To Main Proceedings Document
USING SYSTEM DYNAMICS AND 3-DIMENSIONAL
VISUALIZATION TO EXPLORE THE DYNAMICS OF
FUTURE GLOBAL PROTEIN CONSUMPTION
Steve Sonka, Donna Fisher, and Randall Westgren’
University of Illinois, Urbana-Champaign
National Soybean Research Laboratory
1101 W. Peabody, Rm. 170
Urbana, IL 61801 USA
217-244-1706 (office)
217-244-1707 (fax)
sonka@uiuc.edu
Abstract
Strategic decision making is particularly difficult relative to research investments, where the
uncertainty inherent in research and lengthy time lags requires investments to be made far
before outcomes are known. This paper reports upon the development and evaluation of the
Protein Consumption Dynamics (PCD) system, a tool created to assist managers to improve
their perspective of future protein needs. This research effort was funded by the Illinois Soybean
Checkoff Board to aid them in strategic allocation of research funds.
The PCD system includes a Powersim model, the output of which is displayed using a 3-
dimensional visualization software package, In3D. The system dynamics model component
relates population and income growth to regional protein needs and malnutrition. The model
tracks estimated consumption annually (for the years 2001 to 2025) of six agricultural
commodities that serve as sources of protein for humans in eight regions that encompass the
world. The system dynamics model is designed so that alternative scenarios of the future can be
examined using population and income projections of the World Bank and the UN’s Food and
Agricultural Organization.
The output of the system dynamics model is displayed using 3-dimensional visualization
techniques. The visualization component was developed in collaboration with design experts
from the National Center for Supercomputing Applications at the University of Illinois.
Through formal experiments with actual manager in the soybean sector, the effects of use of the
PCD system are being formally evaluated. This evaluation documents the effects of scenario
modeling and visualization on individual and group decision-making processes.
Introduction
Strategic decision making is particularly difficult relative to research investments, where
the uncertainty inherent in research and lengthy time lags requires investments to be made far
before outcomes are known. This paper reports upon the development and evaluation of the
Protein Consumption Dynamics (PCD) system, a tool created to assist managers to improve their
perspective of future protein needs. This research effort was funded by the Illinois Soybean
Checkoff Board to aid them in strategic allocation of research funds.
Daellenbach (1994) identifies several factors that contribute to today’s complex
environment including rapid technological advances, information explosion, and the widening
gap between the developed and underdeveloped countries of the world. Nowhere is this more
apparent than in the agricultural sector. The seasonal dimension of agriculture means the results
of decisions made today regarding planting and chemical applications often take months to
materialize. Likewise, decisions related to investments, market development, and agri-chemical
research can take years, or even decades, to yield results. Other factors that contribute to
complexity in agriculture include demographic issues (poverty, high population growth, and
income growth rates), dietary and consumer preference changes, government action, agricultural
research, land use, and climatic changes (Pinstrup-Andersen & Pandya-Lorch 1998).
The Illinois soybean industry continually grapples with such complexity. The Illinois
Soybean Program Operating Board (ISPOB) is a public sector organization, which invests in
soybean research and market development for Illinois soybean producers. As such the Board
faces many similar challenges of a private firm regarding technological innovation decisions.
However, as a public entity, ISPOB must also answer to the producers (stakeholders) it is
designed to serve. This includes educating producers regarding the appropriateness and
applicability of ISPOB research and marketing activities. Membership of the ISPOB is elected
and voluntary. Thus the group faces problems inherent to any organization in terms of decision
making and learning, but the situation is exacerbated by the turnover and pluralism inherent in
the organizational structure.
The long-run future of the soybean sector is very promising. However the current actions
of the industry’s decision makers will determine the nature of that future. Through the use of
scenarios, this sophisticated modeling tool assists decision makers to focus on and better
anticipate the future. The ultimate goal of the research is to improve decision makers’
confidence about where to invest research dollars so as to positively affect future success.
This research focuses on how decision support systems can alter perceptions of the
decision making environment in the soybean industry. The research investigates whether group
decision making processes, namely those of the ISPOB, can be improved by using computerized
decision aids. To do so the study examines the decision makers’ cognitive maps (or perceptions)
of the decision environment.
