Development and Demonstration of the Uses of Multivariate Outputs in Integrated Climate Assessment Models

Milind Kandlikar and M. Granger Morgan
Carnegie Mellon University

Objectives: Most climate impact assessment models produce a single output, typically measured in terms of dollar losses. As a consequence they do not display the rich economic, social, and ecological variety that is likely to characterize actual impacts. This is a problem because many different policy actors are involved in climate related policy decisions, and each is likely to assign different degrees of importance to different impact categories and reach very different conclusions about how a given vector of impacts should be valued. Models which integrate across all the separate impact categories and affected parties are likely to be of limited use in exploring climate change policy questions, since they will not be able to differentiate and support the different needs of various actors.

These considerations suggest that there is a need to develop techniques for integrated impact assessment that produce and use multivariate outputs. We are developing, demonstrating, and evaluating such techniques in the context of the CarnegieMellonlCAM-2framework. Multivariate outputs can be used in several ways. Single outputs or simple combinations of outputs can be used to evaluate alternative policy options against specific dimensions. The entire vector of outputs can be used, either directly, or through a utility function, to allow users to make holistic judgments. In addition, multivariate outputs can be used to work backwards to estimate the implicit cost of adopting certain "inviolate constraints".

Through a combination of such techniques, we are exploring the different valuations of specific scenarios for different key actors. This is the first step toward identifying "tradeoffs" that might be made among groups of actors who must cooperate if a particular strategy is to succeed.

Products: 1 ) A parametric analysis using simple constructed utility functions which reflect the values of key actors and demonstrate the usefulness of the approach (completed). 2) A detailed research plan to obtain utility functions from a number of groups of real policy actors in the U.S. and around the world, suitable for use in integrated assessment studies.

Approach: To illustrate the usefulness of the approach we constructed multi-attribute utility functions for two representative actors from the industrialized world - a pro-environment actor ("Environmentalist"), and a pro-growth actor ("Industrialist"). Subjective preferences were defined for the two actors over four climate change metrics - global mean temperature change, global mean sea level rise, changes in income from climate change and abatement measures, and changes in ecosystem prevalence. The Environmentalist was assumed to be risk averse to changes in ecosystems, sea level and global mean temperature, and risk neutral to changes in income. In contrast, the Industrialist had qualitatively opposite preferences. The single attribute utility functions for the metrics were combined additively, where the different actors associate different importance weights to each of the attributes.

The resulting multi-attribute utility function was used to evaluate three greenhouse policy options: a do-nothing option (A), a low carbon tax option (B) and a high carbon tax option (C). The two actors had very different choices for dominant policy strategies. Environmentalists preferred option C in most situations, except in the very short term where they are indifferent (B~C) between the two tax options. Industrialists on the other hand preferred option A in most cases, except when the decision rule is to minimize worst case outcomes in the long term where option B is preferred. The decisions of the two actors are most similar when either the short term expected damages or chances of long term low probability-high consequences events are minimized. These results demonstrate the usefulness of performing multi attribute evaluations of climate change policy options. Details of the work can be seen in Kandlikar and Morgan (1995).

The follow-up to the above analysis is a value construction research plan and protocol. This plan will be used to obtain utility functions from a number of groups of real policy actors in the U.S. and around the world. An important aspect of this approach is that it places climate change squarely in the context of overall global change. A generic value tree with the flexibility of representing the values of various actors is constructed. The end points of the value tree also called the "leaves" correspond to socio-economic and environmental variables that individuals care about. Many of the leaves will be affected by climate change, some may not be affected by climate change, others may be affected but in ways that are hard to quantify. For each of the leaves affected by climate change a single attribute utility function is elicited. As a starting point for the analysis we assume only incremental changes in the attributes a reasonable assumption as long as climate change does not cause major perturbations in the global economic and social landscape.

There are two dimensions to the value construction exercise. The first is a geographical dimension. Individuals may feel differently about climate change which has effects on their own community and other which are socially close to them. They may also feel differently about ecological impacts in different locations. Additionally, individuals may value ecosystems on the basis of their ecological prominence and richness rather than a geographical basis. A set of map based response forms are being develop to address these issues. The second dimension is temporal. As one moves further into the future, the line between generations starts to get blurred. We expect people to draw cleaner distinctions between impacts to those who they feel close socially close, and those who are more distant, for the next few generations that they do for later generations.

The value construction protocol is based on the notion that individuals have well characterized basic values. It is assumed that they may understand tradeoffs and have robust value judgments at a higher level, they may not have well articulated values on all parts of the tree, and may have to construct their values (rather than have them elicited). We are working to develop methods that will allow respondents to work from the top-down and from the bottom up in the tree, in order to construct their multi--attribute utility function.

Once we have completed our development of a utility construction method we will run a experimental studies to empirically evaluate the approach. If after refinement, we conclude that the approach is feasible, in the future we will seek support to ask several groups in the US and elsewhere to construct utility functions for use in integrated assessment studies.

In addition to the value construction exercise we are also working on methods to display multiple impact metrics. For the specific case of the climate problem, we are developing and demonstrating a variety of options including: Displays that summarize all outputs simultaneously; Displays that combine subsets of attributes (through functional combinations and/or parameterizations); A simple system that will allow us to specify multi-attribute utility functions for different actors and display the results as one-dimensional measures of utility.

Results to Date: We have completed the parametric analysis using simple constructed utility functions. The work was presented at the First Open Meeting of the Human Dimensions of Global Change Community at Duke University and has been accepted for publication in Human Dimensions Quarterly. We are now developing the value construction protocol which forms the core of the second stage of the work.

Manuscripts in preparation: Kandlikar, M., M. Granger Morgan (1995), Addressing the Human Dimensions of Climate Change: A Multi-Actor, Multiple Metric Approach, Human Dimensions Quarterly, in press.

Presentations: Kandlikar, Milind (1995) A Multi-Attribute Approach to Climate Change Policy, Seminar at CMU, May 1995.

Kandlikar, Milind.(1995a) A Multi-Attribute Framework for Climate Change Policy Analysis. First Open Meeting of the Human Dimensions of Global Change Community, Duke Univ, Durham, N.C., June 1995.

Student participation: In addition to Milind Kandlikar, the post doctoral fellow who has taken the lead on this work, a number of current Ph.D. students in the integrated climate change group at CMU have participated in the discussions and intellectual debate that has lead to the formulation of the second stage of this research. They include Elena Shewliakowa, Charles Linville. Natesan Mahasenan and Matt Oravetz.