Content uploaded by Hans-Joachim Mosler
Author content
All content in this area was uploaded by Hans-Joachim Mosler on Oct 31, 2017
Content may be subject to copyright.
Journal of Social Issues, Vol. 63, No. 1, 2007, pp. 97--116
Simulating Human Behavior for Understanding
and Managing Environmental Resource Use
Wander Jager∗
University of Groningen
Hans Joachim Mosler
University of Z¨
urich and Eawag
Computer simulation allows for the experimental study of dynamic interactions
between human behavior and complex environmental systems. Behavioral deter-
minants and processes as identified in social-scientific theory may be formalized
in simulated agents to obtain a better understanding of man–environment interac-
tions and of policy measures aimed at managing these interactions. A number of
exemplary agent-based simulation studies is discussed to demonstrate how simu-
lations can be used to identify behavioral determinants and processes underlying
environmental problems, and to explore the possible effects of policy strategies.
Finally, we highlight how agent-based simulation may contribute to our under-
standing of the dynamics of environmental resources, and how to manage them in
a sustainable way.
The often devastating effects of human behavior on natural resources have
been observed since ancient times. For example, Plato in his Critias (360 B.C.)
already discussed the erosion of Attica due to agriculture. Since a few decades,
most notably after Hardin’s (1968) paper on the tragedy of the commons, the social
sciences took up the challenge of identifying the social and behavioral drivers of
environmental resource use, and to develop and test environmental management
strategies. Since then, the experimental tradition within the social sciences has
yielded an abundance of laboratory studies revealing how various personal and
contextual factors influence people’s use of an environmental resource. Results
∗Correspondence concerning this article should be addressed to Wander Jager, University of
Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands [e-mail: w.jager@rug.nl].
The authors thank Wernher Brucks, the issue editors, and several anonymous reviewers for pro-
viding useful comments on earlier versions of this article.
97
C
2007 The Society for the Psychological Study of Social Issues
98 Jager and Mosler
obtained in studies focusing on, for example, attitudes, interpersonal processes,
and social dilemmas, are relevant in understanding how people interact with their
natural environment. However, one of the main problems of translating experimen-
tal findings to real-world situations is the complex nature of reality. Laboratory
studies usually focus on a limited set of variables in a well-controlled environment.
In real life, human behavior is determined by a multitude of constantly changing
and interacting variables. This is especially true in environmental resource use. On
a personal level, factors such as knowledge, attitudes, goals, power, personality,
and the like determine the individuals’ disposition toward performing certain be-
haviors. Whereas the effects of single factors are often known, for combinations of
such factors the effects are unclear, as the effects may interact in a complex manner,
and cannot be simply added or subtracted. At the level of the social system, com-
plexities arise from large populations of heterogeneous people, interacting through
networks that are subject to change. The resulting group processes are very com-
plex and often unpredictable, such as public opinion formation, diffusion of new
behaviors, and the development of fashions. Complexity of the environment arises
from multiple interacting environmental processes (e.g., global warming affects
the climate, which affects chances of floods and storms), the different time-scales
involved, the large time lags of effects, the unknown regeneration rate of resources,
and several feedback and feed-forward mechanisms.
Whereas systematic experimentation is possible by changing a limited num-
ber of factors in experimental settings, for instance, the effect of environmental
information on the valuation of the fuel efficiency of cars, laboratory experiments
are not a suitable tool to study human behavior in complex environmental settings
because effects of interventions may depend strongly on the context in which they
are implemented. For example, people may respond very differently to (combina-
tions of) policy measures, and they may respond to what other people are doing,
causing effects to be sometimes rather unpredictable. Hence, the experimental
laboratory approach does not allow for the a priori evaluation of policy strategies
for the management of complex environmental resources. Case studies offer an al-
ternative tool for studying behavior in complex environments. Whereas these may
provide valuable information on the complexities in given domains of consumer
behavior, due to the lack of experimental control case studies do not provide the
researcher with insights about causal behavioral mechanisms. Rather, a method-
ology is needed that allows for experimenting with behavioral processes within
actors, social processes between actors, and interactions between actors and the
environment. Agent-based simulation is a tool that offers a perspective on simu-
lating human behavior in complex environments, and thus may provide a suitable
tool to experiment with the management of complex environmental resources.
First, agent-based simulation allows for experimenting with the complexities
at individual, social, and environmental levels by formalizing populations of artifi-
cial humans, called “agents” in an artificial world. Many factors can be included in
Understanding and Managing Environmental Resource Use 99
these formalizations, and using computer simulations one can conduct thousands
of experiments in a short time, thus allowing for exploring the effects of many
combinations of factors. Such experiments also reveal to what extent certain com-
binations of factors result in fairly robust outcomes, or, on the contrary, result in
outcomes that are highly susceptible to minor changes in factors.
Second, agent-based simulation allows for the modeling of interactions be-
tween individuals. Assuming that social interaction causes information and norms
to spread, the accumulation of these interactions can be studied on a population
scale. The number of agents to be included in a computer simulation depends
only on the power of the computer, but even with an ordinary PC it is possible
to simulate populations from 10.000 to 1 million. This allows for showing how
population phenomena, for example, opinions on energy issues, may emerge from
interactions at the local level.
Finally, agent-based simulation allows for experimenting with policy mea-
sures without harming people and the environment. Via simulations the mid-term
and long-term effects of policy measures can be studied in scenarios. Different sce-
narios including different forecasts of economic development and environmental
quality could be used to test the effectiveness of policy measures under different
conditions. Experiments can be repeated under the same starting conditions with
different policy measures (and assumptions on their effect on individuals) as many
times as we want. Therefore it is possible to simulate different policy strategies to
examine which policy may be optimal in the specific situation at hand.
What Should Be Simulated, and How?
