6 - evidence on dynamic choice Flashcards
what is the paper
Cubitt Starmer and Sugden
what were the 3 challenges of the experimental design that needed to be solved
- plans
- incentives
- decision trees
what was the plans problem
how can you test dynamic consistency?
problem with observing plans in heads - hard to
- it is hard to directly test dynamic inconsistency
what was CSS solution to the planning challenge
looks at across problem restrictions instead of within (dynamic consistency is a within problem restrcition)
avoids the need to elicit a plan by using timining independence instead of dynamic consistency
what is the difference between timing independence and dynamic consistency
dynamic consistency = test if people plan ahead - final action at n2 (choice node) to be the same at n1 (planning node)
timing independence = the action you make at n2 would be the same regardless of the timing you had to make the decision - you would make that decision before or after the chance node
what was the incentives problem
within subjects design - no incentives
each person does all of the tasks
- problem with how you would incentivise all of the tasks = costly
- what if they already win enough money and then stop attempting the rest of the tasks
what was CSS solution to incentives problem
use a between subject design instead of within
- each subject one task for real money outcomes
- different groups of subjects face each task
* will compare the proportions of each group that make choices
- random assignment - no systematic differences between risk attitudes in each group
what was the Trees problem
theory assumes that frame independence justifies the use of decision trees for analysis on decision making
- but we cant assume that all subjects understand decision trees
- could make errors
what was CSS solution to trees problem
use words to describe the decision problems instead
- would test frame independence by using 2 problems that differ in verbal description but have the same tree
what are the 3 features of CSS that confront the challenges
- uses between-subjects design
- uses timing independence instead of dynamic consistency
- no trees - verbal descriptions of the problems
- tasks for real money
what was the structure of CSS experiment and what was it testing
scaled up problem
prior lottery problem
precommitment problem
2 stage problem
scaled down problem
what does EUT imply about scaled up and scaled down problem
equivalent
by 4 principles
- we should observe same behavuour in each stage
what does CRE suggest about scaled up and scaled down
some subjects will make different choices in each
- violate their equivalence
how did they test separability
- make choice
- make the choice after a lottery
- shouldnt matter what the history is they should make same chocie - same subtree
how did they test timing independence
- choice after lottery
- choose choice that will happen after the lottery - but make the choice before the lottery
- same choice just making it at a different time (ahead in time)
- it shouldnt matter at what time you make the choice
how did they test frame independence
only difference is the way the problem is verbally written
- has the same tree
how did they test reduction of compound lotteries
4.
5.
5 is just problem 4 but with the compound lotteries multiplied out
what did CSS find
results
- only found statistical difference in the proportion of subjects choosing risky option between
stage 2 and stage 3 - violation of timing independence
if EUT principles held what results would you expect to see
statistically similar proportions of subjects choosing risky option in each stage - if random allocation is valid
how was timing independence violated
subjects facing precommitment were twice as likely to choose risky option than those in the prior lottery
* evidence that people take more risks when they precommit
* people precommit to risks they wouldnt take if they werent precommitted
did CSS find results of CRE
- no statistical difference between 1 and 5 - doesnt show CRE
- pooled test - find that there is statistically significant difference between scaled up and scaled down - inconsistent with EUT
what was objective of CSS
to see if there is any violation of the principles
- found violation of timing independence
what does CRE do
- CRE not consistent with EUT
- CRE motivates behavioural models of risky choice
what is the case used to justify that EUT is the correct model of peoples underlying true preferences
- even with CRE
- people sometimes make random errors in their choice - explains why CRE - people switch
what is the Fechnerian error model
peoples preferences are captured by EUT
but they make errors when calculating EU
why would you make error in Fechnerian error model
how does this explain CRE
more likely to make an error the smaller the difference between the EUs
CRE = shrinks the EU difference between the lottery - making it more lil
what are the problems with the fechnerian model
- error has mean zero - doesnt explain why large cohort change from safe to risky - it suggests we would expect it to move to 50/50
- doesnt explain CCE - because CCE there is no change in the differences in EU (which is where we get the error from)
- variance of e independent
what are the 2 classic assumtions of EUT
- people have preferences representable as maximising EU
- always choose in line with their preferences
what is a good thing about the Fechnerian model
- predicts the asymmetry in the direction of the violations