A system dynamics model of global human protein consumption dynamics is the basis
for the project. The uncertain time paths for consumption of protein from animal and vegetable
sources in diets around the world are a cause of decision ambiguity for the soy value chain today.
The model allows decision makers to explore how consumption plays out on uncertain futures
given alternative scenarios of income and population growth over the span of the simulation.
Sophisticated three-dimensional visualization techniques are used to communicate the model
output to decision makers.
The remainder of this paper addresses the research question, theoretical background,
visualization model, data collection experiment, and preliminary results.
Research Question
The primary research question of this study centers on how to assist decision makers to
improve strategic decision making. If we are able to broaden their perspective to include a more
global and long-term outlook, then the quality of their decision making should be enhanced.
Thus, the visualization model that is a part of this research, is designed to impact strategic
decision making. Decisions are based on many things, one of which is the decision maker's
perception (cognitive map) of the decision environment. A goal of this research is to measure
the effectiveness of improved understanding and decision making by exploring how the
visualization model changes the cognitive maps of various soy industry decision makers.
Theoretical Background
This research draws on a number of different literature streams. The study of decision
making has been central to several fields including economics, anthropology, psychology,
computer science and management. | Cognitive maps, or perceptions, are an integral part of
strategic decision making. Scenarios help decision maker’s comprehend the complexity of their
environment. Visualization, then, transforms a multitude of data into information that is easily
utilized by decision makers. Thus, individual learning can take place through the use of
scenarios and visualization. The remainder of the section discusses each of these concepts as
they relate to this study.
Psychology and organizational behavior scientists struggle with how to measure the
decision making process. One important facet of the process is the decision maker's cognitive
map (perception) of their problem environment (Huff 1990). Cognitive maps help decision
makers organize the over abundance of information to which they are exposed. The cognitive
processes associated with strategy formation (and decision making) are based on maps that
individuals have of the world around them. These maps can represent the individual’s
interpretations about the world (Mintzberg et al, 1998).
Mason (1994) asserts that “[a] critical task of planning is to provide tools that adjust
managers’ [cognitive maps] to reflect the rapid changes in their competitive environment,” (p. 7).
Cognitive maps based on outdated information result in bad decision making and focus attention
away from important causal relationships. By making cognitive maps explicit, one can identify
gaps as well as key variables. Thus cognitive maps help structure and resolve problems,
sometimes in a creative manner. This research examines how cognitive maps change as the
result of exposure to a visualization model designed to highlight the relationships between key
variables and make explicit the complexity of the decision environment.
While system dynamics research professes to change mental models, these changes are
generally measured using self-assessment surveys. This type of self-evaluation of cognitive
changes can be problematic due to the participant’s lack of understanding of how he/she has
been influenced. Providing the participant with sufficient detail to understand the experiment,
however, may result in subject biases from knowing too much about the studied behavior (Doyle
1997:256). | Cognitive psychology offers techniques to accurately measuring these changes.
Doyle posits that system dynamics intervention evaluation can gain from the controlled
experimentation techniques of cognitive psychology, specifically, the use of “pre- and post-
measurements of cognitive processes and mental models,” (p. 256). This research uses pre and
post questionnaires to assess the changes in mental models of soybean industry decision makers.
Cognitive maps change through learning (Sterman 1994). However, according to Argyris
(1994), organizations in and of themselves do not learn. Learning takes place at the individual
level. An organization learns either through the learning of its individual members or through
acquiring new individuals with knowledge beyond that already within the organization, (Sonka
et al 1995 point to Simon). Because of bounded rationality (or limits on cognitive capacity),
models and scenarios are needed to help decision makers narrow the scope and therefore better
comprehend the complexity of their environment.
Scenario analysis differs from other forecasting in that it is more descriptive, qualitative
and contextual; and that it identifies plausible possible futures. “Scenarios also provide a
common means for everyone in the company to think about the future that takes into account
many uncertain factors (some of which are qualitative) in a flexible, although estimative, way,”
(Mason, 1994:66). By focusing on only a small number of potential futures, decision makers
will be able to more fully explore the implications of decisions they make today in relationship to
these various futures scenarios.
