Transcript 3 country application of Alberini/Krupnick survey

3 country application of
Alberini/Krupnick survey instrument
– Methodology and Results
Alistair Hunt and Anna Alberini
University of Bath &
University of Maryland
For UK Defra Workshop 21.06.04
• Theoretical basis for valuation of Mortality
risk changes
Life Cycle model
• at age j, max expected utility over
remaining life time:
max  V (c j t ) q (1   ) j t
t
jt
Definition of VSL
dWTP
1
U (Cj  t )
j t
VSLj 

q
jt
(
1


)

dDj
1  Dj t U ' (Cj  t )
• Ambiguous net effect of age j on VSLj
Study Features
• Survey-based; UK, France, Italy
• Directly values mortality risk changes
• Uses framework methodology developed in
N.America
• Targets age group 40+
• Computer-based; self-administered; voiceover
Methodology Adaptation
Testing comprised:
– 10 one-to-one 1-2 hour in-depth interviews
– 3 1-hour focus groups (8 participants)
And aimed to clarify linguistic and comprehension issues
whilst retaining comparability with N.American
instrument
– In UK, 330 people surveyed: recruited in 30-mile radius
around Bath, SW England, using specialist recruitment
company
Sample size and experiment design for the
three-country study.
UK
Italy
France
No.
330
292
299
Locale of the Study
Bath*
Venice, Genoa,
Milan and Turin
Strasbourg
Experimental Design
Wave 1
Wave 1
Wave 1 and wave 2
*
•recruited within 35 Km of Bath.
•Random digit dialing, in-street recruiting and snowballing
•Eligible and contacted: 1350. Cooperative: 355. Finally attended: 330.
Structure of Survey Instrument
• 5 sections
– Personal information
– Introduction to probability concepts
– Causes of death; risk-mitigating behaviours and
associated costs
– WTP for risk reductions
– Debriefing and socio-demographic questions
Introduction to probability concepts
Causes of death; risk-mitigating behaviours
and associated costs
WTP for risk reductions
• Dichotomous-choice approach with two follow-up
questions and final open-ended question
Respondents are asked to value:
a 5 in 1000 risk reduction spread over the next
10 years, with effect immediately;
– a 1 in 1000 risk reduction spread over the
next 10 years, with effect immediately and;
– a reduction of 5 in 1000 over the ten years from
age 70.
–
Initial and follow-up bids in the UK study. (£)
Initial bid
Bid if response to
first payment
question is no
Bid if response to the
first payment
question is yes
45
20
100
100
45
325
325
100
475
475
325
650
Debriefing questions
•
•
•
•
•
•
•
understanding of idea of ‘chance’
accept specific baseline?
specific product in mind? Yes – what kind of product?
Doubts about product? Yes – influence WTP?
Did you think you would suffer any side-effects?
Did you consider whether you could afford payments?
Think of other benefits? Yes - to yourself, others, for you
living longer, improved health
Yes – influence WTP? – raise/lower? Other people
• On WTP 70 did you consider whether
– would live to age 70?
Or your health at age 70?
• Household Income
Health Status data
• Gathered from application of short-form
(SF 36) questions within survey instrument
– Series of questions relating to respondent’s
current and historic physical and mental health
status
• Results of Survey application in EU
Descriptive Statistics of the Respondents’ Sociodemographics. Sample averages or percentages for
selected variables
UK
Italy
France
Age
58.03
57.04
55.35
Male
49.39%
48.63%
47.29%
Income in EUR
Mean
Median
40,096
38,690
40,115
25,000
32,186
32,012
14.10
12.99
11.04
Education
(years of schooling)
Health status of the respondent
• Elicited using three sets of questions:
• -- direct question: “Compared to other people your
age, how would you rate your health?” (Excellent,
very good, good, fair, poor)
• -- direct questions about specific illnesses: “Has a
health care professional ever diagnosed you to
have…” (list of cardiovascular and respiratory
illnesses)
• -- Short Form 36 questions about general health
and functionality
.
Health status of the respondents
Percentages of the sample with specified conditions
UK
Italy
France
Rates own health as good or
excellent relative to others
same age
High blood pressure
60.79
42.12
38.46
28.48
33.33
21.07
Any chronic cardiovascular
disease (CARDIO)
Any chronic respiratory illness
(LUNGS)
Cancer (CANC)
8.18
15.41
12.37
15.45
12.67
18.73
6.36
6.85
6.35
43.33
44.86
39.46
High blood pressure or other
cardiovascular illness, or chronic
respiratory illness, or stroke
(CHRONIC)
Percent of the sample who have various problems
with risk comprehension
Based on complete samples
A. Wrong answer in the
probability quiz
B. Confirms wrong answer
in the probability quiz
C. Probability choice qn:
- prefers person - higher risk
- indifferent
D. Confirms wrong answer
in probability choice
question
A and C (FLAG1=1)
UK
Italy
France
15.33
11.64
22.74
0.91
2.74
4.01
14.29
6.97
11.99
10.96
10.37
22.41
1.52
3.08
1.34
2.45
3.77
2.01
Responses to starting bid values
UK Study:
Percentage respondents willing to pay
5 in 1000 immediate risk reduction
80
70
60
50
Percentage
40
Yes
30
20
10
0
71.11
70.73
48.75
41.03
45
100
S1
325
Bid Amount (British Pounds)
475
Responses to immediate & future risk
reductions
UK Study: Comparison of the % willing to pay for the
immediate and future risk reductions
71.11
80
70.73
70
60
48.75
50
45.45
41.03
36
Percentage Yes 40
30
20
19.51
10
19.05
0
45
100
325
Bid Amount (British Pounds)
475
Percentage of respondents with WTP = 0
Risk reduction
Sample size
Percentage
respondents with
zero WTP
5 in 1000 over the next
10 years (immediate)
330
15.76
1 in 1000 over the next
10 years (immediate)
330
42.12
5 in 1000 between
ages 70 and 80
187*
41.71
* = only respondents up to age 60 were asked to value the future risk reduction
Statistical Model of WTP
 Double-bounded model of WTP
 Weibull distribution of WTP with scale parameter  and shape 
 Log likelihood function:

