ArticlePDF Available

An Experimental Analysis of Optimal Renewable Resource Management: The Fishery

Authors:

Abstract and Figures

This paper experimentally studies the extraction decisions of a sole owner in a fishery, the population dynamics of which behave according to the standard deterministic logistic growth model. Four treatments were implemented which differed in the level of information supplied to the subjects. Compared to the theoretic benchmark, the data reveal that efficiency losses increase as the information on population dynamics and stock size deteriorates. Three common patterns of behaviour are identified. The distribution of these patterns is significantly affected by the informational setting.
Content may be subject to copyright.
L
LS
SF
F
R
Re
es
se
ea
ar
rc
ch
h
W
Wo
or
rk
ki
in
ng
g
P
Pa
ap
pe
er
r
S
Se
er
ri
ie
es
s
N
N°
°.
.
0
08
8
-
-
1
10
0
Date: 2008
Title: An Experimental Analysis of Optimal Renewable Resource Management:
The Fishery
Author(s)*: John D Hey, Tibor Neugebauer and Abdolkarim Sadrieh
Abstract: This paper experimentally studies the extraction decisions of a sole-owner in
a fishery, the population dynamics of which behave according to the
standard deterministic logistic growth model. Four treatments were
implemented which differed in the level of information supplied to the
experimental subjects. The theoretical solution was used to evaluate the
behaviour of experimental subjects. The data reveal high efficiency losses
due to the lack of information on population dynamics and stock size.
Efficiency varied between treatments according to the information
conditions.
Keywords: Experimental economics, renewable resources, dynamic decision making,
decisions under risk and uncertainty, misperceptions of feedback
JEL Classification: C91, D81, Q22
*Corresponding Author’s
Address: Tel. : +352 46 66 44 6800; Fax : 352 46 66 44 6835.
E-mail address: tibor.neugebauer@uni.lu
The opinions and results mentioned in this paper do not reflect the position of the Institution.
The LSF Research Working Paper Series is
available online:
http://www.lsf.lu/eng/Research/Working-
Papers/2008
For editorial correspondence, please contact:
caroline.herfroy@uni.lu
University of Luxembourg
Faculty of Law, Economics and Finance
Luxembourg School of Finance
4 Rue Albert Borschette
L-1511 Luxembourg
AN EXPERIMENTAL ANALYSIS OF OPTIMAL RENEWABLE
RESOURCE MANAGEMENT: THE FISHERY
John D Hey, Tibor Neugebauer and Abdolkarim Sadrieh
University of York and LUISS, University of Luxembourg, and University of Magdeburg
Abstract
This paper experimentally studies the extraction decisions of a sole-owner in a fishery, the
population dynamics of which behave according to the standard deterministic logistic growth
model. Four treatments were implemented which differed in the level of information supplied
to the experimental subjects. The theoretical solution was used to evaluate the behaviour of
experimental subjects. The data reveal high efficiency losses due to the lack of information on
population dynamics and stock size. Efficiency varied between treatments according to the
information conditions.
JEL classifications
C91, D81, Q22
Keywords
Experimental economics, renewable resources, dynamic decision making, decisions under
risk and uncertainty, misperceptions of feedback
This paper is part of the EU-TMR Research Network ENDEAR (FMRX-CT98-0238). The authors
acknowledge valuable comments of Kurt Schnier, and seminar participants at Akureyri, Amsterdam, Barcelona,
Castellon, Clausthal, Fullerton, Magdeburg, Kiel, Salamanca, Tilburg, and York.
1
1 Introduction
Renewable resources are those for which the stock can be continually replenished. Fishery
resources are renewable. However, if (through human activities or otherwise) the population
of some species is drawn down beyond a critical threshold, the species can become extinct. A
recent concern has been with the dramatic decline in the populations of several valuable fish
species such as cod, halibut and haddock. Since the seminal article of Gordon (1954),
difficulties in effective management of fisheries have been attributed to the resource’s
peculiarity of being a common property. However, due to the new law of the sea (established
in 1982) more than 90 percent of fish resources are now under the exclusive jurisdiction of
coastal states and can, in principle, be protected. Distant water fishing fleets are restricted to
cooperative arrangements. The coastal state has to establish a total allowable catch
(henceforth TAC) for each fishery resource in its extended economic zone. The TAC is
allocated among the fishermen; the individual quotas are transferable and can be reallocated
through a market for certificates. In theory, established property rights and the individual
transferable quota system warrant optimal resource management. In practice, however, errors
might occur when the decision-maker determines the TAC, because the size, growth and
population dynamics of the fishery are not exactly known. Since fish is only observable upon
landings, the estimated stock size of the species is likely to be different from the actual one.
The importance of errors in measurement or assessments of the stock level for harvesting
policies has been stressed in the recent literature. Moxnes (2003) pointed out that due to the
quota management, applied in practice, the implications of measurement errors for harvesting
policies are much stronger than, for instance, of natural stochastic variations which have been
extensively studied in the theoretical literature (e.g., Clark 1990).
The primary research question addressed in the present study is to which extent the accuracy
of stock surveys and the knowledge of the population dynamics may alter the decisions of the
planner and affect efficiency of resource management. We study the resource extraction
decisions of a sole owner in absence of the commons problem under different information
conditions in a deterministic laboratory setting. Our experimental results indicate that the
knowledge of both the species’ growth model and to a smaller extent the accuracy of the stock
estimate may produce significant efficiency enhancements in the dynamic decision task. In
fact, these effects are not only a consequence of the different information conditions of
experimental treatments but arise also from subjects’ deficiencies in learning non-linear
2
dynamics. The paper thus contributes also to the growing experimental evidence on
misperceptions in dynamic decision making problems (for a survey see Rouwette et al, 2004).
Within the scope of resource extraction decisions other contributions to the literature have
involved problems of optimal-sized fishing fleets (Moxnes, 1998a, 2000; Schnier and
Anderson, 2006), optimal use of reindeer rangelands (Moxnes, 1998b, 2000, 2004) and
optimal harvesting in a multiple species fishery (Brekke and Moxnes 2003). Most of these
studies involve more complex settings than ours and do not allow an exact measurement of
the efficiency losses due to errors in measurement or assessment. An exact measurement is
usually not possible, because the efficient solutions are ex-ante undetermined.
In contrast, we use a single-species model featuring logistic growth for which the optimal
extraction path is uniquely determined. The model is the standard in lecture books and we
adapt it to laboratory conditions. Subjects are introduced to a finite-horizon, neutrally framed,
deterministic decision problem in discrete time and are motivated by salient rewards to
maximize efficiency under risk and ambiguity. We observe that risk regarding the stock size
and ambiguity about the growth function have both an extensive and significant effect on the
efficiency of the individual extraction decisions. Most of our subjects follow one of three
typical extraction patterns. The decisions of about 34% of our subjects seem to aim at control.
They either try to achieve a desired constant stock level or a desired constant harvest. The
behaviour of 45% displays oscillations that suggest pulse fishing that is characterized by
periods of harvest and periods of recovery. Finally, 17% of our subjects extract only minimal
amounts or resource, possibly due to a linear mental model of growth.
While some of these behavioural patterns have been reported in the literature,1 our results
show that the informational setting significantly affects the distribution of the behavioural
patterns. Patterns of control are mainly observed with full information on the stock and the
growth function, but also play a major role in the cases in which only exact stock information
is given. Pulse fishing is especially frequent in the case with noisy stock information. Finally,
behaviour that is in line with a linear growth model is especially frequent in the ambiguous
environment.
1 Oscillations have been reported by a number of authors, e.g. Moxnes (1998a and 1998b). Schnier and Anderson
(2006) report pulse fishing behaviour that is very similar to our observations. Sterman (1994) reports a number
of studies that find a misperception of non-linear growth. We discuss further findings in dynamic non-linear
decision problems in section 3.
3
The paper is organized as follows. Departing from the classical logistic growth model, we
derive the finite-horizon optimal extraction plan in the subsequent (second) section. In the
third section we highlight research issues and present our experimental design. In the fourth
section we report the results of our study and relate them to the received literature. Finally, the
fifth section concludes.
2 Theoretical Considerations
Consider the standard logistic growth function (as depicted in Figure 1), F(xt) = rxt (1 - xt/K),
where xt denotes stock, r > 0 denotes the species’ intrinsic growth factor and K > 0 denotes
the carrying capacity.2 Assume harvesting costs equal zero, normalise the price to one, and let
the discount factor be denoted by
ρ
= 1/(1 + δ), where r > δ.3 The optimal extraction policy
in the finite-horizon management problem can be determined as the solution to the following
program.
Kxyxz
zFzxts
yV
ttt
ttt
T
tt
t
==
+=
=
+
=
0
1
0
;
)(..
max
ρ
(1)
Here, xt denotes the stock before extraction, zt denotes the stock size after extraction and yt
(the control variable) denotes the extraction in period t {1, 2,.., T}. The optimal solution to
this problem can be calculated by means of Bellman’s (1957) maximum principle. Define
Jn(x) as the maximum total value when only n periods remain, and the state variable at the
outset of these n periods is x. Thus, beginning with the last period, the decision-maker faces
the following problem.
}{max)(
0T
T
yyxJ T
ρ
=
(2)
2 This is the maximum viable (long-run) stock size.
3 If the intrinsic growth-factor r is smaller than the interest rate δ, costless harvesting implies an immediate
extinction of the stock. Thus r > δ is a necessary and sufficient condition for an interior solution of the sole-
owner’s maximization problem. In the open-access fishery this condition is insufficient to prevent extinction of
the resource stock, as the equilibrium level of the resource stock is determined by the ratio of harvesting cost to
the price and thus immediate extinction of the stock follows for any positive price at zero cost.
