Shed light on dark feedback loops! - The consequences of not recognizing
feedback loops for the dominant management logic
35° International Conference of the System Dynamics Society
Cambridge, MA, July 2017
Martin FG. Schaffernicht
Facultad de Economia y Negocios
Universidad de Talca
Talca (Chile)
What you see is all there is
Daniel Kahneman
Abstract
This paper deals with feedback loops in mental models of dynamic systems (MMDS). However, untrained
individuals fail to recognize most of the loops inherent in by the causal structure they articulate: many loops
remain invisible like dark matter. This paper takes the example of recent research conceming the strategic
reasoning of vineyard executives. Out of nine participants, only three recognized any loops, but 90% of the
inherent loops — identified following the logic of the shortest independent loop set (SILS) - remained ‘dark’
to them. By identifying the connections between the inherent loops, it is shown that the impact of the dark
loops on the majority of variables in their mental models was not recognized. Based on these results, it is
argued that decision makers ought to be given qualitative modeling tools for articulation of mental models
which automatically detect and visually feed back the SILS loops. Also, the enhanced ‘distance ratio’
method ought to become able to process ‘dark’ loops. Eventually, the debate concerning loops in mental
models ought to be taken to the larger community of mental model research in management.
Keywords
Keywords: Mental model of dynamic systems, feedback loops, shortest independent loop set, strategic
management
Acknowledgments
This research has been funded by grant # 1140638 of the Fondo Nacional de Desarrollo Cientifico y
Tecnologico (FONDECYT) of the Comision Nacional de Investigacion Cientifica y Tecnologica (CONICYT)
of Chile
Shed light on dark feedback loops! - The consequences of not recognizing
feedback loops for the dominant management logic
1 Introduction
Steering a firm on an economically sustainable path in a competitive market requires executives to
understand and interact with dynamic situations. Multiple decision makers set goals involving the same
variables and simultaneously try to achieve them, each affecting the conditions under which the others take
their decisions. Each executive’s decision policies are based upon how they understand their situation, how
they describe the purpose and form of their industry and their firm, explaining their functioning and
observed behavior, enabling them to anticipate likely future developments. This description of their activity
almost literally conforms to an early definition of “mental model” (Rouse and Morris 1986), as cited in
(Mathieu, Heffner et al. 2000), which is much referred to in general mental model studies in management.
Since the complexity of a situation can only be absorbed by a mental model of equivalent complexity
(Ashby 1991), these mental models should consist of knowledge structures which are able to represent such
complexity. Standard mental model methods lead to maps of attributed causal structures consisting of
variables and causal links (Markdéczy and Goldberg 1995a). However, the behaviors of the business systems
are driven by the feedback loops and their interaction (Forrester 1969; J. Sterman 2000), leading to dynamic
complexity (Ozgin and Barlas 2015). In models, dynamic systems are represented as sets of interacting
feedback loops (S. N. Groesser and M. F. Schaffernicht 2012), which emerge when the change of one
variable depends on the quantity of another variable. Therefore, a mental model of a dynamic system
(MMDS) has three different levels of description: the elements (variables and causal links), the individual
feedback loops and the entire model (S. Groesser and M. Schaffernicht 2012).
Contrary to what might be expected according to this argument, many decision makers appear to understand
their situation in a way which almost completely lacks feedback loops. This is an observation frequently
made by system dynamics modelers working with clients. Of course, if loops are important drivers of a
system’s behavior, executives who fail to recognize them are in danger of not perceiving important drivers
in their decisions (Sterman 1989). This so-called misperception of feedback has been a focus of attention
in the literature, and empirical studies suggest it is a widespread problem (Sterman 2010).
Despite of being considered a relevant problem by researchers close to the field of system dynamics, the
mainstream mental model literature in management and organizational studies has not yet taken notice of
the role of feedback loops (Klimoski and Mohammed 1994; Langan-Fox et al. 2004; Langan-Fox et al.
2000a; Langan-Fox et al. 2000b; Langfield-Smith and Wirth 1992; Markdczy and Goldberg 1995b;
Mohammed et al. 2010; Mohammed et al. 2000). In the recent past, the ‘distance ratio’ method (Langfield-
Smith and Wirth 1992; Markdczy and Goldberg 1995b) has been adapted such as to take loops into account.
Its representation and analysis of MMDS keeps the advantages of representing models as directed graphs
(Langan-Fox et al. 2004), extending the methods for dealing with the three levels (M. Schaffernicht and S.
Groesser 2011), calculating an ‘Element Distance Ratio’ for variables and causal links, a ‘Loop Distance
Ratio’ for every pair of comparable loops and a ‘Model Distance Ratio’ for the entire set of loops
conforming the model. Analysis tools have been developed for the current version of this method (M.
Schaffernicht and S. Groesser 2014).
Data from a recent research project about the mental models of winemakers in the Maule region of Chile
(Fondecyt 1140638) goes along with the misperception of feedback thesis, but also call attention to an
aspect which has not been problematized and discussed as yet: the project finds that the causal structures
of the mental models articulated by executives and owners of vineyards contain many more loops than are
recognized by the executives. In other words, these executives recognize all the variables and causal links
the loops consist of, but they do not recognize their circular nature.
This article focuses on the differences between the ‘dark’ loops inherent in the model structure — as detected
using the logic of the shortest independent loop set (Oliva 2004) - but unrecognized and those recognized
by the executives. The differences which appear when comparing the recognized loops with the inherent
ones show how little of the feedback structure has been recognized by the owners of these mental models.
When an executive makes plans, these will only refer to what the executive’s mental model contains —
variables and relationships which are absent from the mental model are not taken into account, hence the
initial quote (Kahneman 2011). When feedback loops are not seen, many variables may be mistaken as
levers for discretionary management decisions, despite the fact of being subject to the autonomous influence
of the loops inherent in the management situation. The data presented here strongly point that way: each of
the 47 most relevant variables out of 446 belonged to one or several of 50 loops, but only 5 of these loops
have been recognized. Many influential variables are therefore understood in an input-output manner. They
may look as if they can be used as levers, but many ‘dark’(unrecognized) loops will feed back current
decisions into future behavior, and the behaviors will be surprising to the executive who does not see the
loops. Just like dark matter is unseen but influences the behavior of the astrophysical universe, dark loops
are unseen but influence the behavior of business systems.
Therefore, the existence of ‘dark’ loops severely reduces the executives’ ability to take into account relevant
sources of influence, thus diminishing the possibility of basing decision policies on a systematic analysis
of the underlying situation.
The article proposes several conclusions:
1) there is a need for tools to support on-the-spot construction of causal maps including automatic loop
detection and display;
there is also a need to enhance the extended ‘distance ratio’ method in order to assist comparison
between the set of recognized loops and the entire set of inherent looks, so to say to shed light on the
‘dark’ loops;
it is time to open a critical debate in the field of mental model research in management, in order to
establish feedback loops as important component of mental models.
The remainder of the article is organized as follows: section 2 introduces the methodical aspects. The mental
model data used is presented in section 3. Section 4 elaborates on the data discussing its implications and
leading to some new research challenges, before concluding in section 5.
