3.1.1. Incidence versus prevalence

Incidence is the number of new cases in a given time period and prevalence is the frequency of the disease in the population. Prevalence is reduced by reducing incidence. As an example, in the United States, the incidence of childhood asthma in children at risk is 12.5 in 1 000 and the incidence of adult asthma in adults at risk is 4 in 1 000 (Winer et al., Reference Winer, Qin, Harrington, Moorman and Zahran2012). The prevalence of asthma is about 80 in 1 000 in the US population and approximately the same in children and adults. Because the aim of the study is to elicit the value of a statistical case of asthma, the baseline risk presented to respondents in the survey was based on adult asthma incidence of at-risk adults. This ensures that the presented baseline risk, although overestimated, is a realistic approximation and is therefore credible. However, smaller risk values such as 4 in 1 000 are generally more difficult to grasp for most respondents. Therefore, it was decided to present a baseline risk over 10 years equal to 40 in 1 000 or 4.0% and to ask respondents what they are willing to pay per year over 10 years to reduce this risk. The baseline “total” risk is broken down into severity groups (mild, moderate, and severe) in the population. Mild asthma is more prevalent than moderate asthma, which is more prevalent than severe asthma.

3.1.2. Duration

As noted, there is no cure for asthma although childhood asthma may become asymptomatic over time and asthma may develop in an adult, even though they were never asthmatic as a child. In the survey, the baseline risk of developing asthma over a period of 10 years was presented to non-asthmatic adults and parents of non-asthmatic children, and in scenarios, small reductions to that probability if the non-asthmatic uses safer home cleaning and other “SAFETYFIRST” products. Asthma was presented as a non-curable lifetime disease. In the survey, asthmatic adults and parents of an asthmatic child were asked about their willingness to pay to reduce the severity of their asthma during a period of 10 years after, which their severity would go back to its previous severity if they stopped using SAFETYFIRST products (Figure 1).

Figure 1.

Picture used to present original vs SAFETYFIRST household products.

Note: Labels on products were translated in each country. Source: Authors’ own elaboration.

3.1.3. Risk factors

The survey language recognizes that chemicals in household products are risk factors for asthma onset and for asthma attacks. People who buy and use reformulated products are told that these products can reduce their probability of getting the disease or, if they have the disease, reduce its severity. In line with expert views, no scenarios take risks to zero or eliminate attacks for those with asthma, in contrast to some of the literature reviewed earlier.

3.2.1. Asthmatic adults

Starting with the asthmatic adult branch, the survey defines how severity varies, as in Figure 3, and asks respondents to rate their own asthma severity. Although only three levels are provided in the figure, respondents can also rate their asthma as very mild, mild plus (between mild and moderate), and moderate plus (between moderate and severe) to better capture their own severity.

Figure 3.

Visual description of asthma severity levels.

Source: Authors’ own elaboration.

The severity reduction mechanism and payment vehicle are introduced next as a new line of household products specially formulated to contain fewer and safer chemicals, called SAFETYFIRST products. To avoid environmental and other health co-benefits identified in one-on-one interviews, the language specifies that SAFETYFIRST products are as effective as the original products, that both sets of products have the same amount and composition of packaging and that the only difference is in the reduction of attack severity and frequency.

The valuation questions are formulated as DBDCs (Figure 4). For the sake of clarity, the follow-up bid questions and open-ended questions about reasons behind their choices are not presented below, even though they were asked to the respondents.

Figure 4.

Double-bounded dichotomous choice to elicit WTP to reduce asthma severity. Source: Authors’ own elaboration.

There are several elements that need to be pointed out. First, the language used throughout the survey aims to focus people away from valuations for their household and towards valuation for only themselves and (separately) for their youngest child. Thus the $12 per month in Figure 4 is the WTP of the respondent out of their household budget for reducing their asthma severity. The second issue is to focus the respondent away from any savings in medical expenditures from their asthma being less severe so that the estimated WTP will capture the utility associated with a reduced severity. This is easier to assume for European respondents as most do not pay out-of-pocket costs and thus would not think about these savings. However, in countries where patients pay for medical care out of pocket, respondents would think about the savings from reduced asthma severity. Thus, the language telling the respondent to think only of their non-monetary benefits from less severe asthma allows the estimates to be comparable between European countries and other countries. Third, the language stresses the importance that respondents assume they would still buy the same amount of products whether SAFETYFIRST or original products. This type of thinking would bias WTP estimates downward. This problem was checked by using a debriefing question and respondents could also signal their thinking using the open-ended question following their choice.

Respondents were presented with a first bid randomly chosen among five potential values. Table 1 shows the first bid values presented to respondents by country. To enable the comparison of WTP across countries, these values are the same across countries but they were shown in local currencies. Bid values were converted using Purchasing Power Parity (PPP) for actual individual consumption provided by the OECD. The first bid values were tested and updated during survey piloting to aim towards achieving an average yes responses of 50% to the first dichotomous choice. There is a factor of two steps between each first bid value to avoid getting too many “yes”-“yes” responses that do not allow pin down individual WTP. For example, respondents in the United States were presented with first bid values ranging from USD 36 per year to USD 576 per year. Bid values were purposely not rounded to give the impression to respondents that the added costs for SAFETYFIRST products were more realistic.

Table 1.

Bid values for the first dichotomous choice

Note: Bid values as seen by respondents. They were later converted into USD PPP using Purchasing Power Parities for actual individual consumption data for 2019 from the PPPs and exchange rates OECD database. The PPP data was extracted on 22 Feb 2021 at 08:44 UTC (GMT) from OECD.Stat, but has subsequently been revised. The exact series can be provided upon request.

Source: Authors’ own elaboration.

To avoid too many “no”-“no” and “yes”-“yes” responses and tend towards a balance across the four potential outcomes of the dichotomous choice questions, follow-up bids were multiplied by 3 when people responded “yes” to the first dichotomous question and multiplied by 1/3 when people responded “no” to the first dichotomous question. Table 2 provides the bid values that were presented to respondents in the United States.

Table 2.

Bids presented to respondents in the United States, in USD per year

Source: Authors’ own elaboration.

3.2.2. Parents of an asthmatic child

The branch of the parents of an asthmatic child (Figure 2) was identical to the adult asthmatic branch except for minor changes such as replacing the language so that respondents would think about the non-monetary benefits for their asthmatic child and not for themselves.