4
The final extraction yT that maximizes the value function in equation (2) is equal to the
maximal feasible yT, which coincides with the stock remaining in period T, xT. Hence, J0(x) =
ρ
T xT , which, according to (1), is a function of the extraction in period T-1, xT = x(yT-1). Given
J0(x), we can calculate the next term of the maximization procedure, J1(x).
))}(({max)( 101
1
1
1
+=
TT
T
yyxJyxJ T
ρ
(3)
From the first order condition follows the optimality equation F’(zT-1) = r(1 - 2 zT-1/K) =
δ
.
Solving this equation, we obtain the end stock size z*T-1 = K/2(1 -
δ
/r), which is constant as it
does not depend on time. Given initial stock size xT-1 , the optimal extraction in period T-1 is
determined by the optimal end stock size, y*T-1 = xT-1 - z*T-1. Thus, J1(x) =
ρ
T-1(xT-1 - z*T-1) +
ρ
T
(F(z*T-1) + z*T-1). Proceeding by backward induction, the following general expression is
determined,
*
1
0
**
1
)()(
))}(({max)(
zzFzx
yxJyxJ
T
n
j
jT
nT
nT
nTnnT
nT
y
nnT
ρρρ
ρ
++=
+=
=
(4)
where z* = (1 -
δ
/r) K/2 and yt = max{xt z*; 0}. Hence, the first term of the maximization
procedure is JT(x) = x0 - z* +
F(z*)
ρ
T-j +
ρ
Tz*. The first extraction is determined by the
initial stock size x0 = K and the optimal end stock size z*, y*0 = x0 - z* = K - z*. 4 Since the end
stock size is constant for all t < T and growth is deterministic, the initial stock size xt is
constant for t > 1 and, consequently, the extraction yt is constant for all periods 1 < t < T. This
result holds for any finite time horizon T < , and also in the infinite horizon management
problem.5 Hence, the extraction plan in the finite-horizon management problem coincides
with the one in the infinite-horizon case (exclusive of the last period when the resource has to
be extinguished) because at the maximum the marginal productivity of the resource after
extraction F’(z*) must equal the interest rate δ.
4 Note the optimal harvest policy is a “most rapid approach” policy, driving the population toward the optimal
level z* as rapidly as possible.
5 See Clark (1990, Ch. 2) for a derivation of a solution to the infinite-horizon problem and a discussion.
5
Figure 1. Logistic growth function for K=1000 and r=1.5
-400
-200
0
200
400
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
x
F(x)
Note: MSY denotes the maximal sustainable yield.
The graph represents the experimental parameterisation.
3 Laboratory fisheries: Design issues and experimental procedures
Design issues
The model of the previous section (as much as any other theoretical model we can handle) is a
vastly simplified representation of the fishery. The perfect description of the population
dynamics and the knowledge of the exact stock-size in every instance of time are only two of
the idealistic assumptions. If we relax these, the solution to the harvesting-problem -as long as
we can find one at all- becomes more involved. Another serious simplification of the model is
the assumption of unbounded rationality which implies that a decision-maker is able to
determine the optimal catch quota within a system of non-linear dynamics. The literature has
shown that subjects experience significant difficulties in non-linear environments (Sterman
(1989a, b, c), Brehmer (1992), Paich and Sterman (1993), Sterman (1994), Diehl and Sterman
(1995), and Moxnes (1998a, b)). As Sterman (1994) pointed out
… human performances in dynamic (complex) systems is poor … even compared to
simple decision rules. … The observed dysfunction in dynamically complex settings
MSY K
6
arises from misperceptions of feedback.6 People are insensitive to non-linearity and
violate basic rules of probability. The robustness of the misperception of feedback and
the poor performance … result from two basic and related deficiencies in our mental
models of complexity. First, our cognitive maps of the causal structure of systems are
vastly simplified compared to the complexity of the systems themselves. Second, we
are unable to infer correctly the dynamics of all but the simplest causal maps.
This paper addresses the efficiency losses that might accrue in fishery management due to the
decision-maker’s shortcoming in dealing with complexity and due to missing information. For
this study, we have designed and conducted experimental treatments which vary two
information conditions involving the knowledge of the species’ growth model and the
accuracy of the stock estimate. The complexity in the task arises through the non-linearity of
the growth function. We measure the efficiency of subjects’ extraction decisions by
comparing the observed extractions to the maximal possible outcome. In fact, the scope of
this study, in which one sole-owner of the fishery decides on the TAC, is limited to the
examination of the deterministic microworld we described in the previous section. Therefore,
many complications authorities actually face when they set the TAC are missing. Though this
environment is overly simplistic it still captures essential ingredients of a fishery resource’s
population dynamics. Given the (relative) easy tractability of this environment, we put up
with the drawbacks. More realistic settings may be studied in the future.
Still, there are at least three features with respect to the experimental implementation of the
dynamic decision task that should be stressed: First, in the literature the fishery management
problem is typically set in the infinite-horizon. As pointed out in the previous section, the
optimal harvesting policy in the theoretical model is the same whether we consider the finite
or the infinite-horizon setting. Since the infinite-horizon cannot be implemented in the
laboratory, we tackle the fishery management problem as a finite-horizon dynamic decision
task.7 Second, the decision-maker’s presumed objective should be to maximise the present
value of the fishery in every instance. In a world without interest and costless harvest, this
objective involves the most rapid approach to the maximum sustainable yield with every
extraction decision, including a rebuilding of an eventually depleted resource as rapidly as
possible. Naturally, the authorities can not know how well their harvesting-decisions
6 Moxnes (1998a) referred to misperceptions of bioeconomics when he reported from a fishery management
experiment.
7 Given the earth does not exist indefinitely this approach does not seem less plausible, either.
7
approach the maximal economic rent. In the experiment, we implement this ignorance by a
lack of information feedback into the decision-maker’s payoff space.8 Clearly, subjects must
be rewarded according to their extractions. However, the exchange rate between the
experimental currency and the subject’s home currency must not be given before the end of
the experiment.9 Finally, there might be an emotional decision-bias of subjects -particularly of
pity- which might be associated with slaughtering of fish.10 In order to guarantee salience of
the incentive structure the experiment must be neutrally framed. In the experiment we ask a
subject to maximize savings, which are identical to the number of extracted units on the
subject’s account. The procedures are detailed in the following subsection.
Experimental Procedures
In the computerized experiment,11 a subject had to decide one hundred times on the TAC, i.e.,
how much to extract from a privately owned resource stock. The extracted units were saved
on the subject’s account and the logistic growth function was applied to the units that
remained after an extraction. The initial stock size coincided with the carrying capacity
x0=K=1000 units, the intrinsic growth parameter was r=1.5, and the discount rate was δ=0.
The experiment involved four treatments which differed in the level of on-screen information.
In Table 1 an overview is given: the letter G denotes growth, the latter S denotes stock and the
letters No indicates no information on growth or stock. Before every extraction, the subject
received a stock signal revealing information about the number of existing resource units.
This signal was accurate in the treatments GS and S – i.e., equal to the resource stock xt – and
noisy in the other two treatments G and No – i.e., the signal was equal to the resource stock
multiplied by a random draw from the uniform distribution over the interval [0.75,1.25] and
rounded to the next integer. In the treatments GS and G, an on-screen facility (in Table 1
referred to as information about the growth function) was provided by means of which a
8 Apesteguia (2005) finds no behavioural differences in a common pool experiment if payoff information is not
revealed.
9 In the laboratory, the experimenter usually sets the exchange rate of Euros for lab euros in expectation of the
subjects’ average performance. In a deterministic setting like ours, the revelation of the exchange rate would thus
be an indication for the bioeconomic optimum which is the target of the planer subject. If the experimenter gives
such an indication to subjects, behavior should be biased in the direction of efficiency. An unbiased incentive
procedure must ensure that market prices are unrelated to the bioeconomic optimum. (In our setting, the market
price of one extracted unit is set equal to one; while subjects know that more is better, they are not informed
about the number of units they must extract for a Euro). According to the incentive compatibility requirement,
Euro payments are, in fact, determined relative to bioeconomic optimum, i.e., relative to efficiency.
10 See Moxnes (1998b) for a discussion.
11 The software was programmed by means of Abbink and Sadrieh’s (1995) RatImage.
8
subject could anticipate the consequences of any possible extraction for the nearest future
before she/he confirmed an extraction.12 Subjects were instructed accordingly.13
Efficiency in the experiment was defined as the quotient of extracted units and 38125, which
was the maximum number of possible extractions unknown to experimental subjects.14
Efficiency was hence a number between zero and one. The payoff a subject received at the
end of the experiment was the product of efficiency and the premium to be paid in a treatment
which was known to subjects.15
Table 1. Experimental treatments
Accurate
stock size
Noisy signala
about stock size
Growth
Function information GS 25 G 35
No growth
Function information S 31 No 30
a) The noisy signal equals the true stock size multiplied by a random number from the interval [.75, 1.25]
If a subject extinguished the resource before having made 100 extraction decisions the
experiment ended instantaneously, regardless of the number of decisions made to that point.
In order to limit erroneous extractions from the stock, subjects were warned if the extracted
number of units exceeded the stock signal. At the other extreme, an extraction decision of
zero units also triggered a warning. In addition, before the last decision (in period 100) the
subject was informed that no further extraction would be possible thereafter. The preceding
12 Before making an extraction decision, the subject was given an on-screen record of 11 possible extractions in
10 percentiles of the signalled stock in the first column. In the second column the corresponding after-extraction
stock sizes were displayed, in the third column the resulting next stock sizes, in the fourth column the growth of
the resource (i.e., the difference between the third and the second column) was displayed, and finally the savings
were recorded in the fifth column. Additionally, the subject could explore the effects of every possible extraction
at any point in time and before making an extraction decision -between nothing and the maximal available
number of units (i.e., in G the maximum extraction was 4/3 * stock signal). The results of any such enquiries
were displayed in a scroll-box appended to the standard record of possible extractions. Finally, if the subject was
satisfied with the consequences of her/his latest inquiry (displayed at the end of the table) she/he confirmed it as
the harvesting-decision by pressing the “extraction button.”
