Willingness to pay for policies to reduce future deaths from climate change: evidence from a British survey

Study design: as extremely serious (OR 0.04, 95% CI 0.02-0.09). Income was also related to WtP; the highest-income group were twice as likely to be willing to pay as the lowest-income group (OR 2.14, 95% CI 1.40-3.29). Conclusions: There was public support for policies to address future health impacts of climate change; the level of support varied with people ' s perceptions of the seriousness of these impacts and their ﬁnancial circumstances. Our study adds to evidence that health, including the health of future populations, is an outcome that people value and suggests that framing climate change around such values may help to accelerate action.


Introduction
The health impacts of climate change on today's populations are becoming increasingly evident. 1,2 Failure to keep global mean temperatures within 1.5 centigrade of pre-industrial levels will threaten the health of future populations; 3,4 projections point to increases in global temperatures by 2050 that are significantly above this threshold. 5,6 In the UK context, increasing temperatures are associated with an increased frequency and intensity of extreme hot weather and flooding; these climate changeerelated exposures have been highlighted as major risks to the future health of the UK population, 7e10 with older people and children particularly at risk. 7,11,12 The Climate Change Act (2008) sets the framework for climate governance in the UK and mandates action on mitigation and adaptation. 13,14 Both offer significant cobenefits for public health 1,15,16 and are urgently needed to protect the health of future populations. 5,6 The health of future populations turns on the actions of today's population and, in particular, on the speed and magnitude of the policy response to climate change. 1,6,17e19 However, policiesdand the economic evaluations that inform themdgive greater weight to the well-being of current populations. 20e22 A time weight (a discount rate) is applied when assessing the costs and benefits of policies, with future costs and benefits valued less highly than current ones. Discounting is designed to take account of the variable timescales over which costs and benefits are distributed. 23 It is also seen to be in line with public's preference to receive benefits now and 'to defer costs to future generations'. 20 It is increasingly recognised that standard discounting approaches are less appropriate for policies with long-term effects, including those with impacts on the climate system. 22 However, little is known about public preferences for reducing climate change risks for those living in the future. One measure used to assess people's preferences is their willingness to pay (WtP) to help meet the cost of such policies. WtP captures the value that an individual places on an outcome (for example, reduced health risks from climate change) in monetary terms and reflects both the strength of their preference for the outcome and their ability and willingness to give up consumption of other goods to pay for it. We searched Web of Science up to December 2018 for WtP studies framed around the health risks of climate change (using the terms 'health', 'climate change' and 'WtP'). We found very few studies and no UK studies. For example, studies investigated people's WtP to reduce climate changeerelated health risks from flooding in Switzerland, 24 from variations in temperature in Taiwan 25 and from air pollution in Pennsylvania; 26 they indicated that communicating both the public health and the climate change benefits of interventions increased people's WtP. Using a representative British survey, we estimate people's WtP for policies to reduce future deaths in the UK that may result from climate change.

