Last month, I attended the Society for Benefit-Cost Analysis' annual research conference. It is the yearly gathering of the top minds in cost-benefit research from across the world, many of whom were instrumental in the development of the field over the past half century.
This year’s conference opened with a presentation by Amitabh Chandra of the Harvard Kennedy School of Government and Harvard Business School about Medicare recipients, how they respond to gaps in their coverage, and how we can try to elicit better estimates for people's willingness to pay for additional years of life. The points the speaker brought up made me want to explain a little more how we get the estimates we use throughout cost-benefit analysis.
Stated preference vs. revealed preference
The two main ways policy analysts estimate willingness to pay for goods are stated preference studies and revealed preference studies. As their names imply, in stated preference studies participants are directly asked what they’d be willing to pay for something, while revealed preference studies try to find out how people actually react to changing prices to determine their willingness to pay for goods.
Of the two, many researchers prefer results from revealed preference studies. This is because stated preference studies can be subject to a number of biases that are hard to control. For example, people may be influenced by social pressures if asked about their willingness to pay for some goods with stigmas attached. For example, they might overstate their willingness to pay for cancer research or understate their willingness to pay for illegal drugs.
Revealed preference studies are more difficult to set up, but if done well they can circumvent many of these biases that stated preference studies need to control for. Someone may say they’d only buy junk food if it was cheaper than a healthy option, but if we observe them buying it at a higher price then we can better understand their true willingness to pay.
Challenges with estimation of willingness to pay
Neither stated preference nor revealed preference studies work when people don’t really understand the value of the thing we are interested in. One area that people tend to do a bad job understanding values is in healthcare.
Healthcare is full of decisions about low probability, high cost events and humans are notoriously bad about thinking probabilistically. It is really hard to measure how much someone is willing to pay for something like a new drug that reduces the risk of some very rare but serious disease.
We tend to solve this problem by relying on our estimates for the value of statistical life. People can show their willingness to trade off earnings for changes in the riskiness of their job. We might discover that for example, mortality rates are 1% higher for welders working on active construction sites vs. welders working in shops. If welders working on active sites get paid more, we can take that as an estimate for how much a welder is willing to trade the risk of death for income, from which we can determine the value of statistical life.
People don’t do their own cost-benefit analysis
The main point Chandra made was that his research found that Medicare recipients acted in ways that were not consistent with our understanding of how much people value risk of death reductions when faced with budget constraints caused by gaps in their coverage. In particular, people’s noncompliance with drug regimens implied people valuing mortality risk reduction much lower than we see in, for instance, job market wage risk premiums.
This may lead us to believe that our estimates of the value of statistical life are not always accurate or useful. Indeed, there are lots of different estimates for the value of statistical life. One argument against our current values is that because they are based on revealed preferences of people making decisions about where to work and for what salary, that they might not apply to people not in the labor force (say retired recipients of Medicaid).
The vignette approach to estimating willingness to pay for mortality risk reduction
The main method our keynote speaker used to get around this problem of our revealed preference studies not lining up with the behaviors we see people take was to go back to the drawing board and pilot a new way to calculate how much people value reductions in their risk of death.
To do this, he used a dichotomous choice stated preference approach, which is a wordy way of saying survey respondents were given two options and had to choose the one they prefer. So instead of being asked “how much would you pay for a hot dog?” people were asked “would you prefer option A that costs $5.00 or option B that costs $9.00?” If you ask enough people to make these choices and you randomly vary the prices people see when they are asked, you can accurately determine how much people are willing to pay for certain goods.
The big innovation our speaker made was that instead of asking very specific questions, he gave his survey respondents longer vignettes about peoples’ lives to choose between. Each vignette told the story of a person’s life, where they lived, whether they married and had children or not, and importantly what their income was and how old they were when they died. His goal was to use those two facts to determine how much people were willing to trade off money for extra years of life.
This approach has two main advantages. First, it allows the researcher to control for a large set of preferences between people. With a large enough sample size, you can extract the importance of income and years separate from the other characteristics revealed in the vignettes. Second, it makes the question easier to understand for people. There isn’t some esoteric question about how much you’d pay for an extra year of life, you just see annual incomes and the age at which someone dies. Those are much easier for people to wrap their heads around.
In the end, this week’s keynote highlighted how difficult and how important it is to understand how people value changes in their health and longevity. Traditional revealed preference methods remain useful, but they can fall short when people face complex risks they do not fully grasp. The vignette approach offers a promising alternative by grounding abstract tradeoffs in clear and relatable life stories. As the field continues refining how we estimate willingness to pay for added years of life, innovations like this show how cost-benefit analysis evolves as we learn more about real human decision making.