13 Instructions and the computer-screen (for G) are depicted in the Appendix.
14 The maximum is easily calculated by applying the results from Section 2: First, extracting 500 units to reach
the steady state (the maximal sustainable yield since the interest rate is zero); then, extracting 375 units (equal to
the growth in the steady state); and finally, extinguishing the resource in the last decision.
15 The premium (i.e., the maximal payoff) in GS was €15, in G and S €17.50, and in C €20 (1€ 1$). The
average payoff was €11; the experiment took about an hour.
9
extractions and the on-screen information, including the stock signal before and after
extraction as well as the resulting savings, were recorded in a history-window that subjects
could access at any time during the experiment.
In total 121 subjects participated in the experiment. The set of decisions made by each subject
represents an independent observation for our statistical analyses. The number of subjects
participating in each treatment is displayed in Table 1. The experimental sessions were
conducted on two occasions, one at the ESSE laboratory at the University of Bari (12 subjects
per treatment) and the other at the CentERlab at Tilburg University (13-23 subjects per
treatment). Each subject participated in only one treatment.
Theoretical Benchmarks
In section 2 above, we derived the optimal extraction strategy in the full information GS
treatment. This strategy is clearly not applicable in the other treatments. Moreover, it is not
clear in these other treatments what the optimal strategy is. However, in this section we
propose ‘reasonable’ strategies in these other treatments, and justify their ‘reasonableness’ by
showing that the implications of following these strategies are close to the implications of
following the optimal strategy (if it were known). Of course, the subjects in our experiment
could not know what was the optimal strategy, but we, the experimenters, know, and can use
that knowledge to justify these ‘reasonable’ strategies. In what follows, we refer to these as
our ‘theoretical benchmarks.’ That for the GS treatment coincides with the optimal strategy;
those for the other treatments are justified in what follows.
The development of the stock on the optimal extraction path for the treatment GS is shown in
the top panel of Figure 2. On first sight, it may seem inadequate to use this perfect
information benchmark for calculating efficiencies in the imperfect information settings.
However, the benchmarks for the two treatments G (top right in Figure 2) and S (bottom left
in Figure 2), are so close to this simple benchmark that using more elaborate comparisons
would not yield any substantially different results. Even the benchmark for the treatment No
that shows substantial stochastic variation centres with a median path around the perfect
information benchmark (bottom right panel in Figure 2). In other words, given subjects follow
a path of action that uses the information input consistently, they are likely to come close to
the perfect information optimum rather quickly in an early phase of the experiment.
In order to avoid the problem of extinction, the suggested theoretical benchmarks for the
treatments without perfect information are all “prudent”, i.e. extraction choices that do not
10
lead to the extinction of the resource with certainty are preferred to those that risk extinction.
While this requirement may be too conservative in general, it seems useful, because it defines
the most cautious benchmarks, below which no reasonable extraction path should fall, no
matter how risk-averse the decision-taker is. Interestingly, prudence does not really pose a
major source of inefficiency in any of the settings. In treatment S, the prudent benchmark
behaviour achieves 98.6 percent efficiency. While average efficiencies of 99.7 and 85.4
percent are achieved in the treatments G and No, correspondingly. These average efficiencies
have been computed on the basis of the theoretical benchmarks and the individual realization
of the stock signals.
In the treatment S, in which the stock size is known, but not the growth function, the decision-
maker must use the information on stock size change (i.e. growth) to infer the best possible
plan of action. Given that the range of possible actions is finite and given that the growth
function is unknown, but single-peaked and fixed, the task can be reduced to a parameter
search problem. The goal of the search algorithm is to identify the stock level inducing the
maximum growth. Since the stock size information is perfect and the growth function well-
behaved, a hill-climbing algorithm can be used that searches the parameter space employing
systematic experimentation and consistent adaptation of the choice variable to achieve higher
and higher values of the goal variable. The only major difficulty that the process must deal
with is the lock-in hazard that is due to the missing information on the growth function. A
lock-in situation arises when experimentation entails the risk of “being stuck” at such a low
level of growth that a return to the optimal stock size is no longer feasible within the decision
horizon. In the case of an extremely skewed growth function, for example, even relatively
cautious experimentation may lead to lock-in situations, on the one hand, while perfectly
conservative stock preservation will obstruct the optimisation process, on the other. Hence,
the decision-maker will have to trade-off the efficacy of the search process against the risk of
being locked in at a sub-optimal stock level.
The benchmark we present in the lower left panel of figure 2 uses a simple hill-climbing
search algorithm with an exponentially decreasing experimentation rate εt. In any period t, the
decision-maker extracts an amount that leaves 1 – εt of the last period’s post-extraction stock
level. As long as the observed absolute growth in t is greater than in t – 1, the process is
continued with the same experimentation rate εt, i.e. εt+1 = εt. As soon as, a decline of the
resource growth is observed in a period τ, extraction is adjusted to restore the previous stock
size, before continuing experimentation at an halved rate, i.e. at the rate ετ = ετ – 1/2. At what
11
speed the process converges and how difficult it is to recover from local lock-ins, does not
only depend on the growth function, but also on the initial conditions, i.e. the initial size of
the stock and the initial experimentation rate ε0.16 Using the experimental parameters and an
initial experimentation rate of .1, we can show that after only 9 of 100 periods, the process
converges to stock levels that are within 10 points around the perfect information benchmark.
By period 20 experimentation ends and the processes rests exactly at the optimal stock level.
Figure 2. Theoretical benchmarks
Things are a bit more complicated in the G treatment, in which the growth function is known,
but not the exact stock size. In this case, using the growth function information, the decision-
maker can calculate the optimal target stock size, which is identical to the target stock size in
the perfect information treatment. However, due to the stochastic nature of stock size
feedback, extraction decisions that perfectly hit the targeted optimal stock size cannot be
made. Instead, to improve the quality of the extraction decisions, the information arriving
16 If ε0 is small the risk of overshooting the optimal growth stock level is small, but the speed of convergence is
low. If ε0 is large then the contrary is true. The path displayed in Figure 2 is derived for an initial stock size of
1000 and an initial experimentation rate of ε0 = .1.
0
100
200
300
400
500
600
700
800
900
1000
0 102030405060708090100
round
Stock-size (after extraction)
Optimal Extractions in GS
0
100
200
300
400
500
600
700
800
900
1000
0 102030405060708090100
round
Stock-size (after extraction)
simulation median
simulation maximum
simulation minimum
Prudent Extractions with Rational Updating in G
0
100
200
300
400
500
600
700
800
900
1000
0 102030405060708090100
round
Stock-size (after extraction)
Prudent Extractions with Hill-Climbing in S
0
100
200
300
400
500
600
700
800
900
1000
0 102030405060708090100
round
Stock-size (after extraction)
simulation median
simulation maximum
simulation minimum
Prudent Extractions with Empirical Growth Estimates in N
12
after each decision must be used to increase the precision of the stock size estimate. In any
period, the extraction history and stock size signals can be combined with the growth function
information to tighten the lower und upper boundaries on the initial stock size. As more and
more observations are made, the range of possible initial stock sizes is reduced, ultimately
making a perfect estimate possible. Once the initial stock size can be pinpointed, the current
stock size can be calculated by reconstructing the history of extractions and applying the
growth function, correspondingly.
While the inference logic described above is unique, neither the realised path of information
disclosure nor the level of extractions up to the point of perfect inference are unambiguous.
The path of inference is not unique, because of the stochastic nature of the stock size
feedback. Obviously, certain sequences of random draws will enable a quicker perfect
inference than other sequences. Furthermore, the optimal extraction behaviour before the
perfect inference is achieved depends on how the threat of pre-mature extinction is treated.
We have chosen a benchmark that deals with the extinction issue by assuming “prudence” in
the sense that any pre-mature extinction of the stock is excluded. The prudent extraction xt in
period t is limited to being no greater than the minimum estimated stock size st at time t (i.e.
given all information collected in the previous periods), hence xt = max(0, st – 500), where
500 is the optimal stock size (derived from the growth function information).
The top right panel in Figure 2 displays the development of the stock in a small Monte-Carlo
sample of runs with prudent extraction and rational updating. The “median path” shown in
the panel, describes which development of the stock size we should be expecting, if subjects
are prudently extracting and rationally updating. The “minimum” and “maximum” paths show
the extremes of the simulated distribution17. As can be seen, the stock size quickly converges
to the optimum of 500 (i.e. the exact initial stock size is quickly inferred from the history),
even though the assumed behaviour is very cautious concerning the threat of pre-mature
extinction. On the median path, the maximum sustainable yield at a stock size of 500 is
reached after only 20 of 100 periods. Even in the worst case observed in the simulation, no
more than 40 periods were needed for full convergence.
What is perhaps even more important than the point of full convergence is the fact that the
path comes close to optimum very quickly and hence induces only minor losses due to the
17 It should be noted that these do not represent actual paths, but just the upper and lower envelopes of the
various possible paths.