3.2.3. Non-asthmatic adults

This branch was different from the asthmatic adults’ branch in two main ways. First, the setup was a choice experiment where the respondent was asked to choose among the baseline scenario and two alternatives, lower-risk scenarios. Because asthma can be mild, moderate, or severe, and the expected probabilities of getting asthma with different severities would likely influence WTP, the risks for the different severity levels were included as attributes. Specifically, the observed incidence rates over 10 years in the United States were used to define these three levels of severity for the baseline and reduce those probabilities for the scenarios. The questions were preceded by a practice choice screen, which made it clear, after working extensively with one-on-one interviewees, that the sum of the probabilities across severity levels is equal to the total probability of developing asthma. The choice screen is reproduced in Table 3. Five consecutive choice screens were presented to respondents. Two sets of five choices were developed. One of two choice sets was randomly attributed to each non-asthmatic adult.

Table 3.

Example of choice in the discrete choice experiment to elicit WTP to avoid developing asthma

If the respondent chooses Mix B (resp. Mix C), he will have to pay added annual costs of 252 (resp.36). Source: Authors’ own elaboration.

Because of the use of probabilities, the standard approach to teaching respondents about this concept was introduced at the beginning of this branch using a coin flip, a die-cast, a grid with 100 and then 1 000 people in it colored orange and blue, followed by two probability tests. The first test question showed two grids, each with 100 persons, with most colored in blue and a few colored in orange indicating the probability of getting asthma. The two grids had different numbers of “orange people” and respondents were asked which grid showed the highest number of asthmatics. Figure 5 shows the probability test taken by respondents.

Figure 5.

Question to test respondents’ understanding of probability using visuals.

Source: Authors’ own elaboration, following Krupnick et al. (Reference Krupnick, Alberini, Cropper, Simon, O’Brien, Goeree and Heintzelman2002).

The second test question was formulated as follows: “Which of these two probabilities (risk) of developing asthma is higher? 40 in 1 000 or 30 in 1 000”.

The first test measures the ability to understand the concept of probability, while the second test measures the ability to read a probability as it is then used in the DCE tables.

3.2.4. Non asthmatic child

If non-asthmatic adults have children who are not asthmatic, they responded to the discrete choice set that was not attributed to them when they had responded as adults.

3.2.5. Debriefing questions and final questions and information

The debriefing section was identical for all pathways through the survey. It contained questions on the respondent: (i) attitudes towards the information provided in the survey e.g. did they believe it or act as if they did, was there enough information to make a choice; (ii) behavior in answering the choice questions e.g. did the respondent consider the probabilities and the cost, (iii) considerations when answering the choice questions e.g. did the respondent think about financial consequences to his or her wages or medical bills, understand that the payments for SAFETYFIRST products carried indefinitely into the future to gain the added protection; (iv) yea-saying and protests, and (v) attitudes towards chemicals, e.g. frequency of exposure and adequacy of government regulation.

The final questions covered the respondent’s health status, their socioeconomic characteristics, and their experience with COVID-19 following Mourato and Shreedhar (Reference Mourato and Shreedhar2021). These questions were followed by a statement correcting any wrong impressions the survey may have caused and giving all the facts about the ideas in the survey, e.g. there is actually a high degree of uncertainty about the role of chemicals in asthma.

4.2.1. Very slow respondents

Very slow respondents were identified as taking an inordinately long time to finish the survey and therefore may forget part of it and lose the logic presented. Sometimes, respondents leave the survey to do other things. In the extreme, they may start one day and finish the next. There is no consensus on the definition of a very slow responder. A definition was agreed to be a respondent taking more than 12 h to complete the survey. 2.0% of the non-asthmatic adults (182) and 2.3% of non-asthmatic children (61) were screened out as a result of eliminating parents who were determined to be very slow responders. None of the asthmatic adults or parents of an asthmatic child had to be eliminated from the sample.

4.2.2. Speeders

Speeders are respondents who run through the survey so quickly that they could not possibly be reading the questions. Speeders cannot provide informed responses so they need to be screened out. Two types of speeders were defined in this analysis. First were speeders who completed the entire survey too quickly. Second were speeders who completed the valuation questions too quickly. Respondents who belong to either of these categories were screened out. A standard recommendation from Survey Sampling International (2013) and Mitchell (Reference Mitchell2014) is to filter out respondents that took 48% less time than the median respondent. However, depending on their circumstances, some respondents answered only one section of the survey rather than two – one for themselves and one for their child. Therefore, respondents’ completion time was re-computed and the median was taken for each of the four-valued health effects based on similar sets of survey items. This allows, for example, for the comparison of completion time for childhood asthma risk of non-asthmatic adults to the completion time for childhood asthma risk of asthmatic adults. A unique median per group was computed for each country to take into account differences in language, IT proficiency, and other cross-country differences. The median values were computed based on the sample that excludes very slow respondents.

The speeder criteria screened out 12.3% of non-asthmatic adults (1 123), 10.2% of non-asthmatic children (270), 15.6% of asthmatic adults (406) and 5.3% of parents of an asthmatic child (75). In an informal literature search of speeding in online surveys, 10% was a common finding (Rao et al., Reference Rao, Wells and Luo2014).

4.2.3. Non-attendance of all the attributes of DCE

Following the five consecutive choices in the DCE, non-asthmatic respondents were asked whether they took into account or neglected some of the attributes during their choices. Respondents who indicated they never considered any attribute during their 5 choices were removed from the sample. This criterion screens out 4.5% of non-asthmatic adults (407) and 3.4% of non-asthmatic children (90).

4.2.4. Not understanding probability

The selection criteria used consists of keeping only respondents who passed both probability tests. This rather strict screening strategy can be justified because the initial test measures the ability to understand the concept of probability, while the second test measures the ability to read a probability as it is then used in the DCE tables. Since a random response could result in a correct answer, this aimed to minimize risks of falsely accepting responses from an individual that really did not understand probabilities. This screening criteria excludes 31.6% (2 882) of non-asthmatic adults and 33.0% (872) of parents of non-asthmatic children. Most of them failed the first test, which was more difficult. For example, in the case of non-asthmatic adults, 25.6% failed this first test, while 11.2% of them failed the second test. As a consequence, only 5.3% of non-asthmatic adults failed both tests. There is a significant difference at the 1% level of risk in the average probability test failure rate between countries for non-asthmatic adults and for parents of non-asthmatic children. Table 6 summarises the difference between the full and screened samples in terms of the number of survey responses.

Table 6.

Full and screened sample composition

Note: *Observations from the pilot were not used because the DCE varied too significantly between the pilot and the main stage. Asthmatic adults and parents of an asthmatic child were not presented with a probability test and did not participate in the DCE.

