Abstract:

Extracting insights from data is a lot like trying to solve a murder based on circumstantial evidence. We create one or more "theories" or narratives and then look to the evidence for confirmation of our hypotheses. This all works reasonably well when the evidence is consistent and well-behaved, but we can run into trouble when the facts contradict one another. This presentation looks at a real murder case with many contradictions among the different bits of evidence as a way to demonstrate the use of Bayes' Theorem to sort out the discrepancies and inconsistencies and arrive at "true" insight–the narrative that is most probable based on the evidence.