8.22 - Experience introduction

Anomaly

Observed behaviour that violates an orthodox economic view of what is rational

How is ‘max something else’ still irrational & what to do

Max something else used as way to model deviation from standard theory and show people are still ‘rational’

But the new model still violates ST assumptions of rationality → model irrational

2 options on what to do:

• Either non-ST objective function that is maximised is the true preference set

  • Therefore the preferences aren’t irrational

  • Model is of rational behaviour with subjects with non-standard preferences

• OR Context leads to subjects attention drawn to things that make them become irrational – perception distorted

  • E.g. overweighting rare events

Internal validity

Design warrants reliable conclusions about an effect that did operate in the experiment

External validity

Design warrants reliable conclusions about an effect that would operate outside the experiment

Validity in experiments

Well controlled experiments have high internal validity

• BUT external validity may sometimes be questionable

Scepticism about external validity is more convincing if reasons provided for why inference is valid

External validity in experiments

Since early generalised scepticism about artificiality of lab environments waned (took ages), external validity concerns about experimental findings have become:

• more precisely targeted (at particular claims / findings)

• Better rationalised by hypotheses about why conclusions of some experiments may not generalise

• More investigable - possibly more experiments

Experience and External validity

Critics argue experiments use inexperienced subjects and this undermines any external validity

• unfamiliar / complex tasks

• limited opportunity for reflection / practice / discussion

• only modest incentives - less reason to be truthful

• young / less experienced students used

Critics and experiments

Critics argue that ST shouldnt have to apply everywhere (not got general applicability)

• testing outside its domain proves nothing

• Experiments only test outside of where ST should operate and so findings arent applicable

ST can only ‘reasonably be expected’ to apply in the lab if:

• Decision problems are simple enough

• Incentives adequate

• Time for learning / experience sufficient

Preference discovery

Individuals have a consistent set of preferences, but such preferences only become known to the individual with thought and experience

Preferences independent of learning process

Plott (1996) - Theory

Choices go through a 3 stage process

• Inexperienced - impulsive behaviour

• Experienced - After experience there is deliberation + reflection (especially if incentives sufficient)

• Converged - Beliefs about other players converge to reality

Discovered preferences are those revealed by choices at end of 3-stages

• inexperienced subjects may not conform to it - thats why they violate ST

• ST only applies to discovered preferences

Plott & Binmore critique

Both experimenters and are critics of tests with too low incentives or inexperiences subjects

• ST shouldnt apply to these domains

Critique of anomalies is doubled edged:

• defending ST against low stakes / inexperience in the lab also means it should hold in the field

• Very common for field to either use low incentives or infrequent tasks

How to test Plott & Binmore critique

Empirically testable as it predicts prevalence of anomalies to fall as:

• incentives get stronger / more realistic

• More experienced subjects gets

Esponda et al (2024) - OV

Tests effect of experience on judgment of probability

• testing for Bayes rule application

In each round subject given question with same underlying base rate and signal probabilities

• asked prob of success or failure

Incentivised to give what subject thinks true answer is

200 rounds + 2 treatments

• Primitives - subjects told underlying probabilities at start + at end see result (success or failure) & record of signals / previous outcomes

  • Could work out in rd.1 and give correct answer in all rounds after

• No Primitives - Subjects never told underlying probabilities

  • only have experience to rely on

Design provides signal & outcome experience in both treatments

Esponda et al (2024) - prediction

Predicts many subjects will answer wrongly the first time

• Perefect base-rate neglect → think test reliability is probability

• Base-rate neglect → thinking prob < reliability BUT > true value

• Perfect signal neglect → Thinking probability is random chance as reliability not 100%

Previous studies suggest base-rate neglect strong when inexperienced subjects face this kind of problem

Esponda et al (2024) - RESULTS

Both treatments not near true Bayes answer in rd.1

Judgement of prob moves in right direction for both BUT learning flattens off after rd.100

No-Priv learns a lot faster in both cases

No-Priv more accurate at end of study

• more explicit information not always helpful - incorrect mental models cause persistent deviation