Notes from meeting on Mortality Risk Valuation Survey, Ashdown House, London, 21st February 2000.

Present: Brigitte Desaigues, Rainer Friedrich, Alistair Hunt, Alan Krupnick, Pamela Mason, Riccardo Scarpa

Background to the Project - NewExt: the latest ExternE project:. Innovations are in:

(i)Valuing impacts to ecosystems and biodiversity. Due to the difficulty with using the impact pathway approach to ecosystem valuation, the project will use the avoidance cost approach.

(ii)Extending the analysis from air pollution to impacts to soil and water

(iii)Valuing accidents associated with non-nuclear energy production, e.g. gas explosions, oil fires.

(iv)Monetary valuation of mortality impacts of air pollution.

VSL vs. VLYL

The first part of the discussion focussed on the issue that had arisen in discussion over the previous two days, namely the question of whether the benefit of reduced risk should be measured in terms of risk, or in terms of life-expectancy. Rainer pointed out that during the ExternE project, the team concluded that for air-pollution issues, the appropriate measure was Life Years Lost, and furthermore that these should be measured directly, rather than being converted from an accidental-fatality VSL measure. Alan pointed out that there is no disagreement that the VLYL is preferable to the accidental-fatality VSL, but that in this context, an age-dependent VSL is the best of the three.

Moreover, as we saw from Hammitt’s presentation, a decrease in the risk of dying (of, say, 5 in 1000) between 70 and 80 translates to an increase in life expectancy (of about 5 months), due to the fact that the survival function shifts out. Therefore, the values that we elicit for risk changes can be interpreted as being equivalent to increases in life expectancy. We feel that this is the best way to elicit these values, even if the final objective is to value an increase in life expectancy, since it is not appropriate to describe to respondents a situation in which time (e.g. an extra year) is “tacked on” to current life expectancy – any change in risk will change the risk of dying over a period of time.

To underline this point further, it was noted that when respondents value a decrease in mortality risk, it is not the case that one is valuing an increase in life of three weeks while another values an increase of three years, even though this could be the ex post result. They value a decrease in the risk of dying over a period of time, which implies a single increase in life expectancy. The relative values of different increases in life expectancy could be measured by valuing two risk changes.

It was acknowledged that the Bath team will prepare a note on the translation of values for age-dependent VSL into values for increased life expectancy.

The Contractual Requirements

The description of work for the valuation workpackage of the NewExt project was discussed briefly – the basic remit is to apply the methodology as developed (international comparability was cited as an advantage of doing this) in order to value decreases in risk and life expectancy. As per the previous section, the results of the methodology can be used to value changes in both risk and life expectancy.

The Study Methodology

(i) Focus Groups

-Focus groups should be between 1.5 and 2 hours, taking perhaps half an hour at the beginning to take the survey (perhaps supplemented by handouts for the new parts if you can’t put in the visual basic).

-In the beginning the most important thing is to get a feel of how the issues play in the particular country, randomness isn’t such an issue, and therefore we can use “convenience sampling”, e.g. a group of support staff from the university. Brigitte noted that it is also helpful to allow individuals to work through the study on their own, and debrief them. It’s important to have their answers during debriefing – Serge has worked out how to download the answers quickly and could let the other partners have this tool.

-Alan will let the teams have tapes of some of their focus groups.

(ii) Pre-test

-For pre-testing, randomness is more important – this would be a pre-requisite for publication

(iii) Full Test - sampling

-In France the standard sampling practice is “representative sampling” rather than random-digit dialling; the former is apparently cheaper.

-The only restrictions in terms of sampling are that all respondents must be over 40, and a third of respondents must be over 60.

-One question, given the limited funds, is the possibility of not using the professional recruitment company, but rather to see if we can allow (non-random) respondents to access the survey from their homes. Alan’s team have used Knowledge Networks in the US, which provides randomly sampled respondents via web TV for $50 per respondent. This service almost certainly isn’t yet available in Europe. Riccardo however has used Neilson in Europe, which provides representative respondents with computers in their homes. This is likely to be more expensive, but we could look into this as a possibility. Riccardo is also conducting a web-based survey (on food safety and humane animal production) for which the survey was supported on their server. This could also be a possibility for our study.

-A key decision, to be discussed with Anil, is given current funds, and if it is feasible, are we willing to trade off decreased representativeness (through use of the web) for increased accuracy through larger sample sizes?

(iv) Full Test – Bid Design

-One issue for the US survey was that the bids should be designed so as to allow for the possibility that the implied VSL was as great as, or greater than, the $6 million currently used in regulation. We should think about any such necessary restrictions in bid design.

-Riccardo suggested that the efficiency of the study could be increased if the bids are redesigned after preliminary results (e.g. from the first 100 responses) are analysed. Alan’s team used the pretest for this, but it’s possible that we could get better results by using some of the full study.

-We should start with the bids used in the US study (converted using PPP) and then if we get reasonable distributions, continue with those. If not, use Bayesian updating. Further information on the need to do this will be provided by pre-testing.

The US and Canadian Studies were carried out as follows:

TimeWave 1Wave 2

t0 – t10WTP 5WTP 1

t0 – t10WTP 1WTP 5

t70 – t80WTP70 5WTP70 5

Our interests are (a) How older/ill people value decrease in contemporaneous risk

(b) How people value decrease in future risk. This question is asked

only to the 40-60 year olds in the current protocol.

The ordering effects of having the WTP70 5 question last have not yet been tested, since this question was third on both surveys. Alan’s team is currently looking into re-developing the survey to deal with this by valuing risks specifically from cancer. However, to do a non-specific, generalisable study with the WTP5_70 question first is problematic and would require survey development and might not be successful.. This is something the EU studies could explore if resources allow.

