Lecture 3 - Why has econ not always included experimental

Standard theory vs Endowment effect

ST

• preference captured by indifference maps

• Constraints captured by feasible separate sets of consumption bundles

• Choice will be same if same feasible set offered

Endowment effects

• People more attached to something if they already have it

• Preferences depend on endowment + feasible set not only aspect driving choice

Endowment effect example - design

Random assignment into groups

-              Group 1 given mugs but allowed to swap for choc

-              Group 2 given chocolate but allowed to swap for mugs

-              Group 3 allowed to choose either

Endowment effect example - Standard Theory prediction

Proportion of subjects who end up with a mug should be the same in each group, but for random sampling variation

Endowment effect example - Result

Observed effect known as endowment effect (Goes against ST that preferences do not vary with endowment)

Benefits of experimental econ

-              Isolation from extraneous influence

-              Different treatments can be identical apart from 1 aspect

-              Random assignment of treatment – allows for causal interpretation

-              Experiments can be replicated to reduce chance that treatment effect is by chance

3 uses of experimental control

• Holding things constant across treatments - nothing varies except initial endowment / treatment difference

• Manipulating systematically - presence / absence + nature of endowments/treatment

• randomising - Random assignment of treatment reduces SB

Methods of control often used in tandem

Internal validity

Design warrnats reliable conclusions about an effect that occurred in the experiment

External validity

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

Experiments & validity

Well controlled experiments have high internal validity but external validity is always more questionable

Artificiality concern of lab experiments

experimenters look at differences between treatments rather than treatment vs no treatment – makes it more valid

Incentives in experiments

-              Not just to gain/retain attention of subjects.

-              To create close relationship between lab objects and theoretical concepts

-              To create real mini-economic systems in lab, showing that “artificial” does not imply “unreal”

Induced values overview

Vernon Smith’s approach to find out S&D curves

• Hard to know in real markets (econometrics helps) – we know p & Q but not curves

Induced value - experiment

By using induced values, experimenter can create a market in the lab where D&S curves known to experimenter before trading begins

Then compare observed p and Q in this market to theoretical prediction

• given by intersection of induced demand and supply curves.

Method uses financial incentives attached to intrinsically worthless objects, so subjects have no unobserved reason to want the tokens

Induced value results

Initial subjects don’t know the induced S&D curves so often announce p initially not at actual convergence levels

However experiments show they converge to induced equilibrium relatively quickly (even with few traders on either side of market)

• Against ST

Experiments in Econ

Huge growth since 80s

• even in top 5 journals

Both lab & field (lab more)

Artificiality concern in Econ

artificial lab environments simply irrelevant to economics

• Friedman (1953) view of prediction success (not including in artificial environments)

artificial (lab) environments have internal validity but do not generalise and so liable to be misleading about other contexts

• Lab subjects may behave differently than agents outside lab would

• Unrepresentative samples + Unfamiliar tasks + Behaviour distorted by experimenter demand effects OR effect of being observed

Artificiality concern in Econ addressed

Experimental research often focussed on treatment differences rather than treatment vs no treatment

Factors held constant across treatments should notaffect treatment differences

F(53) view fell out of favour in modern econ - experiments showed artificial(lab) markets can be real markets.

F53 vs Hausman

Hausman uses a parody in the same form as F53’s argument to undermine conclusion of the latter

Vernon Smith & experiments

“The experimental laboratory .... consists of a …. more complex set of circumstances than is parameterised in our theories. Since the abstractions of the laboratory are orders of magnitude smaller than those of economic theory, there can be no question that the laboratory provides ample possibilities for falsifying any [such] theory”

If lab environment has more complex realistic detail than is described in a (given) theory, then the lab environment is realistic enough to test that theory

“Lab richer than theory” requirement often met (esp. in his market experiments)