Source: Authors’ own elaboration.

4.2.5. Debriefing questions

Open-ended responses: Respondents had multiple opportunities to give open-ended (OE) answers to explain why they chose to pay for a reduction in asthma severity. The responses to open-ended questions in English and in French from the screened sample were analyzed in two steps for Canada, France, the United Kingdom, and the United States. First, the number of characters of OE responses was computed to get a sense of how seriously respondents were taking this opportunity. A minimal number of characters typically indicates that the respondent did not care much about providing an open-ended answer. Second, all of the open-ended responses for adult and childhood asthma severity in English and in French were analyzed to find a set of keywords that could indicate a problematic response, such as protest, altruism, people who thought they could save on their current medication and people who would pay anything to reduce their risk even by a small amount. Table 7 provides examples of responses for the different categories for asthmatic adults and parents of an asthmatic child.

Table 7.

Examples of responses to open-ended from English speaking asthmatic adults and parent of asthmatic child

Source: Authors’ own elaboration.

For asthmatic adults, 4.3% of responses were related to altruism, 0.8% to protest, 2.4% to people who thought that they would save on asthma medication, and 0.6% related to people saying they would pay anything (Table 8). The remaining 91.8% were short, normal, or excellent responses.Footnote ⁶ For parents of an asthmatic child, 4.3% of responses were protest, 1.1% related to altruism, 1.0% related to people saying they would pay anything, and 0.3% thought they could save on asthma medication by choosing reduced asthma severity. The remaining 93.3% were short, normal, or well-aligned responses. Overall, there were very few problematic responses to open-ended questions. Therefore, respondents were not screened out on this basis.

Table 8.

Share of problematic responses to open-ended for asthmatic adults and children

Note: Computed on all responses from Canada, France, the United States, and the United Kingdom after the screen out.

Source: Authors’ own elaboration.

Closed-ended debriefing questions: The questionnaire included 14 debriefing questions at the end of the survey. Table 9 illustrates the questions for asthmatic adults and parents of an asthmatic child. No respondents were dropped based on their responses to close-ended debriefing questions. This avoids being too conservative by preserving degrees of freedom. It also allows keeping a representative sample since responses to these debriefing questions could be correlated with gender, age, and education. However, responses to these questions were analyzed to check if the results were sensitive to this methodological choice. Debriefing questions are grouped into three groups – those whose problematic response would lead to an overestimation of WTP e.g. “I would pay anything”… (8 questions), those that would lead to an underestimation, e.g. “I thought the probabilities were lower than those presented” (4 questions), and those that were neutral or with an undetermined directional bias, e.g. “I did not have enough information to make an informed choice” (4 questions). The number of overestimating, underestimating, and non-directional problematic responses was computed for each respondent.

Table 9.

Bias direction of problematic responses to debriefing questions for asthma severity

Note: *Only parents of an asthmatic child were asked this question.

Source: Authors’ own elaboration.

The utilization of these variables is illustrated with asthmatic adults and parents of an asthmatic child. For asthmatic adults, 13 debriefing questions were analyzed: 4 can indicate poor non-directional responses, 7 help the identification of potential overestimation of WTP, and 4 help the identification of potential underestimation of WTP. The frequencies of poor answers to these debriefing questions from asthmatic adults are included in Table 10.

Table 10.

Frequency of poor answers to debriefing questions, asthmatic adults

Source: Authors’ own elaboration.

It was found that 20.0% of asthmatic adults respond “poorly” at least once to the non-directional debriefs. 66.6% of asthmatic adults have at least one response that could indicate overestimation of WTP and 79.4% have at least one response that could indicate underestimation of WTP. However, these high percentages are not necessarily problematic considering the large number of debriefing questions. For example, 826 asthmatic adults responded poorly to only 1 out of the 7 debriefing questions, which can help identify overestimation of WTP.

For parents of an asthmatic child, 14 debriefing questions were analyzed: 4 can indicate poor non-directional responses, 8 help the identification of potential overestimation of WTP, and 4 help the identification of potential underestimation of WTP. The frequencies of poor answers to these debriefing questions from parents of an asthmatic child are included in Table 11. It was found that 24.0% of parents of an asthmatic child respond “poorly” at least once to the non-directional debriefs. However, 91.3% of them have at least one response that could indicate overestimation of WTP and 83.7% have at least one response that could indicate underestimation of WTP. Similar to the asthmatic adults, it is not necessarily problematic considering the large number of debriefing questions considered. For example, 61.7% of parents of an asthmatic child responded poorly to less than 3 out of the 8 debriefing questions that can help identifying overestimation of WTP. Comparing with the frequencies reported in Table 11 for asthmatic adults, it was found that there are slightly more poor responses for parents of an asthmatic child.

Table 11.

Frequency of poor answers to debriefing questions, parents of an asthmatic child

Source: Authors’ own elaboration.

To make sure that these “poor” answers do not unduly influence the results, robustness checks were performed, where the model was augmented to include the number of poor responses in these three categories: non-directional, overestimation, underestimation as additional control variables.

4.3.1. Composition of the final sample

After screening, there are 2 194 asthmatic respondents, between 288 and 343 per country (Table 12). There are fewer parents of an asthmatic child, 1 339 in total, between 188 and 194 per country. The final sample includes 5 384 non-asthmatic adults, between 620 and 849 per country, and 1 556 parents of non-asthmatic children, from 168 to 321 per country.

Table 12.

Number of respondents in the screened sample by country and by group

Source: Authors’ own elaboration.

4.3.2. Asthmatic respondents

After applying the screening criteria described in the previous section, the number of asthmatic adults is 2 194, i.e. 85% of the initial sample and the number of parents of an asthmatic child is 1 339, i.e. 95% of the initial database. Supplementary materials A.1 and A.2 show the respective descriptive statistics for the key demographics by country.

4.3.3. Non-asthmatic respondents

After applying the screening criteria described in the previous section, the number of non-asthmatic adults is 5 384, i.e. 59% of the initial sample without the pilot surveys, and the number of parents of non-asthmatic children is 1 556, i.e. 58.8% of the initial sample excluding respondents included in the pilot surveys. A large majority of non-asthmatic adults (70.5%) do not have children under 18 living with them (Supplementary material A.3). Respondents from the Czech Republic account for the lowest share of respondents (11.5%) because many are screened out due to the probability tests criterion. The percentage of adults aged between 18 and 29 is higher in Poland reflecting a younger population. The percentage of non-asthmatic adults surveyed with a low level of education is much higher in the Czech Republic than in other countries (Supplementary material A.4). This difference is only due to the way education achievement has been classified for the Czech Republic and not due to lower education achievement in the Czech Republic.

