Content uploaded by Bob Walrave
Author content
All content in this area was uploaded by Bob Walrave on Aug 25, 2014
Content may be subject to copyright.
1
Model documentation for:
“How to counteract the suppression of exploration in publicly
traded corporations”
This document is made available to the reviewers. In case of publication of the article, this appendix
will be made available on a widely accessible website. The purpose of the model appendix is to
provide all details of how the model, developed by Walrave et al. (2011), was adjusted in order to run
the simulation experiments outlined in the paper “How to counteract the suppression of exploration
in publicly owned corporations”.
The system dynamics (SD) model developed and calibrated in Walrave et al. (2011) is adopted
here. The model was originally developed, and also adjusted for the purpose of this study, in
Ventana’s software package ‘Vensim’. Adopting this model and its empirical setting allows for
experimentation with the relevant variables by means of socalled ifthen simulation experiments.
Here we focus on the adjustments made to this model, which were required for testing the
propositions presented in the paper. In this respect, we also refer to the model appendix developed for
the Walrave et al. (2011) paper, for a complete description and validation of the original model. For
your convenience, a copy of the latter appendix has been attached to this document.
The theoretical background of the model can be summarized as follows. First, the model
considers the dynamic effects of aligning exploitation and exploration with environmental aspects.
Second, exploitation and exploration activities are assumed to be two ends of a single continuum that
are constrained by a shared set of (limited) resources. Third, the model focuses on the capabilities of
top management to signal environmental changes and translate these into a balanced portfolio of
exploitation and exploration projects. In this respect, we assume the existence of an ‘optimal’ (i.e.,
most profitable) exploitationexploration balance. This managerial capability arises from the
interaction between the executive team and the board of directors. Fourth, myopic forces limit the
speed in which strategic changes are made. Finally, the firm in our model is assumed to be technically
fit; this assumption serves to focus the model on the firm’s evolutionary fitness and, as such, on top
management’s capability to align the exploitationexploration mix with the changing environmental
context.
Adjustments to the original model (in Walrave et al. 2010)
Opportunity costs
Although the original model considers the operating results, it is not possible to calculate the
opportunity costs arising from shifting the exploitationexploration balance toward more explorative
R&D. In this respect, a specific intervention to the exploitationexploration balance by the executive
team could be successful in terms of realizing a shift in the exploitationexploration balance (toward
more exploration), but may also highly unrealistic due to the high opportunity costs associated with
such a change. As such, in order to consider the financial viability of changes to the exploitation
exploration balance intended to counteract the suppression process, we need to consider the
opportunity costs.
The opportunity costs are the difference between what is (i.e., in terms of financial performance
after the executive team brings about change to the exploitationexploration balance – denoted by the
‘Operating Result’ or ‘OR’); compared to what could have been had the development of the
2
exploitationexploration balance remained ‘untouched’ (i.e., the results of the original historyfriendly
model – denoted by the ‘Replicated Operating Result’ or ‘OR_Rep’). In this respect, we assume the
fixed costs of adjustments made to the exploitationexploration balance are insignificant compared to
the opportunity costs that a firm may incur as the result of such change. As such, the formal model
was extended with the ‘Opportunity Costs’ (OC) variable. The OC are calculated by subtracting the
OR_Rep from the OR; starting from the moment that the exploitationexploration balance is shifted,
for as long as the former is higher than the latter (until the end of the simulation run). By doing so, we
capture the opportunity costs associated with changing the exploitationexploration ratio. This results
in function 1:
𝑑
OC
𝑑𝑡 =𝐼𝐹!𝑇𝐻𝐸𝑁!𝐸𝐿𝑆𝐸
OR_Rep&
–
OR
<0,
OR_Rep
–
OR
,0
!
(1)
Experimental setup
Determining the effectiveness of strategies for countering the suppression process requires a
further specification of success and failure. In essence, successful managerial actions to the
exploitationexploration balance should result in an ‘External pressure to exploit’ (EP) variable that is
as low as possible. In that case, the ‘Attempt to explore’ loop will turn dominant, facilitating escape
from the suppression process – as was described in the manuscript. The original model ran for 800
weeks, and the EP ultimately grew to 1 (i.e., 100 per cent external pressure to exploit). We call an
intervention successful if it achieves an EP < .5 at t = 800; which is highly similar to the ‘stable’
historydivergent simulation (see the original model documentation attached). Effectively, this
denotes the situation in which top executives remain in control of resource distribution. Note that we
investigate the results of period A, B, and C of the suppression process, implying t = 0 till t = 450.
The experiments are directed toward identifying tipping points. We assume that the executive
team can adjust the exploitationexploration balance with a quarter percent per week toward
exploration (not considering the influence of the normal system’s dynamics). This resembles a rather
slow, but therefore also realistic rate of change. In the context of this study, the tipping points
represent the minimum shift required in the exploitationexploration balance – in weeks – in order to
achieve an EP lower than .5 at t = 800. Effectively, this implies that for every t under consideration
(from 0 till 450) the tipping point that achieves this specific goal is calculated, by means of ifthen
experimentation. When the managerial push toward more exploration is shorter than the tipping point
indicates, the firm gets caught in the unfolding suppression process, and vice versa. Note that the
actual Visual Basic for Applications (VBA) programming code in order to determine the tipping
points is given in the section ‘Determining the tipping points by means of ifthen experimentation’.