Richardson (1996) identifies several issues for future system dynamics research. Those
relevant to this paper include understanding model behavior and widening the base of system
thinking in other fields. He suggests the development of computer-based tools that facilitate
“understanding the connections between model structure and behavior,” (p. 142).
Visualization enables understanding and communicating research results to other
researchers and the general public. It helps shape public policy by improving understanding
regarding potential outcomes and the relationships between multiple variables (Orland et al,
1997). “Visualization— combining computer graphics, computation, communication, and
interaction— is invaluable for changing data into information, designing products and supporting
complex decision making,” (Brown 1997:1; also see Rheingans & Landreth 1995).
The three-dimensional representation of important variables is one way to emphasize the
relationship between actions and future consequences, and to illustrate the lack of effects that
exogenous factors have on the system. This combination of the power of system dynamics and
visualization should aid in understanding the interrelationships of the simulation model variables
(Richardson 1996). The three-dimensional representation highlights the relationships between
several variables simultaneously. The understanding gained from seeing the interrelationships
among variables will enable soybean decision makers to more full comprehend their
environment.
The multidisciplinary nature of this study makes it difficult to know the variables and
theories related to the analysis a priori, therefore a qualitative research methodology is used.
Qualitative research exhibits the following characteristics:
1. Data source is in a natural setting with the researcher as the key instrument
2. The research is descriptive in nature
3. Process is more important than outcome or product
4. Induction is used to analyze the data
5. Major focus is on meaning or participant perspective (Bogdan & Biklen, 1992)
Accordingly, this research describes how the cognitive maps of soy industry decision makers are
influenced with the use of sophisticated visualization of information (2). It is concerned with the
nature of these decision makers’ perceptions (5), which are captured during interaction with
subjects (1). Content analysis (4) is used to evaluate the changes in perceptions (3).
Visualization Model
Visualization provides a sophisticated means of characterizing information to enable
decision makers to more easily perceive the interrelationships between the model drivers, and the
resulting appetite for the various commodities. The system dynamics model component of this
research (the model underlying the visualization) relates population and income growth to
regional protein needs and malnutrition. The model tracks estimated human consumption
(potential demand) annually from 2001 to 2025, for six agricultural commodities (beef, fish,
pork, poultry, fats & oils, and vegetable protein) in eight regions that encompass the world.
Population and income growth information are based on secondary data taken from the World
Bank and the United Nations Food and Agriculture Organization. The visualization makes it
easier to see and understand the interrelationships between the variables in this multitude of
information (8 regions x 4 scenarios x 25 years x 6 commodities x income x population). The
understanding gained from seeing the interrelationships among variables will enable soybean
decision makers to more full comprehend their environment."
Figure 1 is a photograph of the visualization model. Regional population and GDP totals
are positioned on the back wall of the visualization. Each color-coded region on the floor of the
visualization contains a tri-colored bar which represents the (potential) demand for the various
commodity groups. As the model animates through time the bars change to reflect how different
population and income growth scenarios affect potential demand on a region-by-region basis.
The visualization also allows for the commodity groups to be explored in more detail (the area
on the left).
Figure 1. Protein Consumption Dynamics Model
Experimentation
The experimentation using the visualization model is conducted with soy industry
decision makers. Following Doyle’s suggestion (1998), data are collected through before and
after questionnaires that elicit the strategic issues map from participants. Subjects also engage in
a group discussion regarding what they have learned from the exercise. The discussion takes
place following the second questionnaire.
Model exposure is a scripted exercise focused on learning from the future that encourages
participants to think about the key factors influencing their industry. Content analysis software
(Nud*ist VIVO) is used to evaluate variations between the before and after questionnaires. The
analysis looks at how the individual’s maps change, as well as how maps within and between
groups change. Transcripts of the group discussions are also analyzed.
Following Creswell (1994), we are more concemed in this study with expert perceptions
than in statistical accuracy. Therefore, the experimentation is with a number of hand-selected
subjects who have special knowledge of key issues within the soybean industry. The subjects
received a treatment that combines the tabular and visualized information.