A very large majority of parents of non-asthmatic children are themselves non-asthmatic (86.4%) (Supplementary material A.5). Regarding the parents of non-asthmatic children, the proportion of people over 60 is (as expected) much lower compared to that of non-asthmatic adults (Supplementary material A.6).

Finally, the issue of serial status quo is explored for choice experiments. A serial status quo respondent is defined as a respondent who for their five discrete choices systematically chooses the status quo, i.e. the original products. While such responses may be legitimate, there is reason to question them. The share of serial status quo respondents is on average 21.5% for adults and 17.8% for children (Supplementary material A.7). These magnitudes are close to those observed in the literature. In terms of countries, important differences emerge, especially between respondents from the United Kingdom and those from the Czech Republic There are 11.6% of serial status quo adults in the Czech Republic sample and 33.1% for the United Kingdom. There are 13.3% of serial status quo respondents for parents of non-asthmatic children in the Czech Republic sample against 24.9% in the United Kingdom.

5.1.1. Baseline estimation strategy

Estimating WTP using contingent valuation. The first aim of this article is to derive mean and median WTP for a reduction in asthma severity from severity $ {s}_0 $ to $ {s}_1 $ where $ {s}_0 $ denotes baseline asthma severity using standard products and $ {s}_1 $ the reduced asthma severity level using SAFETYFIRST products, such that $ {s}_1\succ {s}_0 $ , other things equal. Denote $ y $ the income and $ V\left(s,y\right) $ the indirect utility. Assuming a Random Utility Model, one can write indirect utility of individual $ i $ as follows:

$$ v\left(s,{y}_i\right)+{\epsilon}_i, $$

where $ {\epsilon}_i $ is the idiosyncratic error term. The WTP corresponds to the maximum monetary amount that a person is willing to spend in order to have at least the same utility level as the situation with the baseline severity and unchanged disposable income.

$$ V\left({s}_1,y- WTP\right)=V\left({s}_0,y\right) $$

To estimate WTP, it is possible to ask a sample of the population if they would pay a certain amount of money to reduce their asthma severity. This contingent valuation method is called a single-bounded dichotomous choice approach. An individual who responds yes when asked if he is willing to pay the amount $ b $ for reducing asthma severity from $ {s}_0 $ to $ {s}_1 $ implies that

$$ v\left({s}_1,{y}_i-b\right)+{\epsilon}_{i1}\ge v\left({s}_0,{y}_i\right)+{\epsilon}_{i0} $$

and that $ b\le {WTP}_i $ . Therefore, the probability that individual $ i $ chooses yes when presented $ b $ and that $ b\le {WTP}_i $ . Therefore, the probability that individual i chooses yes when presented b can be written as follows:

$$ {\displaystyle \begin{array}{c}\mathit{\Pr}\left\{{Yes}_i|b\right\}=\mathit{\Pr}\left\{b\le {WTP}_i\right\}\\ {}=\mathit{\Pr}\left\{{\epsilon}_{i0}-{\epsilon}_{i1}\le v\left({s}_0,{y}_i\right)-v\left({s}_1,{y}_i-b\right)\right\}\\ {}=\mathit{\Pr}\left\{{\epsilon}_{i0}-{\epsilon}_{i1}\le g\left(b,{y}_i,{s}_0,{s}_1,\theta \right)\right\}\\ {}=1-F\left(b,{y}_i,{s}_0,{s}_1,\theta \right),\end{array}} $$

where $ F $ is the cumulative distribution function of the error term $ {\epsilon}_{i1}-{\epsilon}_{i0} $ and $ \theta $ the parameter of the distribution. Assuming that the $ n $ observations are independent and identically distributed, $ \theta $ can be estimated by finding the maximum likelihood, which is the joint probability that respondents choose the reduced risk option.

$$ L\left(b,y,{s}_0,{s}_1,\theta \right)=\mathit{\Pr}\left\{{Yes}_1,\dots, {Yes}_i,\dots, {Yes}_n|b\right\}=\prod \limits_{i=1}^n\mathit{\Pr}\left\{{Yes}_i|b\right\} $$

The mean WTP can then be estimated by integrating the probability of choosing the reduced severity option over the interval from 0 to infinite cost.

$$ E(WTP)={\int}_0^{\infty}\mathit{\Pr}\left\{\mathrm{Yes}|b\right\} db $$

The median WTP is the bid level for which the $ \mathit{\Pr}\left\{\mathrm{Yes}|b\right\} $ equals 50%.

DBDC estimation. In this questionnaire, people were asked if they were willing to pay for a reduced asthma severity using a DBDC. This elicitation method allows several of the estimated individual WTP to be bounded between two values, which is not possible using a single-bounded dichotomous choice. Denote $ {b}_i $ as the first bid level proposed to respondent $ i $ . Denote $ {b}_i^U=3{b}_i $ the follow-up bid level proposed to respondent $ i $ if he responded yes to the first valuation question. $ {b}_i^L=\frac{b_i}{3} $ is the follow-up bid level proposed to respondent $ i $ if he responded no to the first valuation question.Footnote ⁷ This elicitation provides four outcomes per respondent: $ {d}_i^{YY} $ , $ {d}_i^{YN} $ , $ {d}_i^{NY} $ , and $ {d}_i^{NN} $ . Denote $ {d}_i^{YY} $ a dummy variable equal to one when respondent $ i $ chooses yes to both valuation questions. When $ {d}_i^{YY} $ equals 1, $ {WTP}_i\ge {b}_i^U>{b}_i $ where $ {b}_i $ is the first bid level proposed to respondent $ i $ and $ {b}_i^U $ is the higher follow up bid level proposed to respondent $ i $ . Denote $ {d}_i^{YN} $ a dummy variable equal to one when respondent $ i $ chooses yes to the first valuation question and no to the follow-up valuation question. When $ {d}_i^{YN} $ equals 1, $ {b}_i\le {WTP}_i<{b}_i^U $ . Denote $ {d}_i^{NY} $ a dummy variable equal to one when respondent $ i $ chooses no to the first valuation question and yes to the follow-up valuation question. When $ {d}_i^{NY} $ equals 1, $ {b}_i^L\le {WTP}_i<{b}_i $ . Finally, denote $ {d}_i^{NN} $ a dummy variable equal to one when respondent $ i $ chooses no for both valuation questions. When $ {d}_i^{NN} $ equals 1, $ {WTP}_i<{b}_i^L $ .