The first experiment concerns propositions 1a, 2a, and 3a and are about the executive team
directing their firm toward more explorative R&D – at different moment in times – in an attempt to
counteract the suppression process. In light of the original SD model, this implies increasing the
‘Perceived Need to Explore’ PNE variable. This variable captures the need perceived by the executive
team to increase explorative R&D activities, as the result of a perceived misalignment between the
environmental context and the current exploitationexploration balance. Systematically increasing the
PNE, to uncover the tipping points per t, then allows for testing propositions 1a, 2a, and 3a. In the
manuscript, we renamed the PNE variable to the ‘decision to shift the exploitationexploration
balance’, as this better reflects the operation of this variable in the context of our study.
Function 2 denotes the change made to the PNE variable to be able to run the first experiment.
P1_start denotes the starting time of the change targeted at PNE and ranges from t = 0 till t = 450. The
value of P1_duration constitutes the tipping point (calculated in weeks) and is determined by means of
3
ifthen experimentation (through the Visual Basic for Applications script that is discussed later in this
appendix). Finally, P1_size is a constant that is fixed at .0025 (and denotes the amount of change to
the PNE that is possible per week). Sensitivity analyses, outlined in the section ‘Sensitivity analyses’,
demonstrated that the findings discussed are robust relative to different adjustment rates. Note that the
final part of equation 2, that is (1 – PAE – PNE) / AT_Myopia, captures the normal system’s dynamics
(see the original model appendix).
𝑑𝑃𝑁𝐸
𝑑𝑡 =𝑃𝑈𝐿𝑆𝐸
P1_start
,!
P1_duration
∗
P1_size
+
1−𝑃𝐴𝐸 −𝑃𝑁𝐸
AT_Myopia
!
(2)
The second experiment pertains to propositions 1b, 2b and 3b regarding the influence that the
board of directors can have, by adjusting the external pressure to exploit, on the success chances of
the first experiment. The second experiment therefore tests, while conducting the first experiment
again, the following interventions: any ‘too early’ intervention (executed in period A) increases the
pressure to exploit; while any ‘too late’ intervention (executed in period C) decreases the external
pressure to exploit – for the same duration as the length of the tipping point indicates. That is, as long
as the executive team keeps pursuing more exploration, the board of directors acts in the (anti
cyclical) manner described. We assume this behavioral pattern to be temporary in nature, because it is
the result of negotiations and agreements.
Function 3, regarding the newly added variable ‘Adjusted EP’ (A_EP), captures the adjusted
behavior of the board of directors. This variable is calculated by taking the current EP (following from
the normal system’s dynamics) and (a) in case of a too early executive adaptation of the exploitation
exploration balance (period A), is increased by A_EP_size; or (b) in case of a too late executive action
to adjust the exploitationexploration balance (period C), is decreased by A_EP_size. A_EP_size
equals +/ .1 here. This variable was subjected to sensitivity analysis, that demonstrated our findings
are robust relative to different A_EP sizes (see section ‘Sensitivity analyses’). In both cases, the
adjusted behavior starts at P1_start and lasts as long as P1_duration (the tipping point) indicates.
𝐷𝑢𝑟𝑖𝑛𝑔!𝑝𝑒𝑟𝑖𝑜𝑑!𝐴→𝐴_𝐸𝑃_𝑠𝑖𝑧𝑒 =.1!
(3)
𝐷𝑢𝑟𝑖𝑛𝑔!𝑝𝑒𝑟𝑖𝑜𝑑!𝐶→𝐴_𝐸𝑃_𝑠𝑖𝑧𝑒 =−.1
!
𝑑𝐴_𝐸𝑃
𝑑𝑡 =𝐸𝑃 +𝐼𝐹!𝑇𝐻𝐸𝑁!𝐸𝐿𝑆𝐸 (𝑇𝐼𝑀𝐸 <𝑃1_𝑠𝑡𝑎𝑟𝑡,0,𝐼𝐹 !𝑇𝐻𝐸𝑁!𝐸𝐿𝑆𝐸 (𝑇𝐼𝑀𝐸
>𝑃1_𝑠𝑡𝑎𝑟𝑡 +𝑃1_𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛,0,𝐴_𝐸𝑃_𝑠𝑖𝑧𝑒))!
!
Determining the tipping points by means of ifthen experimentation
In order to determine the tipping points (for the P1_duration variable), VBA (Microsoft Excel) was
utilized in combination with Vensim. The programming code below illustrates how the tipping points
are calculated by means of Dynamic Data Exchange (DDE) between Microsoft Excel and Ventana’s
Vensim. The given VBA programming lines – if used in conjunction with the original SD model
including the adjustments described earlier – serves to calculate the tipping points and related
opportunity costs of the two experiments. (Manual adjustment of the constants P1_size and
A_EP_size is required in the Vensim model.) Note that in the given code, all text starting with an ‘ are
comments, which explain the purpose of the subsequent programming lines, but do not execute any
commands.