Methods
Stated preference (SP) methods provide a survey-based method for estimating the economic value that people attach to outcomes for which there is not a direct market value, 27e29 such as protecting the health of future populations and improving the natural environment. One widely used SP method is contingent valuation (CV). This asks people how much they would be willing to pay for a benefit in a hypothetical scenario, for example, if they would be willing pay a specified monetary cost (e.g. £1 a month for 10 years) to support a policy that yielded a specified benefit in X years' time. 28 The use of SP methods in such contexts is supported by official guidance on how to conduct appraisals of public policy initiatives such as climate change policy. 20 For our study, we used deaths, the most widely used health measure and one suitable for a general population survey. 30 We derived our measure from the UK's national assessment of future UK deaths from increasing global temperatures. 7 This estimated an additional 7000 climate changeerelated deaths per year from heatwaves by the 2050s, with additional, but uncertain, deaths from other climate changeerelated exposures, including from flooding, worsening air quality and increased pathogens. 7 Based on a representative British survey, we used CV to estimate people's WtP to reduce future deaths that may result from climate change. The Opinions and Lifestyle Survey (OLS) is a cross-sectional national survey governed by the UK Government's Statistics Authority Office code of practice and overseen by the National Statistician's Data Ethics Advisory Committee. 31,32 Our study was approved by the Department of Health Sciences Research Governance Committee at the University of York.
The OLS is based on a random probability sample of private households in Britain, drawn from the Postcode Address File (PAF) containing approximately 26m addresses. 32 Updated every three months, the PAF is the most complete address database in the UK. The OLS sample is stratified by region and sociodemographic profile. Sociodemographic stratification is based on the proportions of households with no car; households where the household reference person has an occupation in a higher socio-economic group; and people aged >65 years. The survey runs eight months a year; a new sample is drawn each month, and one adult per household is randomly selected for a face-to-face computer-assisted personal interview. The survey consists of a standard set of sociodemographic questions, together with commissioned modules.
We commissioned a CV question in the January and February 2016 OLS on people's WtP for policies to maintain climate changeerelated deaths at their current level. Survey participants were asked whether they would be willing to pay ('yes'/'no') an ongoing monthly payment of a randomly assigned 'bid amount' of £1, £2, £5, £10 or £20 to hold climate changeerelated deaths at current levels rather than see such deaths increase (see Box 1); evidence indicates that binary choices result in a more reliable revelation of preferences. 28 The range of bid values was determined from a pilot study, also using the OLS, where participants were asked to choose how much they would pay per month from a choice of £0, £1, £2, £5, £10, £20 or £50.
The question provided additional contextual information on the number of additional annual deaths expected by 2050 and one of two subsequent years (2080 and 2115) if no additional action was taken; and on the population groups most at risk (the elderly and the very young). This contextual information was provided in one of the three forms (see Box 1), which were randomly assigned to respondents. Participants were also asked, after the question on WtP, 'how would you rate the seriousness of the impacts of climate change?' (hereafter 'perceived seriousness of climate change'). Response options were extremely serious, fairly serious, not very serious or not at all serious.
In line with the OLS average, the response rate across the months of our study was 50% (n ¼ 1859). Analyses were performed using Stata, version 15. An estimate of average WtP was obtained from the CV responses by fitting a univariate logistic regression to the binary responses (willing or not willing to pay) with the bid amount as the sole independent variable. To make the results representative of the general population, data were weighted to account of non-response bias and the probability of each respondent being included in the survey. 32 Mean and median WtP were calculated from the coefficient associated with the bid amount in the logistic regression model. 33,34 A sensitivity analysis was also performed, in which all participants with unknown WtP (refused to answer or stated that they did not know) were assumed to be unwilling to pay.
A multivariate logistic regression was then used to assess the association of WtP with other factors. Candidate independent variables were the bid amount; the respondent's perception of the seriousness of climate change; their generation of birth (before 1946; 1946e1965; 1966e1985; after 1985); their individual gross income (split into five bands with approximately equal numbers of participants in each: less than <£7280, £7280e£12,479, £12,480e£18,719, £18,720e£28,599 and £28,600); education (degree level or above, below degree level, none); health status (presence or absence of longstanding limiting illness); gender; parenthood status (presence or absence of a child younger than 16 years in the household); and the context offered in the question. Variables were added in turn and retained if they reduced the Bayesian information criterion (BIC) by more than 3. Multiple imputation with ten iterations using chained equations 35 with an ordered logistic model was used to handle missing data for income band and perceived seriousness of climate change. In addition to the candidate-independent variables listed previously, car ownership and employment status were also included in the imputation.

Results
The majority of respondents were willing to answer questions on their perceptions of climate change (98%) and their WtP to reduce climate changeerelated deaths (97%); the remaining responses were refused or answered 'do not know'. Among the other variables, only income had missing data (7%).