2
3
2 Methods
2.1 Elicitation and coding
For inter-subject comparison of mental models, a number of methods and tools have been developed
(Langan-Fox et al. 2000a; Mohammed et al. 2000). Oftentimes, researchers elaborate a reference list of
available concepts in order to maintain the total number of variables in a range between 15 and 25 (Lim
and Klein 2006). However, only in-depth open interviews are able to not preclude participants to articulate
causal beliefs only because they are not contained in the previously established concept list (Mohammed et
al. 2000) — at the price of increasing the number of elements contained in the elicited mental models.
Therefore, in-depth interviews of one hour have been the data source of this study. The interviews were
framed by the question “what do you see coming in the regional wine industry over to coming 5 years or
so, and how do you steer your vineyard in that context?” The future-oriented question avoided post-fact
rationalizations, but otherwise left the choice of topics up to the interviewee.
The interviews were taped, transcribed and then coded in two iterations - first iteration open, second
iteration axial. The coding scheme provided instructions to look out for variables, causal links and feedback
loops (M. F. Schaffernicht and Groesser 2015). The resulting data ware mapped as a causal map.
2.2 Representation
A MMDS is represented by a set of variables, causal links (or links) and feedback loops (or loops). A
variable is represented by its name. A causal link has a direction and a polarity, which can be either positive
or negative (M. Schaffernicht 2010; J. Sterman 2000). Additionally, a link can have a delay mark, when it
is significantly slower than the other links to transmit an effect. A loop is a closed sequence of links
connecting a series of variables; it has a polarity, which is the consequence of constituting links’ polarity
and can be either positive or negative. To avoid confusion between loops and links, positive loops are called
reinforcing loops and negative loops balancing loops. A reinforcing loop generates exponential behavior,
whereas a balancing loop creates goal seeking behavior; failure to recognize feedback loops usually leads
to flawed expectations concerning a system’s behavior (J. D. Sterman 2000).
The general methodic approach has been published elsewhere (M. F. Schaffernicht and S. N. Groesser
2011) and is partially automated (M. F. Schaffernicht and S. N. Groesser 2014). A specific method has been
developed to assure that the aggregation level of the respective MMDSs is comparable. This method selects
variables which satisfy one of the following criteria: (1) a minimum number of interviewees mention the
variable, (2) the variable is characteristic for an individual interviewee, (3) the variable is an input or output
variable in at least one of the MMDSs, describing the business environment or an outcome variable (by
aggregating it away, some essential aspect of the interviewee’s mental model ceases to be taken into
account), or (4) the variable is needed to avoid that a loop contains only one single variable.
Variables which do not satisfy any of these criteria become aggregated: incoming links are redirected to the
following variable. For instance, if in the sequence innovation — new wines — differentiation, only the
first and the last variables are selected, then new wines will be hidden and the MMDS is represented as
innovation — differentiation '. The connection between these two variables is called a ‘path’ because it is
a sequence of links; its polarity is obtained by multiplying the polarities of the constituting links, positive
in this case. The path length in the example is equal to 2, and the path will be assigned a weight of 1/length
= 0.5. This method therefore redefines sequences of unselected intermediate variables and links as paths;
all relevant structural information is maintained because all connections between the selected variables in
the original MMDSs and all feedback loops in the original MMDSs are conserved.
2.3 Shortest independent loop set
In models of dynamic systems, most of the variables are contained in one or several loops. However, the
data in the current project contradict this statement: only 5 of the 50 loops have been recognized, suggesting
an input-output type of reasoning. This also means that the MMDS contain all variables and links to
constitute 45 more loops than the recognized set. Since by definition, the interviewee did not indicate the
feedback loops he or she did not recognize, a replicable way to identify these unrecognized feedback loops
is needed.
Usually, there is a hierarchy of loops, and several relatively short loops combine into larger ones. The total
number of loops quickly rises to unmanageable quantities, but if one can determine a loop set such as to
assure that each variable and link which belongs to any number of loops are taken into account at least
once, then none of the possible behavioral implications of any of the loops is eliminated by the fact of not
taking into account the combined loops. Such a set is called an independent loop set (Kampman 1996). The
so-called shortest independent loop set (SILS) selects the shortest loops first and thereby assures that
' The names of variables which belong to an MMDS are printed in italics to make clear these are not just words.
4
unnecessary complexity is avoided (Oliva 2004). Such a loop set is elaborated starting with one variable on
a short loop and searching the shortest possible path to come back to this variable; along the path, each link
used is marked. Upon returning to the initial variable, if there are any other loops with links unmarked so
far, the steps are repeated. If there are no more possible loops, another variable selected and the procedure
repeated.
2.4 Sample
The study was located in the Chilean wine industry, where many vineyard executives feel challenged: the
country’s wine industry has displayed a rapid surge starting in the mid ‘90, when labor and energy costs
were low, but two decades later, the steady rise of costs combined with a market category as “best price”
has put growing pressure on margins (Berrios, 2012 #47). According to data gathered by Vinos de Chile
(V. d. Chile 2015), the target prices aimed at by Wines of Chile for 2020 (W. o. Chile 2010) appear to be
impossible to reach. Under such conditions, two distinct archetypical kinds of business policy come to mind
for defending profits: increase the price of wine or decrease the unit costs — by increasing the sales volume
2. Leaning more towards one or the other of these policies depends on how the executive represents his
vineyard in the context of the wine industry, in other words: on his mental model.
The study was driven by the question ‘what do the mental models of these vineyard executives contain?’
of the approximately 300 vineyards in Chile, roughly 100, mostly small to medium sized, are based in the
Maule Region. A sample of these regional vineyards has participated in a study striving to measure and
analyze their owner’s or executive’s mental model. An interview-based study is time intensive; therefore,
asmall sample of vineyards was sought for. The University of Talca in the Maule Region is deeply involved
with the regional wine industry by means of its Wine Center. Leading experts from this Center identified a
sample of 9 vineyards along two dimensions: innovation oriented versus traditional and small versus
medium size. The resulting selection includes vineyards of different sizes (from very small to medium) and
ages (from start-up to several generations of family ownership), as shown in the following table:
Table 1: The vineyard sample
Type Size
Small Medium
Innovative 2,6,8 17
Mainstream 3 4,5,6
? Words referring to the behavior of variables are printed in fat letters to distinguish them from the variables
themselves and to make the causal relationships between variables more salient.