4
VBA code used for calculating the tipping points
Sub run_model_experiment()
'Starts Vensim
Dim DDE_channel As Integer
DDE_channel = Application.DDEInitiate("VENSIM", "System")
'No interaction (messages and warnings will be suppressed)
Application.DDEExecute DDE_channel, "[SPECIAL>NOINTERACTION1]"
Application.DDEExecute DDE_channel, "[SETTING>SHOWWARNING0]"
'Start loop from 0 till 450 weeks
‘Set the variables
Dim P1_start As Long
Dim P1_duration As Long
Dim Cell As Long
Dim operating_result As Long
Dim returnList As Variant
Dim ch1 As Long
Dim ch2 As Long
Dim ch3 As Long
Dim ch3a As Long
Dim ch3b As Long
P1_start = 0
‘Set the starting output row in Excel
Cell = 2
Do While P1_start <= 450
'Reset variables for the next Time_step
P1_duration = 0
operating_result = 0
‘Start loop searching for the required P1_duration (ITERATION 1)
Do While P1_duration <= 200
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 duration=" & P1_duration & "]"
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 start=" & P1_start & "]"
'Run the model and execute delay
Application.DDEExecute DDE_channel, "[MENU>RUNO]"
Application.Wait (Now + Sheets(1).Cells(3, 2))
‘Get and assess the output variable OR
varstr$ = " ""External pressure to exploit (EP)""@800"
test = Application.DDERequest(DDE_channel, varstr$)
external_pressure = test(LBound(test))
If (external_pressure < 0.5) Then Exit Do
P1_duration = P1_duration + 100
Loop
'Start loop searching for the required P1_duration (ITERATION 2)
P1_duration = P1_duration  100
Do While P1_duration <= 200
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 duration=" & P1_duration & "]"
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 start=" & P1_start & "]"
'Run the model and execute delay
Application.DDEExecute DDE_channel, "[MENU>RUNO]"
Application.Wait (Now + Sheets(1).Cells(3, 2))
'Get and assess the output variable OR
varstr$ = " ""External pressure to exploit (EP)""@800"
test = Application.DDERequest(DDE_channel, varstr$)
external_pressure = test(LBound(test))
If (external_pressure < 0.5) Then Exit Do
P1_duration = P1_duration + 50
Loop
'Start loop searching for the required P1_duration (ITERATION 3)
P1_duration = P1_duration  50
Do While P1_duration <= 200
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 duration=" & P1_duration & "]"
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 start=" & P1_start & "]"
'Run the model and execute delay
Application.DDEExecute DDE_channel, "[MENU>RUNO]"
Application.Wait (Now + Sheets(1).Cells(3, 2))
'Get and assess the output variable OR
varstr$ = " ""External pressure to exploit (EP)""@800"
test = Application.DDERequest(DDE_channel, varstr$)
external_pressure = test(LBound(test))
If (external_pressure < 0.5) Then Exit Do
5
P1_duration = P1_duration + 10
Loop
'Start loop searching for the required P1_duration (ITERATION 4)
P1_duration = P1_duration  10
Do While P1_duration <= 200
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 duration=" & P1_duration & "]"
Application.DDEExecute DDE_channel, "[Simulate>SETVALP1 start=" & P1_start & "]"
'Run the model and execute delay
Application.DDEExecute DDE_channel, "[MENU>RUNO]"
Application.Wait (Now + Sheets(1).Cells(3, 2))
'Get and assess the output variable OR
varstr$ = " ""External pressure to exploit (EP)""@800"
test = Application.DDERequest(DDE_channel, varstr$)
external_pressure = test(LBound(test))
If (external_pressure < 0.5) Then Exit Do
P1_duration = P1_duration + 1
Loop
‘P1_duration is known > Generate output from here
'Return Time of increase (P1 start)
varstr$ = "P1 start@" & P1_start
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 1).Value = returnList(LBound(returnList))
'Return Size of increase (P1 duration)
varstr$ = "P1 duration@" & P1_duration
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 2).Value = returnList(LBound(returnList))
'Return Operational Result (OR) at t=800
varstr$ = " ""Operating result (OR)""@800"
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 3).Value = returnList(LBound(returnList))
'Return Operational Result Cumulative (ORC) at t=800
varstr$ = " ""Operating result cumulative (ORC)""@800"
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 4).Value = returnList(LBound(returnList))
'Return Need to Explore (NE) at t=800
varstr$ = " ""Perceived need to explore (PNE)""@800"
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 5).Value = returnList(LBound(returnList))
'Return External Pressure (EP) at t=800
varstr$ = " ""External pressure to exploit (EP)""@800"
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 6).Value = returnList(LBound(returnList))
‘Opportunity costs at t=800
varstr$ = " ""Total investment costs (TIC)""@800"
returnList = Application.DDERequest(DDE_channel, varstr$)
Sheets(1).Cells(Cell, 7).Value = returnList(LBound(returnList))
Cell = Cell + 1
P1_start = P1_start + 1
Loop
'Stop DDE communication and end experiment
Application.DDETerminate DDE_channel
End Sub
Sensitivity analyses
As explained earlier, the PNE variable is rampwise increased through the P1_size variable, with an
additional .0025 ‘units’ per week. This implies that in equation 2 the value for P1_size is fixed at
.0025. This assumption was subjected to sensitivity analysis to test its robustness. More specifically,
we ran the first experiment again with adjustment rates ranging between .0015 and .0035 ‘units’ per
week. Figure 4 present the results of this robustness test: 4a presents the tipping points while 4b
depicts the opportunity costs related to these points. The A_EP_size was also subjected to sensitivity
analysis. More specifically, the second experiment was conducted again with A_EP_size values of
6
ranging between .15 and +.15. The results of this robustness test are depicted in Figure 5: 5a denotes
the tipping points, while 5b presents the opportunity costs related to these points. All findings imply
that the results are rather robust to changes to these newly introduced variables.
The result of the experiment conducted with P1_size = .0025 is given in black. The same experiment
was conducted with adjustments to the P1_size of: ! +/ .0005 and ! +/ .0010.
Figure 4: Sensitivity of the tipping point analyses, experiment 1: P1_size.
Fi gur e 5a
Fi gur e 5b
0
0,05
0,1
0,15
0,2
0
52
104
156
208
260
312
364
416
0
20
40
60
80
100
0
52
104
156
208
260
312
364
416
Shift required in exploitation
exploration balance (weeks)
Opportunity Costs
in Eu r o s (x 1,000,000)
Change to the exploitationexploration balance started in week
Fi gur e 4a
Fi gur e 4b
C
A
B
80
60
40
20
7
The result of the experiment conducted with A_EP_size = .10 is given in black (P1_size = .0025). The
same experiment was conducted with adjustments to the A_EP_size of: ! +/ .03 and ! +/ .05.
