2.6 - Evidence on dynamic choice

Ways to test if people plan & how it relates to DC

Ask for plan and force them to follow it

• incentivised but doesnt test dynamic consistency as plan turned into pre-commitment decision

Ask for plan and allow departure from it for a cost

• incentive to stick to plan distorts test of Dynamic Consistency

Randomly select whether initial plan or final decision implemented for payoffs

• Changes decision problem by adding a new chance node / randomisation

ALL FLAWED

Bone et al. (2009) vs Cubitt et al. (1998)

Bone

• Test if people plan at all - Not what the specific plan is

• Set up a decision problem with certain behaviours that no sensible plan could rationalise.

  • Then if we see those behaviours, we'll know there was no plan.

Cubitt

• Reconceptualise theoretical framework so DC not a ‘within’ problem but now an ‘across’ problem

• Across problem restriction used instead (Timing Independence used instead of DC)

Timing Independence vs Dynamic Consistency

DC - Requires final plan / action to match initial plan

TI - Requires final action to match what subject would do in a different problem where they precommit before 1st chance node to both possibilities

Within-subject design flaws

Demand effect - Can lead to subject figuring out what relationship is and then doing results that obey that consistency

Incentive structure design flaws

Income effect - Paying out after every task can lead to income effects as subjects richer than at start (need randomisation of payoff round)

Randomisation of payoff - Subject could see whole experiment as 1 big dynamic choice problem with a chance node at the end (that determines payoff)

• Not Testing same dynamic choice principles that want to test

Decision tree design flaws

Can’t be sure all subjects understand decision-trees

• have to test frame independence as its not true by design

CCS approach - use words not trees to present decision problems

• phrase same thing in 2 different ways but have same tree to test

Between subject design for DC testing (CCS approach for within subject flaws)

Each subject faces 1 task with real money outcomes

Different groups of subjects face each task - compare groups rather than individuals

• compare proportions of subjects making same choice in a group

Random assignment of subjects to groups - no systematic differences in risk attitudes between groups (unless task causes it which we’re testing)

CRE example - CSS 1998 OVERVIEW

q - probability

r - Common ratio

When problem scaled down preference flips from certainty to riskier option

different versions of the same problem presented to subjects to test each different rational dynamic decision making assumptions

• EUT says scaled UP vs Scaled down identical BUT CRE argues there is a difference

CRE example - CSS 1998 test of rational dynamic decision making

Test of separability - Scaled UP vs Prior lottery problem

• Decision-maker is not asked to choose between Options A and B until (and unless) the outcome of the 1st stage lottery requires her to do so.

Test of Timing Independence - Prior lottery vs pre-commitment problem

• same problem but required to pre-commit herself before resolution of 1st stage lottery (timing shouldn’t affect decision as same probability distribution of prizes)

Test of Frame independence - Pre-commit vs 2 stage problem

• subject faces 2 2-stage lotteries (decision tree identical) BUT now 1st stage written into the options

Test of Reduction of compound lotteries - 2 stage vs Scaled DOWN

• same problem but compound lotteries reduced to simple ones

CRE example - CSS 1998 RESULTS table

Subjects became more risk averse after the resolution of prior uncertainty

• once subjects know the prior risk has fired and their decision is live and consequential, they become more cautious - safe option chosen

• Pre-commitment

CRE example - CSS 1998 RESULTS explained + implication

Classic test of CRE - P1 vs P5 → Difference has direction of CRE but not sig at 5%

• maybe from small samples

Dynamic CRE - P1 vs P4 → Difference has direction of CRE + stat sig

• EUT violated

Pooled test - (P1 + P2) vs (P3 + P4 + P5) → Difference has direction of CRE + stat sig

People are taking the risky option more frequently when they can pre-commit to it, than when they can't - Violates TI principle

• Robust evidence from other literature showing TI violated

CRE & CCE as violations of EUT

Violates EUTs property of linearity in probabilities

Need a model that is non-linear (e.g. prospect theory)

• max something else

Timing independence definition

The final action taken by an individual should align with what they would choose in a scenario that allows for pre-commitment before encountering any uncertainty

• Preferences over lotteries are unaffected by when uncertainty resolves.

The outcome of a decision should not vary based on the timing of when that decision is made

Bone et al. (2009) - OV

Test if people plan

subjects asked to choose at choice nodes for 4 different decision trees

• If plan then will pick dominant strat - if not then pick best out of available

Individual No precommit vs I precommit vs Pairs

• IPC only had precommit based on what nature did - doesnt include counterfactual so not forced to view all payoffs at once

Bone et al. (2009) - RESULTS

People pick best option at final stage BUT majority dont pick dominant strategy from the start

• no improvement over time (experience)

• Minority did plan - got all right BUT majority didnt plan

INPC & IPC have similar levels of no planning + Pairs has same as well ~ 30%

Reject null that chose at random - actively made incorrect choice