Bias check III — decision biases and statistical traps

Biases are not fallacies

A fallacy is a bad argument on the page. A decision bias is a habit of judgement that produces bad arguments without anyone noticing — it operates before the sentence is written. The bias check's third pass reads the analysis for six of them, and for three statistical traps that come from the same place: a mind that finds patterns faster than evidence supports them. Each has a counter-check — a procedure, not a resolution to be careful.

Six decision biases, in Lakeshore's words

• Sunk cost — "We've put $38,000 into this press this year; we can't walk away now." Money already spent is the same under every alternative and cannot be recovered by any. Counter-check: compare futures only; strike every past cost from the matrix.
• Anchoring — the vendor's first quote for the new press sets the scale, and every later figure is judged as "more" or "less" than it. Counter-check: get independent figures before reading the first quote, or score cost against an external benchmark (L5.6's information need).
• Status-quo bias — "It's always worked." (Priya.) The current state feels safe because it is familiar, not because it scored well. Counter-check: score the status quo in the matrix like any other alternative — Lakeshore's scores 2.80, and the reason is written in the cell.
• Confirmation bias — once the lease looks good, only lease-friendly sources get read. Counter-check: search for the disconfirming case on purpose ("digital press lease problems"), and score the source.
• Availability — the one spectacular breakdown, the day the government catalogue was late, drives the reliability score more than the log does. Counter-check: use the log — fourteen events — not the memorable one.
• Groupthink — the team stops questioning the option everyone likes. Counter-check: assign a devil's advocate for one meeting, whose job is the case against the leader.

Three statistical traps

• Small numbers — one bad month at a supplier is not a trend (L5.5). Counter-check: ask how many observations sit under the figure.
• Base-rate neglect — fourteen breakdowns sounds alarming until you know how many an ageing press of this type typically has. Counter-check: find the base rate before scoring the signal.
• Regression to the mean — an extreme result is usually followed by a less extreme one, whatever you do. "Choose last month's best operator to run the new press — she was best last month." One month's best is partly luck; next month she will likely be nearer the average, and the choice will look wrong for no reason. Counter-check: judge on a run of periods, not the latest one; expect an extreme to come back toward the mean.

Name four in the clinic case

"We've kept paper for eleven years" — status quo (and a hint of sunk cost, though nothing was spent). "The first vendor quoted $150 a month, so $60 is a bargain" — anchoring; $60 is judged against the quote, not against the value of a reminder. "Last week no one missed an appointment, so no-shows aren't really 12 %" — small numbers. "Everyone at the practice meeting liked the full system" — groupthink; nobody was asked to argue for the hybrid.

List the biases cold

Anchoring, confirmation, sunk cost, status quo, availability, groupthink. Then the three traps: small numbers, base-rate neglect, regression to the mean. And for each, the counter-check: compare futures; independent figures first; score the status quo; seek the disconfirming case; use the log; a devil's advocate; count the observations; find the base rate; judge on a run.