  WTP L
i
log L   logexp  
  

i 1


n





U


  exp   WTPi

  








 ,


where WTPL and WTPU are the lower and upper bound of the interval around the
respondent’s WTP amount.
1

 Mean WTP =     1 , where  is the gamma function


 Median WTP is equal to
   ln( 0.5) 
1

.
UK Study: Annual WTP Figures
Immediate 5 in 1000 Risk Reduction
In Euro
(s.e.)
In £
(s.e.)
Implied annual
VSL
Mean WTP
672
(86.02)
460
(60.27)
€ 1.344 million
or
£ 0.920 million
Median WTP
354
(34.23)
242
(23.89)
€ 0.708 million
or
£ 0.484 million
*cleaned data (FLAG1=1 deleted); n=322
Internal validity of the WTP responses
 Can be checked by letting the scale parameter of the Weibull be
 i  exp( x i β ) ,
where xi is a 1p vector of regressors, and  is a p1 vectors of coefficients.
 In other words,
log WTP  x i β   i
distribution with scale .
, where  follows the type I extreme value
Pooled data interval-data regressions for WTP.
Immediate 5 in 1000 risk reduction.
Intercept
Household income (thou. Euro)
Age 50-59
Age 60-69
Age 70 or older
Male
Education
Chronic resp or cardio illness
visited ER < 5 years – cardio/ resp
Has or had had cancer
France dummy
Italy dummy
Weibull Shape parameter ()
Coefficient
St. error
5.8024**
0.0098**
0.0245
0.2056
-0.0748
-0.1842
0.0072
0.076
0.5944*
0.4397
0.8636**
0.6705**
0.7400
0.386
0.0031
0.190
0.204
0.256
0.142
0.024
0.152
0.282
0.315
0.214
0.162
0.044
Respondents with FLAG=1 excluded.
* = significant at the 5% level; ** = significant at the 1% level.
Pooled Data: Annual WTP Figures
Immediate 5 in 1000 Risk Reduction
In Euro
In £
Implied annual
VSL
Mean WTP
988
677
€ 1.977 million
or
£ 1.354 million
Median WTP
478
328
€ 0.956 million
or
£ 0.656 million
Summary of results
• UK sample is very small: no statistically significant
association between WTP and age or health.
• Pool data to increase sample size, but account for different
cultural factors and sampling procedures through country
dummies
• Age is not significant associated with WTP, although the
oldest respondents tend to have lower WTP
• Of the health status dummies, dummy for hospital
admission or ER visit in the last 5 years is strongly
associated with WTP
• Income is significantly associated with WTP
• Gender and education not important
Relating WTP with predictions from epidemiological
studies
Regressions of WTP on proportional risk
reduction (5 in 1000 immediate risk reduction)
(cleaned data)
coefficient
Standard error
Intercept
6.3047**
0.1049
France dummy
0.7788**
0.2041
Italy dummy
0.4400*
0.1892
Proportional risk
reduction
(=5 / baseline risk)
Weibull shape
parameter ()
0.9851*
0.4862
1.3809**
0.0816
Relating WTP with predictions from
epidemiological studies
• study values redns in risks  VSL but can couch in terms
of  in remaining life expectancy (or loss/gain of
days/months of life spread over the population)
• Rabl (2001) derives  in remaining L.E. associated with 5
in 1000 risk change over next 10 years
– averages 1.23 months (37 days) for our sample.
Derived VOLYs
UK
EU (Pooled)
Annual WTP
Median
Mean
VOLY
242.22
460.20
328
677
Median
Mean
22,080
41,975
30,203
64,788
Latency
WTP



WTP
j , j 
j , j  e
q
j , j 
2 Step estimation of discount rate
• Immediate 5 in 1000 Risk Reduction → predict
WTP70,70
• Regress log WTPj,70 on log WTP70,70 (coefficient
restricted to 1); log ρj,70 (coefficient restricted to 1)
-Δ=j-70 → coefficient is δ
RESULTS
• UK δ≈ 10%
• France δ≈ 5%
• Italy δ≈ 6%