Description
Enterprise decision management, commonly abbreviated "EDM" (also known as Business Decision Management Solutions, BDMS) entails all aspects of managing automated decision design and deployment that an organization uses to manage its interactions with customers, employees and suppliers.
Presentation on Bringing Managerial
Decision Models into Models of
Management
There Be Models There
©?? Virtually all major aspects of management have embedded decisions
models, whether we recognize this fact or not
©?? However, we invariably short circuit the loop
Environment Manager Firm
Choice Outcomes
©?? How do we short circuit this?
©?? Fit Models: Environment <÷> Firm => Outcomes
©?? RBV: Firms with Better Resource Portfolios Achieve Better Outcomes
©?? Governance and TMT Models: Specific Structures and Mixtures of TMT Members Achieve
Better Outcomes
©?? Fundamentally nearly all of our theories fail to model the process,
particularly the parts of the process that are excluded
1
20/10/12
Why Is This Relevant?
©??Bad Science. Good science requires that we account
for not only what we are studying but what we are not
studying that can influence what we study
©?? We make extraordinarily strong assumptions about what "managers
are doing" when they make the decisions that lead to the outcomes
that we measure
©?? The burden is on us to show that these assumptions do not materially
influence what we are studying
©??Bad Theory. Unless we structure the phenomena we
are investigating correctly we can erroneously believe
that our theoretical reasoning has validity
©?? e.g., who would ever believe that mindless random processes drive
stock prices?
Why Is This Relevant?
©??Bad Analysis. More than theory, empiricism requires
"realism". Data is generated by a process that must be
aligned well with the theory
©?? E.g., what is the mental process that leads a managerial to answer
"4" or "5" on your survey scale? Or what process drives ROA? Or
the choice of market entry? How does data get in the database?
©??Bad Advice. We invariably make statements about
"managerial implications" that assume that what we
have examined is relevant to the decisions that
managers make (and how they make them)
2
20/10/12
Why Managers Matter:
Thinking About This Conceptually
©??A purely "Deterministic" formulation
Managers do not matter:
Market A Firms going to Market i
can at best located at i*
A*
B* Market B
Local Responsiveness
Why Managers Matter:
Thinking About This Conceptually
©??A "Partially Deterministic" formulation
Managers possessing same
"firms" operating in market i
with difference "preferences"
Ma choose different strategic
pre nag
X*
fere eria orientations
Iso
pro
fitf
ron
t
iers
nce l s
Y*
Local Responsiveness
3
G
l
o
b
a
l
I
n
t
e
g
r
a
t
i
o
n
G
l
o
b
a
l
I
n
t
e
g
r
a
t
i
o
n
20/10/12
What Are the Implications?
©??The "deterministic" framework, which is essentially
Barlett & Ghoshal + "strategic fit" is empty
©??Managers do little but engage in alignment. Firms differ only
to the extent that they "differ" or managers make "strategic
mistakes"
©??The "partially deterministic" framework is quite rich
but we can say nothing without knowing:
©??the profit functions of the firms in a large number of
circumstances (in essence the marginal profit implications of
changes in the parameters of interest)
©??The preference functions of the managers (in essence the
marginal utility implications of the changes in the parameters
of interest)
©??The determination of the equilibrium
Resolving the Issue
©??Indirect Modeling. This is in essence what we do.
However, we basically (a) ignore the manager(s) and
(b) the process by which outcomes are determined.
Resolving this would require us to logically work
through the linkages that are missing (Popper).
©?? It would also require potentially formally modeling what is missing
VS
4
20/10/12
Resolving the Issue
©??Direct Modeling. The correct, and most difficult,
alternative.
©?? Multilevel approaches get at this somewhat (note that we should not
confuse decision modeling with multilevel modeling)
©?? May not be possible with secondary data
©??More reliance on experimental modeling
©?? Still requires simplifying assumptions as (a) there may be a lack of
stationarity (as the decision making process may be time dependent)
and (b) the process is fundamentally unidentifiable
©??E.g., specific assumptions about the decision rules will imply that the
model estimated can be characterized either as a multinomial logit or
ordered probit
What Are Some of the Fundamental Issues?
©?? In placing the manager at the center of our theoretical empirical
models we open up a plethora of (a) interesting questions, (b)
theoretical implications and (c) empirical complexities
How do we integrate
environmental
opportunities and
constraints?
How do we integrate
individual level
covariates? e.g., are
some managers better?
What is the influence of organizational routines? How
do these work to influence the structure of
decision models and outcomes?
