed: A white shoe Concluded: All ravens are black. Logic: All observed non-black things are non-ravens (the Contrapositive) 3
Incarnations of Confirmation Bias • • • Evidence interpretation Evidence censoring Restriction of alternative hypotheses Seeking confirmatory evidence Reification – over-estimating the salience of a taxonomy • Illusory Correlation • Primacy – becoming biased by the beginning of a stream of evidence, early hypothesis building 4
Wasson(1966) • Four cards presented. Cards have a number on one side and a letter on the other. • Confirm or deny: All vowels have even numbers on the other side. A B 4 7 • Most subjects (~90%) flip “A” and/or “ 4” • People seek to confirm the hypothesis, they do not seek countervailing evidence. 5
Taylor and Smyth (1860’s) • Egyptologists and Numerologists, they discover “too strange” relationships in Pyramid dimensions. – – – (2 x base)/height = p base/casing stone = 365 height * 109 = distance from earth to sun density of the earth period of the procession of the earth’s axis mean temp of the earth’s surface • Conclusion: Space aliens directed the construction of the Pyramids 6
Einstein and the Expanding Universe • Universe assumed static until mid-50’s • Expanding universe model could have been discovered by any post-Newton astronomer (S. Hawking, 1988) • Einstein actually had evidence of expansion as a by-product of General Relativity, but covered it with his now-infamous Cosmological Constant. • Called this the biggest mistake of his career. 7
Type 1 Error • recall P[type 1] = P[reject Ho|Ho] = a • Typical 1 -tailed test runs on a = 0. 025 • In 40 independent tests, you EXPECT one false rejection • Sequential testing is a great temptation in simulation analysis • Ability to make STDERR arbitrarily small means that meaningful indifference zone needs to be determined a priori. 8
Confirmation Bias • Our leaders and customers have it • Our families and neighbors have it • We have it, but we can protect ourselves 9
- Slides: 9