13
imperfect information. On the median path the total extraction is just slightly below 38000
compared to the 38125 in the optimum of the perfect information setting. Hence, the median
efficiency loss due to application of the prudent extraction rule would be less than 0.4% and
even in the worst case only 1%.18
Finally, defining a convergent benchmark in the treatment No involves using blending the two
methods used for the benchmarks in S and G, because both growth function information and
perfect stock size information are missing. The bottom right panel in Figure 2, displays the
median, the minimum, and the maximum path that were observed in a small Monte-Carlo
simulation using such a combined procedure. In this procedure, the hill-climbing process (i.e.
the search for the stock size that induces maximum growth) cannot be controlled by simply
comparing resource growth at different levels of stock, because the stock size information is
imperfect. Instead, the success measurement has to be based on the distribution of empirically
estimated growth numbers. Since the growth function information is also unavailable,
achieving the same precision of the empirical estimation of the growth numbers as in
treatment G requires having many more observations. Hence, the low level of information in
No leads to a high dispersion in the speed and path of convergence. As the minimum and
maximum paths we observed in our simulation show, often 100 periods will not entail enough
empirical observations as to allow a convergence of the process to the maximum growth
point. Nevertheless, the median simulation path converges well within the first half of the
experiment, indicating that the distribution of experimentally observed post-extraction stock
sizes in the second half of the experiment should be located around 500, the stock level that
induces maximum growth.
4 Experimental Results
This section is organized corresponding to the optimal extraction plan. First, we survey the
efficiency of initial extraction decisions; second, we consider the overall efficiency and the
evolution of extraction decisions; and, third we report on the efficiency of subjects’ last
extractions. As a benchmark we refer to the decisions on the optimal path. These imply a
stock-size after extraction at the maximum sustainable yield (i.e., 500 units) until pen-ultimate
decision and extinction of the resource with the last decision. We conclude the section by
classifying observed individual behavioural pattern.
18 Since subject payments in the experiment were rounded up to multiples of 50 Cents, even in the worst case
simulation subject payments would have coincided with the maximal payoff.
14
The First Extraction Decisions
The first extraction induced significant under-harvesting in all treatments (two-tailed
Wilcoxon signed ranks test at α=.01): subjects extracted less than the optimal 500 units. Table
2 records the statistics on stock after the first extraction.
Table 2. Stock size after first extraction
treatment # minimum Maximum average std. error Wilcoxon-test
H0: average=500
GS 25 300 975 653 165 -3.40**
G 35 280 999 670 197 -3.98**
S 31 100 1000 842 232 -4.34**
No 30 200 1000 932 166 -4.69**
**significant at 1%, one-tailed; *significant at 5%, one-tailed
The deviations from the optimal extraction increased from treatment GS through No (two-
tailed Jonckheere test of ordered alternatives at α=.01). Equation (5) represents a pooled
dummy regression of the distance of stock after the first extraction from the optimum on G
and S. The variables DG and DS denote dummy variables that take a value of one if a subject
receives growth information and accurate stock information, respectively, and zero otherwise.
In accordance with the Jonckheere test, the regression results reveal that growth and accurate
stock information both had a significant influence on the efficiency of subjects’ first
extraction decisions.
|Stock1 – 500| = 438** - 223 DG** - 57 DS* 2
ˆ
R
=.35 (5)
23.96 27.64 27.71 [std. error]
18.30 -8.08 -2.07 [t-ratio]
**significant at 1%, *significant at 5%, one-tailed
Average Efficiency
The subjects’ presumed objective in the experiment was to maximise efficiency, which we
define as the ratio of the actual extraction to the maximal possible one. Table 3 records the
minimum, maximum, and average efficiency attained in the experiment. Standard deviations
15
are reported in the last column.19 The maximum of the observed efficiency levels is close to
the efficiency level proposed by the theoretical benchmarks in every treatment. However, the
deviation of the average observed efficiency from the benchmark differs substantially across
treatments, which is due to the differences in the variance across treatments.
Table 3. Efficiency
observed
treatment # benchmark
prediction maximum average minimum SD
GS 25 1.000 0.997 0.874 0.613 0.108
G 35 0.997 0.946 0.727 0.091 0.254
S 31 0.986 0.922 0.660 0.057 0.252
No 30 0.854 0.853 0.398 0.018 0.287
Efficiency increased across treatments from No to GS. Differences between treatments are
significant at 1% for all pair-wise comparisons, except for the comparison of G to S (Mann-
Whitney test, two-tailed).20 In treatment G (and only in treatment G), three subjects extinguish
the resource within the first 19 extractions (see Table A in the appendix). If we do not take the
extinction observations into account, efficiency in treatment G is significantly greater than in
S. The dummy regression of efficiency on the treatment dummies growth and accurate stock
reported in equation (6) indicates that both treatment variables had a significant effect on
efficiency.21 The knowledge of the growth function implied an increase of average efficiency
by 27.6%, the accurate stock size information by 21%.
Efficiency = .427** + .276 DG** + .210 DS** 2
ˆ
R
=.31 (6)
.038 .044 .044 [std. error]
11.17 6.26 4.66 [t-ratio]
**significant at 1%, one-tailed
19 Table A in the appendix records individual efficiency levels.
20 If we consider only the data from the second half of the experiment the two-tailed Mann Whitney test rejects
the null hypothesis of same efficiency for all pair-wise comparisons at the 5% level of significance. The
theoretical solutions suggested that efficiency should be indistinguishable between treatments in later rounds.
However, this suggestion is not supported by the data.
21 Conducting the regression with two additional dummies, one to determine the cross-effect of growth and stock
information and one to determine the subject pool effect, we find no further significance. Obviously, the two
types of information have no synergies and the results are robust across subject pools.
16
The Evolution of Extractions and Stock
Figure 3 contrasts the evolution of average stock levels after extraction in all treatments with
the optimal path represented by the 500-units line.22 While the overall average distance of
observed end stock from the optimum is rather small, especially towards the end of the
experiment, the spread of the outcomes is large enough to substantially impact on the overall
efficiency. Yet we find that efficiency increases across periods. To show this increase we
compare the observed extraction with the most rapid approach to the optimal path. Payoff
maximization involved in each but the 100th decision an extraction of the maximum of zero
units and the difference of the actual stock size and 500 units. In case the resource was
depleted below 233 units the stock would have to be rebuilt by zero extraction.
Figure 3. Evolution of average end stock (after extraction) compared to optimal path
Note: horizontal line indicates efficient extraction path, oscillating lines indicate average stock-size after
extraction and standard deviation bands
22 An anonymous referee suggested that the gradual decline of the stock-sizes towards the end of the experiment
could be due to a relaxed ‘risk aversion,’ as intuition may suggest that there is less to lose in the case of an
irreversible mistake. If so, such ‘risk aversion’ may have contributed to a certain elevation of the stock prior to
the last few rounds. Indeed, unless the resource is wiped out, due to our design, the same mistake usually leads to
the same loss whether made at the beginning or at the end of the experiment. But in the treatments where the
growth–function was unknown, subjects obviously were not given this information, so they had to find it out by
trial and error.
0
250
500
750
1000
1250
0 102030405060708090100
round
Stock-size (after extraction)
No
optimal
stdev
0
250
500
750
1000
1250
0102030405060708090100
round
Stock-size (after extraction)
GS
optimal
stdev
0
250
500
750
1000
1250
0 102030405060708090100
round
Stock-size (after extraction)
G
optimal
stdev
0
250
500
750
1000
1250
0102030405060708090100
round
Stock-size (after extraction)
S
optimal
stdev
GS G
SNo
17
To check whether the stock converges to the optimal level over time, i.e. over the sequence of
extraction decisions, we run fixed effect regressions to estimate the parameter for time. The
regression results recorded in Table 4, in particular the negative coefficients of the time
variable, confirm that efficiency increases in the course of the experiment. The size of the
coefficients indicate that the effect of time on efficiency was greater in treatments S and No
than in the treatments GS and G, in which subjects received information about the growth
function. This result does not say that subjects in the treatments without growth information
were more efficient in the end than those who had this information. Since the subjects in the
treatments S and No started further away from optimum than those in the treatments GS and
G, they faced a greater potential for convergence over time. Not counting for the last
extraction, the average distance of the observed extraction and optimal extraction did not
decrease below 177 and 241 units in any period of the treatments S and No, respectively. In
contrast, in the treatments GS and G, the average deviation from the optimum never exceeded
175 and 218 units, respectively.
Table 4. Distance of observed and the optimum extraction across time
Coefficient std. err. t-ratio
GS constant 148** 2.998 49.30
period -0.313** 0.052 -6.08
G constant 186** 3.730 49.92
period -0.502** 0.064 -7.83
S constant 266** 3.543 75.29
period -0.915** 0.061 -15.03
No constant 413** 3.971 104.01
period -1.343** 0.068 -19.70
**significant at 1%, two-tailed
Greater efficiency also corresponds to smaller stock sizes across treatments. On average, our
experimental data suggest a rather equal spread of over- and under-harvesting in all but the No
treatment. As can be seen in Table 5, using the binomial test on the distribution of over- and
under-harvesters, we only observe a significant bias in the No treatment, in which 90% of the
18
subjects under-harvest.23 Over-harvesting was most heavy in treatment G, where three
subjects extinguished the resource within 19 decisions. In all of these three observations, the
last extraction before extinction did not exceed the signalled stock, but in fact it did exceed
the actual stock size.
The propensity to harvest more in GS and G than in S and No suggests that subjects are more
confident with their extraction decisions when they receive exact information on the stock
dynamics. In the No treatment, subjects have little information and seem to be reluctant to
exploit the resource too much. The level of extraction they choose is perhaps related to the
fact that initial condition in the experiment is close to the carrying capacity.24 Brekke and
Moxnes (2003) show that subjects that start off above the target capacity tend to have higher
levels of stock throughout. Even though we do not vary the initial condition, our finding may
be related to that result.
