Bioeconomic modelling with end-users in mind:
The MIDAS experience in Western Australia
Amir K. Abadi Ghadim and David J. Pannell
Agricultural and Resource Economics, University of Western Australia, Nedlands 6907
Key words: bioeconomic model, sensitivity analysis, whole-farm model, linear programming
Abstract
MIDAS is a whole-farm linear programming model with a joint emphasis on biology and economics. The first MIDAS model was built in 1983 and since then several other versions have been developed in other regions, states and countries. The MIDAS models have been applied to a diverse range of issues including supply elasticities of farm commodities, evaluation of new technologies and trade-offs between economic and environmental objectives. Key features of the models have been the continuity and pragmatism of the core group of model developers and the continuous input from scientists and extension professionals. To promote multi-disciplinary impact and influence, several innovative tools have been developed to ease the task of generating results and communicating them. Examples of such tools are MARG (MIDAS Automatic Run Generator) and MID (MIDAS Interactive Database). The main strength of MARG is the ease with which it allows the modeller to carry out and summarise a large number of sensitivity analyses. MID utilises the output from this process to create a database of results. An end-user with no modelling or economic analysis background can easily use MID to query the results and compare various scenarios. In this paper we outline how we have used MIDAS, MARG and MID for research and extension. We will also discuss how we have been able to successfully integrate scientific and economic knowledge from a variety of disciplines in our bio-economic modelling work to provide key insights in accessible forms.
Introduction
The MIDAS (Model of Integrated Dryland Agricultural Systems) models are a suite of whole-farm linear programming models developed for several agricultural regions of Western Australia (Kingwell and Pannell, 1987). Over the years MIDAS modellers have shared their knowledge and experience in the development and extension of these models with colleagues from around the world. Whole-farm bioeconomic models like MIDAS provide a vehicle for better understanding important issues and decisions facing farmers and policy makers.
In Western Australia, MIDAS has been an influential tool for research prioritisation, extension, education and other uses. It has had its biggest impact through its influence on the bio-physical research activities of the Western Australian Department of Agriculture. This has been achieved by (a) bringing together researchers of various disciplines and extension agents who otherwise would interact little; (b) allowing scientists and extension agents to assess the economic significance of particular biological or physical information; (c) influencing the thinking of researchers and extension agents about the whole-farm system; and (d) highlighting data deficiencies and allowing prioritisation of research to overcome them.
Experiences with MIDAS point to a number of strategies and practices which may be beneficial to others undertaking similar bioeconomic modelling efforts. We believe that one of the key features of the success of our bioeconomic modelling effort in WA has been the appreciation for the role of sensitivity analysis (Pannell, 1997).
The focus of this paper is to share with other modellers our experience in developing tools that make easier the job of creating, compiling, sifting through, interpreting and communicating the results of complex sensitivity analyses with bioeconomic models. However, before we can do so, a brief history of modelling with MIDAS is helpful to illustrate the reasons why use of sensitivity analysis is such a feature of its application. This is followed by a discussion of the key principles of designing and conducting sensitivity analysis. We then describe specific software tools and approaches for conducting and communicating results from large and complex sensitivity analyses.
MIDAS
There are several versions for representative farms in different regions of Western Australia, but all include components for crops (cereal and legume), pastures, sheep, feeds (crop residues, grain, pastures), machinery and finance. They are detailed in their representation of soil types and potential enterprise rotations, with different production figures for each phase of each rotation on each soil type. The objective function is profit maximisation. The linear programming matrix varies in size between versions. As an example, the eastern wheatbelt version includes approximately 300 constraints, 450 activities and 5000 non-zero coefficients. MIDAS has been described in detail elsewhere (e.g. Morrison et al., 1986; Kingwell and Pannell, 1987).
The initiative for the model's initial development came from David Morrison (an economist in the head office of the Western Australian Department of Agriculture) and Mike Ewing (a pasture researcher in the Merredin regional office). Early model development was by Ross Kingwell, David Morrison and David Pannell. Throughout its history, primary responsibility for development and management of the MIDAS models has rested with a group of agricultural economists. In 1998 there are 12 agricultural economists who are using one or more of the MIDAS models for some aspect of their research or extension.