Based on the previous section, the probability of these four outcomes can be written as follows:

$$ {\displaystyle \begin{array}{c}\mathit{\Pr}\left\{ YesYes|{b}^u\right\}=\mathit{\Pr}\left\{{b}^u\le WTP\right\}=1-F\left({b}^u,\theta \right),\\ {}\mathit{\Pr}\left\{ YesNo|b,{b}^u\right\}=\mathit{\Pr}\left\{b\le WTP<{b}^u\right\}=F\left({b}^u,\theta \right)-F\left(b,\theta \right),\\ {}\mathit{\Pr}\left\{ NoYes|{b}^L,b\right\}=\mathit{\Pr}\left\{{b}^L\le WTP<b\right\}=F\left(b,\theta \right)-F\left({b}^L,\theta \right)\\ {}\mathit{\Pr}\left\{ NoNo|{b}^L\right\}=\mathit{\Pr}\left\{ WTP<{b}^L\right\}=F\left({b}^L,\theta \right).\\ {},\end{array}} $$

In this setting, the log-likelihood function for the sample of $ n $ respondents can be written as follows:

$$ \ln L\left(b,\theta \right)={\displaystyle \begin{array}{l}\sum \limits_{i=1}^n\Big[{d}^{YY}\mathit{\Pr}\left\{ YesYes|{b}^u\right\}+{d}^{YN}\mathit{\Pr}\left\{ YesNo|b,{b}^u\right\}+{d}^{NY}\mathit{\Pr}\left\{ NoYes|{b}^L,b\right\}\\ {}+\hskip2px {d}^{NN}\mathit{\Pr}\left\{ NoNo|{b}^u\right\}\Big].\end{array}} $$

Maximizing $ \ln L\left(b,\theta \right) $ permits an estimate $ \theta $ and derives the mean WTP and median WTP more efficiently than with a single-bounded dichotomous choice approach, although the professional literature debates the incentive compatibility of the double-bounded approach and its empirical significance (Bateman et al., Reference Bateman, Langford, Jones and Kerr2001).

Spike configuration with Weibull distribution of the error. So far, it has been assumed that people will always choose the reduced severity option when it costs them nothing or almost nothing. In other words that $ \mathit{\Pr}\left\{ Yes|b=0\right\}=1 $ . In reality, a small share of the population might still choose the status quo because they do not care enough about reducing their asthma severity. This creates a spike near zero that could be significant in the case of people having mild asthma that can be more easily controlled. Carson and Hanemann (Reference Carson and Hanemann2005) argue that failing to include a spike parameter can in some cases lead to an overestimate of WTP. This spike near zero can be measured using the responses to the open-ended question that followed the DBDC: “What would be the most you would be willing to pay, for the SAFETYFIRST products?”. Denote $ {d}_i^{NNY} $ a dummy variable equal to one when respondent $ i $ chooses no to both valuation questions but provides a positive value to the open-ended questions and $ {d}_i^{NNN} $ a dummy variable equal to one when respondent $ i $ chooses no to both valuation questions and responded 0 to the open-ended questions. The probability of these two events are

$$ {\displaystyle \begin{array}{c}\mathit{\Pr}\left\{ NoNoYes|{b}^L\right\}=\mathit{\Pr}\left\{0< WTP<{b}^L\right\}=F\left({b}^L,\theta \right)-F\left(0,\theta \right),\\ {}\mathit{\Pr}\left\{ NoNoNo|0\right\}=\mathit{\Pr}\left\{ WTP\le 0\right\}=F\left(0,\theta \right).\end{array}} $$

These two events can be added to the likelihood function to improve information as follows:

$$ \ln L\left(b,\theta \right)={\displaystyle \begin{array}{l}\sum \limits_{i=1}^n\Big[{d}^{YY}\mathit{\Pr}\left\{ YesYes|{b}^u\right\}+{d}^{YN}\mathit{\Pr}\left\{ YesNo|b,{b}^u\right\}+{d}^{NY}\mathit{\Pr}\left\{ NoYes|{b}^L,b\right\}\\ {}+\hskip2px {d}^{NN}\mathit{\Pr}\left\{ NoNo|{b}^u\right\}+{d}^{NN Y}\mathit{\Pr}\left\{ NoNo Yes|{b}^L\right\}+{d}^{NN N}\mathit{\Pr}\left\{ NoNo No|0\right\}\Big].\end{array}} $$

To derive the mean WTP and median WTP, it is necessary to estimate $ \theta $ and therefore to be able to compute the log-likelihood for various values of $ \theta $ . Hence, it is necessary to assume a distribution $ F $ for the utility error. In this article, a Weibull distribution is assumed as the baseline because it generally has a shorter right tail than the log-normal and, in its “spike” configuration, usually performs well (Kriström, Reference Kriström1997; Carson & Hanemann, Reference Carson and Hanemann2005).

Control variables and use of post-stratification weights. A Weibull distribution $ \theta =\left\{k,\lambda \right\} $ is characterized by a shape parameter $ k $ that measures the slope of the function and a scale parameter $ \lambda $ that measures the spread of the distribution. All estimations assume a shape parameter equal to 1. In the baseline, specification of the scale parameter when $ b>0 $ is

(1) $$ {\lambda}_{ic}(b)={\alpha}_0+{\alpha}_1{MediumSev}_i+{\alpha}_2{HighSev}_i+{\alpha}_3\ln {b}_i+\sum \limits_c{\delta}_c\left({d}_{ic}\times {\omega}_i\right) $$

and the spike parameter when $ b=0 $ is

(2) $$ {\eta}_{ic}={\alpha}_0+{\alpha}_1{MediumSev}_i+{\alpha}_2{HighSev}_i+\sum \limits_c{\delta}_c\left({d}_{ic}\times {\omega}_i\right), $$

where $ {MediumSev}_i $ is a dummy variable equal to 1 when respondent $ i $ has mild plus or moderate asthma, $ {HighSev}_i $ is a dummy variable equal to 1 when respondent $ i $ has moderate plus or severe asthma, $ \ln {b}_i $ is the logged cost or bid proposed to respondent $ i $ , $ {d}_{ic} $ is a country dummy equal to 1 when respondent $ i $ lives in country $ c $ and $ {\omega}_i $ is the post-stratification weight of respondent $ i $ . Including $ {\omega}_i $ as a control captures the fact that some categories of people were slightly under or over represented in the sample compared to the actual population. The greater respondent $ i $ is underrepresented in the sample, the higher their weight $ {\omega}_i $ . It is necessary to interact country dummies with the weight because weights are defined at the country level.