5
3 Data
3.1 Overview
The original MMDS consisted of 446 variables and 2087 links, and the aggregation process lead to
aggregated MMDS with a total of 219 variables and 551 links. Table 2 shows the mean numbers of variables
and of links for each of the MMDS:
Table 2: Global characteristics of the MMDS
MMDS_Vars_ Links _Links/Vars ratio
1 34 62 1.82
2 36 55 1.53
3 26 44 1.69
4 35 63 1.80
- 36 48 1.33
6 52 72 1.38
4 36 51 1.42
8 41 55 1.34
9 53 101 1.91
Mean 38.78 61.22 1.58
SDev_ 8.70 17.16 0.23
The aggregated MMDS are approximately half the size of the original ones in terms of variables and links,
but the average of 38 variables is still greater than what many mental model studies use when they define
default variable lists {Lim, 2006 #20}. The number of links is roughly 60% higher than the variable count,
indicating than most but not all variables participate in several links. The analysis of the intersections
between the variable sets of the respective MMDS lead to the recognition of two groups of MMDS: group
1(G1) contains MMDS 1, 3 and 5, while group 2 (G2) assembles MMDS 2, 4, 6, 6, 7, 8, and 9. Table 3
contains three subsets of variables which are contained in at least half of the MMDS belonging to the
respective groups:
Table 3: Variables in the two groups of MMDS
Variables mentioned by both groups (18 variables):
price of wine, costs, production costs, sales, profits, differentiation, production, dominance of large vineyards,
price of grape, local partnerships between viney , revenues, costs, efforts abroad,
market power of large vineyards, average quality demanded, fair trade, personnel costs, alternatives to wine
Only group 1 (34 variables):
vineyard size, wine category, economic
growth in major markets, distributors,
identification with the
associativity, degree of
country brand,
j of
Only group 2 (47 variables):
demand for wine, importance of volume in the majority
business model, energy costs, organic vine production,
innovation, mechanization, territorial rootedness, production
company size, international per capita
consumption, liquidity, purchase of grapes
from other producers, authentic narrative,
number of artisan vineyards, wine culture,
domestic per capita consumption,
identification with the region brand,
professionalization
area, in ig capacity, quality,
consumption per capita, innovative wines, value added,
production per hectare, artificial inputs, demand for cellaring,
consumer desire for novelties, demand for wine with
sustainable production, frauds, natural conditions, leasing of
wine cellar, recognizability of the wine, taxes, failures, labor
supply, sustainability, rural desertion, time to market
In terms of the 551 links and paths, only a relatively small fraction is shared by the MMDS of both groups:
138 (25%) are only part of G1, 381 (69%) belong to G2 exclusively, and 32 (6%) links are shared across
the groups. An analysis of links and paths is certainly interesting, but beyond the scope of this paper, which
focuses on the loops. One relevant finding was that even if the MMDSs of both groups include the price of
wine, G1 MMDS tend to take it as exogenously driven (by market forces), whereas G2 MMDSs use diverse
intra-vineyard variables to influence it.
A look at the loops inherent and recognized in each of the MMDS (Table 4) reveals that 4 of the 9 MMDS
have five or more inherent loops:
Table 4: Number of inherent loops per MMDS
MMDS Loops Recognized Dark
1 1 0 1
2 3 0 3
3 5 0 5
4 13 2 11
5 2 0 2
6 14 1 13
7 0 0 0
8 2 0 2
9 10 2 8
It also becomes clear that the inherent loops are mainly ‘dark’ ones, remaining invisible for the interviewee.
Consider now the sequences of variables, the polarity (Pol) and delayed links (Del) of the loops. Table 5
also indicates if the loops have been recognized or not (if Rec is 1, a loop has been recognized). The loops
are ordered by MMDS:
Table 5: Inherent feedback loops detected by SILS logic
Loop
MMDS ID Rec Var Names E Pol Del
1 101 0 distributors ++ degree of adjustment of company size > B 0
2 201 0 price of wine > innovation > differentiation > price of wine B 0
202 0 demand for wine + shortage + demand for wine R 0
203 0 demand for wine — production — shortage + demand for wine B 0
3 301 0 sales — price of grape — distributors — sales R 0
302 0 sales — liquidity + distributors + B 0
303 0 sales — total debt with producers — liquidity distributors + B 0
304 0 total debt with producers — liquidity + total debt with producers B 0
305 0 liquidity debt with third parties — liquidity R 0
4 401 1 price of wine > revenues — profitability + quality + R 0
402 1 price of wine > revenues — profitability + production quality R 0
403 0 price of wine > production > B 0
404 0 price of wine > sales + revenues — profitability + quality > B 0
405 0 price of wine sales + costs — profitability + quality > R 0
406 0 price of wine > sales > marketing costs + costs — profitability R 0
— quality >
407 0 price of wine costs — profitability + quality + B 0
408 0 price of wine > personnel costs —> production costs + costs > 1
profitability + quality >
409 0 price of wine > personnel costs — mechanization production R 1
410 0 price of wine sales + personnel costs —> production costs > B 1
costs — profitability + quality +
411 0 personnel costs — mechanization — personnel costs B 0
412 0 costs — energy costs + production costs + B 1
413 0 costs + personnel costs > production costs > B 1
5 501 0 costs + marketing efforts abroad > B 0
5 502 0 market power of large vineyards — growth rate of big vineyards + R 0
6 601 1 profits + production > B 0
602 0 profits profitability + production > B 0
603 0 profits + production + production costs > B 0
604 0 profits + price of grape + quality > price of wine > R 0
605 0 profits + production > quality > price of wine > R 0
606 0 profits + frauds + quality > price of wine > R 0
607 0 profits + frauds — production — quality > price of wine > R 0
608 0 = quality + categorization on the market importance of volume in R 1
the majority business model — price of grape >
609 0 = quality + categorization on the market importance of volume in R 1
the majority business model > production +
610 0 = quality categorization on the market > importance of volume in R 1
the majority business model > frauds >
611 0 categorization on the market > importance of volume in the R 1
majority business model >
612 0 categorization on the market > importance of volume in the R 1
majority business model — territorial rootedness — differentiation
613 0 territorial rootedness — price of wine — profits + production > R 1
quality categorization on the market + importance of volume in
the majority business model >
614 0 territorial rootedness — differentiation + price of wine — profits R 2
— production + quality > categorization on the market >
importance of volume in the majority business model +
8 801 0 production — artificial inputs — production costs + production B 1
802 0 vineyard size + growth target > R 0
9 901 1 profits + profit sharing + mutual trust with producers > R 0
902 1 purchase of grapes from associated producers — marketing for R 0
the company — estimated marketing capacity >
903 0 profits + profit sharing + mutual trust with producers costs B 1
904 0 profits + profitsharing + mutual trust with producers grape B 1
producers related to each other — purchase of grapes from other
producers >
905 0 profit sharing — mutual trust with producers + grape producers R 2
related to each other > purchase of grapes from other producers
906 0 mutual trust with producers + grape producers related to each R 2
other — purchase of grapes from other producers >
907 0 investment in winemaking capacity + storage capacity + B 0
908 0 investment in winemaking capacity + storage capacity + R 1
vinification for own vineyard >
909 0 investmentin marketing skills + marketability + B 0
910 0 opportunity to sell wine to other vineyards — sales of wine to other R 1
vineyards >
It turns out that each of the variables contained in one or several loops belongs to the set of 219 variables
in the aggregated MMDSs. This means that each of them is relevant because it is mentioned in a sufficient
number of MMDS and/or because many causal links or paths originate in it. Of course, these variables are
not all equally relevant. Also, the number of MMDS where they belong to loops, as well as the number o
loops they are contained in show some variation.