There have been three major developments in the software used for MIDAS. All three have increased the accessibility and usability of MIDAS. The first was the development of software capable of reliably solving large linear programming models on microcomputers. We use AESOP, a linear version of MINOS but there are now many packages available (Murtagh and Saunders, 1983). The second was the development of MARG (Pannell, 1990), a menu driven system for generating a series of model solutions and summarising results into tables. This has made the model much more accessible to non-specialist users and has allowed specialists to much more easily generate large series of runs, saving time and making practical much more comprehensive analyses of any given issue. The third development, MID (MIDAS Interactive Database), is a system that allows users to interrogate a database of MIDAS results addressing a particular issue. This allows rapid response to "what-if" questions and has allowed MIDAS results to be easily accessed by users with limited computer or modelling skills (Pannell, 1996).
Factors contributing to the success of MIDAS
There are a number of distinguishing features in the way MIDAS has been used in Western Australia. Part of its impact is attributable to the way in which analyses have been conducted collaboratively, with interaction between different disciplines and between researchers and extension agents. For this reason, the MIDAS team relies just as much on communication and interpersonal skills as on computer skills. Maintaining the goodwill and the communication channels are time consuming, but essential. It is surprising how quickly a group's confidence in a particular MIDAS model can diminish if the level of contact is too low.
The model is primarily used to put issues in perspective, rather than to provide definitive or precise numerical results. Our philosophy is that our role is to provide inputs to the judgements of decision makers (whether they be farm managers or research managers) rather than to be prescriptive. Consistent with this view, little emphasis is placed on detailed interpretation of any single model solution. The approach is to conduct extensive sensitivity analyses to investigate how the optimal farm strategy and profitability vary in different plausible scenarios. For any given issue, the number of model solutions generated may be very large (up to 10,000). We will now discuss the important role of sensitivity analysis in bioeconomic modelling and how the results of such analyses are communicated to non-modellers.
Sensitivity analysis
There are many reasons why sensitivity analysis is valuable to a bioeconomic modeller. In Table 1 the various uses of sensitivity analysis are grouped into four main categories: decision making or development of recommendations for decision makers; communication; increased understanding or quantification of the system; and model development.
In all models, parameters are more-or-less uncertain. The modeller is likely to be unsure of their current values and to be even more uncertain about their future values. This applies to things such as prices, costs, productivity, and technology. Uncertainty is one of the primary reasons why sensitivity analysis is helpful in making decisions or recommendations. If parameters are uncertain, sensitivity analysis can give information such as:
This information is extremely valuable in making a decision or recommendation. If the optimal strategy is robust (insensitive to changes in parameters), this allows confidence in implementing or recommending it. On the other hand if it is not robust, sensitivity analysis can be used to indicate how important it is to make changes to management to best suit different circumstances. Perhaps the base-case solution is only slightly sub-optimal in the plausible range of circumstances, so that it is reasonable to adopt it in all cases. Even if the levels of variables in the optimal solution are changed dramatically by a higher or lower parameter value, one should examine the difference in profit (or another relevant objective) between these solutions and the base-case solution. If the objective is hardly affected by these changes in management, a decision maker may prefer a simple strategy over a slightly more beneficial one involving complex changes in strategy.
If the base-case solution is not always acceptable, maybe there is another strategy which is not optimal in the original model but which performs well across the relevant range of circumstances. If there is no single strategy which performs well in all circumstances, sensitivity analysis identifies different strategies for different circumstances and the circumstances (the sets of parameter values) in which the strategy should be changed.
Even if there is no uncertainty about current parameter values, it may be completely certain that they will change in particular ways in different times or places. In a similar way to that outlined above, sensitivity analysis can be used to test whether a simple decision strategy is adequate or whether it is worth the trouble of having a more complex strategy that is conditional on the circumstances faced.