The before questionnaire contains 4 questions. Question 1 asks for questions the
respondents have regarding the future of the industry. Question 2 asks for the key issues to be
worked on in the industry. In addition, both of these questions contain a part ‘b’ which asks the
respondents to provide a ranking of his/her responses. Question 3 requires the respondent to
make an explicit decision regarding research funds allocation, (similar to Wilson, et al’s [1989]
on-line judgment). Question 4 asks for a self-evaluation of how confident the respondent is
about the previous decision. The questionnaires are number identified for internal tracking
purposes, with the before and after questionnaires having the same id number for a given subject.
The after questionnaire has the same questions 1, 2 and 3. Question 4 solicits the
decision criteria that influenced the previous decision (following Wilson et al, 1989). Question 5
of the after questionnaire is identical to Question 4 on the before questionnaire. Finally, a few
demographics are collected.
Three pretests were conducted— one to test the questionnaire, one to test the exercise
methodology, and then a final test of both the methodology and the questionnaire. The first
pretest was with the Executive Veterinary Medicine Program, of the University of Illinois,
Urbana-Champaign. The 29 subjects were Midwest veterinarians involved in the swine industry.
The pretest indicated that there were too many open-ended questions. In addition, the process
needed to give something back to participants. The second pretest was conducted with the
Illinois Soy Leaders group. The ten subjects were producers, processors, and personnel involved
in the soy industry. This pretest looked at the experimental exercise process. While no data
were collected via questionnaire, there was positive anecdotal evidence that the subjects
benefited from the group discussions of the experimentation process. The final pretest was
conducted with three faculty members from the College of Agricultural, Consumer, and
Environmental Sciences, University of Illinois, Urbana-Champaign.
As a result of the pretest, the nature of the questions did not change significantly. Two
questions were found to be redundant and the wording was changed on the others. A self-
confidence evaluation question was added after the pretest. Therefore in place of a second
pretest of the questionnaire, it was taken to the Survey Research Institute, University of Illinois,
Urbana-Champaign and received positive comments with only a few additional changes. Initial
findings from the first pretest show that the results are as anticipated. There was a shift from a
local to a more global focus. In the before questionnaire, 55% of the subjects had some mention
of a global perspective. In the after questionnaire, 100% had a strong global emphasis.
Preliminary Analysis“
Data collection took place from January to March of 2000. Primary data are collected
through the use of a before and an after questionnaire designed to solicit subjects’ cognitive
maps (or perspectives on the soybean industry). Table 1 describes demographic information
related to the subjects in part two of this experiment. The gender mix is heavily male, which is
representative of the industry. Most of the subjects have at least some post secondary education,
with nearly 65 percent having college degrees. The subjects are spread across a number of
sectors including producers, researchers, service providers and agribusiness students
(undergraduates and graduates)."”
Table 1. Subject Demographics for the Protein Consumption Dynamics Experiments
Demographic Category Sub-category Number of Participants
Gender
Male 93
Female 28
Education
High School 4
Vocational/A ssociates 13
Some College 26
Bachelor's Degree 23
Master's Degree 26
Doctoral Degree 29
Age Average 38.5
Occupation
Producer 26
Researcher 28
Service Provider 33
Agribusiness Student 34
Total 121
Results show a shift in the research allocation decisions as a result of seeing the
visualization model, as seen in Table 2. In the before questionnaire, respondents focused more
on new product development and developing new markets. In the after questionnaire, the group
directed even more resources toward developing new markets and shifted away from new
product development and genetics research. In the after questionnaire, the subjects still
recognized the importance of the local issues, but this perspective expanded to include more
global and long-term issues.
Table 2. After Questionnaire responses to Question 3 on Research Budget Allocation
Change from
Before
Area Average Questionnaire
Production Research 15.99 -0.04
New Product Development 19.28 -2.54
Marketing Research: Strengthen Existing Markets 18.98 +1.07
Marketing Research: Develop New Markets 25.08 43.33
Genetics Research 16.43 -2,27
Other 4.25 40.45
Total 100.00 0.00
Summary
The primary goal of this research is to analyze changes in cognitive maps of soy industry
decision makers to determine the effectiveness of using the visualized representation of
information from the protein consumption dynamics simulation model. The model uses various
scenarios based on income and population growth projections to determine the future appetite for
a number of commodities related to the soy industry. Preliminary results indicate that the model
is effective in shifting decision makers’ perspectives to a more global and long-term focus, thus
influencing their budget allocation decisions.