We currently have enough money for one treatment. We could do Wave 1 as this treatment, and then do Wave 2 if extra funds are available. Alan’s team is currently working on putting the latency question first in the cancer context, and will send us some tapes from focus groups at a later date.

Also, given that we would like to test different reductions in future risk (thus implicitly different increases in life expectancy) we could test WTP70 5 and

WTP70 10. We could do:

TimeWave 1Wave 2

t0 – t10WTP 5WTP 1

t0 – t10WTP 1WTP 5

t70 – t80WTP70 5WTP70 10

However, Alan’s team is not happy with the second question in each wave – the internal scope test is of limited use due to ordering effects. It would be nice to make the survey shorter, but there’s room to do more. Therefore as the first wave, we could do:

TimeWave 1Wave 2

t0 – t10WTP 5WTP5 (or 1 – decide

later)

t70 – t80WTP70 5WTP70 10

t70 – t80WTP70 10WTP70 5

WTP70 10 would represent a very big reduction in pollution, but it’s probably better than WTP70 1, which is too small to be valued properly. An alternative would be to do

WTP65 5, but as it is, we only have 2/3 of the sample answering these questions, the rest being already over 60, so we can't lower the cut-off age any further. There would then be an overlap between the payment period and the risk reduction period. We could move the risk reduction period to 75-85, or make it from 70 onwards, but the latter is problematic. The general feeling was that WTP70 10 is the best option. Alan recommends that we do Wave 1, drop WTP 1 and add WTP70 10 to the end. We should lose very little in terms of internal scope testing (although Alan will check with Maureen and Anna on this), though there will be ordering effects.

The current protocol asks respondents how likely they think they are to survive to 70. A nice extension that we could make is to ask people how likely they are to reach 80 (and 75 and 85 if we were to shift the risk period).

- Brigitte noted that in her pre-testing, respondents were very unclear about the notion of future risk. Alan pointed out that using the voiceover is extremely important for the difficult concepts. For the web version, due to the size of the voice files, only the difficult bits had voiceovers.

- Riccardo noted that in the food safety study they are using Multivariate Probit to check ordering effects. He will talk with Anna about the possibility of using this and other econometric techniques.

(v) Design of the future health question

We should think about adding questions on expected future health status to use in explaining responses to WTP5_70. Currently there is only a question about comparing current health to future health. The survey also contains questions on respondents' subjective probabilities of reaching 70, and questions on the health of their parents and siblings—additional variables that would be useful in seeing if WTP5_70 varies with expected future health. One possibility is to redesign some of the standard questions on current functionality in order to ask respondents to rate what they think their future functionality will be. This could come immediately before the future health question. This would be a nice extension of the existing methodology, and shouldn't affect comparability.

-Race cannot be mentioned by law in the French study, but is a significant issue in the US study; survival probabilities are different for African Americans, and they also have a higher WTP for risk reduction. This probably won't be such an issue for the UK and Italian studies.

(vi) Data Analysis

-There is money for the US team, they can use this to do some econometric analysis. All the data files should be sent to Alan, they will duplicate the econometric procedure of their protocol for the EU data. After that, the country teams can do what econometric analysis they wish.

-Each data set will have its own peculiarities, e.g. the way that income, age, current health, and certainty of responses are affecting WTP; the country teams should get a good feel themselves for these issues, and share the results.

-Very few of the completed responses are perfect in all respects. However, in the studies conducted by Alan's team, the biases (e.g. those who get the probability questions wrong, those who are less certain of their answers etc.) seem to more or less cancel out. Imperfect responses don't need to be dropped, they can be adjusted for in the econometric analysis e.g. by including dummies.

-Riccardo noted that thinking about the conceptually relevant econometric issues, and developing techniques, would increase chances of publication. Anna is already thinking about a couple of such issues, including endogeneity, and may be willing to share her thoughts.

Funding Issues

-The possibility of finding further sources of funding was briefly discussed. The WHO might be a possibility but they are not particularly well disposed towards valuation.

-Further EU funding will require a justification as to what's being done that's extra, but we can do this. Involving a transitional/accession country might help.

-Funding for further studies should be relatively easy to come by once we have a standardised product.

Comparability

One of the most important issues is that we should minimise the changes made to the survey instrument. All changes to the wording of the instrument should be sent to Alan so he can check that they won't constitute a substantial change. In the international comparison, any differences in methodology will be picked up as being responsible for differences in the results. Pressures for changes will arise (in particular, all of the country teams will need to alter the insurance part of the survey) but Alan asks that the burden of proof of the need for changes be high, since the costs of lost comparability are high.

-Brigitte noted that they have been planning to use monthly bids rather than annual bids as in the current survey, since French people pay for health products on a monthly basis. Alan felt that this could constitute a substantial change; a compromise is to state the annual figure first and put the monthly equivalent in brackets, although even this might be an important change. Ideally one would test for the effects of expressing the bid in annual or monthly amounts by doing a split sample. There should also be no suggestion as to the nature of the product.

-Brigitte noted that people need to imagine the nature of the risk in order to conceptualise it. She'll report on her debriefings to illustrate this.

Publications

- The US and EU teams agree to joint authorship on articles coming out of the EU country applications of the survey and expect that all members of the teams will be on an article that examines cross country values. Joint authorship would not necessarily apply to all uses of the data, e.g. the use of the data set to motivate some new econometric test or estimator. All parties agree to address these types of authorship issues on a good faith, case by case basis.