Determining the parameters in a social welfare function using stated preference data: an application to health

Outline

• Abstract
• I. Introduction
• II. Defining the Social Welfare Function
• III. Estimating the Marginal Rate of Social Substitution Given a Social Welfare
• Function Specification
• IV. the Questionnaire Design
• V. the Empirical Study
• VI. Discussion

VI. Discussion

This study has sought to determine the shape of a health-related SWF from people’s stated preferences over various equality–efficiency trade-offs. While a CES was used as the baseline specification, similar results are derived from the hyperbolic and the parabolic specifications.

Overall, the results seem plausible, suggesting that there is aversion to inequalities in life expectancy, but its extent is sensitive to the groups across which the inequalities exist. However, the study also raises a number of methodological issues that warrant further discussion. In the first part of each question, the information regarding the size of the health gains of the two Programmes was easy to understand and, in the second part of each question, the implications of choices were made clear through changes in the size of the bars on the graph. Nevertheless, to facilitate this visual representation, the scales on the graphs did not start at zero (Appendix), and this could have led some respondents to perceive that the relative difference between the two groups was larger than it really was.

In general, it has been shown that very subtle changes in the framing of a question can sometimes have a dramatic effect on responses (for an excellent review, see Rabin 1998). This study was designed to minimize the effects of certain framing effects but it is impossible to remove every potential bias. For example, we were aware of the evidence from other studies that suggests that respondents might be reluctant to give all the benefit to one individual or group (see e.g. Cuadras-Morato et al., 2001). We went further, thought, and asked respondents who chose not to target if they would have targeted if there had instead been a 1-year benefit to the better-off group (and hence a 3-year benefit to the worse-off group). None of these respondents chose to revise their answers.

It is now well established that respondents may give greater weight to the losses of one group as compared to an equivalent gain to the other group (Schweitzer, 1995). Therefore, the questions were designed so that neither programme in the two questions involved any losses, and so that neither programme was presented as representing the status quo. However, it is possible that loss aversion may also be present when considering potential as well as actual losses from a particular reference point (Dolan and Robinson, 2001). Therefore, if some respondents adopted the potential gains available to both groups in Programme A as their reference point, then Programme B would involve a ‘loss’ to the better-off group. It would be interesting, and policy relevant, to test with further research how sensitive the degree of inequality aversion is to variation in the initial situation.

There is a status quo bias of a different kind that might have made respondents more inclined to stick with Programme B if they chose it initially. This relates to the fact that respondents were always presented with response categories in the same order; that is, Programmes A and B start out being equally effective and then B becomes incrementally less effective. This ordering was chosen to make the equality–efficiency trade-off as transparent as possible and was informed by the results from the pilot interviews which suggested that the trade-off questions would have been cognitively too difficult if the ordering of the response categories was randomized.

However, there is the possibility of a status quo bias whereby some respondents get ‘locked into’ choosing B throughout (Samuelsen and Zeckhauser, 1988). On the other hand, there is some limited evidence that shows there may be a ‘left-hand side’ bias: when respondents are asked to choose between two options laid out next to each other, the default choice is the option on the left-hand side, and the right-hand side option will be chosen only when it is significantly more preferable than the default option on the left-hand side (see e.g. McIntosh and Ryan, 2003). Thus, there are two potential biases working in opposite directions.

Despite these concerns about the data, we believe that this study represents an advance in terms of both the methodology used and the implications for future research that seeks to enhance the policy usefulness of stated preference data. It suggests that people are willing to forego overall health in order to reduce differences in average life expectancy between the social classes. On the other hand, differences in the average life expectancies of men and women did not seem to matter much at all, with the median respondent unwilling to sacrifice any overall gains in life expectancy in order to target men. Tsuchiya and Williams (2005) try to get behind some of the reasons for the very different attitudes towards health inequalities by sex as compared to those by social class.

In conclusion, this study has demonstrated that, using carefully designed questionnaire instruments, the SWF can develop from being a theoretical construct to becoming a potentially powerful practical policy tool. A survey instrument can be designed so that elicits meaningful trade-off responses from the general population can then be used to determine the shape of the SWF. We therefore believe that the study indicates a promising new avenue of economic enquiry that is relevant to important policy questions in health care and other areas of public policy.