For the simulation of interactions between human behavior and environmen-
tal systems, the main question is: What kind of behavior should we model, and
what determinants and processes should be formalized? People do not behave pro-
or anti-environmentally as if environmentally relevant behavior were a special
category of behavior (see also Lindenberg & Steg, this issue). Rather, they per-
form a wide range of behaviors for various reasons, and all these behaviors have
a multitude of consequences, including positive or negative1environmental con-
sequences. Hence, many theories on general human behavior and its underlying
determinants and processes can guide the development of agent rules to simulate
man–environment interactions. Jager and Janssen (2003) propose that theories on
needs, decision-making processes, and processes of (social) learning comprise key
components to be modeled because they describe the motivation to perform be-
havior, diverse choice processes in selecting behavior, and storage of positive and
negative experiences after performing behavior, respectively.
1The distinction between positive and negative environmental consequences is a simplification,
as these consequences are multidimensional by nature.
100 Jager and Mosler
Various needs may underlie the many interests and motivations people have to
behave in a certain manner (e.g., Maslow, 1954; Max-Neef, 1992). Some behav-
iors satisfy several needs simultaneously, while other behaviors may satisfy one
need at the cost of another. For example, in order to be able to pay their children’s
education or to buy a car, people may willingly accept to work and live in un-
healthy environmental conditions. In simulation models, human needs constitute
an important concept for capturing the basic drivers of behavior. The formalization
of needs may be of particular importance when (direct) satisfaction of personal
needs harms the environment, which in turn may affect need satisfaction in the
long run.
Decision processes refer to the way people make choices between various
behavioral options, and they determine the storage of (new) information and the
formation of attitudes. Relevant processes vary from rather complex weighting
strategies to very simple heuristics and habitual behavior (e.g., Gigerenzer &
Goldstein, 1996). In conditions of relatively low involvement, people often use
simple heuristics to save cognitive effort, which generally results in satisfactory
outcomes. However, as a consequence, better or more optimal alternatives may be
overlooked. Besides individual decision strategies, people may also employ social
decision strategies. Here, people may ask other people for assistance and informa-
tion, or consider others’ behavior as a good example, thus addressing informative
as well as normative strategies (see Cialdini & Goldstein, 2004, for an overview).
Social processes play an important role in the diffusion of new behavior and
practices (Rogers, 1995). Network effects, describing how norms and information
spread through a population, may play a pivotal role here, as the connectivity and
structure of a social network have considerable effects on the degree and speed with
which social information travels (e.g., Delre, Jager, & Janssen, 2006). Decision
strategies are related to learning processes, as previously learned information and
behavior will translate into knowledge and attitudes that affect current decision
making.
Obviously, the output of an agent-based simulation depends on the input2; the
formalizations and parameter settings determine the final outcomes of the model.
However, the essential point is that the output—in terms of simulated behavior—
has often not been hypothesized beforehand. Even in relatively simple situations,
for example, where four factors are interacting, we are simply not capable to pre-
dict the effect. Computer simulation models, however, may provide us with the
exact effects. In more complex situations the researcher may observe that behav-
ioral patterns emerge that have not been programmed in the model, which means
that a macrolevel phenomenon grows from the simulation. Such simulations reveal
how collective outcomes emerge from behavioral determinants and processes at
2Also, in experiments and surveys the output is largely determined by the input, as the experimental
conditions restrict the subject’s behavioral freedom and the possible answers to a questionnaire.
Understanding and Managing Environmental Resource Use 101
the microlevel. For example, Delre et al. (2006) were capable of replicating the
diffusion curve of Rogers’ innovation diffusion theory, including the distinction
between the social susceptibilities of innovators, early adopters, majority, and lag-
gards. This is a typical example of the emergence of a macrolevel phenomenon
that originates from microlevel dynamics. Hence, whereas the argument that “you
get out what you put in” is valid, agent-based simulation can contribute to explain-
ing how collective outputs emerge from individual determinants and processes as
formalized in the model.
In the next section, we present a number of studies aimed at exploring be-
havioral processes underlying environmental problems. These studies illustrate
the value of agent-based simulation in revealing how behavioral determinants and
processes generate collective outcomes. Following that, some simulations will be
presented that more explicitly address the issue of policy making aimed at changing
behavior.
Simulating Behavioral Processes in Environmental Management
In this section, we briefly discuss three simulation studies aimed at identi-
fying effects of dynamic behavioral processes, such as the social contagion of
environmental over-harvesting. These studies have been selected because they
employ agent rules that are grounded in behavioral theory, for example, about
social decision making, and they address issues that are relevant in the context of
studying behavior–environment interactions, for example, over-harvesting. These
studies illustrate: (1) How agent-based simulation contributes to the explanation
of well-known empirical phenomena, (2) How agent-based simulation allows for
replicating multiple experimental results by allowing for several factors to interact,
and (3) How agent-based simulation may be integrated with models of environ-
mental systems.
How Uncertainty Stimulates Over-Harvesting
Introduction. A large number of laboratory experiments has been conducted
to explore why people tend to over-harvest from a collective resource under con-
ditions of uncertainty (e.g., Messick, Allison, & Samuelson, 1988; Wit & Wilke,
1998). Explanations for this effect relate to overestimation of the resource size
by participants and an outcome-desirability bias tending to overvalue the resource
size ratings. Jager, Janssen, and Vlek (2002) hypothesized that over-harvesting due
to the desire for higher short-term outcomes would be more socially contagious
and would more quickly result in the development of a consumption habit. To test
this idea, they conducted a simulation study to explore the dynamics of the way
people interact in a common-pool situation, trying to get a deeper understanding
of the role social contagion may play in such over-harvesting.