Perceived seriousness of climate change
Climate change was considered extremely or fairly serious by 86% of respondents (Table 1). Some variation was seen Box 1 Willingness-to-pay question.
Each participant was presented with a variation of the following question, in which text in square brackets was varied: Climate change is expected to have a negative impact on health and well-being, and scientists expect that climate change will cause more deaths in the UK. Scientists from the UK Committee on Climate Change predict that if we continue to use energy in the way we currently do, there may be 7000 more deaths per year in the UK by 2050 (over 3 times higher than current levels) and 12,000 more deaths per year in the UK by [A] (6 times higher than current levels).
[B] are two groups that are particularly at risk. It will cost money to put in place environmental policies to reduce these risks. We would like to know what you feel would be an acceptable amount of public spending to reduce these impacts.
Would you be willing to pay an additional [X] in your taxes each month for the next ten years to reduce climate-related deaths to current levels?
The varied texts were randomly assigned so that WtP to reduce climate changeerelated deaths Overall, 61% of people were willing to pay the amount asked to reduce future climate change deaths ( Table 2). As the bid value increased, the probability that people were willing to pay this amount declined. A large majority (78%) were willing to pay if asked for £1 a month, compared with 46% of those asked for £20. The mean amount people were willing to pay was £19.17 (95% confidence interval [CI] £15$64 -£22$70, Table 3). The final multivariate model included the bid amount, perceived seriousness of climate change, birth generation, income band and level of education. Health status, parenthood status and context offered in the question (timescales for increased deaths and the elderly/very young as high-risk groups) did not improve model fit as measured by the BIC (nor did they have a statistically significant association with WtP) and were excluded. Income was associated with WtP to reduce climate change deaths (Table 4); compared to people in the lowest income band, people in the highest income band were more than twice as likely to be willing to pay (odds ratio [OR] 2$14, 95% CI 1$40e3.29). Perceptions of climate change were also associated with WtP, with respondents who considered it not at all serious being far less likely to pay than those who thought it was extremely serious (OR 0$04, 95% CI 0.02e0.09).
The findings were robust to sensitivity analyses assuming non-response to the WtP question indicated unwillingness to pay (Table 4) and also to modelled scenarios in which nonresponse to the income question was assumed to be missing not at random (Appendix). Multiple imputation changed associations significantly only in relation to generation of birth; without multiple imputation, post-1965 generations were more likely to be WtP than the 1946e1965 generation. This may reflect some bias introduced in the nonimputed model by the missing data. The lack of an association between WtP and birth generation may be due to mediation through perceptions of the seriousness of climate change and income, both included in the models. Uncertainty around perceptions of the future impacts of climate change has been identified as needing further investigation in WtP studies. 24  p u b l i c h e a l t h 1 7 4 ( 2 0 1 9 ) 1 1 0 e1 1 7