Figure 5: Sensitivity of the tipping point analyses, experiment 2: A_EP_size.
0
0,05
0,1
0,15
0,2
0
52
104
156
208
260
312
364
416
0
20
40
60
80
100
0
52
104
156
208
260
312
364
416
Shift required in exploitation
exploration bal an c e (we e k s )
Opportunity Costs
in Eu r o s (x 1,000,000)
C
Fi gur e 5a
Fi gur e 5b
A
B
80
60
40
20
Shift required in exploitation
Opportunity Costs
Change to the exploitationexpl oration balance started in week
8
Original model documentation for the system dynamics model in:
Walrave, B., Van Oorschot, K. E., & Romme, A. G. L. 2011. Getting trapped in the
suppression of exploration: a simulation model. Journal of Management Studies,
48(8): 1727–1751.
The model was developed in VENSIM software. The full model, in terms of stock and flows, is given
on the next page (Figure 1). The model, grounded in the literature, was subjected to sensitivity
analyses and also served to run historyreplicating and historydivergent simulations (see the
manuscript for the main results).
The theoretical background of the model can be summarized as follows. First, the model considers the
dynamic effects of aligning exploitation and exploration with environmental aspects. Second, we
assume exploitation and exploration activities are two ends of one continuum that are constrained by a
shared set of (limited) resources. Third, the model focuses on the capabilities of top management to
signal environmental changes and translate these into a balanced portfolio of exploitation and
exploration projects. In this respect, we assume the existence of an ‘optimal’ (i.e. most profitable)
exploitationexploration balance. This managerial capability arises from the interaction between top
management and the Board of Directors. Fourth, myopic forces limit the speed in which strategic
changes are made. Finally, we assume the firm in our model is technically fit; that is, the model
focuses on the firm’s evolutionary fitness and, as such, on top management’s capability to align the
exploitationexploration mix with the environmental context.
In Figure 1 (next page), the three feedback loops are given in different colors with the variable names
written in black. The ‘External pressure’ feedback loop is depicted in blue, the ‘Stick to exploitation’
feedback loop in red, and the ‘Attempt to explore’ feedback loop in orange. Please note that the
External pressure and Stick to exploitation loops overlap (from RIE to Change in investment
exploitation). Moreover, the Attempt to explore feedback loop overlaps a critical part of the External
pressure loop (from Inv_Explore to RIE). The blue variables denote exogenous influences. The green
variables indicate adjustment times (delays). The unit of time in the model is weeks and the total
simulation time was 800 weeks (slightly more than 15 years). The simulation algorithm was Euler’s
method with a step size (dt) of 0.25 weeks.
The next section of this document describes all equations of the formal model in detail. Subsequently,
we provide an overview of the model settings and the sensitivity of the calibrated variables. The last
section explores whether the model should be deterministic or stochastic.
9
Figure 1: Overview of the complete model
Model description
Capabilities are often a matter of degree (Winter, 2000), and can therefore be modeled as continuous
variables. In our model, the balance between exploration and exploitation (comprising organizational
ambidexterity) is determined by the distribution of the available resources (AR) over the two ends.
Following our assumption described in the previous section, the amount of AR, an auxiliary variable,
is finite: it is calculated as a certain percentage (POR) of the operating result (OR) in a current period.
Nevertheless, we assume a minimum amount of resources (MAR) that will be available even when
the OR is negative or very small. MAR, an exogenous constant (set to 0.5), prevents negative amounts
of AR and thus simulation errors. In order to achieve this, the ‘MAX’ function is used. This function
assesses if the calculated AR is greater than the MAR and then returns the calculated value (if true) or
an assumed fixed minimum amount of resources (MAR) (if false). (Note that MAR does not influence
the process theory as outlined in the paper because a negative OR will only occur at the very end of
the described sequences of events.)
(1) AR = MAX (OR * POR, MAR)
The percentage of the AR invested in exploration is captured by the variable ‘Resource investment
in exploration’ (RIE) (see function 17). The stock ‘Investment in exploitation’ (Inv_Exploit) refers to
the amount of resources invested in exploitation in the current period. On the other end of the
continuum, the stock ‘Investment in exploration’ (Inv_Explore) denotes the level of resources
10
allocated to exploration in the current period. Recent studies show that implementing new innovation
strategies and thus routines is not simple; moreover, it takes considerable time and effort before these
strategies and routines become effective (e.g. Durmusoglu et al., 2008). The desired resource
adjustment is therefore subject to an adjustment time (AT) (exogenous constants). The AT is shorter
for exploitation (AT_Exploit) than for exploration (AT_Explore), since it involves more radical
changes to the routines. This gives the following equations:
Change in investment exploitation:
(2) d (Inv_Exploit) / dt = ((1  RIE) * AR – Inv_Exploit) / AT_Exploit
Change in investment exploration:
(3) d (Inv_Explore) / dt = (RIE * AR – Inv_Explore) / AT_Explore
The exogenous variable ‘Environmental competitiveness’ (EC) represents the level of
competition in the firm’s environment and captures the number and strength of competitors in the
current period. This exogenous variable ranges from 0 (monopolistic) till 1 (highly competitive). The
EC variable was estimated by calculating the Herfindahl index for the case firm. This index is
calculated by subtracting the sum of the squared market shares from one. This is captured by the
following equation, where si is the market share of firm i in the market, and N is the number of firms:
(4) ECi = Herfindahl index = 1 
∑
=
N
i
i
s
1
2
‘Environmental dynamism’ (ED) is an exogenous variable representing the level of dynamism in
the market in the current period. It ranges from 0 (extremely lethargic) to 1 (extremely dynamic). This
variable was estimated by rescaling the S&P 500 index (from the beginning of 1994 till the ending of
the 3rd quarter of 2009). More specifically, the S&P 500 growth rate was calculated for every t (with t0
= 1) and the result subtracted with 1. (This latter is done because the initial situation is assumed stable
and the starting values of ED should therefore be close to 0, rather than 1.) This operation is captured
by gr. The resulting data set (ranging from 0.0 to 2.3) was then divided by x to ensure fit with the
given range for ED. Lastly, the moving average over 26 weeks was taken in order to smooth out any
nonsystematic changes. This results in the following algorithm, where x will equal 3:
(5) 𝐸𝐷!=!