102 Jager and Mosler
The model. Agents were equipped with a need for subsistence, which could
be satisfied by fishing for food, and a need for leisure, which could be satisfied
by not-fishing. The trade-off between the subsistence and leisure need prevented
agents from simply maximizing one need (e.g. subsistence), which would be fairly
unrealistic. Agents’ satisfaction was formalized as a weighted sum of the subsis-
tence and leisure needs, which, for matters of model simplicity, were assumed to
be equally important. Agents experienced uncertainty: the larger the difference
between expected and actual catches of fish, the higher the experienced uncer-
tainty. Expected catch was calculated using data on own previous catch. Agents
could use four different strategies in deciding how much to fish, depending on their
satisfaction and their uncertainty. These four strategies are a simplified formaliza-
tion of a more sophisticated classification of decision strategies on the dimensions
of cognitive effort and social orientation (see e.g., Jager, 2000; Vlek, 2000). Un-
satisfied and certain agents were assumed to deliberate, that is, to consider the
consequences of all possible decisions given a fixed time horizon in order to max-
imize need satisfaction. Unsatisfied and uncertain agents were assumed to engage
in social comparison. This implied comparison of own previous behavior with the
previous behavior of agents having roughly similar fishing abilities, and selecting
the behavior yielding a maximum expected level of need satisfaction. Satisfied
and uncertain agents would simply imitate the behavior of other similar agents,
thereby avoiding the cognitive effort of determining the maximum outcome as
in case of social comparison. Finally, satisfied and certain agents would simply
repeat their previous behavior, because this was satisfactory and the environment
seems to be stable. When agents engaged in reasoned behavior (deliberation or
social comparison), they would update information in their memory to store in-
formation on (fishing) abilities, behavioral opportunities, and characteristics (e.g.,
satisfaction) of other agents. When they engaged in automated behavior (imitation
or repetition), they used their memory without updating the information. Thus the
outcomes of previous behavior determined the decision-making process the agent
engaged in.
Results. The simulation experiments revealed some remarkably simple dy-
namics that may be largely responsible for the over-harvesting effects observed. In
a first simulation experiment (one deterministic run with two agents) it was demon-
strated how uncertainty stimulates an imitation effect promoting over-harvesting.
Whereas imitation (and other decision rules) was formalized in the model, it
was not clear beforehand that this rule would lead to over-harvesting behavior.
Over-harvesting appeared to be “socially contagious” in the sense that imitating
over-harvesting behavior yielded more often satisfactory outcomes—in the short
run—which stimulated repetition and hence often resulted in the formation of
a “bad habit.” When under-harvesting behavior was imitated, however, the lower
outcomes generally caused agents to become dissatisfied, causing them to increase
harvesting.
Understanding and Managing Environmental Resource Use 103
Next, experiments were performed with a stochastic3resource-growth func-
tion, where the growth of the fish stock varied on every time-step, which had an
effect on the expectations of agents. Here, 10 different settings for agents’ need
satisfaction and 10 settings for uncertainty tolerance were used, thus systematically
exploring populations of agents having different tendencies to engage in certain
decision strategies—“easily satisfied” and “highly uncertain” conditions resulting
in a tendency to imitate. Conducting 10 experiments per condition resulted in a to-
tal of 1,000 simulation runs. Again, an “optimism effect” was found, showing that
agents could get stuck in an over-harvesting habit. During a negative fluctuation
of resource growth—when the renewal of the fish stock was temporarily at a low
level—agents decided to harvest less to safeguard future outcomes. Consequently,
agents experienced a lower needs satisfaction, causing them to deliberate, thus
keeping track of developments in the fish stock. If a positive fluctuation of the
resource growth appeared, agents’ expectations concerning the future fish stock
rose, to which they responded with increased harvesting. As a consequence, their
satisfaction increased, and they started to repeat their own behavior. Because during
repetition the agents only considered their short-term outcomes, negative fluctua-
tions of the fish stock growth were not observed and thus did not influence their
fishing behavior. Hence, the agents habitually continued fishing to the moment
that the fish stock was depleted to a level where their satisfaction dropped and the
agents switched back to deliberation, only too late to restore the fish stock. Finally,
an “adaptation effect” was identified, indicating that the more agents engage in
social processing, the less likely it is that a new—preferably resource-protecting—
behavior is being discovered.
Implications for understanding behavioral processes. The simulation exper-
iment revealed relatively simple behavioral dynamics that could only become
manifest in a more complex setting. The consequences of switching between de-
cision strategies may have a significant impact on resource use, but due to the
complexities in the real world such effects are hard to identify using standard lab-
oratory experiments. The difference between the two explanations as suggested
by laboratory research—overestimation of resource size and outcome-desirability
bias—and the effects found in the simulation experiments is that the latter explains
the relation between uncertainty and harvesting in terms of behavioral processes.
Rather than identifying fixed factors that often hardly provide points of application
for policy strategies because they are less manageable—for example, group size,
the simulation identified behavioral processes that—if validated empirically—
could provide a starting point for the development of policy measures aimed at
changing these processes. For example, the results suggest that at times of “good
environmental news” people may be more likely to develop an “environmentally
3Here no stochastic variables are incorporated, implying that equal starting settings generate equal
results.
104 Jager and Mosler
bad habit,” which would necessitate a precise timing of promotional strategies
about “proper behavior.” When the basic dynamics of such processes have been
identified using simulation studies, empirical follow-up studies could be focused
on identifying such effects in the field, and on testing actual policy measures.
Because the complexity of the real-life situation makes experimentation very
difficult, empirical studies may focus on qualitative descriptions of (longitudinal)
cases. Additionally, laboratory experiments may focus on switches in decision
strategies so as to experimentally validate the processes as identified by simu-
lation. Such case studies and experiments could provide an empirical basis for
the suggestions derived from this simulation study that social influence strategies
(e.g., viral marketing targeting specific small groups of connected people), other
strategies directed at habitual behavior, and the provision of “good” behavioral
examples are important to address the processes behind over-harvesting behavior.
A Simulation of Decision Making for the Sustainable
Use of Environmental Resources
Introduction. When people decide how much to use of an environmental re-
source, their decisions will be influenced by many factors such as own goals,
the size of the resource, and the assessment of the way other people will use the
resource. These factors interact and influence decisions simultaneously. As al-
ready mentioned, many laboratory investigations have been conducted in which
these factors were analyzed in an isolated way or included a very limited num-
ber of interactions. This is an appropriate first step, but for understanding the
real decision processes it is necessary to use a method that takes some of the
most relevant factors and processes into account. Therefore, Mosler and Brucks
(2003) developed a general dynamic model of resource use by means of computer
simulation.
