Eliciting Measures of Value for Health and Safety

1
Eliciting Measures of Value for Health and
Safety

• Michael Jones-Lee and Graham Loomes
• Universities of Newcastle and East Anglia, UK

2
The Central Question

• When innovations/interventions impact upon
• (risks to) length and/or quality of life,
• how do we balance those impacts against other
• costs and benefits?
• Conventional economists answer
• Be guided by the values of the people affected

3
The Stylised Individual Decision

• Balancing benefits from different goods/services
• All converted to the same currency utility
• Individuals maximise utility within wealth
  constraint
• Health/safety goods assumed to be like any other
• i.e. to be balanced against other uses of
  resources

4

• At the individual level, such trade-offs can be
• expressed in monetary form as
• Willingness To Pay - WTP
• (Or sometimes, Willingness To Accept WTA)

5

• So if some measure reduces your risk of death
• by 1 in 100,000 each year, what are you WTP?
• Ask a cross-section, then extrapolate
• If 1 million people WTP average of A X p.a.,
• they raise a total of A Xm to prevent 10
  deaths
• So if X is, say, 30, VOSL/VPF A 3m
• (In UK, currently 1.25m inc.)
• Then plug this value into cost-benefit analysis

6

• Persistent Practical Problems
• Observed patterns of responses raise concerns
• OVERsensitivity to things that SHOULDNT matter
• UNDERsensitivity to things that SHOULD matter

7

• Oversensitivity To
• Things That Should Not Matter
• For example, the particular elicitation
  procedure...

8

• To allow for difficulty of giving precise values
  for
• things which are unfamiliar and complex, elicit
• point estimates within confidence intervals
• Ask what people are SURE they WILL pay, until
• they stop being sure establish a MINIMUM value.
• Ask what they are SURE they WONT pay, until
• they stop being sure establish a MAXIMUM value.
• Ask for BEST point where it is hardest to
  decide.
• But are the points influenced by the procedure?

9

10

11

12

13

14

• Hard to defuse
• NZLTSA Computerised, repeated
• weaker with repetition, but persisted
• DEFRA Random card sort
• first card effect
• Range effects 0-500 vs 0-1500
• Too much attention to irrelevant cues
• but at the same time

15
Insufficient Sensitivity ToThings That Should
Matter

• Within-person insensitivity to
• SIZE OF RISK REDUCTION
• S12 gt S4 S12 S4 S12 lt S4
• 180 178 37
• (45.5) (45.1) (9.4)

16

• Within-person insensitivity to
• SEVERITY OF INJURY
• Of the 360 who ranked R strictly worse than S
• R4 gt S4 R4 S4 R4 lt S4
• 191 131 38
• (53.1) (36.4) (10.5)

17
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 theyd 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
• Which figure should policymakers use (if either?)

18

• Not just a problem of probabilities
• Numbers of deaths prevented in own area
• Extra months of life in normal health
• and so on
• What is the root of the problem?
• What if anything can we do about it?

19

• Conventional model works best for things that are
• real
• experiential and/or uncomplicated
• repeatedly tried chance to learn, reflect,
  refine
• But we want to know about things which are
• hypothetical
• abstract and complex
• one-off little chance to learn or reflect
• Are we asking too much?
• Can we ask it differently?

20
Some Alternative Approaches

• Dichotomous choice (market-like)
• but taps into SP effects plus yea-saying
• De-cue simple open-ended
• but people look for their own anchors
• Decompose and chain
• but only as strong as the weakest link

21

• Relate to existing peg
• but different questions give different
  relativities
• DCE/Conjoint some apparent success
• but not thoroughly stress-tested
• Contingent Ranking under investigation
• but maybe set-specific analytically complex

22
Implications for Policy?

• If responses are so vulnerable to effects
• how well can they inform policy?
• We need to
• always build in checks
• aim to use more than one procedure/variant
• try to understand directions of bias
• be prepared to exercise (explicit) judgment
• and
• remember the alternative

23
Eliciting Measures of Value for Health and
Safety

• Michael Jones-Lee and Graham Loomes
• Universities of Newcastle and East Anglia, UK