How do organizations mitigate 'cognitive bias'?
Environment
What is the form of the model?
What is the most parsimonious
characterization? (e.g.,
Compensatory, Lexicographic?)
What is the most effective empirical
model (e.g., random theoretic?)
Manager
Choice
What is a choice?
Firm
Outcomes
What isWhatcis aechoice? ces)? Out
om = g(Choi
5
20/10/12
Where are the Sources of Variance?
©??Direct Variance
©??Environment
©??Manager
©??Firm
©??Outcomes
©??Process Variance
©??Environment ÷ Manager
©??Firm ÷ Manager
©??Manager ÷ Firm
©??Manager ÷ Choice
©??Choice ÷ Firm ÷
Outcome
©??Choice ÷ Outcome
6
20/10/12
Where Do We Research?
©?? Historically, the dominant
phenomena were associated with
things like:
©?? The MNE (location & structure)
©?? Cross-country studies
©?? Alternative institutional
structures
©??Firm (e.g., JV & WOS)
©??Society
©?? Historically, the dominant
methods were:
©?? Standard and limited
econometric analysis
©??Panel and secondary data
©?? Standard survey and
instrument development
©??Self report with limited
validation
©?? What limitations do we seem to accept (or ignore)?
©?? We accept the less than efficient 'natural experiments' handed
to us (without much questioning)
©?? We make strong assumptions about the domain of our problems
(range of effects)
©?? We know little, if anything, about causality (as opposed to
developing 'causal models' using Lisrel or PLS)
©?? We assume that managers know the best options for them at
any point in time
Less than Efficient Natural Experiments
©??FDI Choice FIRM B
FIRM A X Y Z
X
A. Suppose Firm A and B both chose X Y
but A considered X and Y and B considered X and Z
Z
B. Suppose Firm A chose X and Firm B chose Y
but (1) Firm A considered X and Y and firm B considered Y and Z or (2)
Firm A considered X and Y and firm B considered X and Y
With A and B1 we can say nothing about the population
With B2 we can say something about the population but the
range of our data limits the value
7
20/10/12
What Are Some Sources of a Modeling Dilemma?
©??Do firms consider the same options?
©??The answer is most likely no? (or at least we cannot assume it
is yes)
©??Why don't they consider all the same options?
©??If it is cultural then you have a bias associated with examining
'similarly situated' firms (e.g., all Swedish); i.e., you would have no
real variance and would inflate the role of minor factors
©??They are engaged in a 'location' game
©??If it is bounded rationality what are the boundedness conditions?
©??What is the decision rule? Firms? Boards? Managers?
©??What would they do if they had to consider the same options?
©??Extract out the bias from the fundamental determinants
What Does This Imply (about FDI Models)?
©??At its simplest, we are not measuring any real decision
model
©??Which begs the question of exactly what we are measuring
©??Panel data may not be giving us anything like a natural
experiment (in fact it is very 'unnatural')
©??In other words, our data do not reflect the process of the
decisions being made (we can't extract the rule)
©??Qualitative case analysis is not picking up the
complexity in a form that is relevant
©??What is the fundamental problem?
©??We are assuming a process rather than determining the rules
that drive the data
©??We assume that managers have access to all options and
simply select amongst the plethora available
©??We assume that singular models are the best characterization
8
20/10/12
The Importance of Differences:
Modeling Heterogeneity Preemptively
Thinking About Heterogeneity
©??Heterogeneity is at the heart of many of our theories
©??Cultural differences
©??Institutional differences
©??Institutional distance
©??Path dependence
©??Firm differences
©??RBV/KBV
©??Dynamic Capabilities
9
20/10/12
Traditional Thinking
©??Observed Heterogeneity
©??Individual Level
©??Clusters of individuals displaying specific "cultural" traits in
some pattern that can be represented
©??Country based or "representational" (e.g., power-distance)
©??Organizational Level
©??Characteristic effects that can be accounted for through
observable fixed or characterizable random effects
©??Firm Size, Industry, Firm specific dummies, etc.