The model. The integrative element of the agent model is a theoretical concept
called “social-ecological relevance” which is derived from Festinger’s theory of
social comparison processes (1954). In this theory Festinger introduces the dis-
tinction between social and physical reality. Accordingly, the decision making of
an agent can be based upon information from either the physical or the social
environment. The hypothesis is that an individual in an environmental resource
dilemma simultaneously weights the importance of social factors such as others’
behavior, and ecological factors such as the availability of resources. In a next
step, the two weightings are merged into one dimension called social-ecological
relevance. The latter defines the extent to which social versus ecological factors
will affect individual decisions on the use of the resource. Ecological factors in the
model include state of the resource and resource uncertainty. Social factors include
attributions (who is responsible for the state of the resource?), social values (how
do people want to share the resource between themselves and others?), and others’
Understanding and Managing Environmental Resource Use 105
behavior (how are the others behaving?). Together, these factors determine the
individual consumption of the environmental resource. Different consumers (viz.
agent models) weight these factors differently. For example, a competitive person
gives a heavy weight to others’ consumption behavior because (s)he does not want
to be outperformed. At the same time, resources may be abundant which results in
a low weighting of importance of resource size. In such a situation of abundance,
a competitive person will attach much more importance to others’ behavior than
to the state of the resource, and will be likely to over-consume as a consequence
of others consuming a lot as well.
Results. The model was tested by seeking to replicate findings of laboratory
experiments within the commons dilemma paradigm that used real participants.
For these tests, the relevant information on the study’s methods (e.g., number
of participants, nature of experimental manipulations, personality measurements,
number of decisions) was taken from the method section of the original publication.
For example, when the real experiment used groups of five participants who made
12 consecutive consumption decisions from a common pool, then the simulation
comprised five agents who interacted for 12 runs. To date, tests have shown that
the model replicates various experimental findings quite successfully (see Mosler
& Brucks, 2003).
Having checked the validity of the agent model by making the aforementioned
comparisons, it is now possible to let variables work together whose interaction
has not yet been investigated in real experiments. For example, the joint effect of
causal attributions, ecological uncertainty, and resource availability has never been
studied in the laboratory with real participants. In this case, the simulation revealed
that when uncertainty rises, people attributing the state of the resource to ecological
factors (e.g., to a natural fluctuation of resources) increase their consumption to a
greater degree than people attributing the resource’s state to the group’s behavior.
This is because the latter people attach more importance to others’ behavior and
less importance to uncertainty about the resource. Consequently, they are less
susceptible to the negative influence of resource uncertainty and also consume
less. Ideally, these simulation results may serve as basis for hypotheses that can
be tested in laboratory experiments in turn.
Implications for understanding behavioral processes. With the simulation
model, specific interactions of variables so far examined separately—such as the
one reported above—could be demonstrated which means that at least parts of
the complex processes for decision making in environmental resource dilemmas
could be captured. As a logical next step in research, these interactions have to
be replicated with robust empirical methods in the laboratory or in the field. If
this succeeds, it implies a big step toward more realistic investigation of the be-
havior of resource-using individuals. As a consequence, policy measures can be
addressed more adequately and will be more effective in changing people’s use of
an environmental resource toward its sustainability.
106 Jager and Mosler
Transitions in a Virtual Society
Introduction. Whereas the previous simulation studies used a resource with a
very simple growth function, many social—economic–environmental systems are
much more complex. Whereas more complex models of environmental systems
have been developed, human behavior is usually represented here in simple ag-
gregate functions, not capturing the dynamics of human behavior. In a series of
experiments, Jager, Janssen, De Vries, De Greef, and Vlek (2000) explored how
psychologically more realistic agents would manage an “artificial world.” The
basic question was if such formalizations of behavior would result in different
human–environment interactions compared to standard economically optimizing
agents.
The model. An artificial world called Lakeland was constructed based on a
simple integrated model comprising two natural resources: a fish stock in a lake and
a nearby gold mine. Mining would pollute the lake and have a negative effect on the
fish stock. The model also included an economical submodel, allowing the agents
to sell fish and gold, and to buy food and status-enhancing products. The 16 agents
were equipped with four needs: (1) subsistence, to be satisfied with fish or gold, (2)
identity, expressed as the relative amount of money an agent owns in comparison
to a subset of agents having about similar abilities, (3) leisure, referring to the share
of the time spent on leisure, and (4) freedom, associated with the total amount of
money owned. The decision strategies the agents could employ were deliberation,
social comparison, imitation, and repetition, formalized in the same manner as dis-
cussed in the section on how uncertainty stimulates overharvesting. Allowing the
agents to shift between these four strategies due to changes in satisfaction and un-
certainty resulted in formalization of the behaviorally rich “homo psychologicus,”
whereas limiting its decision strategy purely to (utility-maximizing) deliberation
formalized the “homo economicus” (see Jager et al., 2000).
Results. For both the homo psychologicus and the homo economicus condi-
tions, 100 simulation runs were performed. Remarkable differences were found
concerning the behavioral dynamics of the homo economicus versus the homo psy-
chologicus. The transition from a fishing to a mining society was more complete
for the psychologically realistic agents. Due to processes of imitation and social
comparison, many more agents started working in the mine than was optimal from
an economical point of view. This instigated extra pollution of the lake, which led
to a decreasing fish stock. As a consequence, the relative harvest from the mine
was larger, thereby propagating the completion of the transition. These results
confirmed the idea that macrolevel indicators of sustainability, such as pollution
and fish harvest, are strongly and predictably affected by behavioral processes at
the microlevel.
