Decision Making III: Good Application of Heuristics Flashcards
When Should We Use Heuristics?
- Heuristics can save time
- But also increase bias
- If we make bad decisions, it’s imporant to make adjustments to out process
- Hindsight bias can get in the way
Hindsight Bias definition?
+
Classic paradigm:
1. Ps read assertion (e.g. “NASA defines space as >50 miles above Earth”)
2. Ps report confidence assertion is true (or false)
3. Truth is revealed (it’s true)
4. Ps asked to recall their prior confidence (before reveal)
Result?
Tendency to overestimate our ability to have foreseen an outcome
Result: Participants recall confidence higher than it really was
Hindsight Bias can cause lots of problems…?
When bad things happen: “I knew it was coming”
People take too much risk, don’t adjust when bad things happen (“I’ll just be more careful next time”)
Hindsight bias makes it difficult to appreciate uncertainty
Less is More
Under some conditions, heuristics provide more accurate decisions than complex models
Which conditions?
People adopt heuristics over time based on experience
Uncertainty: when parameters are unknown
“Rational” models of decision making are best when parameters are known
Heuristic approach is better under uncertainty
Advertising Decisions, Database of customers, Need to decide which customers are active
Which ones to target?
Statistical approach:
Regression or other mathematical model
Heuristic approach:
Hiatus heuristic: if customer made purchase within last N months, they are active
Results (Wübben& Wangenheim, 2008)
– Pareto/negative binomial dist. model trained over 40 weeks of customer data: 75% correct
– Hiatus heuristic (no training required): 83% correct
– Complex model required much more information, time, effort. Still worse performance
Similar to satisficing heuristic
Recognition Heuristic
Which city is larger by population?
New York OR Park City
New York!
Recognition heuristic: Recognition indicates higher value on particular criterion
Recognition Heuristic
Predicting sports:
– Serwe& Frings (2006) asked people to predict Wimbeldon(2004 tournament) tennis matches
– People who only knew half the players: 72% correct
– ATP entry ranking: 66% correct
– Wimbeldo nexperts seeding: 69%
Recognition Heuristic
Investment decisions:
– Ortmannet al. (2008) asked people to choose a set of stocks based on recognition
Recognition portfolio outperformed:
- Fidelity Growth Fund
- Total market
- Random portfolios
- Other expert picks
Recognition portfolio did not outperform when chosen by college students! (Boyd, 2001)
Fluency Heuristic?
Definition: if all choices are recognized, choose the one that is recognized faster
Extension of recognition heuristic
Fluency also predicts stock performance
Fluency Heuristic
- Fluency can help decide among self-generated options
- “take-the-first” heuristic: do the first thing that comes to mind
– Hepler (2008) showed short clips of basketball plays to professional players
– Pass or shoot?
– First option to mind was best (matched decision when given more time to decide)
Tracking Criminals (Snook et al., 2005)
Need to find where serial offenders live
- “Geographic profiling”
- CrimeStatsoftware generates geographic probability distribution
- Largest distance heuristic: take two furthest apart crime scenes, draw a circle through them
Largest distance heuristic more accurate than 10 other strategies (until >9 crimes involved)
Health-Care Decisions (Green & Mehr, 1997)
- Which people with chest pain need to go to the ICU? (possible heart-attack)
- Michigan doctors sent 90% to the ICU, just to be safe, but this caused overcrowding
Statistical solution: huge checklist of symptoms, probabilities, requires calculator
Fast-and-frugal decision tree
Result?
Results:
– Decision tree reduced false-alarms by 50%
– Decision tree also reduced patients with real heart attack sent to regular nursing bed!
– More accurate than statistical method
– Much faster, doctors liked it better
Health-Care Decisions (Kattahet al., 2009)
Dizzynessusually isn’t a problem, but can be caused by rare cerebellar stroke
- Is it really a stroke or not?
MRI scan method: expensive, exclusive, time-consuming, and only 88% sensitivity
Bedside exam: horizontal head impulse test (measures vestibulo-ocular reflex), and a few other tests that take less than 1 minute, 100% sensitivity; 96% specificity
A Less is More Effect :)
Using a subset of the data (simplification), and make more accurate predictions
Investment Portfolio Selection
Many alternatives to choose from
Markowitz mean-variance optimization
– Computationally intense
– Requires lots of data for correct parameter estimation
1/N Heuristic:
– Allocate investment equally (1/N % each) = Look at all possible options, allocate money evenily -> Ignore all data girle-pop :*
Using a subset of the data (simplification), and make more accurate predictions
The number of error increases for each estimation
A Less is More Effect :)
When is Less Really More?
When the risks are certain, using all the data to make the decsion is good. But in the real world, we never know all the data and need to estimate all the paremeters -> inrese number of error
- People adopt heuristics over time based on experience
- Uncertainty: when parameters are unknown
- “Rational” models of decision making are best when parameters are known
- Heuristic approach is better under uncertainty
