9/12 Probability Estimation Flashcards
Probabilty it will rain given that it is cloudly or probabilty enjoy a movie given my friend saw it is what?
Conditional probability
History of probability estimation research
Thought we sucked at it
But! Heuristics proposed.
Tversky & Kahneman’s Heuristics:
The Availabilty Heuristic
Ease of retrieval is indicative of…
frequency
(words with r as first letter, vs. 3rd letter) estimate more begin with r
Availability Heuristic
Gabrielcik & Fazio (1984) Experiment
What did they manipulate?
Manipulate Accessbility
Subliminal priming words containing letter t (tattoo, tentative) for milliseconds led participants to judge t to occur more frequently
Do more words in English begin with blank or t? Judge the t.
Availability Heuristic:
Schwarz et al. 1991: Directly Manipulated the subjective experience
Ease of retrieval paradigm, easier to list 6 instances of past assertiveness than 12, greater ease resulted in higher ratings of one’s assertiveness
You might have listed more experiences of assertiveness, but because the retrieval tasks is difficult, at some point you start viewing yourself as less assertive due to the difficulty
Know the difference between thse terms:
Available vs. Availability Heuristic
Available: Does the knowledge exist in memory?
Availablility Heuristic: Notion that we infer frequency from ease of retrieval
What do we mean when we say something is accessible?
How readily can available construct be accessed?
Salience: What does this mean?
Some situational factor FORCES attention to an entity in the environment or a construct in memory
Not memory, it’s the situational factor forcing attention
Tversky & Kaheman’s Heuristics:
Anchoring and Adjustment Heuristic
Is X higher or lower than randomly determined number? Insufficient adjustment from anchor.
Example: UN number wheel. People fail to adjust sufficiently from anchor
Tversky & Kaheman’s Heuristics:
Selective Accesibility Model
(Mussweiler & Strack, 1999)
Anchor provides intital hypothesis
Tested in hypothesis-consistent manner (search our memory to see if hypothesis is good, but we tend to look for evidence consistent with hypothesis: Knowledge consistent with hypothesis is rendered accessible and forms basis for judgement)
Tversky & Kaheman’s Heuristics:
Incidental Anchors (Critcher & Gilovich, 2008)
Anchoring bias does not require an initial, overt comparison
Estimates of how much you’d spend at restaurant named “studio 17” vs. “Studio 97”
Tversky & Kaheman’s Heuristics:
Representativeness Heuristic
Similarity = likelihood
Can result in insensitivity to base rates
How we misinterpret questions: decide if X is engineer, how similar are they to an engineer stereotype.
Lawyer-Engineer Problem
Ginossar & Trope (1987)
30% Engineer
70% lawyer
What was the experiment? What was the twist at the end.
Higher estimate say engineer than expected if given engineer sounding paragraph.
However, new experiment: Ps given 2 nondiagnostic and 2 diagnositic
nondiagnostic descriptions either with no or mixed info. Particpants do well on these.
Diagnosistic descriptions clearly lawyer or engineer.
Results: Nondiagnostic do well, but for diagnostic info, forget about the base rates. (more 40/60, less 30/70).
Twist: Ask about a new person.
Results: No prior problem and diagnostic prior problem were fine, back to using base rates, BUT now non-diagnostic base rate of engineers when up. Based on new information.
THIS IS REPRESENTATIVE HEURISTIC
Krosnick, Li, and Lehman (1990)
Previous research: base-rate info before individuating.
Why is this a problem? (hint: Gricean rules of communication)
This could all be the recency effect
Gricean rules: assume speaker intends to communicate relevant and important info last)
Krosnick, Li, and Lehman (1990)
To combat possible recency effect, what did they manipulate in their base-rate info before individuating experiment?
Order of base-rate vs. individuating info when giving info AND whether order could reflect coversation, by telling participants order was random.
Divorce experiment
Results:
Standard: Base rate first is highest divorce probabilty, lower for base rate second
Randomly: Base rate first still higher, but less different.
Conditional Probabilities via Bayes Theorem
Given symptom, do you have dieases? Given disease, do you have the symptom?
P (D | S) = prob disease given symptom
= P(S | D) x P(D) / P(S)
Do we do this in daily life? No.
Gavanski & Hui (1992)
The shape experiment
how do they use Bayes Theorem?
20 shapes. 2 black circles, 6 line circles, 4 line squares, 8 black squares.
Prob black given round.
You form smaple spaces that only include specific objects. “Black given round” is a whole equation BUt you can notice: 8 found objects, 2/8. Much easier with sample space.
Directly assess prob. from revised sample space.
Gavanski & Hui (1992)
Remember what this experiment is.
Cute little alien fella.
Divide critters into groups: 81% by head shape not nose. Round and square head more natural categories than noses. Easier sample space to form.
36 faces. 4 stimluus configurations that vary base rates.
Easier if up nose given Plam, because I was thinking about head shape.
But Plam given up turned nose…well I wasn’t thinking about that.
Participants varied in how many of each kind they saw. High base rate: Lots of Plams with up noses, low base rates not a lot of plams with up noses.
Results: Who failed?
P(Category | Feature) at the high base rate 80%). At the low base rate (50%), more accurate because you guess around 50%.
P (Feature | Category) were good on both. GIVEN category I was try to remember, do fine.
Sherman, McMullen, and Gavanski (1992)
Compared probabilities conditioned upon a natural categoy (Gender) to those conditioned upon an ad hoc category (Blue and Green)
Natural: 100 random women, prefer blue to green, reasonably accurate
Ad hoc: 100 random people who like blue > green, how many are women. Inaccurate.
natural sample space = more accurate estimation.
McMullen, Fazio, & Gavanski (1997)
What did they do to the cute little critters in 1997 and why did they do it?
Which conditions were easier for what group.
**Hair and teeth instead of the nose. To test MOTIVATION. **
Critters take or give points.
Attend to Hair (side part +1000, center -1000) Teeth (one tooth +20, two teeth -20 points)
Attend to Teeth, the teeth are worth more points.
Categorization follows attention. From similarity judgements, understand how they were categorized.
Similarity index (2 pairs w/common hair - 2 pairs with common teeth). similarity inputs in multidimensional scaling program.
Did the same base rate thing: low was 50%, high was 80%
Results:
Judge Hair GIVEN tooth easier for teeth
**Judge Tooth GIVEN Hair easier for hair group **
