Bayesian Life Analysis

A powerful explanation of why some people see racism (or other forms of sinister discrimination) in cases where others dismiss or overlook it — Doug Glanville writes in a recent New York Times opinion essay "I Was Racially Taunted on Television. Wasn't I?" about how evidence accumulates and properly influences interpretation of what seem to be ambiguous events:

... The communication breakdown here can be illustrated by imagining a coordinate graph on which you plot what you understand to be the racist episodes you experience or hear about during your life. The x-axis represents the passage of time and the y-axis represents the degree of racism of an episode — from someone's assumption that you're a valet when you're parking your own car to the burning of a cross on your lawn. For each experience, you mark a dot.

Over time, the dots accumulate, and you start to see a pattern. You draw a curve that connects the dots and you develop a keen sense of things that happen to you because of your race. The pattern allows you to notice correlations, to make predictions. You are learning from evidence, in part for your self-preservation.

Now imagine someone plotting a graph who encounters such episodes from a more privileged or isolated perspective. Maybe this person hears about them only if they are sensational enough to make the news. He sees evidence of racism only from time to time, and when he does, it tends to be stark and unambiguous — the use of racial slurs, an explicit avowal of hate. ...

... and yes, as with all events throughout history, interpretation should be Bayesian. Prior experience establishes the baseline probability for a hypothesis (e.g., "Such an Act is X% likely to be racist."); the current situation is judged in that context; and the estimated probability distribution for future events is updated. Reported observations by others contribute to the constellation of evidence, appropriately weighted by their reliability and ambiguity.

... and when something truly surprising happens, the posterior probability estimate changes a lot more more than when events are ho-hum as-expected. In brief, Bayes Rule simply says to update the odds:

before * likelihood = after

... and that's how to be a better thinker!

(cf Statistics - A Bayesian Perspective (2010-08-13), Introduction to Bayesian Statistics (2010-11-20), Change of Heart (2011-06-21), Adventure of the Bayesian Clocks - Part One (2013-12-04), Forecasting Lessons from Systems Dynamics (2017-07-05), Be Skeptical of Bluster (2018-04-02), ...) - ^z - 2019-05-30