Aggregating Welfare Measures

Aggregation of individual utilities in empirical applications is based on certain pre-assumptions, which have to be clarified first. These pre-assumptions are related to people' s preferences and behavioral models. In addition, we have to deal with the question about the choice of correct summary statistics and appro-priate statistical estimation techniques of WTP.

As we know, utility or welfare cannot be measured directly. There are no measurable units that could be identified as "utils". The tradition among econo-mists is to assume that money is a close measure of utility and that an individual expresses his search for ever-greater satisfaction through the consumption of goods and services. However, it is not likely that ali people have similar atti-tudes towards money. For many people money and utility can be surrogates, but this need not be the case in general. Thus, the marginal utility of income can vary considerably among people.

Monetary gains and losses can be summed up across individuals in a theo-retically correct way Only when the distribution of welfare in the society is at the optimal level. This, in turn, implies that ali individual marginal utilities of income have to be equal (Johansson 1993). This kind of situation occurs only when the society is in the Pareto optimal state. Normally, not ali the criteria of Pareto efficiency are fulfilled. Thus, certain preconditions have to be satisfied before measures of welfare change (for instance, equivalent version of willing-ness-to-pay) can be added up across individuals or households. First, it is necessary that the marginal social utility of income is identical for ali individu-als. This is possible only in two cases. The first alternative is to assume that the marginal utility of income is the same for ali individuals. This is true if ali individuals have identical ordinal preferences, i.e. the same indifference curve maps. If individual indifference curves do not coincide, there is no unambiguous way to make individual marginal utilities of income equal across individuals.

The second alternative is to assume that some institution (e.g. government) continually redistributes income in order to maintain the equality of marginal social utilities (Boadway and Bruce 1986). However, every action of redistribu-tion is likely to create transacredistribu-tion costs. In other words, an attempt to develop a society that makes it possible to aggregate individual utilities in a theoretically correct manner is in itself a costly project that has redistribution and realloca-tion effects.

Aggregation across people can also be difficult because it is hard to know what the target population should be. This is a relevant question especially when significant nonuse values are involved. Jakobsson and Dragun (1996) conducted a CVM survey examining the value of the preservation of endan-gered species in the State of Victoria, Australia. The population surveyed was adult Victorians, but the authors emphasize that it is quite likely that Australians in other states and even people overseas may have existence values for Victo-rian endangered species. Thus, restricting the survey to VictoVicto-rians may underes-timate the value of protecting species in the State of Victoria.

Rubin et al. (1991) carried out a CVM survey that was aimed to investigate the value of the preservation of the northern spotted owl in the Pacific North-west in the USA. The target population consisted of residents of the Washington State. The authors reasoned, however, that due to the public good nature of

preservation, protecting spotted owls in the Washington and Oregon states would provide benefits to people throughout the nation. Therefore, they ex-trapolated the results to cover the whole country by assuming that WTP de-creases about 10% for every 1000 miles in distance. Even though the authors remarked that their extrapolations are more illustrative than conclusive, their approach underlines the potentially important role of nonuse values.

If the theoretical reservations conceming the aggregation of money metric measures of utility are left aside, it seems that aggregation across individuals depends heavily on the magnitude of nonuse values involved and on the scope of locality of the good being valued. Some natural resources create nonuse values on a global scale, for example, rain forests of the Amazon area in Brazil.

In this case, a random sarnpling from the target population would be extremely difficult and costly. Obviously, there is no definite rule where to draw the line in the aggregation issue. In practice, nevertheless, some Iines must be drawn.

Consider the valuation of the Finnish rural environment. It can be regarded as a national public good because the deterioration of the Finnish countryside does not have much impact on environmental quality and life-supporting systems on the world scale, unlike the Amazonian rain forests. Thus, the target population is those who are able to enjoy the rural environment and who in essence are responsible for the costs of producing the public good in question. This is obviously important from the social point of view because people usually con-sider it fair that potential beneficiaries of public goods participate in their production costs.

Another important aggregation issue is the use of summary statistics, which usually means the choice between the mean and the median. In the case of dichotomous contingent valuation data, effects of different statistical estimation techniques should also be considered. The choice of summary statistics depends on the formulation of the valuation scenario and the nature of desired policy recommendations. If the ultimate goal is to aggregate costs and benefits in a benefit-cost analysis framework, then the mean value is the appropriate measure because it makes it possible to apply Pareto concepts in the welfare change analysis (Harrison and Kriström 1995). This indicates that, if the objective is to find out whether the proposed change is a potential Pareto improvement, the mean value should be used. However, if the goal is to interpret the results in terms of the outcome of a referendum, the median is a more appropriate sum-mary statistics, although the use of the median value does not in general produce a Pareto efficient outcome (Johansson et al. 1989).

From the point of view of statistical estimation techniques, the median has some advantages over the mean. First, it is a more robust measure of central tendency. The mean is more sensitive to perturbations caused by errors in the data or by unusual observations (Hanemann 1984). It is also true that the mean is sensitive to the specification of the cumulative distribution function used in

the dichotomous data estimation, while the median is robust in this case, too (Kriström 1990a). Furthermore, the mean is more sensitive to methodical alter-natives in the structural model estimation, such as generalized least squares or maximum likelihood (Hanemann 1989). However, robustness of the median can also be a problem. Boyle et al. (1988) argue that, although the median may be less sensitive with respect to skewness in the distribution, one can expect the distribution of WTP values to be skewed, and thus the median may not be the ideal measure of central tendency. Consequently, the choice between the mean and the median should be made in a theoretically well-grounded manner, al-though issues related to statistical estimation should not dictate it.