Table 5. Over- and under-harvesting
treatment # over-harvester # under-harvester
GS 10 (40.0%) 15 (60.0%)
G 18 (51.4%) 17 (48.6%)
S 11 (35.5%) 20 (64.5%)
No** 3 (10.0%) 27 (90.0%)
** difference is significant at 1%, two-tailed
The Final Extraction Decision & Extinction of the Resource
With the final extraction, subjects were expected to extinguish the resource. However, only
about one half of them did so as shown in Table 6. 60 subjects (50%) ended the experiment
with a zero stock, 9 subjects in treatments G and No did not extinguish the resource, but
extracted all units signalled to them in the last decision. Apparently they had forgotten that the
23 We classified a subject as over-harvester, if the subject’s end stock was below the optimum in more periods
than above. Since we observed no tied cases, all the other subjects were classified as under-harvesters.
24 At the carrying capacity the non-linear growth curve implies that growth first increases with extraction before
it decreases. If the initial conditions are close to the maximum sustainable yield, however, growth decreases with
extraction. In the former case, constant harvesting can lead to a stable equilibrium while in the latter case it may
accelerate depletion.
19
signal was most likely incorrect.25 There were also 12 subjects who extinguished the resource
too early, i.e. before they reached the 100th decision: eight subjects did so in the 98th and 99th
decision in treatment No and one subject in the 93rd decision of GS. As already pointed out
above, we observed three cases of apparently unintentional extinction in treatment G within
the first 19 decisions.
Table 6. Distribution of final extinction and non-extinction
Treatment zero-stock left extinction
attempted
minimal stock
left non-extinction
GS 15 - 1 9 (36%)
G 14 4 - 17 (49%)
S 14 - 4 13 (37%)
No 17 5 - 8 (27%)
Total 60 9 5 47 (39%)
Behavioural pattern: Control Theory, Linear World and Misperceptions of Feedback
In agreement with Edwards’ (1962) classical description, the present work contributes to the
laboratory studies on dynamic decision making.26 Brehmer (1992) suggests that experiments
on dynamic decision making are particularly valuable since real world problems such as
company management or even everyday life involve many dynamic tasks, and field data is
difficult to obtain. As a general framework for the study of dynamic decision making,
Brehmer (1992) proposed control theory (although not the mathematical term).27 He pointed
out, subjects’ overall goal in a dynamic decision task should be one of “… achieving control:
that is, that decisions are made to achieve some desired state of affairs, or to keep a system in
some desired state.”
As we observe literally no incidence of individual decision making in support of our above
outlined theoretical benchmarks, we establish the alternative research hypothesis that subjects
either try to hold the stock signal constant or the extraction level (through the extractions 2-
99). This hypothesis is based on the idea that subjects try to take control over the dynamic
25 In total, 23 subjects in treatments G and No extracted the signalled stock-size in the last decision. In the other
14 cases the signalled stock-size exceeded the actual stock-size.
26 A dynamic decision problem implies that 1) a series of decisions is required to reach the goal, 2) the decisions
are not independent, and 3) the state of the decision problem changes. See Brehmer (1992) for a discussion.
27 This was noted before; see for instance Rapoport (1975).
20
system. Actually, we can find support for both extraction policies. In Figure 4, we have
plotted the individual stock development of four subjects, who exhibit behaviour that can be
identified as “typical” for the control theory hypothesis. The number of observations that
follow similar patterns is stated in table 7. For instance, 15 individual charts or 60% of the
observations in GS display straight lines similar to that presented in the top left panel of
Figure 4.
In treatments GS and S, in which subjects received accurate stock information, it is difficult to
tell whether subjects were maintaining stock or extraction as both variables depend on each
other. However, these questions can be addressed by examining the plots of treatments G and
No, where such information was not supplied. Next to the optimal path that is indicated with a
dotted line, these plots exhibit two further lines: The unbroken line represents the movement
of end stock (after extraction) and the dashed line represents the noisy end stock signal (after
extraction). The displayed plots represent 34% and 7% of similar patterns in the treatments G
and No, respectively. In the representative plot of treatment G, the dashed line is straight
indicating that the extractions were meant to maintain a constant stock signal. In contrast to
this, the straight segments in treatment No exhibit a constant end stock, which hints at a policy
of constant extraction.
Figure 4. Behavioural pattern – control theory
21
In fact, more plots of individual extraction decisions indicate a constant stock size in
treatments S and No, but not in support of the control theory story. Samples of these are
displayed in Figure 5. The striking pattern is that half of the subjects in treatment No and 19%
in S extracted almost nothing. They held their stocks near the biological equilibrium size of
1000 units where growth was very close to zero. This odd behaviour can hardly be
rationalized if not in the light of Brehmer’s (1980) observation that people tend to believe in a
linear model rather than in other models. If subjects actually believed in a linear relationship
between stock size and growth they might have taken for granted that growth increases with
stock. From this perspective it would make sense to let stock grow and extract at the end the
profit maximizing stock size.
Figure 5. Behavioural pattern – linear world
Such misperception of linearity in non-linear dynamic systems has been reported in earlier
experimental research (see Sterman (1994) for a survey)), as we pointed out in section 3.
Another behavioural pattern, which Sterman (1989a, b, 1994) called the misperceptions of
feedback, must be seen in the fluctuations of the stock-sizes after extractions (see Figure 6).
Such fluctuations in stock can be due to “pulse fishing,” a behaviour that makes sense in some
fishery environments (Schnier and Anderson 2006),28 but not in our experiment. Our subjects
were informed that they are facing a deterministic system. In such settings, the optimal
harvesting behaviour is non-pulsing. It seems particularly surprising that even in the
transparent setting of treatment GS (in which subjects experienced feedforward information)
cycles and oscillations of stock occurred. Paich and Sterman (1993), who observe comparable
28 Pulse fishing refers to a behaviour that alternates between harvesting and not-harvesting from a stock (Schnier
and Anderson 2006). Note that seeing the periods with zero harvesting in our graphs is not possible, because our
graphs display the end stock of each period and not the harvest.
22
patterns, claim that subjects’ learning in complex environments is poor. This argument could
in fact explain the persistency of these oscillations in the data.
Figure 6. Behavioural pattern – pulse fishing
The behavioural patterns that classify our subjects almost perfectly are summarised in Table
7. About 34% of subjects tried to achieve control over the complex system by holding either
stock or extraction levels constant. Another 45% of subjects managed their stocks by pulse
fishing and 17% misperceived the non-linearity of the environment and extracted almost
nothing. None of these behavioural patterns could be used to classify the remaining 5%.
Testing for differences in distributions with pairwise chi-squared tests, we find that the
distribution of patterns significantly differs between No and any other treatment, as well as
between G and any other treatment.29 The distribution of patterns across GS and G, however,
are statistically indistinguishable.
5 Summary and Conclusions
In this paper we have considered the fishery management problem under a finite-horizon
condition. We established the benchmark solution (in the full information treatment) that is in
29 The effects are all significant at the 1% level, two-tailed, expect for the difference between distributions in GS
and G that are only significant at the 5% level, two-tailed.
23
line with the infinite-horizon solution in all but the last extraction period. Hence, the
theoretical benchmark basically does not differ from that used in other fishery experiments,
where “infinite” horizon tasks are simulated by either providing a residual resource value in
the final period or unexpectedly cutting sessions short that were ostensibly longer. For
laboratory studies, our finite-horizon design seems more compelling and transparent than the
residual payment and indefinite length settings.
Table 7. Individual pattern: summary
Treatment # Control theory Pulse fishing Linear world Unexplained
GS 25 15 (60%) 9 (36%) - 1 (4%)
G 35 12 (34%) 19+3 (63%) - 1 (3%)
S 31 12 (39%) 10 (32%) 5 (16%) 4 (13%)
No 30 2 (7%) 13 (43%) 15 (50%) -
Total 121 41 (34%) 54 (45%) 20 (17%) 6 (5%)
In line with the experimental literature on resource extraction (Mason and Phillips, 1997;
Moxnes 1998a, b, 2000, 2004; Brekke and Moxnes, 2003; Schnier and Anderson, 2006),
extinction of the resource before the end of the experiment was generally not a problem in our
study. We only observed a few early terminations in the treatment, in which subjects received
information on the growth function, but only a noisy signal of the stock.
In the treatments with some information, we find over-harvesting and under-harvesting in
almost equal frequencies. We find a significant under-harvesting problem only when subjects
neither have exact information on the growth function nor on the stock size. This result
diverges from most of the findings in the literature, where over-harvesting is generally more
prevalent than under-harvesting. Our result for the no information treatment may be due to the
specific experimental characteristics: in the our study, subjects had a direct control over the
fishery resource and the initial stock levels were set to the carrying capacity. As Moxnes
(2004) shows, direct control may have an effect on behaviour. It, however, seems difficult to
compare those findings with our results, due to the numerous other design differences.
Whether the initial stock size in our setting has an effect on harvesting behaviour also remains
an open issue for future research. There is evidence suggesting that subjects facing a stock
that is well above the target will tend to maintain a higher stock level throughout (Brekke and
24
Moxnes, 2003). Obviously, this is in line with our finding. It, however, cannot explain the
strong treatment differences that we observe.
Our study also adds to the evidence on the behavioural relevance of pulse fishing, which was
experimentally first reported by Schnier and Anderson (2006). It seems that using a total
moratorium to help resource recovery is not only a path often chosen by subjects in patchy
environments (e.g. Schnier and Anderson 2006), but also in a single resource pool. In fact, the
instrument is also often used by political administrations that seem to prefer a total ban on
harvesting for a limited time over other recovery methods. Perhaps it is easier to control a
moratorium both on an individual self-control level and in a public law-enforcement setting.