The following Table 6 shows these variables, ordered by (1) the number of feedback loops they belong to,
(2) their degree of relevance in terms of MMDSs and links or paths. For the sake of displaying the table in
a single page, the column names have been abbreviated: M = number of MMDS including the variable; LP
= number of causal links or paths originating in this variable; FL = Number of feedback loops containing
the variable; R = number of the loops which have been recognized; MMDSs = Number of MMDS
containing these loops; MMDS IDs = where the loops appear; G1 = some loops are in MMDS of Group 1;
G2 =some loops are in MMDS of Group 2.
Table 6:The most relevant variables are contained in dark loops
Variable M LP FL R MMDS MMDS G1 G2
quality 3 1 2
price of wine 9 12 ~#«217 2 3
production
profits
profitability
costs
categorization on the market
importance of volume in the majority business model
sales
production costs
mutual trust with producers
personnel costs
profit sharing
distributors
liquidity
grape producers related to each other
frauds
purchase of grapes from other producers
revenues
territorial rootedness
differentiation
price of grape
shortage
total debt with producers
investment in winemaking capacity
storage capacity
mechanization
demand for wine
debt with third parties
growth target
marketability
sales of wine to other vineyards
degree of adjustment of company size
energy costs
estimated marketing capacity
marketing costs
purchase of grapes from associated producers
artificial inputs
BRR BRE
Bee
B
NB Oe BW eee RoW ee or ON en We oo Bo Ww ou
B
WRNNNN EE EHOW RR HH OOO RR UH RW RW ROO ROW OS
Be REN NN WlWlwlulWlw wa ela Uo SOO DE
OHoKM OOD COCO COCO OCOD OND OOOOH OH OCC OONNN
Fa Fa frre ra rae ry rar Fe ry FFF) FP Ca FO Fa OPE PN
DORO ARH OWOOWN ROW WN
a
BeBe
growth rate of big vineyards 1 3 1 0 1 5 1
investment in marketing skills 1 3 1 0 1 9 1
market power of large vineyards 4 3 1 0 1 5 1
marketing for the company 1 3 1 1 1 9 1
opportunity to sell wine to other vineyards 1 3 1 0 1 9 1
vinification for own vineyard 1 3 1 0 1 9 1
marketing efforts abroad 1 4 1 0 1 5 1
innovation 3.7 1 0 1 2 1
vineyard size 3.12 1 0 1 8 1
Visual inspection of Table 6 reveals that where the number of feedback loops (FL) is high, the number of
links or paths (LP) is high, too: the more links or paths originate in a variable, the more feedback loops it
is contained in. This relationship is shown in Figure 1:
18
e e
16
14
e
ry
g 12 ®
te)
ZY 10
rey e e
8s
@
$5 wf
ms e
4 e
a dae e e
2 ° °
e e e
0
0 2 4 6 8 10 12 14
Links or paths originated
Figure 1: The relationship between the number of links or paths of a variable and the number of loops it is
contained in is positive
Given the observation that only 5 out of the 50 feedback loops have been recognized by the interviewed
executives (see Table 5), this allows the following proposition:
Most of the relevant variables are influenced by multiple ‘dark’ loops, and the MMDS
owner did not recognize this.
Keeping in mind that the MMDSs of group 1 (G1) tend to consider the price of wine as driven by factors
they cannot govern, whereas group 2 MMDSs tend to believe the price of wine can be influenced by internal
decisions, it is also interesting to observe the columns G1 and G2. It turns out that only 8 variables are
contained in loops belonging exclusively to G1 MMDSs, whereas 36 are affected by loops only contained
in MMDSs of G2 (and 5 belong to both groups). Only the case of distributors affects more than one single
MMDS (two MMDS out of the three which contain the variable). In comparison, the price of wine belongs
to each of the 9 MMDSs — so each of the interviewees in each of the groups referred to it - but the 17 loops
10
it is contained in belong to 3 MMDS from G2. This reinforces the observation that G1 executives did not
see the a reasonable chance to influence the price of wine, but the G2 executives do — even if they did not
recognize the circular nature of their corresponding arguments. Another very relevant variable — costs —
belongs to 8 MMDS, but its 9 loops belong to both groups.
Consider the following two sets of loops shown in Table 7, which either contain the price of wine, the costs
or both. Only 2 of the 17 loops containing the price of wine have been recognized, and all 9 lops containing
costs have remained dark (unrecognized). Both variables are printed fat to make them more salient:
Table 7: Loops containing price of wine and loops containing costs
Rec i Pol Del Price Costs
0 __ price of wine — innovation — differentiation > BO 1
1__ price of wine — revenues — profitability quality — R 0 a
1__ priceof wine — revenues — profitability ~ production > quality > R 0 1
0 price of wine — production + BO 1
0 price of wine — sales — revenues — profitability + quality BO el
0 price of wine — sales — costs — profitability quality price of wine R 0 a 1
0 price of wine — sales — marketing costs + costs — profitability quality R 0 1 1
0 price of wine — costs — profitability quality + price of wine BO 1 1
price of wine — personnel costs — production costs + costs — profitability +
0 quality > R11 1
0 price of wine — personnel costs — mechanization — production > R 1 1
price of wine — sales — personnel costs — production costs + costs >
0 profitability quality + Bol el 1
0 costs — energy costs — production costs > Bool 1
0 costs — personnel costs — production costs + B 1 1
0 costs — marketing efforts abroad > BO 1
0 profits + price of grape > quality > price of wine — R 0 1
0 profits + production — quality price of wine — R 0 2
0 profits + frauds + quality > price of wine — R 0 1
0 profits + frauds + production — quality > price of wine — profits RO 1
territorial rootedness — price of wine — profits + production > quality >
categorization on the market + importance of volume in the majority business
0 model Rol 1
territorial rootedness — differentiation — price of wine — profits — production
Q = quality — categorization on the market > R 2 1
0 profits — profitsharing — mutual trust with producers > costs > Boo 1
Costs is a variable under the influence of 9 unrecognized lops: while the MMDS owners reason about costs
in an input-output manner, any change they manage to achieve in costs will be fed back and have future
consequences they apparently do not consider. In other words, some of the costs behavior may surprise
them because it has been unintendedly caused by past decisions. In a very similar way, the same holds for
production costs, personnel costs, marketing costs, production, sales, demand for wine and profits - to
mention only the most shared ones (refer to Table 5 and Table 6).
For the MMDS of G1, the price of wine is under external influence and therefore one may agree that in
such cases, the MMDS display an input-output architecture without loops. However, in the MMDSs of G2,
changes in the price of wine will trigger 15 unrecognized loops, while the vineyard executives (of G2)
expect the price of wine to be to be open to discretionary management measures: this is again input-output
11
reasoning, but this time in the presence of as many as 15 more feedback loops than what has been recognized
by the executives.
But the problem of not seeing dark loops is not limited to isolated variables. As Table 7 reveals, the price
of wine and costs are connected to one another by feedback loops which contain both variables. To illustrate
this, we will analyze the case of the two MMDSs with the greatest number of loops: MMDSs 4 and 6,
respectively. Since this analysis focuses exclusively on the loops, all variables which do not belong to the
loops are suppressed — even if the belong to the MMDS.
