Chapter 18: Taming Intuitive Predictions
- · forecasting is ubiquitous — economists, financial analysts, military experts, venture capitalists, etc.
- · predictive judg’ts — some based on reason, analysis, statistics, etc.
- · others intuition & S1:
o
(a) skill & expertise acq’
by repeated experience
o
(b) heuristics (often subst.
easy q. for the harder one asked)
- · some judg’ts combine both
Nonregressive Intuitions
- · predict;n taken as = eval’n of current evidence, i.e. predict’n matches eval’n — another example of subst'n — will always à predict’ns wh. are systematically biased, completely ignore regression to the mean
- ·
- · Several steps: (1) Look for causal link betw. evidence (her reading) and predict’n (GPA) — wd not accept everythg as evidence, S2 recog.’s irrelevant evidence — but S1 canno adjust for weaknesses in evidence, intuitive predict’ns almost completely insensitive to real predictive quality of evidence — (2) evidence compared to a norm, reference group (how precocious to read well at age 4?) — (3) substit’n + intensity matching
A Correction for Intuitive Predictions
- · e.g. Julie is currently a senior in a state university. She read fluently when she was four years old. What is her grade point average (GPA)? — Why do we predict 3.7 or 3.8?
- · e.g. Julie’s reading & future GPA — some causal factors are common to both, but each also has its own causal factors — maybe <30% are common, correlated
- · how to correct intuitive predict’n for Julie’s future GPA:
o
1. Start w. an estimate of avg
GPA — (i.e. baseline, avg student’s GPA, w/o any info abt Julie)
o
2. Determine the GPA that
matches yr impression of the evidence — (i.e. intuitive predict’n, matching eval’n
of evidence)
o
3. Estimate correlation betw.
yr evidence & GPA
o
4. If corr. is .30, move 30% of
distance fr. average to matching GPA
- · begin w. intuition, then moderate it, regress toward the mean
A Defense of Extreme Predictions?
- · intuitive predict’ns usually too confident, too extreme
- · how to overcome these 2 biases: (1) predict’s expressed on a scale, e.g. GPA, co’s revenue (2) judging prob’s of outcomes
o
start w. a baseline predict’n,
wh. you wd w/o any knowledge abt current case — for categ’l case, use base rate
— for numer’l case, avg outcome in relevant categ.
o
both (1) and (2) hv intuitive
predict’n, wh. comes to mind (e.g. probability, GPA)
o
aim for predict’n betw.
baseline & intuitive response
o
if no useful evidence, use the
baseline
o
opp. Of baseline is yr predict’n — use it only if completely
confident in initial predict’n, after
critical review of supporting evidence
o
usually you will find that corr.
betw. intuitive judg’t & truth is not perfect, so judge somewhere betw. the
two poles.
- · sometimes not want to correct intuitive predict’s — req’s effort fr. S2 (find relevant reference categ., estimate baseline predict’n, eval. quality of evidence), so when correcting intuitive predict’s stakes shd be high, v. keen not to make mistakes — may want try predict a long shot — corrected intuitive predict’s will not predict rare or extreme outcomes (e.g. excellent law student becomes Sup. Court judge), because evidence is limited —absence of bias not always highest priority, e.g. venture capitalist try to pick next Google
A Two-Systems View of Regression
- · it is S1 that makes extreme predict’ns, willing to predict rare events fr. weak evidence — intuitions deliver predict’ns that are too extreme (extreme evidence à extreme predict’n, i.e. substit’n) — also, we put too much faith in them — S1 creates overconfident judg’ts, because confidence ß coherence of the best story arising fr. evidence at hand
- · S2 finds concept of regr. to the mean difficult, hard to underst. —matching predict’ns to evidence even seems reasonable
Speaking of Intuitive Predictions
- · “That start-up achieved an outstanding proof of concept, but we shouldn’t expect them to do as well in the future. They are still a long way from the market and there is a lot of room for regression.”
- · “Our intuitive predict’n is very favorable, but it is probably too high. Let’s take into account the strength of our evidence and regress the predict’n toward the mean.”
- · “The investment may be a good idea, even if the best guess is that it will fail. Let's not say we really believe it is the next Google.”
- · “I read one review of that brand and it was excellent. Still, that could have been a fluke. Let’s consider only the brands that have a large number of reviews and pick the one that looks best.”
No comments:
Post a Comment