Discussion
The majority of study participants considered climate change to be a serious problem, a finding consistent with evidence that most adults in the UK are worried about climate change. 36e38 Perceptions of the severity of climate change may be a result of the increase in extreme weather and flooding in the UK, and a consequent heightened sensitivity to the adverse impacts of climate change. 39 The majority (over 60%) were willing to pay to reduce the risk of future climate changeerelated deaths. Altruistic motives may partly determine WtP, including the value that people place on protecting the natural environment and future populations. 26,36,40 Our CV question used additional deaths by 2050 as its baseline scenario and included future timescales ranging from 2080 to 2115; the majority of study participants were also alerted to the very young being at particular risk. None of these contextual factors were significant predictors of WtP, suggesting that there was no significant additional discounting for avoidance of deaths further into the future. The presence of children in the household was also not a significant predictor of WtP, suggesting that WtP was not tied specifically to concern about the future welfare of one's children.
WtP was significantly associated with perceptions of climate change. Those who thought climate change to be a more serious problem were more willing to pay for policies to reduce to address it, a finding consistent with other WtP studies. 18,24 The connection between perceptions of climate change and WtP has been attributed to a sense of responsibility for the environment and future generations as well as a desire to reduce one's own exposure to climate change risks. 24,26,36 People were less willing to support policies to address climate changeerelated risks to health as the cost to them increased. There is evidence that the public support action on climate change at the national and international levels but oppose tax policies that would affect them directly. 41 WtP was lower for those in the lowest income band compared to the highest, a pattern found in most, 24,25,36 but not all, 26 studies. Studies of the relationship between household income and stated WtP typically reveal rather low elasticities (less than unity). 42 Lower WtP among those on lower incomes may not reflect lower concern, but simply lower ability to pay. Meeting the costs of climate change through progressive systems of   taxation (e.g. income tax) could provide an equitable approach to funding these policies. Some limitations of the study should be noted. Firstly, people's perceptions of climate change and therefore their WtP to reduce its health impacts may be related to factors beyond the scope of our study, including an individual's experience of extreme weather events 39 and concerns about nonehealth-related impacts of climate change such as damage to property and infrastructure. 43,44 Context is important in encouraging concern and WtP for policies to address climate change. 45,46 However, in our study, variations in the contexts presented were not associated with a significant change in WtP. WtP was not affected by whether the very young were said to be at particular risk in addition to the elderly or by the timeframes and severity of the projected impacts. This may reflect a low discount rate among respondents, such that avoided deaths in 2050 were not viewed as significantly more desirable than avoided deaths in 2115. Across such long time periods, intragenerational time preferences (an individual's preference for the timing of benefits across generations) may be a factor. 18,47 The low discount rate may also reflect a lack of information on how the risk reduction would be achieved in practice. An individual's WtP may be policy sensitive, influenced, for example, by perceptions of and preferences for climate change mitigation and adaptation, 48 as well as by views on the effectiveness and fairness of such policies. 49 Including such potential influences on WtP was not possible using our chosen study design (questions inserted into a general survey) but would merit further investigation. We would also seek to explore non-WtP in greater depth. Of those asked to pay £1 per month, one in five were unwilling, a proportion rising to four in five among those who considered the effects of climate change to be not at all serious. Finally, our cross-sectional study predates the recent increased public debate about climate change action; given increasing public concern about climate change, 50 a follow-up study may find a larger majority of adults willing to pay to reduce its future health impacts.
Secondly, SP methods may elicit responses that are not in line with real choice behaviour. In particular, WtP is estimated from questions based on hypothetical scenarios; study participants may therefore not regard their responses as consequential (as having real-world consequences). Our question on the perceived seriousness of climate change provided an indication of whether the individual believed in and cared about the scenario; these affective attributes have been found to be associated with respondents regarding the scenario as one likely to be implemented. 51 Nonetheless, a study where participants believed they would be required to pay if the proposed policy secured popular supportdfor example, a study linked to a public consultation on policy optionsdwould be an important next stage in establishing people's WtP to reduce the future health risks of climate change because we know from other work that consequentiality is an important aspect of demand revelation in SP methods. 51 SP methods are the only ones available for estimating economic values for which there is not a market value or other data based on people's behaviour 28 (for example, a policy experiment enabling comparisons to be made between exposed and non-exposed populations). We followed standard SP guidelines for survey design, including the identification of a representative sample of the potentially affected population, a single binary-choice question (yes/no) that included a baseline 'status quo' condition (the current number of deaths), designation of who pays ('you'), payment amounts and frequency, a change mechanism (a policy intervention) and a defined benefit (reduced climate changerelated deaths at specific points in time) that, because it was based on official estimates, would be likely to be deemed credible by survey participants. The question was also piloted in an earlier ONS survey to ensure it was understood by the study participants.
Thirdly, our study explored associations between WtP and income, generation and education. All three factors were included in our final model, and model fit supported their inclusion. While it is possible that there are interactions between the factors, sample sizes were insufficient to fully explore this. However, such possible interactions do not impact on the main aim of our study: to assess populationlevel WtP for a defined future public health benefit.
Our study had a number of further strengths. As well as being based on a representative survey of adults in Britain using individual home-based interviews, it collected a range of sociodemographic data, including educational attainment, income and parental status, enabling analysis of their association with WtP. There were few missing data for most variables used in the models. As missing data in variables used in the multivariate model may bias estimation, mean and median WtP were estimated from the univariate model; using data imputation, mean and median WtP were also estimated from the multivariate model and the 95% CIs of these estimates included the point estimates from the univariate model provided in Table 3. The sensitivity analyses, which treated all missing data on WtP as refusal to pay, set a lower bound on the WtP estimates that were not significantly different to those from the main model. Across a range of scenarios, respondents expressed a WtP to reduce future health impacts of climate change, suggesting that health, including the health of future populations, is an outcome that people value. 1,18,24,40 Such a health framing is supported by evidence of the adverse health impacts of climate change and the opportunities for health co-benefits of mitigation and adaptation, particularly for future populations, 1,15,16 However, although the Paris Agreement to strengthen the global response to climate change recognised the right to health, 52 health remains marginal to climate change policies and public debate. 1,53 In this context, public support for climate change policies that yield health benefits underlines the important role to be played by the public health community in accelerating national and global action on climate change. 1,54 Author statements the ONS Opinions and Lifestyle Survey for their help with the study.

Ethical approval
The Opinions and Lifestyle Survey (OLS) is a cross-sectional national survey governed by the UK government's Statistics Authority Office code of practice and overseen by the National Statistician's Data Ethics Advisory Committee.

Funding
This work was supported by the UK's Economic and Social Research Council (ESRC), grant number ES/L003015/1.