©??Unobserved heterogeneity assumed to fall into the error
term, with the DV characterizable by one "dominant"
model
Standard Modelling
Y
i
= o + |
v
j
X
v
j
+ I
v
k
Z
v
k
+ c
i
Where some Z
k
are "controls" for
"observed" heterogeneity
c
i
includes "actual" variance and
"unobserved" heterogeneity
Independent Variable j
10
D
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
20/10/12
A Different Characterization
©??There may not be one dominant model but many
possible models
©??Each individual/firm is a "composite" or "mixture" of
these many possible models
©??Heterogeneity is now
©??Direct and Observed
©??Indirect and Observed
©??Unobserved (some of which we will be able to make Direct and
Observed via modeling)
Modeling Heterogeneity (one example)
M1: Y
1
= o
1
+ |
1
v
j
X
v
j
+ c
1
Each i is characterized as:
e
1i
M1 + e
2i
M2 + ?
i
M2: Y
2
= o
2
+ |
2
v
j
X
v
j
+
c
2
Independent Variable j
11
D
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
20/10/12
A "Better" Characterization
©?? The standard model is simply a special case of the mixture
model with e
1i
=1 for all firms
©?? We can distinguish between controls that need to be in the
model (e.g., as part of the X
v
j
rather than as outside the
model)
©?? We must theoretically know what is a X
v
j
and what is a Z
v
k
©?? Unobserved heterogeneity is revealed through the mixture
equation (e
1i
M1 + e
2i
M2 + ?
i
)
©?? Note that this separates the model variance (c
1
and c
2
) from
true heterogeneity variance ?
i
(hence the individual models
are "better" fitted)
©?? Observed heterogeneity is modeled directly; e.g., [either
separately or more efficiently within the mixture equation]
as e = , + ì
v
k
Z
v
k
+ u
Why Do We Need to Do This
©??We do not model heterogeneity as it appears in reality
but put very hard restrictions on our models that we do
not readily acknowledge
©??Our theories have a very limited and potentially incorrect
characterization of differences
©??Bayesian modeling approaches are increasingly de
rigueur in many fields
©??Classical approaches can be considered special cases
©??IB is a field where "heterogeneity" matters
significantly and our contribution can be in helping to
integrate this further into thinking, theory and modeling
12
20/10/12
What Does This Imply?
©?? Econometric models assume a very strict structure without
recognizing the implications of that structure
©?? Decision = f(Criteria) + c
©??All of the excluded process is assumed into the error
©?? They do not account for individual managerial/firm level
heterogeneity. For example, assume that managers have
different levels of experience. What might that imply?
©?? Decision = f(Criteria) + c
©??All of the heterogeneity is assumed to be in the error
13
doc_309647095.docx
Enterprise decision management, commonly abbreviated "EDM" (also known as Business Decision Management Solutions, BDMS) entails all aspects of managing automated decision design and deployment that an organization uses to manage its interactions with customers, employees and suppliers.
Presentation on Bringing Managerial
Decision Models into Models of
Management
There Be Models There
©?? Virtually all major aspects of management have embedded decisions
models, whether we recognize this fact or not
©?? However, we invariably short circuit the loop
Environment Manager Firm
Choice Outcomes
©?? How do we short circuit this?
©?? Fit Models: Environment <÷> Firm => Outcomes
©?? RBV: Firms with Better Resource Portfolios Achieve Better Outcomes
©?? Governance and TMT Models: Specific Structures and Mixtures of TMT Members Achieve
Better Outcomes
©?? Fundamentally nearly all of our theories fail to model the process,
particularly the parts of the process that are excluded
1
20/10/12
Why Is This Relevant?
©??Bad Science. Good science requires that we account
for not only what we are studying but what we are not
studying that can influence what we study
©?? We make extraordinarily strong assumptions about what "managers
are doing" when they make the decisions that lead to the outcomes
that we measure
©?? The burden is on us to show that these assumptions do not materially
influence what we are studying
©??Bad Theory. Unless we structure the phenomena we
are investigating correctly we can erroneously believe
that our theoretical reasoning has validity
©?? e.g., who would ever believe that mindless random processes drive
stock prices?
Why Is This Relevant?
©??Bad Analysis. More than theory, empiricism requires
"realism". Data is generated by a process that must be
aligned well with the theory
©?? E.g., what is the mental process that leads a managerial to answer
"4" or "5" on your survey scale? Or what process drives ROA? Or
the choice of market entry? How does data get in the database?