The efficiency of extraction decisions in our experiment was an estimated 21% higher if the
stock signal was accurate and 27.6% higher if subjects had knowledge on the growth function.
Both results seem to suggest that research can significantly help to increase extraction
decisions. This corresponds well to the findings by Brekke and Moxnes (2003), who report
that decision support systems (i.e. “better” information) can enhance outcomes. Our study
adds to these finding, because we can identify different behavioural patterns that are more or
less likely to evolve, depending on the informational setting. In cases in which there is ample
information on the growth function and the stock size, we can expect to find many subjects
using control heuristics to either keep the stock size or the harvest constant. Whenever stock
information is noisy, pulse fishing behaviour is especially frequent. If additionally the growth
function is not known, we can expect a substantial number of decision makers, whose
behaviour indicates a linear misperception of the growth dynamics. Obviously, identifying the
behavioural patterns that are most prevalent in a certain resource extraction environment may
help design institutions that are especially effective.
However, it should be noted that we considered here a highly simplified, deterministic model
in which the precise growth function is given or not. In a real world resource management
problem the decision maker faces an inaccurate growth model, non-deterministic stocks and
positive market parameters as interest rates, costs, and prices. Furthermore, we are aware that
political influences may affect the decisions of the authority as well (e.g., lobbyism) but we
left these imperfections in the decision making process out of focus. Yet, all these
complications can conveniently be accommodated within the presented framework. The
standard logistic growth model which we considered in the experiment seems to be ideally
behaved to provide the experimenter with a rich research environment.
25
6 References
Abbink, K. and A. Sadrieh, 1995, “Ratimage, research assistance toolbox for computer-aided
human behavior experiments,” Discussion paper B-325, University of Bonn.
Apesteguia, J., 2005, “Does Information Matter in the Commons?” Journal of Economic
Behavior and Organization (in press).
Bellman, R., 1957, Dynamic Programming, Princeton University Press, Princeton.
Brehmer, B., 1980, “In One Word: Not From Experience,” Acta Psychologica 45: 223-41.
Brehmer, B., 1992, “Dynamic Decision Making: Human Control of Complex Systems,” Acta
Psychologica 81: 211-41.
Brekke, K.A., and E. Moxnes, 2003, “Do numerical simulations and optimization results
improve management? Experimental evidence,” Journal of Economic Behavior and
Organization 50, 117-131.
Clark, C. W., 1990, Mathematical Bioeconomics, second edition, John Wiley & Sons, New
York.
Diehl, E., and J. D. Sterman, 1995, “Effects of Feedback Complexity on Dynamic Decision
Making,” Organizational Behavior and Human Decision Processes 62 (2): 198-215.
Gordon, S. H., 1954, “The Economic Theory of a Common Property Resource: The Fishery,”
Journal of Political Economy 62: 124-142.
Mason, C. F. and O. R. Phillips, 1997, “Mitigating the tragedy of the commons through
cooperation: an experimental evaluation,” Journal of Environmental Economics and
Management 32: 148-172.
Moxnes, E., 1998a, “Not Only the Tragedy of the Commons: Misperceptions of
Bioeconomics,” Management Science 44: 1234-1248.
Moxnes, E., 1998b, “Overexploitation of Renewable Resources: The Role of
Misperceptions,” Journal of Economic Behavior and Organization 37: 107-127.
Moxnes, E., 2000. Not only the tragedy of the commons: misperceptions of feedback and
policies for sustainable development, System Dynamics Review 16(4), 325-348.
Moxnes, E., 2003. Uncertain measurements of renewable resources: Approximations, harvest
policies, and value of accuracy. Journal of Environmental Economics and Management
45(1), 85-108.
Moxnes, E., 2004. Misperceptions of basic dynamics, the case of renewable resource
management. System Dynamic Review 20 (2), 136-162.
Paich, M. and J. D. Sterman, 1993, “Boom, bust, and failure to learn in experimental
markets,” Management Science 39 (12): 1439-1458.
Rapoport, A., 1975, “Research Paradigms for the Study of Dynamic Decision Behavior,” in
Wendt, D. and C. Vlek (eds.), Utility, Probability and Human Decision Making: 349-
369, Reidel, Dordrecht.
Rouwette Etiënne A.J.A., Größler, Andreas and Vennix, Jac A.M., 2004. Exploring
Influencing Factors on Rationality: a Literature Review of Dynamic Decision-Making
Studies in System Dynamics. Systems Research & Behavioral Science 21 (4), 351-370.
26
Schnier, K.E., Anderson, C.M., 2006, Decision Making in Patchy Resource Environments:
Spatial misperception of bioeconomic models. Journal of Economic Behavior and
Organization 61(2), 234-254.
Sterman, J. D., 1989a, “Deterministic Chaos in an experimental economic system,” Journal of
Economic Behavior and Organization 12: 1-28.
Sterman, J. D., 1989b, “Misperception of feedback in dynamic decision making,”
Organizational Behavior and Human Decision Processes 43 (3): 301-335.
Sterman, J. D., 1989c, “Modelling Managerial Behavior: Misperceptions of Feedback in a
Dynamic Decision Making Experiment,” Management Science 35 (3): 321-339.
Sterman, J. D., 1994, “Learning In and About Complex Systems,” System Dynamics Review
10 (2-3): 291-330.
27
Table A. individual efficiency
Note: Subjects’ results are arranged according to their performance.
In G, three subjects extinguished the resource within the first 19
extractions. The last extraction was two or five units smaller than the
signalled stock but exceeded the actual stock size.
# No S G GS
1 0,018 0,057 0.091† 0,613
2 0,036 0,066 0,095† 0,618
3 0,043 0,168 0,151† 0,706
4 0,056 0,192 0,396 0,741
5 0,059 0,358 0,437 0,803
6 0,084 0,484 0,488 0,810
7 0,087 0,552 0,504 0,822
8 0,157 0,561 0,507 0,851
9 0,161 0,606 0,521 0,877
10 0,178 0,643 0,580 0,879
11 0,192 0,648 0,649 0,882
12 0,218 0,657 0,656 0,891
13 0,221 0,658 0,782 0,903
14 0,267 0,666 0,803 0,904
15 0,276 0,697 0,814 0,917
16 0,388 0,740 0,836 0,920
17 0,487 0,757 0,853 0,939
18 0,557 0,771 0,860 0,942
19 0,611 0,773 0,866 0,950
20 0,613 0,791 0,866 0,964
21 0,623 0,824 0,868 0,970
22 0,639 0,836 0,874 0,974
23 0,646 0,843 0,890 0,980
24 0,648 0,851 0,896 0,993
25 0,686 0,861 0,902 0,997
26 0,695 0,866 0,904
27 0,755 0,897 0,907
28 0,836 0,901 0,916
29 0,846 0,906 0,927
30 0,853 0,914 0,927
31 0,922 0,929
32 0,936
33 0,936
34 0,937
35 0,946
28
Instructions
In the experiment you are asked to make saving decisions. With every decision you determine
how many units you extract from a fictitious resource stock. Every extracted unit is credited
to your savings account, which is displayed on your screen (in a window labelled “status”).
Your objective in the experiment is to maximize savings. You begin with zero savings.
With each extraction you transfer units from your stock to your savings account. After the
decision, the stock will be subject to deterministic growth. That is, the resource stock grows
by an amount that is unequivocally determined by the number of units that remain after
extraction. If the stock is zero, growth is zero. Unless you extract the entire stock you are
asked to make 100 extraction decisions.
The stock size information
At every time before you make an extraction decision, the stock, i.e., the number of units from
which you can extract, will be revealed to you on the screen. [subjects in G and No read: Yet,
this information is biased. Your information reflects the product of a random number in the
range 0.75-1.25 and the actual stock. In other words, the number of units you have in the
stock is multiplied by a randomly determined number between 75 percent and 125 percent.
The computer determines a new random number after each of your decisions. Consequently,
you never know whether the actual stock is greater, smaller or equal to the revealed one.]
[subjects in GS and G read: The growth function
You are given information about the relation of stock size and growth through an onscreen
tool in a window titled “result calculation”. It is easy to handle: Insert a potential number of
units to be extracted (how to do it is detailed below). The corresponding stock after extraction
and the resulting stock from which you can extract at your next decision will be stated in the
second and the third column. The fourth and the fifth column record the corresponding growth
and the savings after extraction, respectively. By default, this information is recorded for all
potential extractions involving 10 percentiles (i.e., 10%, 20%,…, 100%) of the [subjects of G
read: revealed] stock, as recorded in the first column of the result calculation window.]
Your payoff
There is an optimal extraction plan, though you will not be told any details about it. However,
your payoff relates to the maximum possible amount of savings as follows. At the end of the
experiment your payoff will depend on the quotient of your actual savings and the maximum
29
possible savings (i.e., the quotient corresponds to the result of dividing your savings by the
maximum possible ones). This quotient will be taken times {(subjects in GS read 15),
(subjects in G and S read 17.50), (subjects in No read 20)} Euro to determine your payoff,
which will be paid to you in private as soon as you have taken your last decision in the
experiment.
The software
To make your decision you proceed in 2 steps: First, insert a potential number of units to be
extracted with the keyboard or the mouse, and confirm it with the enter key. The number will
be highlighted in the display of the “decision” window on the bottom right of your screen.
Second, to make your extraction decision final you press the button labelled “extract”. Note,
unless you press the extraction button with the mouse you can insert other numbers as often as
you like without any consequences.
The history
From the menu bar at the top left of your screen you can retrieve the “history” window. The
history records all information you have received and the decisions you have taken in the
experiment.