The model is also estimated when the scale parameter includes additional explanatory variables as follows:

(3) $$ {\displaystyle \begin{array}{c}{\lambda}_{ic}(b)={\alpha}_0+{\alpha}_1{MediumSev}_i+{\alpha}_2{HighSev}_i+{\alpha}_3\ln {b}_i\\ {}+\sum \limits_c{\delta}_c\left({d}_{ic}\times {\omega}_i\right)+{\alpha}_4{Female}_i+\sum \limits_a{\tau}_a{Age}_{ia}+{\alpha}_5\ln {y}_i+{\alpha}_6{HighEduc}_i\end{array}} $$

where $ {Female}_i $ is a dummy variable equal to 1 when respondent $ i $ identifies as a female, $ {Age}_{ia} $ is a country dummy equal to 1 when respondent $ i $ belongs to age category $ a $ , $ \ln {y}_i $ is the logged monthly income for the household of respondent $ i $ and $ {HighEduc}_i $ is a dummy variable equal to 1 when respondent $ i $ achieved a high education outcome.

The model is also estimated when the scale parameters include information on whether respondents have to pay health costs out of pocket, whether they perceive their health below or above average people of their gender and age, whether they are diagnosed with any other chronic disease and whether they or a relative was diagnosed with COVID-19.

Deriving mean and median WTP based on individual WTP. The mean WTP for a one-step reduction in asthma severity is computed as a simple average of the individual mean WTP as follows:

$$ \hat{WTP}=\frac{1}{n}\sum \limits_{i=1}^n{\hat{WTP}}_i. $$

The individual mean WTP is computed by integrating the probability of responding yes to the valuation question over the interval from 0 to maximum bid with adjustment:

$$ {\hat{WTP}}_i=\int_0^{b_{max}}\frac{f\left({\lambda}_{ic}(b),k\right)}{1-f\left({\lambda}_{ic}\left({b}_{max}\right),k\right)} db, $$

where $ f $ is the density function of the Weibull distribution. Truncation at maximum bid level $ {b}_{max} $ is necessary since the right tail is not null when the cost goes to infinity. The adjustment of the denominator compensates for the fact that the support of $ f\left({\lambda}_{ic}(b),k\right) $ does not stop at $ {b}_{max} $ . The median WTP is also computed as a simple average of individual median WTP, computed as follows:

$$ {\overset{\sim }{WTP}}_i=\frac{\ln 2}{\left|{\alpha}_3\right|}{e}^{\eta_{ic}\left(\frac{1}{\left|{\alpha}_3\right|}\right)}, $$

where $ {\alpha}_3 $ is the parameter for the logged bid value as indicated in Equation (1).

5.1.2. Robustness checks

Several robustness checks were performed. Overall, the baseline estimation results are robust to various methodological choices. First, the model was estimated assuming different distributions for the utility error including lognormal or log-logistic. Second, the model was estimated without allowing for a spike. Third, the model was estimated without post-stratification weights to see if rare respondents have an outsized impact on the estimates. Finally, the model was augmented with counts of problematic responses to debriefing questions to examine the sensitivity of the estimated coefficients to respondents who likely underestimated or overestimated their WTP.

5.2.1. Overview of the different econometric models

Regarding the Multinomial Logit (MNL) model, individuals are assumed to assign the same value to an attribute entering into their utility function. The coefficient associated with this attribute is considered to be identical for all individuals (McFadden, Reference McFadden and Zarembka1974), which is a very strong assumption.

The Random Parameter Logit model (RPL) makes it possible to take into account the heterogeneity of individuals’ preferences by allowing coefficients associated with the different attributes to vary randomly according to a specified distribution $ f\left(\beta |\Omega \right) $ . The coefficient associated with the status quo is assumed to be normally distributed and therefore it can be positive or negative depending on whether the individual has utility or disutility from not using SAFETYFIRST products. The non-monetary attributes associated with reductions in the risk of developing asthma follow an exponential Weibull distribution because their coefficients would naturally be positive. The coefficient associated with the added cost is assumed to be deterministic. In the case where individual $ i $ makes $ T $ choices,Footnote ⁸ it is assumed that their preferences for a given attribute do not vary over their choices. The probability for an individual $ i $ choosing alternative $ j $ for choice $ t $ is then calculated as follows:

(6) $$ {P}_i\left({y}_{it}=j\;|\;\Omega \right)=\underset{\beta }{\int }{P}_i\left({y}_{it}=j\;|\;\unicode{x03B2} \right)f\left(\beta\;|\;\Omega \right) d\beta, $$

where $ {y}_{it}=j $ corresponds to the alternative chosen by individual $ i $ for the choice $ t $ and where

(7) $$ {P}_i\left({y}_{it}=j\;|\;\unicode{x03B2} \right)=\frac{\mathit{\exp}\left({\alpha}_i+{\beta}_i^{\prime }{X}_{ijt}\right)}{\sum \limits_{j=1}^J\mathit{\exp}\left({\alpha}_i+{\beta}_i^{\prime }{X}_{ijt}\right)}, $$

where $ \alpha $ is the constant associated with the status quo and $ X $ corresponds to the different attributes. Consequently, the log-likelihood function associated with the various coefficients to be estimated is defined as follows:

(8) $$ \ln L\left(\Omega \right)=\sum \limits_{i=1}^N\ln \left(\underset{\beta }{\int}\left(\prod \limits_{t=1}^T{P}_i\left({y}_{it}=j\;|\;\unicode{x03B2} \right)\right)f\left(\beta\;|\;\Omega \right) d\beta \right)\hskip-0.18em . $$

Since the integral of the log-likelihood must be approximated through simulations, the different parameters of the estimate are calculated from different random samples.Footnote ⁹

Nevertheless, in the model presented above, the sources of heterogeneity are assumed to be random. A refinement of this model following Hensher and Greene (Reference Hensher and Greene2003) is to allow the means of the parameter distributions to be heterogeneous according to the country of individuals. Alternatively, the heterogeneity of individuals’ preferences can be modeled through a discrete distribution using the Latent Class model. This type of model assumes that each individual belongs to a class c, and his membership can be linked to his geographic origin. This model simultaneously divides individuals into classes and estimates the different coefficients $ {\beta}_c $ of the utility function conditional on class membership (Greene & Hensher, Reference Greene and Hensher2003). The probability that individual $ i $ chooses alternative $ j $ for choice $ t $ is then calculated as follows:

(9) $$ P\left({y}_{it}=j\right)=\sum \limits_{c=1}^C\left(P\left( class=c\right)\times P\left({y}_{it}=j| class=c\right)\right), $$

where $ P\left({y}_{it}=j| class=c\right) $ is the probability for the individual $ i $ to choose the choice $ t $ , the alternative $ j $ conditional on their membership to the class c. This probability is calculated as follows:

(10) $$ P\left({y}_{it}=j\;|\;\mathrm{class}=\mathrm{c}\right)=\frac{\mathit{\exp}\left({\alpha}_c+{\beta}_c^{\prime }{X}_{ijt}\right)}{\sum \limits_{j=1}^J\mathit{\exp}\left({\alpha}_c+{\beta}_c^{\prime }{X}_{ijt}\right)}. $$

$ P\left( class=c\right) $ is the probability of belonging to class c, calculated as follows:

(11) $$ P\left(\mathrm{class}=\mathrm{c}\right)=\frac{\mathit{\exp}\left({\theta}_c^{\prime }{z}_i\right)}{\sum \limits_{c=1}^C\mathit{\exp}\left({\theta}_c^{\prime }{z}_i\right)}, $$

where $ {z}_i $ corresponds to the choice-invariant characteristics having a potential effect on the probability of belonging to class c (e.g. geographical origin) and $ {\theta}_c $ corresponds to the coefficients associated with the variables $ {z}_i $ , specific to class $ c $ . Since each individual makes a series of $ T $ choices, the probability of choosing alternative $ j $ by individual $ i $ is

(12) $$ P\left({y}_i=j\right)=\sum \limits_{c=1}^CP\left(\mathrm{class}=\mathrm{c}\right)\underset{P_i\left(j|c\right)}{\underbrace{\prod \limits_{t=1}^T\left(P\left({y}_{it}=j\;|\;\mathrm{class}=\mathrm{c}\right)\right)}}. $$

The log-likelihood function associated with the various parameters to be estimated is then defined as follows:

(13) $$ \ln L=\sum \limits_{i=1}^N\ln \left(\sum \limits_{c=1}^CP\left(\mathrm{class}=\mathrm{c}\right)\prod \limits_{t=1}^T\left(P\left({y}_{it}=j\;|\;\mathrm{class}=\mathrm{c}\right)\right)\right). $$

Due to its structure, the Latent Class model is generally considered to be an easier model to implement and to understand by decision makers, as it allows for the identification of different classes of marginal willingness-to-pay, defined according to the individual’s profile as well as their respective weights in the surveyed population.

5.2.2. Deriving mean and median marginal WTP

The individual marginal Willingness to Pay for each non-monetary attribute is then calculated as follows in the case of an RPL model:

(14) $$ {MWTP}_{i, non\ monetary\ attribute}=-\frac{{\hat{\beta}}_{i, non\ monetary\ attribute}}{{\hat{\beta}}_{i, added\ cost}}. $$

In the case of a Latent Class model, the calculation of the individual marginal willingness to pay for each non-monetary attribute is calculated in the same way after calculating the coefficients of the different attributes $ {\hat{\beta}}_i $ , as follows:

(15) $$ {\hat{\beta}}_i=\sum \limits_{c=1}^C{\hat{\pi}}_{ic}^{\ast }{\hat{\beta}}_c, $$

with

(16) $$ {\hat{\pi}}_{ic}^{\ast }=\frac{{\hat{\pi}}_{ic}{\hat{P}}_i\left(j|c\right)}{\sum \limits_{c=1}^C{\hat{\pi}}_{ic}{\hat{P}}_i\left(j|c\right)}. $$

The unweighted individual marginal willingness to pay ( $ {MWTP}_i $ ) for the non-monetary attributes are calculated via Equation (14). They are then weighted to compute willingness-to-pay statistics that are representative of the population. The weighted mean marginal WTP is calculated as follows:

(17) $$ \overline{MWTP}=\frac{1}{N}\sum \limits_{i=1}^N{\omega}_i{MWTP}_i, $$

where $ {\omega}_i $ is the post-stratification weight for individual $ i $ derived from a raking algorithm that corrects for differences between target population quotas and achieved sample quotas. The mean marginal WTP (and the median marginal WTP) by country are calculated from pooled data according to the respondents belonging to the different countries.

7.5.1. Using the value of a statistical case of asthma in cost–benefit analysis

The estimates of a value of statistical case should be used in cost–benefit analyses addressing proposed regulations of chemicals or other pollutants that influence asthma. Presented here is the recommended use.

Assume a policy is appraised over $ T $ years in country $ c $ . Compared to the status quo, this policy is estimated to lead to a reduction of $ {SCA}_{ct} $ statistical cases of adult asthma and to a reduction of $ {SCC}_{ct} $ statistical cases of childhood asthma in country $ c $ in year $ t $ . The discounted benefits of the policy in terms of avoided asthma should be computed as follows:

(18) $$ {Discounted\ benefits}_c={\sum}_{t=0}^T\frac{VSCA_{ct}\times {SCA}_{ct}+{VSCC}_{ct}\times {SCC}_{ct}}{{\left(1+{k}_c\right)}^t}, $$

where $ {k}_c $ is the discount rate used in country $ c $ ,Footnote ¹⁹ $ {VSCA}_{ct} $ is the recommended value of a statistical case of adult asthma and $ {VSCC}_{ct} $ is the recommended value of a statistical of childhood asthma in country $ c $ in year t. $ {VSCA}_{ct} $ and $ {VSCC}_{ct} $ are based on the recommended values $ {VSCA}_{c,2022} $ and $ {VSCC}_{c,2022} $ reported in USD PPP in Table 27 and should reflect increase in prices and in GDP per capita over time such that

(19) $$ {VSCA}_{ct}={VSCA}_{c,2022} \times {PPP}_{c,2019} \times \left(1+\%\Delta {P}_{c,2022-t}\right) \times {\left(1+\%\Delta {Y}_{c,2022-t}\right)}^{\beta }, $$

(20) $$ {VSCC}_{ct}={VSCC}_{c,2022} \times {PPP}_{c,2019} \times \left(1+\%\Delta {P}_{c,2022-t}\right) \times {\left(1+\%\Delta {Y}_{c,2022-t}\right)}^{\beta }, $$

where $ {PPP}_{c,2019} $ is Purchasing Power Parities for actual individual consumption in national currency per USD for 2019 that was used to convert bid levels in the survey, $ \%\Delta {P}_{c,2022-t} $ is the increase in consumer price index from 2022 to year $ t $ , $ \%\Delta {Y}_{c,2022-t} $ is GDP per capita growth from 2022 to year $ t $ and $ \beta $ is the income elasticity.