3.2 MMDS4
InaMMDS, variables are simultaneously interconnected by different causal links; the entire model is easier
to grasp when represented as a causal loop diagram (M. Schaffernicht 2010; J. Sterman 2000). Figure 2
shows MMDS4 as a tightly interconnected set of variables, in which two loops are specifically labeled
because they have been mentioned by the interviewee (the respective causal links are highlighted to make
these loops more salient). Links appear as solid arrows, and paths as dotted arrows:
+
> production <t-
mechanization
revenues =.
Ae *
gyre Se
a
sales
+ marketing oe — Gye
406
Figure 2: Causal map of MMDS¢4 with recognized and ‘dark’ loops
12
Figure 2 allows to see the whole set of loops, but also to examine each loop separately. The very nature of
loops has it that there is no beginning and no end; however, selecting a variable at the crossing of several
loops is a good heuristic to start somewhere. In the case of MMDS4, many loops contain the price of wine.
Loop 401 describes how an increase of the price of wine increases the vineyard’s revenues (note that this
is a path, which in the original MMDS4 passes through revenues from wine sales) and subsequently
increases profitability. This allows them to decrease production over another path (reducing their grape
production per hectare); when there are less grapes per hectare, the quality of the resulting wine increases
and consequently there arises a pressure to increase the price of wine. Thus any increase of the price of
wine will lead to a further increase of the price of wine: this loop is reinforcing — its causal structure will
reinforce an initial change will lead to another change with the same sign.
In loop 402, the same increase in profitability will (also decrease production per hectare and therefore)
lead to a reduced production, which will — all other things equal — lead to an increased price of wine. This
is a reinforcing loop, too. Note that by the same token, any decrease of the price of wine will lead to
pressures towards increasing the price of wine: just like Janus hat two faces, reinforcing loops have two
operational modes: virtuous or vicious.
The remaining variables and causal links have been mentioned by the interviewees without explicitly
connecting it to these two loops are recognizing how they build a house set of other feedback loops. Figure
2 has been laid out in a way which facilitates visual recognition of loops; note that the interviewee does not
receive such visual feedback during the interview. However salient these cyclical visual arrangements may
be, some of these loops are quite long and partially overlap with other loops. The list of variables per loop
was shown in Table 5 (above).
Loop 403 also has to do with the price of wine: when it increases, there will be an increase in production
(because increased incentives to grow grapes lead to an augmented planted area), which in tun will
decrease the price of wine. Thus any increase of the price of wine will lead to a subsequent decrease of
the price of wine. Since the sign of the initial change is inverted, this is a negative or “balancing feedback”
loop (labeled with a “B”), expressing the fact that in the (social) system, there is a process whose influence
on the price of wine limits the decision maker’s ability to influence the price of wine. Loop 404 adds the
inverse relationship between price of wine and sales as a balancing loop. Loops 405 and 406 articulate the
observation that selling costs money and therefore sales increase costs, yielding two very similar balancing
loops (405 including a longer path, 406 mentioning marketing costs explicitly). Loops 407 and 408 state
that a higher price of wine indirectly (path) leads to increased costs, and that a higher price of wine leads
to increased personnel costs, too. Both loop state a decrease of profitability with the ensuing decrease of
the price of wine: balancing loops. Loops 409 and 411 cycle around the increase of mechanization in order
to decrease personnel costs (411 is a balancing loop), which at the same time increases production because
of increased productivity (409 is reinforcing). Loop 410 is balancing and describes that an increase in
personnel costs resulting from an increased price of wine increases costs which reduces profitability and
leads to a decrease of the price of wine. Eventually, loops 412 and 413 are balancing and describe how
energy costs and personnel costs are being controlled.
Itis relevant to note that the first four loops (401 — 404) focus mainly on the price of wine, loops 405 — 411
deal with the price of wine and with costs, and loops 411 — 413 concentrate on costs.
3.3 MMDS6
In the case of the second MMDS, there is one recognized loop. Loop 601 is balancing and represents the
inverse relationship between the profits and production. Loop 602 counters this with the observation that
increased profits increase profitability and attract additional production. 603 adds that increasing
13
production also increases production costs, diminishing profits. Loop 604 states that increased profits tend
to increase the price of grape, which leads to an increase in the quality (of the resulting wine) and increases
the price of wine, thereby further increasing profits. Recall that the same loop will also reinforce a possible
decrease of the price of wine. Loop 605 articulates the reaction of quality to changes in production:
increasing the production (volume) tends to decrease the quality, which decreases the price of wine and
diminishes profits, then leading to a further increase in production: a reinforcing loop. Frauds are
introduced in loop 606: a decrease in profits will increase frauds, which will decrease quality, then
diminish the price of wine and finally further decrease profits: this looks like the vicious side of a
reinforcing loop. Frauds also tend to increase production, which connects loop 607 to the same logic which
characterizes loop 605.
Loops 608-610 build a delayed connection between quality and the categorization on the market. An
increased quality slowly increases the categorization, which reduces the importance of volume in the
majority business model: (sales) volume is less important when you market a highly ranked wine. This
leads to a more generous price of grape, less frauds and less production in general. The consequence will
be a higher quality: a threefold reinforcing process. A gain, this may as well lead into a spiral of decreasing
categorization! Loops 611-614 describe how the importance of the volume in the business model interacts
with the territorial rootedness (how strongly the vineyard and its people feel connected with their terroir),
with differentiation and with the price of wine; these are reinforcing loops, which can therefore either
decrease the importance of the volume or increase it.
categorization on
the market
importance of volume in
the majority business
model
differentiation St :
\_*™ temritorial rootedness
price of grape
Figure 3: Causal map of MMDS6 with recognized and ‘dark’ loops
4 Discussion
4.1 The coverage of inherent loops in MMDS4
For single causal links and paths, the ‘ceteris paribus’ condition is usually applied. However, variables like
the price of wine or costs belong to multiple loops (Table 7), some of them reinforcing any change, others
balancing changes. So ‘ceteris paribus’ has to be dropped when considering the entire business situation.
By not recognizing 8 of the 10 loops directly affecting the price of wine, an executive may think of variables
like personnel costs or mechanization as levers which can be used in an input-output manner. But if a
discretionary decision concerning mechanization has a ‘side effect’ in the price of wine, an individual who
believes such a connection does not exist will not recognize the consequences. Even more so, he cannot
take into account the consequences of the fact that many loops are interconnected by the price of wine and
by costs, and that it is impossible to influence any of their variables without triggering effects in all loops
and their respective variables.
The question arises if all loops in MMDS5 are interconnected or if there are independent groups. Since any
intersection between a pair of loops means that they have at least one variable in common, inspection of
how the variables intersect across the MMDS allows to answer this question. The following Table 8
contains the variables of MMDS1 and indicates their presence in each of its ten feedback loops. The 5
variables in the upper section of the table belong to the loops recognized by the interviewee (loops 1 and
2), while the 7 variables in the lower section do not belong to any recognized loop.