Understanding and Managing Environmental Resource Use 107
Implications for understanding behavioral processes. The simulation experi-
ment described in this section demonstrated that the incorporation of a microlevel
perspective on human behavior within integrated models of the environment yields
a better understanding of the processes involved in environmental degradation. In
particular, the large-scale transition toward mining was explained by processes of
imitation and social comparison. Future studies should establish whether the same
processes are functioning in real-life transitions. One suggestion is that these pro-
cesses have also played a role in the twentieth century transition toward a car-based
transportation system. Whereas policy strategies were not tested in this model, this
simulation model revealed the relevance of behavioral processes as a driver of a
large-scale transition. Policy makers wishing to support an envisaged transition—
for example, the hydrogen transition—could benefit from using such processes,
i.e., by enhancing the social visibility of the desired behavior, and/or by creating
uncertainty concerning the (future) availability of undesired behavioral options, to
mention just two examples.
Exploring the Effects of Environmental Policy Strategies
for Sustainable Management of Environmental Resources
A challenging application of simulation models lies in the experimental eval-
uation of policy measures. Because simulation models allow for capturing a part
of real-world complexity, it is possible to experiment with policy making in a
dynamic context. This implies that the nonlinear effects of policy measures can
also be studied, thus providing a perspective on the robustness versus volatility of
policy effects. This is important, because the effects of policies may be less clear
in a dynamical context, which would require a close monitoring of effects and
adequate policy responses to unforeseen (negative) developments. In using simu-
lation models to explore the effects of policy measures, a clear link between the
empirical context and the simulation contributes to the applicability of simulation
results in practical policy settings. In this section, we discuss three studies in which
effects of policy measures are examined using agent-based simulation tools.
Determining Policy Effectiveness of a Car-Speeding Campaign
Introduction. Car use is one of the most detrimental activities for environ-
mental sustainability. One possible policy measure could aim at reducing driving
speeds, which decreases noise and pollution, and increases traffic safety. In the area
of car use and driving behavior, reliable and effective behavior-change measures
can hardly be found (see also G¨arling & Schuitema, this issue). Policy measures
sometimes have effects on behavioral change and sometimes not and we do not
know why. If we would understand how these measures work at the individual level,
we could develop more reliable, more specific, and more effective measures.
108 Jager and Mosler
In a campaign promoting slower driving speeds that took place in M¨unsingen,
a Swiss municipality, Mosler, Gutscher, and Artho (2001) designed measures that
confronted people with inner (cognitive) contradictions, which resulted in a re-
markable reduction of average driving speeds. The campaign was evaluated and
used to test the validity of a simulation model.
The model. A multi-agent simulation model was developed (Mosler, 2002),
based on the theory of cognitive dissonance (Festinger, 1957). This theory specifies
different ways in which people deal with inner inconsistency, for example, between
their behavior and their attitudes. In the model, these processes occurring in the
agent are described as follows. If agents experience a discrepancy, or dissonance,
between attitude and behavior, they will attempt to reduce dissonance by either
changing attitude or changing behavior. Agents will change the factor showing
the least resistance to change, which is determined by a comparison of people’s
attitudes and behavior with the values they hold. Personal values make up a person’s
general, basic orientation. The smaller the difference between values and attitude
or behavior, the greater the resistance to change. The extent of attitude or behavioral
change is determined by the difference between attitude and behavior, or in other
words, by the magnitude of the actual dissonance. This change in attitude or
behavior is then weighted in terms of self-responsibility, in such a way that if a
person has no feeling of self-responsibility for the behavior (e.g., (s)he is forced
to show the behavior) there will be no change, as no dissonance exists.
Results. In the actual campaign that took place, a number of different inter-
ventions were applied, designed to encourage drivers in M¨unsingen to slow down
(see Mosler et al., 2001). Three types of these measures are pertinent to cogni-
tive dissonance theory. First, personal commitment in writing to drive slowly was
designed to engender dissonance, in that an inner inconsistency would arise be-
tween the behavior people pledged to perform and their previous behaviors and
attitudes. Second, prompts were designed to make existing inner inconsistency
salient, thereby setting off the dissonance process within the person. As prompts,
120 colored flags showing the campaign logo and the “Voluntary Slow-Down in
M¨unsingen: 30 km/h” slogan were hung throughout the town, and key chains and
bumper stickers showing the campaign logo were distributed to serve as daily re-
minders. Third, feedback measures to make inner inconsistency salient were also
used. During the entire campaign, three mobile units measuring driving speeds
were moved from place to place within the municipality. Clocked speeds posted
on the electronic boards gave drivers feedback on their actual speeds and served
to remind them about the campaign.
To assess the effects of the actual campaign, a questionnaire survey was con-
ducted before and after the campaign using the same sample. The questionnaire
was used to determine the values of the relevant variables for each respondent and
to discover people’s reactions to the measures implemented. The campaign ran for
25 weeks.
Understanding and Managing Environmental Resource Use 109
Using assumptions about the way the policy measures would work and the
collected data, it was possible to successfully model the dissonance processes
occurring in people through the course of the campaign. To run the model, the
data of the before-survey were used as starting values and they were processed for
25 runs. In order to test how adequately the model reflected the actual processes
in the real persons, the measured values from the after-survey and the simulated
values for attitude and behavior after 25 runs were compared to find out if they
matched. Several parameters of the simulation first had to be determined. Therefore
a number of simulations with different parameter constellations was run until an
optimal prediction of the end values of attitude and behavior was reached for one-
half of the sample (total N =134). The same parameter values were then used
to validate the simulation with the other half of the sample. For the first half of
the sample (with the optimized end values) the attitude change of 87% and the
behavior change of 65% of the cases could be predicted (afterward) correctly. For
the second half of the sample the attitude change of 84% and the behavior change
of 61% of the cases could be predicted correctly.
Conclusions about environmental policy making. The results reveal that with
the simulation it was possible to replicate the outcomes of the dissonance reduction
processes triggered by policy measures, in many different people over a reasonable
period of time. The simulation improves our understanding of the intra-individual
processes that take place in a campaign designed to produce attitude and behavior
changes through addressing inner contradictions. It is possible to use simulation
with the already determined parameters, into which precampaign data have been
fed, as a support tool for determining in advance which measures would lead
to optimal changes in attitudes and behavior in a given population. For future
campaigns, therefore, it should be possible on the basis of preliminary surveys to
determine in advance the policy measures that would be most effective.
