individual decision experiments and public policy

Individual Decision Experiments and Public Policy

Graham Loomes

University of East Anglia, UK

The Value of Health & Safety

When measures affect risks to length and/or quality of life, how do we balance those effects against other costs and benefits?

Economists’ answer: be guided by the preferences of the people affected

Elicit VPF (VoSL), VPI, VoLY, QALY to feed into CBA or CUA

Different ‘stated preference’ methods:

Contingent valuation (CV / WTP)

Dichotomous choice

Discrete choice experiments (DCE / SP)

Standard gambles / risk trade-offs etc.

Requirements:

1. Within a method, the value should be independent of the particular parameters of the question

2. Orderings over alternative goods/measures should be consistent across methods

But these requirements are systematically violated

Persistent Practical Problems

UNDERsensitivity to things that SHOULD matter

OVERsensitivity to things that SHOULDN’T matter

Insufficient Sensitivity to Things

That SHOULD Matter

Size of risk reduction or gain in life expectancy

Severity of injury

Period of payment

Other opportunities

Why Insensitivity Matters

One safety measure reduces risk by 1 in 100,000

Another reduces the risk by 3 in 100,000

3 out of 10 said they’d pay the same for both

Another 4 would only pay up to twice as much

Extrapolating to 1 million people:

£98m to prevent 10 deaths: VPF = £9.8m

£138m to prevent 30 deaths: VPF = £4.6m

Also for extra months of life:DEFRA Air Pollution study

1 month 3 months 6 months

60.15 67.72 80.87

(25) (30) (40)

Implied VoLY

27,630 9,430 6,040

Also: severity of injury/illness; period of payment;

other opportunities

Oversensitivity to

Things That Should Not Matter

1. Variants within a method e.g.

starting point in an iterative procedure

2. Variations between procedures:

e.g. CV vs SG

CV vs SG

CV SGR:Death 0.875 0.233S(4) 0.640

0.151S(12) 0.262X 0.232 0.055W 0.210 0.020

Other Possible Approaches

Dichotomous choice: market-like; but ‘yea-saying’?

DCE: infer from simple choices; but can these be TOO simple and subject to ‘effects’?

Ranking: extra complexity – and effects of its own?

Study problems and properties experimentally …

Experimental studies using lotteries:

familiar consequences

known & comprehensible probabilities

incentive-linked

Money value (certainty equivalent)

vs

choice

vs

probability equivalent

Eliciting CE (CV)

What value of X makes you consider B just as good as A?

25% 75%

A $80 0

B $X

100%

What value of Y makes you consider B just as good as A?

70% 30%

A $24 0

B $Y

100%

Eliciting PE (SG/RTO)

What value of p makes you consider B just as good as A?

70% 30%

A $24 0

B $160 0

p% 1-p%

What value of q makes you consider B just as good as A?

25% 75%

A $80 0

B $160 0

q%

1-q%

Choose whichever of the two you prefer

25% 75%

A $80 0

B $24 0

70% 30%

Classic preference reversal:

First lottery given higher CE

Second lottery preferred in straight choice

Opposite reversal – value the second higher but choose the first – much less often observed

But what if elicit PE rather than CE?

Methods of elicitation

Open-ended

Iterative choice

Dichotomous choice

All produce classic PR asymmetry for CE (CV)

Evidence of opposite asymmetry for PE (SG)

Choice vs

CE CEP > CE$ CEP = CE$ CEP < CE$

Chose P 10 1/3 9 1/3 44 2/3

Chose $ 2/3

2/3 23 1/3

Choice vs

CE CEP > CE$ CEP = CE$ CEP < CE$

Chose P 10 1/3 9 1/3 44 2/3

Chose $ 2/3

2/3 23 1/3

Choice vs

PE PEP > PE$ PE$ = PEP PEP < PE$

Chose P 59 2/3 4 2/3

Chose $ 16 1/3 8 1/3

CE vs PE PEP > PE$ PE$ = PEP PEP < PE$

CEP > CE$ 10 0 1

CEP = CE$ 8 0 2

CEP < CE$ 57 1 10

Exploring reasons & possible ‘solutions’

Imprecision / error

Market discipline

Embed in broader set, rank and infer values

Cut off one head, two more grow …

Set 1

EV

A 0.5 x £25 12.50

B 0.75 x £15 11.25

C 0.6 x £15 9.00

D 0.85 x £10 8.50

E 0.7 x £12 8.40

F 0.9 x £9 8.10

G 0.95 x £8 7.60

H 0.8 x £9 7.20

I 0.4 x £18 7.20

J 0.55 x £13 7.15

Set 1 Set 2

EV EV

A 0.5 x £25 12.50 K 0.25 x £6 1.50

B 0.75 x £15 11.25 L 0.2 x £9 1.80

C 0.6 x £15 9.00 M 0.15 x £15 2.25

D 0.85 x £10 8.50 N 0.1 x £25 2.50

E 0.7 x £12 8.40 P 0.1 x £60 6.00

F 0.9 x £9 8.10 Q 0.15 x £45 6.75

G 0.95 x £8 7.60 R 0.2 x £35 7.00

H 0.8 x £9 7.20 S 0.3 x £25 7.50

I 0.4 x £18 7.20 I 0.4 x £18 7.20

J 0.55 x £13 7.15 J 0.55 x £13 7.15

Inferred values for I and J

Set 1 Set 2

I 5.80 8.64

J 5.82 8.40

Set 2 > Set 1 Set 2 = Set 1 Set 2 < Set 1

I 121 16 17

J 120 28 6

But if control for other items in Sets, PR goes

Even so, ordering from ranking diverges from ordering from pairwise choices

Especially in area where choice ‘anomalies’ have inspired ‘alternative’ theories

But rankings still don’t conform with standard theory

Implications for Policy?

If responses are so vulnerable to ‘effects’ how well can they inform policy?

We need toalways build in checksaim to use more than one procedure/varianttry to understand directions of biasmaintain two-way traffic between lab & fieldbe prepared to exercise (explicit) judgment, with data from experiments an important input into those judgments and their justifications

Individual Decision Experiments and Public Policy

Graham Loomes

University of East Anglia, UK