An example for a fictional policy that reduces the number of statistical cases of adult asthma by 1 000 every year in France for 2022–2025 is provided in Table 28 for illustration purpose.Footnote ²⁰ Based on a Value of Statistical Case of USD 260 000 in 2022, the discounted benefits of the policy over the 4 years equals EUR₂₀₂₂ 767 million.

Table 28.

Measuring the benefits of policy intervention in France: an illustrative example using the value of a statistical case of adult asthma

Note: This illustrative example assumes a fictional policy that would reduce the number of statistical cases of asthma by 1 000 every year in France from 2022 to 2025. GDP per capita projections for 2022–2024 are provided by the OECD Economic Outlook (2022).

^a GDP per capita for 2025 is computed by the authors based on the linear fit of 2022–2024 values over time.

^b Consumer Price Index data for 2022 comes from the OECD Dataset: Consumer price indices (CPIs) as of January 2022. A 2% increase per year is assumed for the Consumer Price Index for 2023–2025 and is not an OECD forecast. PPP for actual individual consumption data is for year 2019 as used to convert bid levels across countries and comes from the OECD Dataset: PPPs and exchange rates as of January 2022.

^c The discount rate comes from Quinet (Reference Quinet2013) and is what is used in France for short assessment period for which there is no systemic risk in the implementation of the policy. The income elasticity equals 0.07 as estimated for total adult asthma risk in this article.

Source: Authors’ own elaboration.

Finally, the discounted costs of the policy should be subtracted from these discounted benefits to compute the discounted value of the policy.

7.5.2. Using the WTP value for reduced asthma severity in cost–benefit analysis

Similar to the WTP for the reduction of asthma risk, WTP for the reduction in asthma severity can be used in cost–benefit analysis. Suppose a risk management option or a policy reduction leads to a quantified number of severity steps in a given population of asthmatics. There are two potential scenarios. Either the reduction of severity is quantified for each baseline severity or more likely, only the number of severity reductions for the average baseline severity is available. In the former case, discounted benefits should be computed as follows:

(21) $$ {Discounted\ benefits}_c={\sum}_{t=0}^T\frac{1}{{\left(1+{k}_c\right)}^t}\left(\begin{array}{c}{\overline{WTP}}_{mild, ct}\times {RS}_{mild, ct}\\ {}+{\overline{WTP}}_{mild plusmoderate, ct}\times {RS}_{mild plusmoderate, ct}\\ {}+{\overline{WTP}}_{moderateplussevere, ct}\times {RS}_{moderateplussevere, ct}\end{array}\right), $$

where $ {\overline{WTP}}_{baselineseverity, ct} $ is the mean WTP for going from baseline asthma severity to one step lower in country $ c $ in year $ t $ , $ {RS}_{baselineseverity, ct} $ is the number of asthmatic people whose asthma severity is reduced by one step from the baseline asthma severity for 1 year due to the policy intervention in country $ c $ in year $ t $ and $ {k}_c $ is the discount rate used in country $ c $ .

In most cases, data are scarcer. Then, the discounted benefits can be estimated as follows:

(22) $$ {Discounted\ benefits}_c={\sum}_{t=0}^T\frac{{\overline{WTP}}_{ct}\times {RS}_{ct}}{{\left(1+{k}_c\right)}^t}, $$

where $ {\overline{WTP}}_{ct} $ is the mean WTP for going from baseline asthma severity to one step lower, $ {RS}_{ct} $ is the number of asthmatic people whose asthma severity is reduced by one step for 1 year regardless of their current asthma severity due to the policy intervention in country $ c $ in year $ t $ and $ {k}_c $ is the discount rate used in country $ c $ .

$ {\overline{WTP}}_{baselineseverity, ct} $ and $ {\overline{WTP}}_{ct} $ are based on the recommended values $ {\overline{WTP}}_{baselineseverity,c2022} $ and $ {\overline{WTP}}_{c2022} $ reported in USD PPP in Table 26 and should reflect increase in prices and in GDP per capita over time such that:

(23) $$ {\overline{WTP}}_{ct}={\overline{WTP}}_{c2022} \times {PPP}_{c,2019} \times \left(1+\%\Delta {P}_{c,2022-t}\right) \times {\left(1+\%\Delta {Y}_{c,2022-t}\right)}^{\beta }, $$

where $ {PPP}_{c,2019} $ is Purchasing Power Parities for actual individual consumption in national currency per USD for the 2019 that was used to convert bid levels in the survey, $ \%\Delta {P}_{c,2022-t} $ is the increase in consumer price index from 2022 to year $ t $ , $ \%\Delta {Y}_{c,2022-t} $ is GDP per capita growth from 2022 to year $ t $ and $ \beta $ is the income elasticity. An illustrative example for a fictional policy that increases the number of asthmatics people whose asthma severity is reduced by one step for 1 year by 1 000 every year in France for 2022–2025 is provided in Table 29 for illustration purposes.Footnote ²¹ Based on a mean WTP of USD 440 in 2022, the discounted benefits of the policy over the 4 years equals EUR₂₀₂₂ 1.3 million.

Table 29.

Measuring the benefits of policy intervention in France: an illustrative example using the mean WTP to reduce adult asthma severity by one step

Note: This illustrative example assumes a fictional policy that would reduce the number of statistical cases of asthma by 1 000 every year in France from 2022 to 2025. GDP per capita projections for 2022–2024 are provided by the OECD Economic Outlook (2022).

^a GDP per capita for 2025 is computed by the authors based on the linear fit of 2022–2024 values over time and is not an OECD forecast.

^b Consumer Price Index data for 2022 comes from the OECD Dataset: Consumer price indices (CPIs) as of January 2022. A 2% increase per year is assumed for the Consumer Price Index for 2023–2025 and is not an OECD forecast. PPP for actual individual consumption data is for year 2019 as used to convert bid levels across countries and comes from the OECD Dataset: PPPs and exchange rates as of January 2022.

^c The discount rate comes from Quinet (Reference Quinet2013) and is what is used in France for short assessment periods for which there is no systemic risk in the implementation of the policy. The income elasticity equals 0.3 as estimated for total adult asthma risk in this article.

Source: Authors’ own elaboration.