Environmental Technologies for Households: Shared Solar Power Plants
Introduction. Two billion people in the world are living without electricity
and it will be a big challenge to provide them with energy in an environmentally
sustainable way. Shared solar power systems could provide a clean, sustainable,
and autonomous form of energy. But so far, experience has shown that solar elec-
trification can lead to technical problems (due to wrong sizing of the system, for
example) as well as social conflicts (due to power limitations).
The model. To study the dynamics of these two sources of trouble, a model
has been developed that consists of a technical submodel (corresponding to a
shared solar power plant) and a social submodel (corresponding to the user com-
munity; see Brucks & Mosler, 2002). The submodel of the shared solar power
system consists of all components usually found in a real solar power system. The
photovoltaic modules produce electric current according to information about the
110 Jager and Mosler
meteorological situation at a given place at a given time. According to actual elec-
trical current from the modules, actual voltage of the batteries, and the complete
load of the user community, the charge controller determines the input and output
of the batteries which in turn feed back the actual state of charge to any single
household of the community.
The core of the community model consists of agents that represent single
households. Any number of agents can be implemented, and all functions according
to a model of human resource use (see the subsection on decision making above,
see also Mosler & Brucks, 2003). Agents get two crucial pieces of information
that influence their decision making: the actual state of charge of the batteries (the
resource size) and the average use of other agents (use of others). They weight and
process this information according to individual attributes (the social values of the
household) and their perception of the situation (attribution: what to blame for the
actual state of charge, the technical system or the users). Based on the decision
process, agents decide how to change their behavior, that is, increase or decrease
their electricity consumption to a certain degree.
Results. A case study was conducted in Santa Maria de Loreto, a rural com-
munity of 50 households on Cuba. Since a couple of years, this remote village
is equipped with a shared solar power system that usually runs very well and
provides all households with enough energy for lighting, radio and TV, and other
small appliances.
With a simulation run of the complete energy demand of all houses (i.e., 50
“household” agents) during a period of 6 days (Jan. 29–Feb. 03, 2002) as measured
by manually reading hourly each house’s energy meter, it could be shown that the
simulation reproduced the behavior of the whole system for 144 hours (=6 days)
quite well. Knowing that the simulation model was able to sufficiently reproduce
the actual behavior of the villagers, it was then adapted to the following real-life
scenario.
Recently, one component of the solar power system in Santa Maria de Loreto
was damaged and caused a loss of power of about 50%. The reaction of the mayor
was to divide the village into two halves who took turns using the remaining energy.
In other words, a household was allowed to switch on appliances only any other
day. In a cooperative community like Santa Maria de Loreto, probably a solution
could be found that would be more convenient for the people than repeatedly
restricting the use of each household to zero for a whole day. Through simulating
the entire village again, we tried to predict how people would behave in a power
shortage if they knew about the state of the batteries and the consumption behavior
of their neighbors, which is actually not the case. The “household” agents were then
reacting on the feedback about the state of charge and interacting with each other
via a social network. Once again, a simulation run over 144 hours (=6 days) with
50 household agents was conducted where the simulated photovoltaic system had
only half of its usual power output, but no rules were enforced and the people had
Understanding and Managing Environmental Resource Use 111
complete feedback about the state of charge and the behavior of others. In terms of
the system’s performance (e.g., batteries’ state of charge), the simulation showed
that the village performed the same or better when no consumption rules were
enforced. Note that in this case people could use energy every day and voluntarily
committed themselves to reducing consumption.
Conclusions about environmental policy making. Besides the application
shown in the previous sections that was concerned with the management of an
existing solar power system, the simulation tool serves as a help for planning solar
power systems in nonelectrified remote villages. Once enough data have been col-
lected about a candidate village, the size and design of the solar power system can
be planned and tested on the computer. At the same time a social arrangement can
be found on the way the system should be managed by the community. In simula-
tion trials, the size or design of the solar power plant can be modified and social
measures applied. The goal is to simulate optimal management of the plant by a
given community and, from that, to derive important considerations for the instal-
lation and operation of shared solar power plants in such communities. In general,
the example demonstrates that with the help of the simulation method it will be
possible to advise politicians for designing and building environmental–technical
systems which can be used in a sustainable way by the people.
Diffusion of Green Products
Introduction. In promoting the diffusion of green products, one strategy is
taxing the nongreen products. Imposing a tax on nongreen products can be done
abruptly, thus causing a “tax-shock” in the system, or it can be done gradually.
Janssen and Jager (2002) presented a model-based analysis of the introduction
of green products having low environmental impacts, testing how different tax
regimes would affect the speed and the degree to which green products would gain
market share. Because markets may differ concerning the speed of new product
development, two prototypical markets were compared, a stable market and a
market with continuous product development. Because they wanted to explore
how the effects of tax regimes are related to assumptions on behavioral processes,
they compared the ways in which the tax affected economically maximizing agents
versus “behaviorally rich” agents.
The model. Both consumers and firms were simulated as populations of agents
who differ in their behavioral characteristics. For the formalization of the con-
sumers the methodology as described in the section on how uncertainty stimu-
lates overharvesting. Agents had two needs, namely a social and a personal need.
Personal need expressed the personal preferences or taste of an agent for cer-
tain products. The more a product matched the personal preference position (on
an abstract preference dimension ranging from 0 to 1), the higher the personal
need satisfaction. Social need was formalized as agents having a preference for
112 Jager and Mosler
consuming the same products as their neighbors. The more neighbors consumed
the same product, the higher the satisfaction of the social need. Prices influenced
the relative satisfaction rate of using a product, causing a higher price to have a
negative effect on need satisfaction. Overall need satisfaction was thus a weighted
function of both the personal and social need, taking into account the price of a
product.
The decision strategies agents could employ were deliberation, social compar-
ison, imitation, and repetition, formalized in the same manner as discussed earlier.
Two experimental conditions were being used, the “homo psychologicus (HP)”
condition where the agents were allowed to shift between these four strategies,
and the “homo economicus (HE)” condition where agents engaged exclusively in
deliberation.