Table 1. Uses of sensitivity analysis. Adapted from Pannell (1997).
| 1. |
|
| 1.1 |
|
| 1.2 |
|
| 1.3 |
|
| 1.4 |
|
| 1.5 |
|
| 1.6 |
|
| 1.7 |
|
| 2. |
|
| 2.1 |
|
| 2.2 |
|
| 2.3 |
|
| 3. |
|
| 3.1 |
|
| 3.2 |
|
| 3.3 |
|
| 4. |
|
| 4.1 |
|
| 4.2 |
|
| 4.3 |
|
| 4.4 |
|
| 4.5 |
|
| 4.6 |
|
Sensitivity analysis can be used to assess the "riskiness" of a strategy or scenario (use 1.7). By observing the range of objective function values for the two strategies in different circumstances, the extent of the difference in riskiness can be estimated and subjectively factored into the decision. It is also possible to explicitly represent the trade-off between risk and benefit within the model.
Tools for sensitivity analysis
MARG (Mathematical-programming Automatic Run Generator) is a software package which works with AESOP to carry out large and complex sensitivity analyses of LP models. It also provides facilities for generating summary tables of desired variables from the lengthy output of the solver. In this, at least, it has some commonality with GAMS® (GAMS, 1988), but it still employs a matrix-based representation of the model. Some of the matrix generating capability of GAMS is captured by using spreadsheet software to present and store tables of data, and do pre-processing of data to generate matrix coefficients.
Over the years, improvements in spreadsheet technology have made the job of modelling much easier. Spreadsheet software also now provides facilities for solving mathematical models. These facilities range from relatively limited built in solvers to more powerful add-in programs such as Whats Best®. However, there is an important shortcoming of the use of such software. The facilities they provide for doing any more than a single optimal solution is either non-existent or cumbersome. Such facilities for sensitivity analysis can only be created with a considerable effort on the part of the modeller through the use of internal macro commands. The inability to conduct sophisticated sensitivity analyses excludes the modeller from the most important and interesting aspect of analysis with normative economic models (Pannell, 1997).
MARG was designed with the importance of sensitivity analysis in mind. It is very user-friendly, with simple-to-use menus, informative error messages and hundreds of context specific help screen specifically encoded to facilitate the process of conducting sensitivity analysis. It simplifies this process by integrating and streamlining a number of steps in the process of running a sensitivity analysis and summarising an output from it. It does this by first providing several facilities for making any required changes to model assumptions for the particular problem being addressed. Depending on the changes required, this can either be done using the matrix editor, GULP, or from revision files generated by a spreadsheet or text editor. Second, it creates a batch file to control the sequence of programs and model revisions during the sensitivity analysis. Third it runs the batch file and collates the output. Finally, it extracts and summarises the required information from the output files. It is this final step that allows us to look at only those sections of the model output that are relevant to the issue being considered. A neat, fully codified table of results generated in this way, no matter how large, can then be used to import into a spreadsheet or a database for making queries and further filtered viewing of the output of sensitivity analysis. MID is an example of such a database.
One of the most powerful features of MARG is the facility for cross-tabulation of several sensitivity analyses. There are no limits on the number of dimensions within which sensitivity analyses can be performed (i.e. the number of variables which can be cross tabbed) other than the practical ones of hard disk size and time required to generate the solutions. MARG makes it very easy to generate very large numbers of LP model solutions. This creates the risk of the volume of data obscuring the important issues (Eschenbach and McKeague, 1989 ). For example, the analysis and interpretation of output from 1000 solutions of an LP model with 300 activities and 150 constraints or transfer rows can be simply too hard and daunting to contemplate without a facility to condense, encode and sift through this volume of information. For this reason, the modeller must process and summarise the information to allow decision makers to identify the key issues. Pannell (1997) suggests possible layouts for graphs and tables.