Table 8: Presence of variables in feedback loops, MMDS4
Variables Loops
1 2 = 3 #4 5 67 8 9 10 11 12 13
profitability 161 1 4, 214 1
price of wine 11 1 1 & DIkbh 1
production 1 1 1
quality 161 bo ob 111 1
revenues 1 o4 1
costs 1111 1 11
energy costs 1
marketing costs ai
mechanization 1
personnel costs hh Lf a8
production costs 1 1 1 &
sales a: 4 2 1
Inspection of the upper part of Table 8 reveals that loops 1 through 10 heavily intersect: the price of wine
belongs to each of these loops, and each of the other variables is either in loops 1, 2 and 4-10 or in loops 2,
3 and 9. Loops 4 - 10 also have variables in the lower section of the table — these variables have been
mentioned by the interviewee, but the loops they belong to have not been recognized. These variables, in
tum, belong to the loops 4 through 10. This means that if something is changed in a variable belonging to
one of the loops 1 or 2 — which have been recognized by the interviewee, the price of wine will be
influenced; this will also trigger the loops 3 through 10, which will then also trigger loops 11 through 14.
The connections between the loops can be represented as an adjacency matrix, where there is a row anda
column for each loop. If there is a connection from the loop in row r to the loop in column c, then the cell;
is set to 1 (blank or cero otherwise). Each pair of 1 in Table 8 is such a connection; for instance, the first
15
row (variable global supply) connects loops 1, 4, 5 and 6 to one another. Table 9 shows the adjacency
matrix for the loops in MMDS1 in the left section (zeros have been replaced by blanks for visibility):
Table 9: Adjacency and distance matrices for the loops in MMDS4
Adjacency matrix of the loops in MMDS 4 Distance matrix of the loops in MMDS 4
Loop 123456789 10 11 12 13 Loop 1234567 89 10 11 12 13
af liiliiiiid 1 11iiliiiiidi2 2 3
2 1 1ililiiidl 2 {1 1211212121121 2 2 3
3 da iLiad 2 ijk L20422.7 222 3
4 ~LLI 2Lidtad @jliadt ttiiadti2 23
5 LLId Wid. 5 (lth DPHeLI& 2 2 B
6 Mata: a a, a 6 |LI2Lt LLIt 22 3
7 LIDEhh Ya 7 7 |IsIPLIT 21 2 2 8
8 L1litid bd i 1 B ltr ie tadzta 2
9 11iiiiiii 11 1 9 j11111111 1112
lo jl111211111 10 j121111111 2234
11 14,2 1 1142222222111 21
12 1.11 1 4. 1242222111121 2 1
13 Li 1 1 13) (222212112 1 1
Visual inspection of the adjacency matrix suggests that the loops 1 through 10 are massively interconnected,
and that the connections are lose between loops 11 through 14. There is no direct connection from the loops
1-7 to any of the loops 8-13. However, the adjacency matrix contains only 0 and 1, representing only the
direct connections between the loops. Loops which are more than one step away, can be made visible in a
distance matrix; such a matrix has the same structure as the adjacency matrix, but its cells contain the
number of steps to reach the loop represented in column c from the loop represented in row r. To construct
the distance matrix, one can directly use the data in the adjacency matrix. For example, cell:1: in the
adjacency matrix is equal to zero: there is no direct connection from loop 1 to loop 11. But cells in the
adjacency matrix is equal to 1; therefore inspect row 8 (loop 8), column 11, which is equal to 1. This means
that in 2 steps, one can reach loop 11 starting at loop 1, and therefore the cell: 1: in the distance matrix
receives a 2. The distance matrix of the loops in MMDS (right section in Table 9) clearly reveals that all
loops of MMDS can be connected in 1, 2 or 3 steps. This means that any change to any variable in any loop
will affect all variables of all loops.
4.2 The coverage of inherent loops in MMDS6
Analogously, most of the variables in MMDS6 belong to more than one feedback loop, as shown in Table
10. Similarly to MMDS4, the upper block contains the 2 variables which belong to the two recognized
loops, whereas the 10 variables belonging to unrecognized loops are in the lower block. The upper block
variables are mainly part of the loops 1 — 9 and 13-14, but only loops 10-12 are not directly connected to
these loops. However, each of the variables in the lower part of the Table also connects with either one of
the two recognized lops, or with another lower part low which in tum has such a connection.
Table 10: Presence of variables in feedback loops, MMDS6
Variables Loops
1 2° 3 4 5 6 789 10 11 12 13 14
production 111 1 1 1 11
profits 1121212222 141
categorization on the market 1122123122121
profitability al
differentiation 1
frauds 1
importance of volume in the majority business model 112122122212
price of grape 1 1
price of wine 11 11 11
production costs 1
quality 11 a 1 A, +t
territorial rootedness 1 1
The adjacency matrix of the loops of MMDS6 shown in
Table 11 suggests that there are two groups of loops: 1 — 9 and 10 — 14 are tightly interconnected amongst
one another, but with little connection between the groups. However, the distance matrix reveals that the
loops of these two groups are no further than 2 steps apart from one another. Just like in the case of MMDS4,
the second interviewee did recognize only a small minority of the loops inherent in the structure of the
variables and causal links he articulated during the interview.
Table 11: Adjacency and distance matrices for the loops in MMDS
Adjacency matrix of the loops in MMDS 6 Distance matrix of the loops in MMDS 6
Loop 1234567 89 10 11 12 13 14 Loop 1234567 89 10 11 12 13 14
1 TEnwhIYT 1 11 1 PRIS Pista 1 22 2 2
2 | Akbar . 11 2il FIidttidii i 22 2 2
3 11 oa1111 1 11 3 J11 2211111 22 2 2
4 111 111 11 4/111 111211 22 2 2
5 Litd i2 21 11 & jlLl11 Lidd 12:2 2 2
6 111i 1 os. & jTL1I£ 122 Be 2 B
7 LIighigl 4 I 1 BULLI TTL 2 2 2
8 aL 12h @ 1 BLLL2LLL PLkbh bh 2
9 Lit kh Ft LL 2 1 9 JITLITILI 22 2b 2 7
10 11 tadaiit 10 jITUbliiiidi bk b 2
11 111 111 1 6(2221222111 1, 1, 2
12 1111 11 22 (22222221111 Y tt
13; Jl1111111111da1i21 1 13) |(2:22:2222:21 1. 1 1 A,
46$[2i1ili1iiiiiliiil a 14, [2.225222 2511 1 1. 1. 2
4.3 The of not izing dark feedback loops
Each of the executives recognized that the variables belonging to the recognized loops are driven by these
loops, as virtuous cycles or as vicious cycles. Variables not belonging to any of the recognized loops could
then be influenced in a one-off manner. For instance, the first executive could decide to increase
mechanization to increase labor efficiency (this intermediate step is aggregated in the path shown in Figure
2 and Figure 4), thereby reducing the personnel costs to reduce the production costs and diminish costs.
There is a manifest difference between thinking to increase mechanization in order to reduce costs and
increase profitability, like displayed in Figure 4, and keeping in mind that a change to mechanization will
trigger all loops shown in Figure 2 and set off cyclical changes in all variables.
ersonnel costs
»?