Concerning the producers, also two conditions were created. First, a stable
market was formalized as offering a fixed number of products, namely, five “green”
and five “nongreen” ones. A continuous product-development market was formal-
ized by allowing producers to replace at any moment an existing product with a
new product (green or nongreen). If an existing product dropped below a market
share of 10%, producers could launch a new product. Initially, all firms produced
nongreen products, allowing for studying how green products might penetrate
such a market. Taxing was formalized either as “fast,” introducing a full tax at t=
50, or as “slow,” implicating a gradual increase starting at t=25 and reaching a
maximum level at t=75.
Results. Simulation experiments illustrated the influence of different behav-
ioral characteristics on the success of switching to green consumption. It was found
that in a stable market the tax regime makes the largest difference. Although the
HP responded a bit slower than the HE on a tax increase, the major finding was
that a slow increase in tax resulted in a slow increase in market shares of green
products, whereas a fast tax resulted in a fast increase of market share. In the con-
tinuous product development condition, however, it was not the tax regime but the
behavioral characteristics that made the main difference. Here, it was found that
the HE responds much slower to a change in tax regime than the HP. This coun-
terintuitive result can be explained as follows. Due to the optimizing behavior of
the HE, the producers were stimulated to develop products matching the personal
preference of agents. In the HP condition, however, agents could develop habitual
behavior, and thus were more likely to continue consuming an existing product, at
the neglect of more attractive products. In the HE condition, product development
thus was more effective in satisfying the personal needs of agents. Imposing a tax
regime consequently had a smaller effect on the HE, as they consumed products
fitting their personal need more than in the HP condition. Because the HP derived a
lower level of need satisfaction from their existing product consumption, they were
more susceptible to change toward green products when the tax regime changed.
Understanding and Managing Environmental Resource Use 113
Here, it hardly mattered whether the introduction of the tax was fast or slow.
Conclusions about environmental policy making. The results from this sim-
ulation experiment indicate that assumptions concerning the decision making of
agents and related market dynamics are critical in understanding the effectiveness
of policy measures. Moreover, this experiment provided a perspective on the way
policy measures, in this case tax policy, can be tested in a multi-agent simulation.
Obviously the efficacy of policy measures depends on many more factors than
can be captured in a relatively simple multi-agent simulation as discussed here.
However, these experiments reveal some of the dynamics that may determine the
effectiveness of policy measures in the real world, and thus may contribute both
to the understanding of the success of real-world policy measures, and to the
development of more effective policy measures.
Implications for Research and Policy Making
on Environmental Problems
The studies presented illustrate how behavioral determinants and processes
in environmental problem situations may be formalized in agent-simulation rules,
and demonstrate how the operation of these rules leads to aggregate effects. Agent-
based simulation thus offers a tool to explain the behavioral determinants and pro-
cesses responsible for environmental events that happened in the past. Moreover,
this article illustrated how multi-agent simulation can be used to test environmental
polices in a context determined by complexities at the individual, social, and envi-
ronmental level. This is an important advantage of multi-agent simulation, as most
environmental policy making takes place within complex and dynamic environ-
ments, where many different (groups of) stakeholders are interacting. Moreover,
policy measures generate outcomes on the personal, social, and environmental
level, which are being valued differently by different (groups of) people. Due to
such complexities, the effects of policy measures are often difficult to forecast,
since they become manifest only after a long time.
Obviously, in solving environmental resource dilemmas, policy measures have
to address human behavior. However, the course of developments in social sys-
tems is often capricious, and many autonomous developments may take place
beyond the—limited—control of policy makers. Simulations may contribute to
understanding the behavioral dynamics underlying these developments. Rather
than forecasting developments and effects of policies, simulations contribute to
the understanding of the dynamics, and the identification of possible scenarios,
while indicating possible ways to encourage or steer certain developments. Fore-
casting turbulences in a certain behavioral domain may reveal the necessity of
dynamic policy making, involving a close social monitoring of, and fast policy
response to behavioral changes. For example, simulations may indicate that the
114 Jager and Mosler
effect of, for example, a promotional campaign in a given context, is unstable due
to strong autonomous social processes in the field, but that in case of an emerging
failure a timely support, for example, by public figures endorsing the campaign,
may prevent such failure. In practice, this would mean that such campaigns are
started with a supportive strategy standing by, the latter being launched as soon as
field monitoring indicates a potential failure of the campaign.
Simulations may address relatively limited areas of policy making, of which
the car speeding campaign is a nice example. A major challenge, however, is to
apply simulations in exploring the dynamics of large-scale societal transitions. One
example is the forecasted hydrogen revolution (e.g., Dunn, 2002), which would
imply a major transition of society. In modeling such a complex and large-scale
development, one is confronted with many conflicting perspectives from various
stakeholders, such as policy makers, industry, consumers, NGOs, and scientists.
Here, the development of the simulation itself would require intensive discussions
with stakeholders, serving the debate on the crucial issues in understanding the
process. This development process is also called the “agenda-setting function” of
simulations, as during this stage decisions are being made concerning what factors
and processes are relevant for the issue to be simulated. Next, a simulation could
serve as a training tool for various stakeholders in trying to explore policy strategies
that would be effective and acceptable at the same time.
Difficulties with agent-based simulation relate to formalization and validation.
Concerning formalization the challenge resides in simplifying the often complex
theories of social science and the complex reality into simple sets of rules. Ac-
knowledging the sensitivity of models for small changes in this formalization,
this implies that for practical applications it is important to explore how differ-
ent formalizations—and parameterizations—affect the results obtained with the
model. The more robust the outcomes are, the greater the confidence one may have
that the simulation model captures relevant behavioral dynamics. For example, the
social contagion of over-harvesting was a result found over a range of different
parameter settings, which indicates the robustness of this effect.