Further work in this study will look at changes in perspectives at the individual, group
and across group level. It will also test the differences between using the 3-D model and a
tabular representation of the same information. Long-term plans include making the
visualization interactive and including more policy variables in the model.
References:
Argyris, C. (1994). On Organizational Learning, Blackwell Publishing, Cambridge.
Bogdan, R.C. and Biklen, S.K. (1992). Qualitative Research for Education: An Introduction to
Theory and Methods, Allyn and Bacon, Boston.
Brown, J.R. (1997). Visualization and Scientific Applications. In Earmshaw, R. J. Vince, and H.
Jones, 1997, Visualization and Modeling, Academic Press, San Diego, pp. 1-11.
Crano, W.D, and Brewer, M.B. (1973). Principles of Research in Social Psychology, McGraw-
Hill, New Y ork.
Creswell, J.W. (1994). Research Design: Qualitative and Quantitative Approaches, Sage
Publications, Inc., Thousand Oaks, California.
Daellenbach, H.G. (1994). Systems and Decision Making: A Management Science Approach,
John Wiley & Sons, Inc, New Y ork.
Doyle, J.K. (1997). The Cognitive Psychology of Systems Thinking. System Dynamics Review,
13(3):253-265.
Huff, A.S. (ed) (1990). Mapping Strategic Thought, John Wiley & Sons, Inc., New Y ork.
Mason, D.H. (1994). Scenario-based Planning: Decision Model for the Learning Organization.
Planning Review, April/March, 6-11.
Mintzberg, H., Ahlstrand, B., and Lampel, J. (1998). Strategy Safari: A Guided Tour Through
the Wilds of Strategic Management, The Free Press, New Y ork.
Orland, B., Ogleby, C., Bishop, I., Campbell, H., and Yates, P. (1997). Multi-media approaches
to Visualization of Ecosystem Dynamics. Intemet world-wide web page, at URL:
<http://imlab9. landarch.uiuc.edu/papers/>.
Pinstrup-Andersen, P, and Pandya-Lorch, R. (1998). Major Uncertainties and Risks Affecting
Long-Term Food Supply and Demand. In Organization for Economic Co-Operation and
Development (OECD), The Future of Food: Long-Term Prospects for the Agro-Food Sector,
Author, Paris, pp. 53-70.
Rheingans, P., and Landreth, C. (1995). Perceptual Principles for Effective Visualizations. In
Grinstein, G., and Levkowitz, H. (eds), Perceptual Issues in Visualization, Springer, Berlin,
pp. 59-73.
Richardson, G.P. (1996). Problems for the Future of System Dynamics. System Dynamics
Review, 12(2):141-157.
Sonka, S., Cloutier, L.M., and Banik, M. (1995) Organizational Learning Models of Strategic
Alliances, Working Paper presented at the American Agricultural Economics Association
Organized Symposium on Using Firm-Level Models for Strategic Management Research in
Agricultural Economics, August 1995.
Sterman, J.D. (1994). Learning in and About Complex Systems. System Dynamics Review,
10:291-330.
Wilson, M.G., Northcraft, G B., and Neale, M.A. (1989). Information Competition and
Vividness Effects in On-Line Judgments. Organizational Behavior and Human Decision
Processes, 44:132-139.
Footnotes
‘ Authors are respectively Director, National Soybean Research Lab and holder of the Soybean
Industry Chair in Agricultural Strategy; graduate research assistant; and Associate Professor of
Agricultural Management.
“ We collaborated with the National Center for Supercomputing Applications at the University of
Illinois in developing the 3-D visualization. We used Visible Insights’ In3D software, which
provides a three-dimensional, dynamic programming environment for data representation.
™ The analysis of collected data is still in progress at the time of writing this paper. However,
complete results will be reported at the System Dynamics conference, and will be incorporated in
a future version of this paper.
"’ Agribusiness students were included for two reasons. First, many of the students are from
farms where soybeans are produced. Second, many of these students will be future decision
makers of the industry.