The screen
Screen G
... has been surprisingly scant consideration of another potentially fundamental factor in poor resource management: the cumulative impacts of ineffective individual strategies (Moxnes 1998;Hendrickx et al. 2001;Ostrom 2007). This is surprising given the acknowledged potential for individual cognitive and behavioural biases to undermine group-based resource outcomes (Hey et al., 2009;Messick & McClelland, 1983;Moxnes, 1998;Schnier & Anderson, 2006). ...
... In fact, resource outcomes in single-player games tend to be heterogenous (Hey et al., 2009;Messick & McClelland, 1983;Schnier & Anderson, 2006). Individuals often erode or exhaust resources prematurely. ...
... Critically, knowing the maximum number of remaining harvesting opportunities provided no direct information about the resource dynamics or strategic clues, except on the final round of the game when participants know they can harvest the entire resource without penalty (Hey et al., 2009). Up until that point, best play with both known and unknown horizons involved harvesting amounts of rewards that allowed the resource to replenish to its maximum level of 60 (without forfeiting any replenished rewards above this maximum) to support larger harvests over the duration of the game. ...
Preprint
Full-text available
Research into improving poor resource management has tended to focus on social interventions that mitigate so-called 'Tragedy of the Commons' outcomes. Less work has investigated interventions to help individuals better manage their personal resources. Over three studies (N=1,597), we test whether informational cues can improve resource management strategies while individuals harvest from a replenishing but depletable resource, returning monetary rewards. To succeed, individuals need to learn about the dynamics (and vulnerability) of the resource while avoiding the long-term costs of early bad decisions. We find little evidence that outcomes are improved by knowledge of the potential monetary returns of the resource or by prior encounters with its dynamics. By contrast, future-orienting information about the potential availability of a resource; its horizon – operationalized as the maximum number of remaining harvesting opportunities – dramatically improves individual resource outcomes (in Experiments 1 and 2). This future-oriented cue did not provide information about the dynamics of the resource that could be used to infer an improved resource management strategy and its benefits are not transitory: informational 'nudges' about the resource horizon continue to improve outcomes over multiple encounters (Experiment 3). These findings indicate that when individuals try to explore and manage a personal resource in uncertain environments where early missteps have long-term adverse consequences, simple future-orienting cues about potential resource longevity improve harvesting decisions. Our findings also highlight how variability in how individuals approach resource management problems per se might contribute to resource difficulties at the group-level.
... An effect can be expected for these first few periods of the game: subjects may not know the game well enough to understand the consequences of their behaviour until after the first couple of periods. However, in the current game the initial drop of the resource size is not necessarily an artefact of participants misunderstanding the game but rather a behavioural pattern common in CPR games (see for instance [31] and [93]). Thus, as this is a process that is not an unnatural behavioural response to the game but a process that can be expected to be found in any CPR, the models will not control for startgame effects. ...
... This trend is very similar to the trend shown in the CPR game by Janssen, Holahan, Lee and Ostrom [31], under treatments of no communication and punishment and costly punishment. A steep initial drop in resource size is also visible in the first 10 periods of the CPR fishing experiment by Hey, Neugebauer and Sadrieh [93], in a treatment where information resource growth and stock size are available. ...
... With the use of detailed and heavily contextualised games, comes a more detailed-and thus limited-interpretation of the results; the results of this study are valuable for commonpool resources, and especially resources with structures similar to fishing grounds. With the information about the resource size (fish stock) available for players to see at the beginning of every new period, the results may not hold in situations where the total allowable catch cannot be correctly measured because the fish population dynamics is unknown [93]. In addition, whereas players could see the history of other players' actions, this may not always be the case (see for instance Lacomba, Lagos and Perote [113]). ...
Article
Full-text available
The increasing heterogeneity of populations affects cooperation in common-pool resources in a time where the depletion of natural resources is a growing problem. This study investigates the effects of economic and sociocultural heterogeneity on trust and cooperation in common-pool resources using a laboratory experiment. The experiment comprises two Investment Games and a Common-Pool Resource Game, with a sample of 344 subjects from the United Kingdom and the Netherlands. By measuring the effects of economic and sociocultural heterogeneity separately as well as combined, this study disentangles the effects of specific heterogeneity types on cooperation in common-pool resources; something that has not been done before. Higher levels of trusting behaviour are found to have a positive effect on cooperation on the micro- and macro-level over time. While theory suggests negative effects of both forms of heterogeneity on cooperation through decreased levels of trust, the results show a surprising positive effect of economic heterogeneity on cooperation, but a negative effect if economic and sociocultural heterogeneity are combined. This study concludes that economic inequality can promote cooperation in CPRs, unless this inequality is lined up with sociocultural differences.
... A relatively small number of studies have removed the social aspects of common pool games and tested whether individuals struggle to navigate resource dynamics in isolation. The outcomes in single-player resource games, like those of common-pool games, tend to be heterogenous (Hey et al., 2009;Messick & McClelland, 1983;Schnier & Anderson, 2006). Operating by themselves, some struggle to sustain resources, eroding or exhausting them prematurely. ...
Article
Full-text available
Encouraging sustainable use of limited natural, social, and economic resources requires understanding the variety of ways in which people think about how resources work and how they adjust their behaviour (or not) as available resources fluctuate. Previous investigations which have focused on understanding how individuals navigate erodible resources, have tended to use group-based, common pool games. However, such social games make it difficult to disentangle whether resource erosion is linked to difficulty navigating the dynamics of the resource or caused by social factors. Here, in two experiments, we recruited 781 participants to play a single-player resource management game in which individuals were invited to harvest monetary rewards from a fully depletable but stochastically replenishing resource over time. We find that the ability to sustain a resource over successive harvesting opportunities (in order to maximize the total harvested rewards) is reliably worse in individuals reporting elevated psychological distress, the often cooccurring hazardous alcohol use, and elevated rates of delay discounting. The associations between resource outcomes, harmful alcohol use, and psychological distress remained substantial even once we had controlled for elevated discounting rates (as a form of impulsivity and a strong risk factor for these health challenges). By contrast, individuals who reported higher levels of financial literacy and general well-being achieved better resource outcomes. Our observations demonstrate that the capacity to respond effectively to the dynamics of a resource are compromised in individuals at risk of psychological and alcohol-related disorders.
... A relatively small number of studies have removed the social aspects of common pool games and tested whether individuals struggle to navigate resource dynamics in isolation. The outcomes in single-player resource games, like those of common-pool games, tend to be heterogenous (Hey et al., 2009;Messick & McClelland, 1983;Schnier & Anderson, 2006). Operating by themselves, some struggle to sustain resources, eroding or exhausting them prematurely. ...
Preprint
Encouraging sustainable use of limited natural, social, and economic resources requires understanding the variety of ways in which people think about how resources work and how they adjust their behaviour (or not) as available resources fluctuate. Previous investigations which have focused on understanding how individuals navigate erodible resources, have tended to use group-based, common pool games. However, such social games make it difficult to disentangle whether resource erosion is linked to difficulty navigating the dynamics of the resource or caused by social factors. Here, in two experiments, we recruited 781 participants to play a single-player resource management game in which individuals were invited to harvest monetary rewards from a fully depletable but stochastically replenishing resource over time. We find that the ability to sustain a resource over successive harvesting opportunities (in order to maximize the total harvested rewards) is reliably worse in individuals reporting elevated psychological distress, the often cooccurring hazardous alcohol use, and elevated rates of delay discounting. The associations between resource outcomes, harmful alcohol use, and psychological distress remained substantial even once we had controlled for elevated discounting rates (as a form of impulsivity and a strong risk factor for these health challenges). By contrast, individuals who reported higher levels of financial literacy and general well-being achieved better resource outcomes. Our observations demonstrate that the capacity to respond effectively to the dynamics of a resource are compromised in individuals at risk of psychological and alcohol-related disorders.
... This is not surprising, as the shape of the curve describing resource size and appropriation effort over time for this treatment differs from the other treatments in the IND and UKNL studies. Whereas the mixed model does well in predicting most treatments' resource size curves that follow the pattern of a quick decrease in the first 10 periods followed by a flattening of the curve-as we also see in other CPR experiments [59,60]-it does not do well in predicting a resource size curve as visible in the EHSH treatment of the IND study. The fact that the random model has a relatively high fit for the EHSH treatment in the IND study does not necessarily mean that the EHSH subjects acted randomly. ...
Article
Full-text available
Rising migration numbers and the resulting increase in economic and sociocultural heterogeneity in societies all over the world are theorised to put pressure on the sustainable use of common-pool resources [CPRs]. Increased heterogeneity is argued to decrease trust and diversify interests between resource users, leading to overuse and decline of natural and man-made CPRs. The aim of this paper is to understand cooperative behaviour under economic and sociocultural heterogeneity in CPRs, through the analyses of experimental data including 344 subjects from the United Kingdom and the Netherlands, and 144 subjects from India. Multilevel regression, ordinal logistic regression, linear conditional-contribution profiles [LCPs] and agent-based models [ABMs] are used to analyse and replicate experimental outcomes on the micro- and macro-level. Results show that the combination of economic and sociocultural heterogeneity affects cooperation negatively when the decision-situation is perceived as unfair, but that neither economic nor sociocultural heterogeneity on themselves affect cooperation negatively. Economic heterogeneity is even found to affect cooperation positively relative to homogeneity. Player type classification based on LCP scores shows that experimental outcomes can be interpreted with player types, and ABM simulations validate the experimental results by replicating the main outcomes.