) +
mechanization production costs nt production
+
R
costs lity “402
ae wality
price of wine
revenues <---
he
Figure 4: Causal diagram of a plan to increase profitability
Analogous reasoning could be made with costs, energy costs, marketing costs, mechanization, personnel
costs, production costs, sales (MMDS4) and categorization on the market, profitability, differentiation,
frauds, importance of volume in the majority business model, price of grape, price of wine, production
costs, quality, territorial rootedness (MMDS6). Considering that out of the dark loops, 3 and 11
respectively are reinforcing in nature, instability may be caused without the ability to recognized its
endogenous cause. Of course, without quantification and simulation, it is not possible to know how
important the combined effects of these dark loops will be in each case.
This limitation notwithstanding, any change to one of the variables in MMDS4 or MMDS6 will not only
have several side effects, but also come back to the initial variable in several ways. One single decision will
trigger sequences of behavior change, rather than one single event. Taking these variables as possible levers
for management decisions in an input-output way of reasoning will not be able to take this into account.
Mental models are the structure used to reason through different possible decisions (Johnson-Laird 2001),
but if the structure of the mental model is only partially recognized and understood, then many decisions
18
may be derived rather by repeating decisions from past experience than from systematically analyzing the
given possibilities.
Summing up, the interviewees behind MMDS4 and MMDS6 recognized only 20% and 13% of the inherent
loops in their MMDS, respectively. This made 57% and 77% of the variables mistakenly seem to be free
of the influence of feedback loops. The analysis of the two exemplary MMDSs makes a strong case for the
proposition that
Untrained decisi kers and pl need assi to recognize dark loops.
This is not diminished by the observation that the remaining MMDS have less dark loops — even that one
of the MMDS did not have any such loop at all. The fact remains that in almost all of the MMDSs, the
majority of the relevant variables was contained in dark loops.
4.4 Challenges for research
Therefore there is a need for tools helping executives to recognized what their mental models contain. It is
inherently helpful to use an external “boundary object” (Black 2013). The “cognitive mapping” thread has
contributed some tools (Eden 2004), but these are input-output oriented and do not help to recognize
feedback loops. System dynamics is based upon feedback loops (Forrester 1969; J. Sterman 2000), but its
tools are aimed at simulation modelling and demand too much training to be a tool for executives.
Qualitative modeling of feedback-rich situations is prone to misinterpreting behavioral consequences (M.
Schaffernicht 2010); however, not being able to determine plausible behaviors from a qualitative causal
loop diagram is preferable to not even recognizing the loops in the first place. Clearly, a tool able to
automatically detect feedback loops and classify their polarity while a executive or a management team
work through a problem — identifying variables and causal links — would give executives the opportunity
to more thoroughly consider possible strategies, decision policies or plans.
While the main software tools for system dynamics diagramming and modeling include a feature which
detects if a given variable belongs to loos (and which ones), this function has to be invoked by the modeler
each time, it refers only to the selected variable(s) and it is presented as pop-up information which becomes
invisible as soon as the modeler clicks on the diagraming canvas. This is not a trivial user interface problem
to solve because not only are there multiple loops, but on top of it most causal links participate in several
loops at the same time. One way to feed this information graphically back to the user/modeler can be to
define a separate color for each loop and draw each link one per loop, with its respective color. Additionally,
the ID of each loop can be displayed along the causal links. It may even become useful to create a new kind
of diagram, where the loops are represented as nodes in a network (or a directed graph), and arrows showing
the connections between them. Figure 5 shows how this might look in the case of the first 4 loops in MMDS
4, and how interconnected the network of loops of MMDS 4 is.
+ + P.
401 price of wine
404%, Ae)
sales
a) Causal diagram showing loops b) Network diagram of the loops in MMDS 4.
(part of MMDS 4). Dotted arrows Nodes represent loops. Arrows represent
are paths. Color indicates loop ID. connections because of shared variables.
Figure 5: Options for assisting decision makers and modelers
One challenge for the programs running beneath the user interface would be the automatic detection of
loops including the recalculation of the shortest independent loop-set. This may be rendered even more
difficult if the user/modeler wishes to rearrange the loop set because his business logic and his mental model
do not strictly comply to the rules and conditions of a shortest independent loop set.
This is certainly a challenge for software developers, but such a tool would doubtlessly be of substantial
help for planners, decision makers and consultants.
There is also a need for tools supporting researchers investigating the mental models of executives and
other decision makers. The need for taking feedback loops into account has been argued for (S. N. Groesser
and M. F. Schaffernicht 2012) and the so-called ‘distance ratio’ method (Markdczy and Goldberg 1995b)
has been adapted (M. F. Schaffernicht and S. N. Groesser 2011), but the current tools (M. F. Schaffernicht
and S. N. Groesser 2014) only take into account the feedback loops recognized by the interviewees. The
20
“loop distance ratio” (LDR) compares the elements (variables and causal links), the polarity and the delays
of those loops which the interviewees have recognized and the researchers deem to be equivalent. For
instance, neither of the two recognized loops in MMDS4 is equivalent to any of the recognized loops in
MMDS6, and therefore the LDR would be 100% (maximum distance, in other words: completely different
MMDS). However, amongst the unrecognized loops, there are equivalences, like displayed in Table 12:
Table 12: Equivalent loops
MMDS Loop Pol Del Elements
4 402 R 0 price of wine — revenues from wine sales — revenues — profitability >
production +
6 602 R 0 profits — profitability + production >
4 412 B 1 energy costs — production costs — energy efficiency +
6 603 B 0 profits — incentive to decrease costs — production volume — costs
A third challenge refers to MMDS elicitation. As opposed to intervention and consulting, research projects
usually take great care to avoid or minimize the influence of the researcher on the participating individuals.
This has lead to the interview — transcribe — code procedure applied in the current study. It also meant that
no visual feedback was given to the interviewee during the interview. However, this lead to the researcher
making multiple choices regarding causal links and can have some undesired consequences. The
interviewee may not mention a variable or a causal link or even a loop which is completely obvious to him,
leading to the danger that “not mentioned” is mistakenly interpreted as “not recognized”; it is then up to the
researcher to prompt for sufficient elaboration on behalf of the interviewee without priming or directing
him. The fact of constructing and displaying the causal loop diagram (like the ones shown in Figure 2 and
Figure 3) during the interview would allow the interviewee to articulate that which was obvious to him (but
maybe not to the researcher). As far as mental model research is carried out with experienced executives as
participants, the danger that the diagram might influence the interviewee’s mental model is minimal. A
diagramming tool like the one described above would be highly useful to assure that the interviewee
articulates every variable, link and loop he internally recognizes.
5 Conclusions
This paper had the purpose to show that in matters of mental model comparison, there is a significant
difference between those loops which individuals recognize when they talk about the subject and those
loops which are inherent in the structure they articulate but which remain dark. Out of nine executives
interviewed in the study reported here, only three recognized any feedback loops at all: 45 of the 50 loops
have remained dark and unrecognized, implying that the 47 most relevant variables in this set of MMDS
are held to be part of input-output models by the executives, but are really endogenously driven by manifold
feedback loops.
By consequence, an important share of each model’s variables had not been recognized as being under the
influence of loops. It has also been shown that by taking into account only recognized loops, current
methods and tools for mental model comparison may provide biased quantifications of how similar or
distant a pair of mental models is at the level of loops.