!"
!!
!
!
!!!"
!!!
The variable ‘Environmental competitiveness and dynamism’ (ECD) represents the state of the
environment in the current period, which determines the most appropriate exploitationexploration
mix at a specific moment in time. ECD is a continuous variable ranging from 0 (extremely stable) till
1 (extremely instable). The ECD variable is determined by the two exogenous variables EC and ED.
More specifically, the two lookup variables ‘Effect of EC on ECD’ and ‘Effect of ED on ECD’
capture the influence of EC and ED on ECD. Concerning the former, the Scurve (see Figure 2)
represents the situation in which high levels of dynamism bring along the need for exploitation, while
low levels of dynamism need a more balanced portfolio of exploitation and exploration activities.
Concerning the latter, the Scurve (see Figure 2) captures that high levels of dynamism require more
exploration efforts, while low levels of dynamism demand (mostly) exploitation initiatives.
11
Figure 2: Effect of EC and ED on the required exploitationexploration mix
As argued in the main text, the ED variable has more influence on the appropriate mix than the EC
variable which results in the following formula (note the two lookup functions) (where ω is the
weight factor, which is equal to 2/3 in our case):
(6) ECD = ω * Effect ED on ECD (ED) + (1 ω ) * Effect EC on ECD (EC)
The ECD variable thus captures the assumed ‘optimal’ balance between exploitation and exploration
and is the basis for both the reinforcing ‘Stick to exploitation’ and the balancing ‘External pressure’
loop.
Stick to exploitation
In general, perceptions tend to adjust to new circumstances with a certain delay, which can be
modeled in terms of the behavior of a firstorder adaptive system (Sterman, 2000). Top management’s
perception of the environment, denoted by the stock ‘Perceived environmental competitiveness and
dynamism’ (PECD), is thus subject to such a delay. This variable captures the perceived
environmental situation in the current period. The delay is specified by the variable ‘Perception
adjustment time Management’ (AT_Management) (an exogenous constant).
Change in PECD:
(7) d (PECD) / dt = (ECD – PECD) / AT_Management
The operational balance between exploitation and exploration in the current period is captured by
the variable ‘Relative investment in exploitation’ (RI_Exploit). The balance is given in terms of the
relative investment in exploitation. Since both Inv_Exploit and Inv_Explore denote the investments in
respectively exploitation and exploration at a certain time, RI_Exploit is calculated by dividing the
Inv_Exploit by the sum of Inv_Exploit and Inv_Explore.
(8) RI_Exploit = Inv_Exploit / (Inv_Exploit + Inv_Explore)
From the PECD and the RI_Exploit, the ‘Perceived alignment with the environment’ (PAE) can be
calculated. Here, 1 implies a perfect alignment, while 0 means no alignment at all. (Please note that
12
the kind of manufacturing firm modeled typically does not have very low values for RI_Exploit, given
the importance of efficiency.)
(9) PAE =(1  RI Exploit) * PECD
Subsequently, the PAE triggers managerial action – denoted in the stock ‘Perceived need to
explore’ (PNE). This variable constitutes the cognitive aspect of the behavior of top management in
the current period. More specifically, it denotes top management’s perceived appropriate balance in
the current period. Due to inertial forces (AT_Myopia; an exogenous constant), PNE is subject to a
firstorder delay.
(10) d (PNE) / dt = (1 – PAE – PNE) / AT_Myopia
External pressure
The alignment between the exploitation–exploration mix and the environment influences the return on
investment (ROI), and thus the operating result of the firm. In that respect, heavy investments in
exploration, when the environmental situation demands more exploitation, will result in an inferior
return on (exploration) investments. We thus consider two ROIs, one for exploitation and one from
exploration investments. The former one is captured by the stock ‘ROI_Exploit’ while the latter one is
denoted by the stock ‘ROI_Explore’. Both capture the level of ROI in a current period. Moreover, this
sequence of events (from investments to operating results) takes place with a certain delay because
initial investments have to be transformed into (money generating) innovation. This delay is smaller
for returns related to exploitation (exogenous constant RD_Exploit) than it is for exploration
(exogenous constant RD_Explore), since the latter needs significantly more time to generate market
success (Burgelman et al., 2004). Moreover, investments made in exploration that are aligned with the
environmental situation (i.e. the alignment between the exploitationexploration investments and the
ECD; see functions 11 and 12) yield a higher return on investment (Jansen et al., 2006; Uotila et al.,
2009). For example, the identification of a new market can, most likely, make a larger (positive)
financial impact than the incremental improvement of a product in an existing market. Therefore, two
different constants are needed to create a distinction between ROIs from exploitation and exploration:
‘Result factor exploitation’ (RF_Exploit) and ‘Result factor exploration’ (RF_Explore).
