Ian Rutherford
This box discusses the potential uses of a hierarchy of models to link economic, social and environmental indicators, and the results of a Canadian pilot study.
CORRELATION MODELS
One approach that has been used with some success to reduce the number of variables that a modeler or a decision-maker has to deal with is to compute correlation coefficients between all of the candidate variables. Those that are uncorrelated below a certain threshold are discarded as irrelevant and those that are nearly perfectly correlated are discarded as redundant. Correlation models can tell us something about the 'linkages' between a larger number of variables than we might think of in a simple Pressure- State-Response (PSR) framework, but they tell us nothing about why certain variables are correlated, and they deal only with linear correlation at that. In fact a correlation model does not care at all about cause or effect, so it sidesteps the arguments about the validity of cause-effect reasoning engendered by PSR models. Correlation results can stimulate research into new forms of conceptual models or complex system models.
INPUT - OUTPUT MODELS
Input-output models are widely used in economics to relate the input factors of production with the outputs. They have been extended with some success to deal with 'non-productive' outputs such as wastes and pollution and other by-products. They can deal quite readily with outputs that become inputs for other processes, in fact they were designed for precisely this. They can deal with physical as well as monetary flows. Input-output models are essentially linear and although they can handle a large number of variables of different types, they cannot deal with the feedback effects so characteristic of real, nonlinear, systems. They are fully deterministic. Although they can and have been used to construct scenarios for various levels of inputs, there is always the uncertainty that the relations, captured by the model for inputs and outputs in a given range, may not be valid for many different ranges because of the absence of feedback and other non-linear effects that cannot be captured in such a model.
COMPLEX SYSTEM MODELS
Complex system models are based on equations reflecting known relationships between the variables. They can be fully nonlinear, allow feedback, and demonstrate the full range of phenomena such as self-organizing structures, multiple quasi-stable states (attractors), sudden flips from one state to another, etc. Conceptually, what seems to be required is a complete and comprehensive model of the entire world and every living and non-living system in it. If such a model existed, we might be able to predict with reasonable certainty where we are headed, foresee disasters and be guided to take action to avoid them. We would know which are the critical variables to watch for (e.g. the indicators). Unfortunately, no such model exists and one will likely not be built in the near future. The real question is whether there are practical and useful models somewhere in between these ideal and linear single-issue models.
Some major parts or sub-systems of the whole ecosphere (understood to include human economies and societies) have been successfully modelled, notably certain physical systems like the flow of groundwater or the transport of air pollutants, or even the detailed evolution of weather, for which the nonlinear governing equations are reasonably well known. Even in these models, there are always some processes or boundary conditions that have to be approximated and the errors of approximation often turn out to be crucial in the long run. Nevertheless, such models have proven to be extremely useful. By working with them, a lot has been learned about predictability. It is known, for example, that even the simplest nonlinear systems have an element of unpredictability associated with the growth of trivial errors in initial conditions (the butterfly effect) and the existence of multiple attractors (quasi equilibrium points). Such models are not truly deterministic in the sense of being able to predict the detailed evolution of the system indefinitely into the future. Nevertheless, a range of useful predictability can be defined. Even though the details of the evolution of the system may not be predictable beyond a certain point, the statistics of the evolution and their dependence on boundary conditions or changes in external forces may be predictable for quite a bit longer. This is the basis for the modelling of climate (the statistics of weather) for example.
Attempts to model the economic sub-system have also been made, but with mixed success. Econometric models, for example, have many uses but because of the unpredictability of human behaviour, there are crucial processes for which either the form of the equations or their coefficients, or both, remain unknown or which change in unpredictable ways. These elements ultimately lead to uncertainty in the results.
In the field of ecology, some progress had been made in modelling, but there remain many areas where knowledge is simply too inadequate to construct even a conceptual model of how the system functions, let alone a detailed complex systems model. Ecological systems are particularly rich in multiple attractors and exhibit chaotic behaviour and unexpected flips from one set of dynamics to another. Some success has been achieved in building ecological systems models for limited systems such as a river basin or a lake.