The mental model findings support the statement that without an extemal support, the mental models of
executives will be incomplete and the executives’ assessment of possible strategies and policies will not be
based on a thorough analysis of the likely unfolding of their management situations.
21
Based on this, a call is made to develop tools which help decision makers articulate variables and links and
which automatically detect and display loops, so that they be dark loop no longer. Such tools will also be
helpful for mental model researchers. There is also a need to adapt mental model comparison methods and
tools at the level of loops, helping to quantify the impact of the differences between the recognized and the
inherent loops.
Specifically for mental model research, the method and the tools for MMDS analysis and comparison (M.
Schaffernicht and S. Groesser 2014) will be enhanced by incorporating the possibility to detect and utilize
the SILS, make intra-model comparisons based on the different loop set and calculate two sets of ‘loop
distance ratios’ based on the recognized feedback loops and the SILS. This will allow to quantify the
distance between the mental model articulated by the decision maker and the mental model which an analyst
with experience in system dynamics hordes of the decision-makers mental model.
Of course, a study based on as few as nine mental models is limited by the small sample size. This limitation
notwithstanding, under the current conditions the effort to interview, code and analyze is huge and may be
prohibitive for many researchers. Therefore hope is that the availability of the methods and tools called for
would make mental model research less time consuming and more attractive.
References
Ashby, W. R. (1991). Requisite Variety and Its Implications for the Control of Complex Systems. In
Facets of Systems Science (pp. 405-417). Boston, MA: Springer US.
Black, L. J. (2013). When visuals are boundary objects in system dynamics work. System Dynamics
Review, 29(2), 70-86, doi:10.1002/sdr.1496.
Chile, V. d. (2015). Estadisticas Anuales.
http://www.vinosdechile.cl/contenidos/informacion/estadisticas-anuales/. Accessed 11/12/2015
2015.
Chile, W. 0. (2010). Strategic Plan 2020 - International Market. Santiago de Chile.
Eden, C. (2004). Analyzing cognitive maps to help structure issues or problems. European Journal of
Operational Research, 159(3), 673-686, doi:10.1016/s0377-2217(03)00431-4.
Forrester, J. (1969). Principles of Systems: MIT Press.
Groesser, S., & Schaffernicht, M. (2012). Mental Models of Dynamic Systems: Taking Stock and
Looking Ahead. System Dynamics Review, 28(1), 22.
Groesser, S. N., & Schaffernicht, M. F. (2012). Mental models of dynamic systems: taking stock and
looking ahead. System Dynamics Review, 28(1), 22, doi:10.1002/sdr.
Johnson-Laird, P. N. (2001). Mental models and deduction. Trends in Cognitive Sciences, 5(10), 9.
Kahneman, D. (2011). Thinking, Fast and Slow. New Y ork: Farrar, Strauss and Giroux.
Kampman, C. Feedback loops gains and system behavior. In S. D. Society (Ed.), International System
Dynamics Conference,, Cambridge, Mass, 1996: System Dynamics Society
Klimoski, R., & Mohammed, S. (1994). Team Mental Model: Construct or Metaphor? Journal of
Management, 20(2), 34.
Langan-Fox, J., Anglim, J., & Wilson, J. R. (2004). Mental Models, Team Mental Models, and
Performance: Process, Development, and Future Directions. Human Factors and Ergonomics in
Manufacturing, 14(4), 21.
Langan-Fox, J., Code, S., & Langfield-Smith, K. (2000a). Team Mental Models: Techniques, Methods,
and Analytic Approaches. Human Factors, 42(2), 29.
22
Langan-Fox, J., Wirth, A., Code, S., Langfield-Smith, K., & Wirth, A. Analyzing shared mental models.
In HFES 2000 Congress, 2000b (pp. 57)
Langfield-Smith, K., & Wirth, A. (1992). Measuring Differences between Cognitive Maps. The J ournal
of the Operational Research Society, 43(12), 15.
Lim, B.-C., & Klein, K. J. (2006). Team mental models and team performance: a field study of the effects
of team mental model similarity and accuracy. J ournal of Organizational Behavior, 27(4), 403-
418, doi:10.1002/job.387.
Markoczy, L., & Goldberg, J. (1995a). A Method for Eliciting and Comparing Causal Maps. Journal of
Management, 21(2), 28.
Markoczy, L., & Goldberg, J. (1995b). A Method for Eliciting and Comparing Causal Maps. Journal of
Management, 21(1), 28.
Mohammed, S., Ferzandi, L., & Hamilton, K. (2010). Metaphor No More: A 15-Y ear Review of the
Team Mental Model Construct. Journal of Management, 36(4), 876-910,
doi:10.1177/0149206309356804.
Mohammed, S., Klimoski, R., & Rentsch, J. R. (2000). The Measurement of Team Mental Models: We
Have No Shared Schema. Organizational Research Methods, 3(2), 123-165,
doi:10.1177/109442810032001.
Oliva, R. (2004). Model structure analysis through graph theory: partition heuristics and feedback
structure decomposition. System Dynamics Review, 20(4), 313-336, doi:10.1002/sdr.298.
Ozgiin, O., & Barlas, Y. (2015). Effects of systemic complexity factors on task difficulty in a stock
management game. System Dynamics Review, 31(3), 115-146.
Schaffernicht, M. (2010). Causal loop diagrams as means to improve the understanding of dynamic
problems: a critical analysis. Systems Research and Behavioral Science, 27(6), 13.
Schaffernicht, M., & Groesser, S. (2011). A Comprehensive Method for Comparing Mental Models of
Dynamic Systems. European J ornal of Operational Research, 210(1), 10.
Schaffernicht, M., & Groesser, S. (2014). The SEXTANT software: A tool for automating the
comparative analysis of mental models of dynamic systems. European J ornal of Operational
Research, 238(2), 12.
Schaffernicht, M. F., & Groesser, S. N. (2011). A comprehensive method for comparing mental models of
dynamic systems. European Journal of Operational Research, 210(1), 57-67,
doi:10.1016/j.ejor.2010.09.003.
Schaffernicht, M. F., & Groesser, S. N. (2014). The SEXTANT software: A tool for automating the
comparative analysis of mental models of dynamic systems. European Journal of Operational
Research, 238(2), 566-578, doi:10.1016/j.ejor.2014.04.002.
Schaffernicht, M. F., & Groesser, S. N. Capturing Executiveial Cognition in Chilean Wineries: Hardening
a New Method to Elicit and Code Mental Models of Dynamic Systems. In 33rd International
Conference of the System Dynamics Society, Cambridge, MA, 2015
Sterman, J. (2000). Business dynamics: systems thinking and modelling for a complex world: McGraw
Hill.
Sterman, J. D. (1989). Modeling executiveial behavior - misperceptions of feedback in a dynamic
decision-making experiment. Management Science, 35, 321-339.
Sterman, J. D. (2000). Business dynamics: systems thinking and modeling for a complex world. Boston,
USA: McGraw-Hill.
23
Sterman, J. D. (2010). Does formal system dynamics training improve people's understanding of
accumulation? System Dynamics Review, 26, 316-334.
24