Concerning validation the difficulty resides in the comparison of simulation
results to empirical data. If we are dealing with complex behavioral domains, we
have to realize that the current situation is just a single manifestation of the wide
collection of possible outcomes. Obviously, if developments had taken another
course, the empirical data describing the current situation could have been rather
different. This implies that calibration of a computer simulation model against an
empirical data set describing a single event at the macrolevel is a risky business,
unless the macroeffect can be observed in many conditions and represents a kind
of stylized fact. The latter would imply that data are available on a larger set of
comparable macroevents, and that all these data show a (qualitatively) comparable
trajectory of developments.
In sum, agent-based simulation offers a rich methodology that is expected to
contribute significantly to the study of behavior–environment interactions, and
Understanding and Managing Environmental Resource Use 115
to provide a valuable tool for exploring the effectiveness of policy measures
in complex environments. To apply such models effectively in the context of
environmental management, it is crucial that agent rules are grounded in behav-
ioral theory relevant for the issue at stake. This allows for an empirically valid
simulation of the dominant behavioral dynamics, and it provides points of ap-
plication for policy measures addressing these processes. Simulation model runs
have to be capable of replicating empirically observed phenomena, and hence it
is important to collect a large number of cases as reference scenarios, showing a
range of possible developments and behavioral dynamics in real-world systems.
Policy applications of agent-based simulations will both allow for and benefit from
aligning behavioral theory and empirical case studies in modeling exercises.
References
Brucks, W., & Mosler, H.-J. (2002). Applying agents with basic needs and social psychological charac-
teristics for the design of shared solar power plants. In C. Urban (Ed.), 3rd Workshop on Agent
Based Simulation (pp. 99–104). Erlangen: SCS-European Publishing House.
Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: Compliance and conformity. Annual Review
of Psychology, 55, 591–621.
Delre, S. A., Jager, W., & Janssen, M. (2006). Diffusion dynamics in small-world networks with
heterogeneous consumers. Computational and Mathematical Organization Theory, 4, 5–22.
Dunn, S. (2002). Hydrogen futures: Toward a sustainable energy system. International Journal of
Hydrogen Energy, 27(3), 235–264.
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140.
Festinger, L. (1957). A theory of cognitive dissonance. Evanston, IL: Row Peterson.
G¨arling, T., & Schuitema, G. (this issue). Travel demand management targeting reduced private car
use: Effectiveness, public acceptability and political feasibility. Journal of Social Issues.
Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded
rationality. Psychological Review, 103, 650–669.
Hardin, R. G. (1968). The tragedy of the commons. Science, 162, 1243–1248.
Janssen, M. A., & Jager, W. (2002). Stimulating diffusion of green products. Journal of Evolutionary
Economics, 12, 283–306.
Jager, W. (2000). Modelling consumer behaviour. Doctoral thesis. University of Groningen (NL),
Department of Behavioral and Social Sciences.
Jager, W., Janssen, M. A., De Vries, H. J. M., De Greef, J., & Vlek, C. A. J. (2000). Behaviour in
commons dilemmas: Homo Economicus and Homo Psychologicus in an ecological-economic
model. Ecological Economics, 35, 357–380.
Jager, W., & Janssen, M. A. (2003). The need for and development of behaviourally realistic agents. In
J. S. Sichman, F. Bousquet, & P. Davidson (Eds.), Multi-agent based simulation II (pp. 36–49).
Berlin: Springer.
Jager, W., Janssen, M. A., & Vlek, C. A. J. (2002). How uncertainty stimulates overharvesting in
a resource dilemma: Three process explanations. Journal of Environmental Psychology, 22,
247–263.
Lindenberg, S., & Steg, L. (this issue). Normative, gainand hedonic goal-frames guiding environmental
behavior. Journal of Social Issues.
Maslow, A. H. (1954). Motivation and personality. New York, USA: Harper and Row.
Max-Neef, M. (1992). Development and human needs. In P. Ekins & M. Max-Neef (Eds.), Real-life
economics: Understanding wealth creation (pp. 197–213). London, New York: Routledge.
Messick, D. M., Allison, S. T., & Samuelson, C. D. (1988). Framing and communication effects on
group members responses to environmental and social uncertainty. In S. Maital (Ed.), Applied
Behavioural Economics, 2 (pp. 677–700). Brighton, UK: Wheatsheaf.
116 Jager and Mosler
Mosler, H.-J. (2002). Agent-based simulation of an environmental action campaign: Changing people’s
behaviour via their inner contradictions. In A. E. Rizzoli & A. J. Jakeman (Eds.), Integrated
assessment and decision support. Proceedings of the 1st biennial meeting of the International
Environmental Modelling and Software Society (pp. 202–207). Como: IEMSS, 2.
Mosler, H.-J., & Brucks, W. (2003). Integrating resource dilemma findings in a general dynamic model
of cooperative behavior. European Journal of Social Psychology, 33, 119–133.
Mosler, H.-J., Gutscher, H., & Artho, J. (2001). Wie k¨onnen viele Personen f ¨ur eine kommunale
Umweltaktion gewonnen werden? [How to gainmany participants for communal environmental
action?] Umweltpsychologie, 5, 122–140.
Rogers, E. M. (1995). Diffusion of innovations. Fourth edition. London: The Free Press.
Vlek, Ch. (2000). Essential psychology for environmental policy making. International Journal of
Psychology, 35, 153–167.
Wit, A., & Wilke, H. (1998). Public good provision under environmental and social uncertainty.
European Journal of Social Psychology, 28, 249–256.
WANDER JAGER is an associate professor of marketing at the University of
Groningen. He studied social psychology and obtained his PhD in the behavioral
and social sciences, based on a dissertation about the computer modeling of con-
sumer behaviors in situations of common resource use. His present research is
about consumer decision making, innovation diffusion, and market dynamics. In
his work he combines methods of computer simulation and empirical surveys.
HANS JOACHIM MOSLER is an associate professor of social and environmental
psychology at the University of Zurich, Switzerland and head of the research
group “Modelling Social Systems” at EAWAG (Swiss Federal Institute of Aquatic
Science and Technology). His current research is on behavioral change in large
populations using agent-based simulations.