Change in ROI_Exploit:
(11) d (ROI_Exploit) / dt = (Inv_Exploit * (1 – ECD) * RF_Exploit – ROI_Exploit) / RD_Exploit
Change in ROI_Explore:
(12) d (ROI_Explore) / dt = (Inv_Explore * ECD * RF_Explore – ROI_Explore) / RD_Explore
OC denotes the 'Operating costs' (an exogenous constant), and OR (a variable) is a function of:
(13) OR = ROI Exploit + ROI_Explore  OC
Shareholders (the board) also perceive the ORs with a certain delay, implying the use of a first
order adaptive system regarding the trend of the OR. The perceived trend in the OR (captured by the
stock PTOR) is therefore calculated as the average (thus delayed) fractional growth rate (which is
negative for decline). As such, it provides a simple trend estimate for the currently perceived OR.
(14) PTOR = (OR – Average_OR) / (AT_Board * Average_OR)
13
(15) d (Average_OR) / dt = Change in Average_OR = (OR – Average_OR) / AT_Board
The PTOR determines the amount of external pressure to generate shortterm financial results.
This is captured by the stock ‘External pressure to exploit’ (EP) which refers to the level of pressure
in a current period. This effect is determined by the lookup variable ‘Effect of POR on EP’ (see
Figure 3). This lookup captures the process that when top management fails to achieve acceptable
financial returns, this will result in pressure from the owners on top management to generate short
term financial results (i.e. a pressure to exploit). On the contrary, when the board perceives the
financial performance to be adequate, top management will have the possibility to adjust the
exploitationexploration mix as desired (the influence of the EP becomes evident at the ‘Attempt to
explore’ loop).
Figure 3: Effect of PTOR on EP
The increase and decrease of external pressure is also subject to a delay, the pressure adjustment
time (exogenous constant AT_Pressure). This delay arises from the fact that, first, the Board of
Directors operates on the basis of quarterly reports of operating results (reporting delay), and second,
the Board acts on the basis of the trend rather than incidental fluctuations in OR. Therefore, the
following equation was used for the external pressure to exploit (EP) on top management:
(16) d (EP) / dt = Change in EP = (Effect of PTOR on EP (PTOR) – EP) / AT_Pressure
Attempt to explore
The subsequent interaction between the perceived need to explore (PNE) and the external pressure to
exploit (EP) determines the value of the variable RIE and reflects top management’s behavior (related
to the exploitationexploration balance). This variable can range from 0 to 1 (0 implying a sole
investment in exploitation projects while 1 means a mere investment in exploration initiatives).
Because this variable depends on both PNE and EP, it is calculated by multiplying top management’s
desired and the shareholder’s allowed investment in exploration activities. The result of this process is
the actual investment level in exploration as well as in exploitation which constitutes a key component
of the ‘Attempt to explore’ feedback loop:
(17) RIE = PNE * (1  EP)
14
Model settings and sensitivity
This section presents all the values for the constants after conducting historyreplicating simulation
based on the obtained data (see the manuscript for more details regarding data collection). This
implies that certain constants were ‘calibrated’ to fit the model variables with corresponding data
gathered on site. The results can be seen in Table I where the variables are alphabetically ordered and
their set value presented. In this table, a ‘*’ denotes the variables that were taken into the calibration
process. In addition, Table II provides an overview of all the variables in the model and Table III
gives an overview of all the functions.
As can be seen in Table I, certain variables were not estimated during the historyreplicating
simulation, but based on reasoning and case study observations. This can be explained by the fact that
the firm, from which we gathered our data, did not engage significantly in exploration. As such, it
makes no sense to calibrate the delays for exploration. This concerns the variables ‘AT_Explore’ and
‘RD_Explore’. We manually set these variables to two years; in line with the literature that observes
the development of radical innovation is likely to take years (e.g. Burgelman et al., 2004). As
described in the manuscript, these two variables were subject to a multivariate sensitivity analysis.
These variables were given a 5 percent range to vary within (101.4 < 104 weeks < 106.6). The result
(of 200 runs) is reported in Figure 4, which demonstrates that the confidence levels only drop
somewhat in the last 200 weeks of the total simulation period. As such, all simulations up to the 95%
confidence bounds follow the same trend as the historyreplicating simulation. This implies the model
is rather robust.
Confidence level:
Dotted white line represents
the history replicating
simulation.
Figure 4: The sensitivity analysis of the manually estimated ‘exploration’ constants
(AT_Explore and RD_Explore)
Other variables not included in the calibration were the adjustment times (delays) that we could
estimate by means of case observations and reasoning: ‘AT_Management’, ‘AT_Board’, and
‘AT_Pressure’. Data related to these variables become (formally) available to the Board of Directors
and the executive board every quarter. However, only if a certain trend occurs over a period of two
quarters (e.g. negative operating result), the Board of Directors and the executive board are likely to
perceive it as a systematic trend. Therefore these variables were set to 26 weeks (six months). Also
these three variables, including the AT_Myopia variable, were subjected to a sensitivity analysis. All
15
variables were allowed an 8 percent variation. For AT_Management, AT_Board, and AT_Pressure
this resulted in the following range: 24.96 < 26 weeks < 27.04. AT_Myopia had the following range:
438.4 < 456.7 weeks < 474.9). The results (200 simulations) reported in Figure 5 once more indicate
good model robustness.
Confidence level:
Dotted white line represents
the history replicating
simulation.