In our experience it is important that in advance of any analysis the modeller consider carefully their approach to sensitivity analysis. In essence, sensitivity analysis is a simple idea: change the model and observe its behaviour. In practice there are many different possible ways to go about changing and observing the model. It is important to consider what to vary, what to observe and the experimental design of the sensitivity analysis. These issues are explored in detail by Pannell (1997).
Creating a database of sensitivity analysis results
Once tables of results from a sensitivity analysis are produced, they can be placed in a database for future interrogation and analysis. The term database is used generically in this context. Such databases can be maintained with a spreadsheet program or a dedicated database program. MID is an example of such a data base operated through the use of Microsoft Excel®. We prefer using spreadsheets because most modellers and their clients have access to one and are familiar with using it.
As shown schematically in Figure 1, MID can be viewed simply as a database with two main sections. The database section, invisible to the user making the query, is the large amount of sensitivity analysis output data stored with a unique code for each record.
The user-interface screen has an input section and an output section. In the input section the user specifies the scenario of interest, constrained to variables that were subject to sensitivity analysis. A series of query functions are the search facilities that fetch the results requested by the user.
In designing the output screen, the modeller must exercise judgement on which variables to display, guided by discussions with the end-user. Important features of a good output section are the judicious and selective display of only the variables likely to be of interest to the user and brevity and conciseness of the labels for tables and figures. A good design does not disappoint through over-promising. It leaves the user feeling like they got the answers that were possible and tantalises them to ask the modeller for more model runs. This outcome is ideal for the modeller.
Figure 1. A schematic diagram of the creation of a user-friendly database of results from the MIDAS model
Design considerations for the user-interface section of MID
We have become aware of features that help to make computerised communication tools effective and user-friendly. Here are features we have incorporated in the design of MID:
Figure 2 shows an example of the user interface of a MID dealing with grain legumes for the medium- to heavy-textured soils of the WA wheatbelt.
Figure 2. A typical example of the design of MID user-interface screen.
Other advantages to this approach include the following:
In conclusion, if the objective of a modelling project is to make a difference, then effective communication is at least as important as the actual modelling, and perhaps more so. We have outlined some tools and strategies which we have found to be effective in our work with the MIDAS model.
Acknowledgements
We thank the various members of the MIDAS team who have contributed ideas to the MIDAS process. We also thank Grains Research and Development Corporation and Rural Industries Research and Development Corporation for funding various related projects.
References
Eschenbach, T. G. and McKeague, L. S. (1989), "Exposition on using graphs for sensitivity analysis", The Engineering Economist 34, 315-333.
GAMS (1988),GAMS, The Scientific Press, Redwood City, California.
Kingwell, R. S. and Pannell, D. J. (1987), MIDAS, A Bioeconomic Model of a Dryland Farm System, Pudoc, Wageningen.
Morrison, D. A., Kingwell, R. S., Pannell, D. J. and Ewing, M. A. (1986), "A mathematical programming model of a crop-livestock farm system", Agricultural Systems 20, 243-68.
Murtagh, B. A., and Saunders, M. A. (1983), MINOS 5.0 User's Guide. Stanford California: Systems Optimization Laboratory. Technical Report SOL 83-20.
Nordblom, T., Pannell, D. J., Christiansen, S., Nersoyan, N. and Bahhady, F. (1994), "From weed to wealth? Prospects for medic pastures in Mediterranean farming systems of northwest Syria", Agricultural Economics 11, 29-42.
Pannell, D. J. (1990), MARG - MP Automatic Run Generator User Manual, Program version 2.4, Western Australian Department of Agriculture, Perth.
Pannell, D. (1996), "Lessons from a Decade of Whole Farm Modelling in Western Australia", Review of Agricultural Economics 18, 373 - 383.
Pannell, D. (1997), "Sensitivity analysis of normative economic models: theoretical framework and practical strategies", Agricultural Economics 16: 139-152.
Citation
Abadi, A., and Pannell, D.J. (1998). Bioeconomic modelling with end users in mind. Paper presented at Bioeconomics Workshop, University of New England, Armidale, NSW Jan 22 1998.