GAMING MODELS
Some quite sophisticated and realistic simulation models of systems such as a city, a farm or an entire country have been constructed as computer games, for example SimCity by Maxis Corporation. Although intended as entertainment, to the extent that the relationships built into the model are realistic and complete, they can have considerable educational value. They can sensitize people to the complex interactions that go on within and between an ecosystem and its human inhabitants, and their social and economic systems. They can illustrate the consequences of human decision-making and behaviour and capture the complex feedback that either magnify or dampen their effects. Above all, they can get people used to thinking in systems terms and prepare them to make decisions based on holistic considerations, rather than simple linear, cause and effect terms. They can build an understanding of those circumstances in which simulation models can be relied upon and those in which they cannot. They can illustrate the power of looking only at certain key indicators as a surrogate for monitoring everything everywhere and all the time.
A good part of the problem of defining the right indicators is a question of mindset. Mindset limits our perceptions of what is a desirable future and our willingness to look at variables which we may not perceive to be important. We can learn from such models that sometimes the best way to approach that desirable future is not to aim directly at it, by attempting to maximize some variable of primary interest, but rather to follow some other path which arrives at the desirable future just as surely but with less effort and with fewer negative consequences along the way. This is another way of saying that we can learn the value of thinking and acting for the benefit of the long term rather than the short term.
In one such game, being built for the Lower Fraser Valley in British Columbia, the emphasis will be not on prediction but rather on the exploration of the problems involved in arriving at a pre-specified future scenario. The player will be able to explore the influence of such factors as values, the individual versus society, efficiency versus equity, and world views, such as the importance placed on: each of the ecological, economic and social subsystems and perceptions about the ability of the ecosystem to adjust to stress, the economy to foster technological change, and society to adapt. These are aspects that are crucial in the real world and which govern the reactions of societies to events and which are often the determinants of how they evolve. An understanding of these factors is crucial in the selection of appropriate indicators for decision-making.
TESTING SUSTAINABILITY INDICATORS THROUGH SCENARIO MODELLING
A pilot study to test the application of sustainability indicators for decision-making in a regional context, specifically for the Fraser River Basin in central British Columbia, using scenario modelling was completed during 1996 and led by a group at the University of Victoria.
Three types of modelling approaches were used in this study:
1. correlation modelling using pair-wise and multi-variate analyses;
2. input-output modelling using input-output structures with satellite accounts
for environmental impacts;
3. complex system modelling using qualitative policy variables linked to dynamic
and non-linear quantified indicators.
Scenarios and a futuring exercise were conducted with a group of 16 key stakeholders involved with sustainability research in the Fraser River basin. These were used to guide the study team in making projections or hypothesizing scenarios within the framework.
Solrie of the conclusions from the pilot study were:
1. It is important to link sustainability goals to movements of a small slate of indicators. Single indicators can rarely be linked to any specific sustainability goal. Detailed specific indicators are often highly correlated. It is more fruitful to focus attention on a small number of indicators within selected indicator classes.
2. The poor quality, inaccessibility and irrelevance of existing data are more pervasive constraints to reliable indicator modelling than is commonly thought. This is particularly true for smaller areas such as sub-basins within a larger watershed. Modelling is most appropriate at those scales for which reliable and independent data can be obtained.
3. Linking the use of deterministic and qualitative modelling approaches is a useful means for projecting indicators and discerning important policy linkages. Indicator modelling work is most suited to identifying policy trade-offs and implications on large systems, rather than forecasting specific indicators. Modelling efforts should focus on such general policy-related tasks.
4. Greater focus is required on modelling frameworks that can use incomplete data sets or qualitative information, including fuzzy logic models, advanced neural network models and other such non-statistical techniques.
OVERALL CONCLUSIONS FOR FURTHER RESEARCH
A number of conclusions, admittedly very limited, may be drawn from the above discussion of some of the issues surrounding indicator frameworks and the problems of linkages:
REFERENCES
Agriculture and Agri-Food Canada (1966) A National Ecological Framework for Canada. Ecological Stratification Working Group. Agriculture and Agri-Food Canada, Research Branch, Centre for Land and Biological Resources Research and Environment Canada, State of the Environment Directorate, Ecozone Analysis Branch.
Environment Canada (1996) Selection and Modelling of Sustainabilily Indicators for the Fraser River Basin. Final Report. Occasional Paper No.8, State of the Environment Reporting, Environment Canada.