Figure 5: The sensitivity analysis of the manually estimated Adjustment Time constants
(AT_Management, AT_Board, AT_Pressure, and AT_Myopia)
The historydivergent simulations were also subject to sensitivity analyses. For this, the exogenous
ECD variable was (two times) randomly adjusted over 200 runs. The first set of runs randomly
decreased the ECD variable by up to 50%, simulating a decreased level of dynamism and increased
level of competitiveness (stablescenario). The second set of runs randomly increased the ECD
variable by up to 50%, simulation an increased level of dynamism and a decreased level of
competitiveness (unstablescenario). Figures 6 and 7 depict the results of the sensitivity analyses of
the chosen adjustment in the ECD variable, in the stable respectively unstable scenarios. The results
of both exercises further confirm the robustness of the sequences of events described in the
manuscript: for the stablescenario (Figure 6), all 200 simulations end with a notably decreased
external pressure (EP), while for the unstablescenario (Figure 7) all simulation runs result in the
success trap. As such, the sensitivity analysis for the stablescenario underscores the robustness of our
finding that when top management is able to cope with the environmental change, a low level of
external pressure results and the suppression process (and success trap) is avoided. The sensitivity
analysis for the unstablescenario confirms the robustness of the conclusion that if top management is
not able to cope with environmental change, it will trigger the suppression process and eventually
lock the firm in the suppression of exploration. (Note that from period D onwards, it is very likely that
the firm will need to engage in major reorganizations in order to survive.)
16
Confidence level:
Dotted black line represents the
history replicating simulation.
Figure 6: Sensitivity of the history divergent process theory, stablescenario.
17
Confidence level:
Dotted black line represents
the history replicating
simulation.
Figure 7: Sensitivity of the history divergent process theory, unstablescenario.
18
Table I: Overview of all model constants and settings.
Variable name
Setting
Unit
95% CI
Comments/ Explanation of the source of delay
AT_Exploit
37.7085
Weeks
37.6968 
37.7228
Time necessary to bring about changes in the
routines in exploitation activities.
AT_Explore*
104
Weeks

Time necessary to create, or bring about changes in,
the exploration routines.
AT_Myopia*
456.754
Weeks
449.608 
465.622
Time necessary to overcome managerial myopia.
AT_Management*
26
Weeks

Time necessary to perceive a systematic change in
the environmental situation by the executive board.
AT_Board*
26
Weeks

Time necessary to perceive a systematic trend by the
Board of Directors.
AT_Pressure*
26
Weeks

Time necessary to perceive a systematic change in
the operating results by the Board of Directors.
Initial Inv_Exploit
1
Million
Euros

Necessary for starting the simulation. Initial
situation implies a mere focus on exploitation,
which is in line with the investigated firm.
Initial Inv_Explore
0
Million
Euros

Necessary for starting the simulation. Initial
situation implies a mere focus on exploitation,
which is in line with the investigated firm.
MAR
0.5
Million
Euros

Minimum amount of resources available, even when
the operating result is negative. Required to avoid
model errors.
OC
81.9477
Million
Euros
81.9469 
81.9486
Operating costs assumed as constant.
POR
0.0236391
Percentage
0.0236385 
0.0236396
Percent of the operating result that is available for
investment in exploitation and exploration.
RD_Exploit
35.5818
Weeks
35.596 
35.6136
Time necessary to turn investments in exploitation
into moneygenerating products/processes.
RD_Explore*
104
Weeks

Time necessary to turn investments in exploration
into moneygenerating products/processes.
RF_Exploit
127.774
Euros
127.775 
127.776
Factor to differentiate between the results from
exploitation and exploration. Lower for the former.
RF_Explore
1312.29
Euros
1301.6 
1321.02
Factor to differentiate between the results from
exploitation and exploration. Higher for the latter.
* Subject to sensitivity analysis.
19
Table II: Overview of all model variables.
Variable name
Type
Unit
Comments
Time
reference
AR
Auxiliary
Euros
Resources available for both exploration and exploitation
initiatives.
Current
period
PAE
Auxiliary
Percentage
Perceived alignment with the environment. Can range
from 1 (no gap) till 0 (maximum gap).
Current
period
EP
Stock
Percentage
External pressure to exploit. Can range from 1 (only
invest in exploitation) till 0 (invest in exploitation and/or
exploration).
Current
period
ED
Exogenous data
variable
Percentage
Environmental dynamism (S&P 500 index). Can range
from 0 (extremely instable) till 1 (very stable).
Current
period
EC
Exogenous data
variable
Percentage
Environmental competitiveness (1  Herfindahl index).
Can range from 0 (monopoly) till 1 (extremely
competitive)
Current
period
ECD
Auxiliary

Environmental competitiveness and dynamism. Can
range from 0 (implying a sole need for exploitation) till 1
(implying a mere need for exploration)
Current
period
Inv_Exploit
Stock
Euros
Sum of Euros invested in Exploitation.
Current
period
Inv_Explore
Stock
Euros
Sum of Euros invested in Exploration.
Current
period
PNE
Stock
Percentage
Perceived need to explore. Can range from 0 (only invest
in exploitation) till 1 (only invest in exploration).
Current
period
OR
Auxiliary
Euros
Sum of exploitationexploration ROI’s minus the OC.
Current
period
PECD
Stock

Perceived environmental competitiveness and dynamism.
Can range from 0 (extremely instable) till 1 (very stable).
Current
period
PTOR
Auxiliary
Euros
Average fractional growth rate of OR.
Current
period
RI_Exploit
Auxiliary
Percentage
Percentage of total invested Euros in exploitation
compared to the sum of exploitation and exploration. Can
range from 0 till 1.
Current
period
RIE
Auxiliary
Percentage
Result of the interaction between management (PNE) and
the Board of Directors (EP). Can range from 0 (only
invest in exploitation) till 1 (only invest in exploration).
Current
period
ROI_Exploit
Stock
Percentage
Return on investment exploitation (considering
RF_Exploit and RD_Exploit).