Article
Open access resource problems and harmful pollutants from manufacturing activities are common in resource management practices. Nevertheless, their implications have only been studied in different and separate frameworks that are not covered within the same structure. Previous studies suggest that resource management enforced by one country can increase welfare levels and rebuild resource conservation, compared to the case where no country imposes resource management policies. However, in real-life examples, the harvesting and manufacturing industries exert simultaneous pressure on fishery resource stocks, thereby changing the nature of the supply curve of renewable resources. This study investigates the effects of trade liberalization under unilateral resource management regimes in a two-country, two-sector model, in which both production sectors can detrimentally affect renewable natural resources by generating two interacting environmental burdens: excessive harvesting and industrial pollution. It is demonstrated that unilateral resource management applied by a country in which the resource-good sector is relatively less damaging to fishery stocks is welfare-reducing for both countries compared to the situation where neither manages its resource sector. This result is identified as “immiserizing resource management.” Notably, however, unilateral resource management by one country in which the resource-good sector has a more significant negative impact than the manufacturing industry can benefit both trading partners in welfare terms; this is referred to as “improving resource management.” Policymakers in international organizations should consider the relative dominance of externalities in the presence of weak property rights before requiring resource management as a condition for participating in international trade.
Article
We conduct a laboratory experiment to test a continuous-time model that represents a dynamic groundwater extraction problem in an infinite horizon. We compare the observations to the equilibrium path of the usual behaviours, for the case where the player is alone in extracting the resource (optimal control) and when two players extract the same resource simultaneously (differential game). We use a within-subjects design. This allows us to identify individual profiles of players playing alone and then characterize groups based on their composition with respect to these individual behaviours. We find that approximately a quarter of the players and groups succeed in playing (significantly) optimally, and none behave myopically. Moreover having an agent that behaved optimally in the control in the pair increases the likelihood that the group cooperates. We also identify other categories of players and groups that allows us to classify an additional 50% of the observations.
Article
Full-text available
We study the impact of discrete versus continuous time on the behavior of agents in the context of a dynamic common pool resource game. To this purpose, we consider a linear quadratic model and conduct a lab experiment in which agents exploit a renewable resource with an infinite horizon. We use a differential game for continuous time and derive its discrete time approximation. In the single agent setting, we fail to detect, on a battery of indicators, any difference between agents’ behavior in discrete and continuous time. Conversely, in the two-player setting, significantly more agents can be classified as myopic and end up with a low resource level in discrete time. Continuous time seems to allow for better cooperation and thus greater sustainability of the resource than does discrete time.
Article
Full-text available
Complex dynamic systems such as common-pool resource systems can undergo a critical shift at a given threshold, the so-called tipping point, which potentially requires substantial changes from the management system. We present in this research a framed laboratory experiment design to examine how the threat of economic sanctions influences the strategic management of a common-pool resource. We use the context of the East Atlantic bluefin tuna international fishery as it has been the archetype of an overfished and mismanaged fishery until a dramatic reinforcement of its regulations followed the threat of a trade ban. We consider endogenous threats and examine their effects on cooperation through harvest decisions taken in the context of non-cooperative game theory in which cooperation could be sustained using a trigger strategy. Our experiment results show that the threat of economic sanctions fosters more cooperative behaviors, less over-exploitation, and a more precautionary management of resources, reducing the economic rent dissipation. This result is exacerbated when the location of the tipping point that triggers the economic sanction is uncertain. In order to avoid free-riding behaviors and foster the emergence of a self-enforcing agreement, we suggest to introduce economic sanctions, such as trade restrictions, associated with uncertain biological limit reference points.
Article
This study addresses one of the most basic questions in renewable resource management: the ability of economic agents to exploit a renewable resource in an efficient and sustainable manner. In a laboratory experiment, subjects are presented with renewable resource extraction problems, where optimal management will lead to a stable steady state. A test of sustainability of the extraction practices shows that extraction behaviour results in steady states only 56% of the time. The mode of the steady state distribution coincides with the optimal steady state extraction. The trade-off between accruing a higher payoff in the present and sustaining the resource for future exploitation leads to suboptimal behaviours such as initial overextraction of the resource compared to the optimal extraction policy, costly downward adjustment of the extraction later in time, and settling down for lower long-run resource and extraction. The suboptimal behaviours lead to 17% loss in efficiency on average in terms of the accumulated payoff. We further look at extraction behaviour in terms of the degree of impatience it projects and find, based on their extraction decisions, that most of our subjects seem more impatient in managing their resource than is justified by the decision-making problem presented to them.
Article
The article summarizes key insights from four laboratory experiments to study renewable resource management. The commons problem, which is widely held to be the cause of mismanagement of common renewable resources, was ruled out by the design of the experiments. Still the participants overinvested and overutilized their resources. The explanation offered is systematic misperceptions of stocks and flows and of nonlinearities. The heuristics that people apply are intendedly rational for static, now resources, but not far dynamic, stock resources. Simplifying and reframing the management problem, by focusing on net growth rates, is suggested as a means to foster the use of more appropriate heuristics. Copyright (C) 2000 John Wiley & Sons, Ltd.
Article
This paper reviews research on dynamic decision making, i.e., decision making under conditions which require a series of decisions, where the decisions are not independent, where the state of the world changes, both autonomously and as a consequence of the decision maker's actions, and where the decisions have to be made in real time. It is difficult to find useful normative theories for these kinds of decisions, and research thus has to focus on descriptive issues. A general approach, based on control theory, is proposed as a means to organize research in the area. An experimental paradigm for the study of dynamic decision making, that of computer stimulated microworlds, is discussed, and two approaches using this paradigm are described: the individual differences approach, typical of German work in the tradition of research on complex problem solving, and the experimental approach. In studies following the former approach, the behavior of groups differing in performance is compared, either with respect to strategies or with respect to performance on psychological tests. The results show that there are wideinterindividual differences in performance, but no stable correlations between performance in microworlds and scores on traditional psychological tests have been found. Experimental research studying the effects of system characteristics, such as complexity and feedback delays, on dynamic decision making has shown that decision performance in dynamic tasks is strongly affected by feedback delays and whether or not the decisions have side effects. Although neither approach has led to any well-developed theory of dynamic decision making so far, the results nevertheless indicate that we are now able to produce highly reliable experimental results in thelaboratory, results that agree with those found in field studies of dynamic decision making. This shows that an important first step towards a better understanding of these phenomena has been taken.
Article
This paper reviews experimental approaches to dynamic decision behavior like, e.g., Toda’s fungus eater game, Edwards’ extension of SEU theory into dynamic decision making, storage control problems, the multistage betting game, and optional stopping problems. The respective merits and shortcomings of these approaches are discussed as well as some general problems of research on dynamic decision making.
Book
The author presents an introduction to the theory of biologial conservation, including a wealth of applications to the fishery and forestry industries. The mathematical modelling of the productive aspects of renewable-resource management is explained, including both economic and biological factors, with much attention paid to the optimal use of resource stocks over time. This book includes chapters on the theory of resource regulation and on stochastic resource models, sections on irreversible investment, game-theoretic models, and dynamic programming.
Article
"Research in progress on information seeking, intuitive statistics, sequential prediction, and Bayesian information processing is reviewed… . The relevance of mathematical developments in dynamic programming and Bayesian statistics to dynamic decision theory is examined. A man-computer system for probabilistic processing of fallible military information is discussed in some detail as an application of these ideas and as a setting and motivator for future research on human information processing and decision making." (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Previous laboratory experiments, using quite complex resource simulators, suggest that renewable resources are over-utilised because of a general tendency for people to systematically misperceive the dynamics of bioeconomic systems. Here, similar experiments with simplified simulators involving the management of reindeer rangelands are carried out. Sufficient information is given for the subjects to construct perfect mental models. Misperceptions persist for a simulator containing only the basic building block of all dynamic systems: one stock and two flows. Results deteriorate in a second treatment where a two-stock model is used. Compared to earlier studies using questionnaires, where subjects do not benefit from repeated outcome feedback, the experiments show that, even in these simple systems, information feedback is not sufficient to make up for misperceptions. Simulations are used to test two hypothesised decision rules: the optimal policy is rejected; a simple feedback rule is not. Altogether, the experiment and the simulations provide both a motivation for and an introduction to studies of system dynamics. Copyright © 2004 John Wiley & Sons, Ltd.
Article
Simulators and experiments studying dynamic decision making offer a way of finding out about factors enforcing and inhibiting human rationality. This literature review of 51 studies from the system dynamics field helps to identify various factors that influence decision making. The factors are classified into model, simulator and player characteristics. In the paper the effect of these variables on gaming performance is reviewed and synthesized. Model characteristics such as presence of delay and increase of feedback strength seem to lower performance, while changes in exogenous conditions lead to mixed results. With regard to simulator characteristics, the decision interval does not seem to influence performance. Model transparency has a positive relation to performance, similar to decision information (in conjunction with player characteristics). Lastly, with regard to player characteristics, there is some evidence that a long-term goal increases performance. Although problem-solving style and mental model characteristics impact performance, no relation between general personality types and performance is found. Also, there are no consistent differences between individuals and pairs with regard to performance. The paper closes with brief comments on future research directions. Copyright © 2004 John Wiley & Sons, Ltd.
Article
The article summarizes key insights from four laboratory experiments to study renewable resource management. The commons problem, which is widely held to be the cause of mismanagement of common renewable resources, was ruled out by the design of the experiments. Still the participants overinvested and overutilized their resources. The explanation offered is systematic misperceptions of stocks and flows and of nonlinearities. The heuristics that people apply are intendedly rational for static, flow resources, but not for dynamic, stock resources. Simplifying and reframing the management problem, by focusing on net growth rates, is suggested as a means to foster the use of more appropriate heuristics. Copyright © 2000 John Wiley & Sons, Ltd.