Current
period
ROI_Explore
Stock
Percentage
Return on investment exploration (considering
RF_Explore and RD_Explore).
Current
period
20
Table III: Overview of all functions.
Variable name
Function
AR
MAX (OR * POR, MAR)
PAE
(1 – RI Exploit) * PECD
Change in EP
d (EP) / dt = Change in EP = (Effect of PTOR on PE (PTOR) – EP) / AT_Pressure
ED
(for period t)
EC
(for period t)
1 
∑
=
N
i
i
s
1
2
ECD
(ω = 2/3)
ω * Effect ED on ECD (ED) + (1 ω ) * Effect EC on ECD (EC)
Change in
Inv_Exploit
d (Inv_Exploit) / dt = ((1  RIE) * AR – Inv_Exploit) / AT_Exploit
Change in
Inv_Explore
d (Inv_Explore) / dt = (RIE * AR – Inv_Explore) / AT_Explore
Change in PNE
d (PNE) / dt = (1 – PAE – PNE) / AT_Myopia
OR
ROI Exploit + ROI_Explore – OC
Change in
PECD
d (PECD) / dt = (ECD – PECD) / AT_Management
PTOR
(trend)
(OR – Average_OR) / (AT_Board * Average_OR)
d (Average_OR) / dt = Change in Average_OR = (OR – Average_OR) / AT_Board
RI_Exploit
Inv_Exploit / (Inv_Exploit + Inv_Explore)
RIE
PNE * (1  EP)
Change in
ROI_Exploit
d (ROI_Exploit) / dt = (Inv_Exploit * (1 – ECD) * RF_Exploit – ROI_Exploit) / RD_Exploit
Change in
ROI_Explore
d (ROI_Explore) / dt = (Inv_Explore * ECD * RF_Explore – ROI_Explore) / RD_Explore
Deterministic versus stochastic
An important characteristic of exploration projects is their uncertain nature. That is, employing a
deterministic model, as described in the manuscript, might seem to bias the results (e.g.
ROI_Explore). Therefore, the effect of a stochastic return on exploration investment (ROI_Explore)
was investigated. In order to do so, a Pink Noise (PN) structure was adopted and its outcome
multiplied with the ROI_Explore variable.
Change in ROI_Explore (stochastic):
(18) d (ROI_Explore) / dt = (Inv_Explore * ECD * PN) / RD_Explore
PN is formed by firstorder exponential smoothing of White Noise (WN) and is often referred to
as firstorder autocorrelated noise (Sterman, 2000). The main difference between the two is that the
former has a ‘memory’, and, therefore, the output of t + 1 is not independent from t. For example, if at
a certain t, the investment in exploration initiates is not as profitable as desired (e.g. 90 percent), it is
unlikely that at t + 1 the package projects will generate above expected returns (e.g. 110 percent). As
21
such, PN provides a more realistic noise process than white noise. The following formulas were used
to generate PN (CT equals correlation time). See Sterman (2000) for more details concerning (pink)
noise generation.
Change in PN:
(19) d (PN) / dt = (WN  PN) / CT
(20) WN = Mean + SD * ( ( 24 * CT/dt)0.5) * UNIFORM(0.5, 0.5, Noise Seed)
Following the argumentation in the main text we assume that, effectively, failures will be
counteracted by successes. Therefore, the mean value was set to 1. The SD was set to 0.3, giving the
PN variable a likely range from about 0.95 till 1.05 and a possible range from slightly less than 0.9
and somewhat more than 1.1 (see Figure 8). The overall result of the PN process is depicted in Figure
3 which illustrates the different confidence interval levels for this variable (based on 200 simulation
runs). Figure 9 and 10 illustrate the behavior of the EP and OR variables in this stochastic model. The
influence of PN on the ROI_Explore variable can be seen in Figure 11.The results of the stochastic
model (captured by the confidence interval levels) can now be compared with the deterministic model
(denoted by the doted white lines). We concluded that the stochastic process (PN) does not alter the
results of this study in a noteworthy manner. As such, the model was kept deterministic, for reasons of
readability.
22
Figure 8: Confidence interval levels for the Pink Noise (PN) variable.
Confidence level:
Figure 9: Confidence interval levels for the ROI_Explore variable
(stochastic model).
Confidence level:
Dotted white line represents the
history replicating simulation.
23
Figure 10: Confidence interval levels for the OR variable (stochastic
model).
Confidence level:
Dotted white line represents the
history replicating simulation.
Figure 11: Confidence interval levels for the EP variable (stochastic
model).
Confidence level:
Dotted white line represents the
history replicating simulation.
24
References
Burgelman, R.A., Christensen, C.M. and Wheelwright, S.C. (2004). Strategic Management of
Technology and Innovation. New York: McGrawHill.
Durmusoglu, S.S., McNally, R.C., Calantone, R.J. and Harmancioglu, N. (2008). ‘How elephants
learn the new dance when headquarters changes the music: three case studies on innovation
strategy change’. Journal of Product Innovation Management, 25, 386403.
Jansen, J.J.P., Van Den Bosch, F.A.J. and Volberda, H.W. (2006). ‘Exploratory innovation,
exploitative innovation, and performance: effects of organizational antecedents and
environmental moderators’. Management Science, 52, 16611675.
Sterman, J.D. (2000). Business Dynamics: Systems Thinking and Modeling for a Complex World.
New York: McGraw Hill.
Uotila, J., Maula, M., Keil, T. and Zahra, S.A. (2009). ‘Exploration, exploitation, and financial
performance: analysis of S&P 500 corporations’. Strategic Management Journal, 30, 221
231.
Winter, S.G. (2000). ‘The satisficing principle in capability learning’. Strategic Management Journal,
21, 981996.