©??Bad Advice. We invariably make statements about
"managerial implications" that assume that what we
have examined is relevant to the decisions that
managers make (and how they make them)
2
20/10/12
Why Managers Matter:
Thinking About This Conceptually
©??A purely "Deterministic" formulation
Managers do not matter:
Market A Firms going to Market i
can at best located at i*
A*
B* Market B
Local Responsiveness
Why Managers Matter:
Thinking About This Conceptually
©??A "Partially Deterministic" formulation
Managers possessing same
"firms" operating in market i
with difference "preferences"
Ma choose different strategic
pre nag
X*
fere eria orientations
Iso
pro
fitf
ron
t
iers
nce l s
Y*
Local Responsiveness
3
G
l
o
b
a
l
I
n
t
e
g
r
a
t
i
o
n
G
l
o
b
a
l
I
n
t
e
g
r
a
t
i
o
n
20/10/12
What Are the Implications?
©??The "deterministic" framework, which is essentially
Barlett & Ghoshal + "strategic fit" is empty
©??Managers do little but engage in alignment. Firms differ only
to the extent that they "differ" or managers make "strategic
mistakes"
©??The "partially deterministic" framework is quite rich
but we can say nothing without knowing:
©??the profit functions of the firms in a large number of
circumstances (in essence the marginal profit implications of
changes in the parameters of interest)
©??The preference functions of the managers (in essence the
marginal utility implications of the changes in the parameters
of interest)
©??The determination of the equilibrium
Resolving the Issue
©??Indirect Modeling. This is in essence what we do.
However, we basically (a) ignore the manager(s) and
(b) the process by which outcomes are determined.
Resolving this would require us to logically work
through the linkages that are missing (Popper).
©?? It would also require potentially formally modeling what is missing
VS
4
20/10/12
Resolving the Issue
©??Direct Modeling. The correct, and most difficult,
alternative.
©?? Multilevel approaches get at this somewhat (note that we should not
confuse decision modeling with multilevel modeling)
©?? May not be possible with secondary data
©??More reliance on experimental modeling
©?? Still requires simplifying assumptions as (a) there may be a lack of
stationarity (as the decision making process may be time dependent)
and (b) the process is fundamentally unidentifiable
©??E.g., specific assumptions about the decision rules will imply that the
model estimated can be characterized either as a multinomial logit or
ordered probit
What Are Some of the Fundamental Issues?
©?? In placing the manager at the center of our theoretical empirical
models we open up a plethora of (a) interesting questions, (b)
theoretical implications and (c) empirical complexities
How do we integrate
environmental
opportunities and
constraints?
How do we integrate
individual level
covariates? e.g., are
some managers better?
What is the influence of organizational routines? How
do these work to influence the structure of
decision models and outcomes?
How do organizations mitigate 'cognitive bias'?
Environment
What is the form of the model?
What is the most parsimonious
characterization? (e.g.,
Compensatory, Lexicographic?)
What is the most effective empirical
model (e.g., random theoretic?)
Manager
Choice
What is a choice?
Firm
Outcomes
What isWhatcis aechoice? ces)? Out
om = g(Choi
5
20/10/12
Where are the Sources of Variance?
©??Direct Variance
©??Environment
©??Manager
©??Firm
©??Outcomes
©??Process Variance
©??Environment ÷ Manager
©??Firm ÷ Manager
©??Manager ÷ Firm
©??Manager ÷ Choice
©??Choice ÷ Firm ÷
Outcome
©??Choice ÷ Outcome
6
20/10/12
Where Do We Research?
©?? Historically, the dominant
phenomena were associated with
things like:
©?? The MNE (location & structure)
©?? Cross-country studies
©?? Alternative institutional
structures
©??Firm (e.g., JV & WOS)
©??Society
©?? Historically, the dominant
methods were:
©?? Standard and limited
econometric analysis
©??Panel and secondary data
©?? Standard survey and
instrument development
©??Self report with limited
validation
©?? What limitations do we seem to accept (or ignore)?
©?? We accept the less than efficient 'natural experiments' handed
to us (without much questioning)
©?? We make strong assumptions about the domain of our problems
(range of effects)
©?? We know little, if anything, about causality (as opposed to
developing 'causal models' using Lisrel or PLS)
©?? We assume that managers know the best options for them at
any point in time
Less than Efficient Natural Experiments
©??FDI Choice FIRM B
FIRM A X Y Z
X
A. Suppose Firm A and B both chose X Y
but A considered X and Y and B considered X and Z
Z
B. Suppose Firm A chose X and Firm B chose Y
but (1) Firm A considered X and Y and firm B considered Y and Z or (2)
Firm A considered X and Y and firm B considered X and Y
With A and B1 we can say nothing about the population
With B2 we can say something about the population but the
range of our data limits the value
7
20/10/12
What Are Some Sources of a Modeling Dilemma?
©??Do firms consider the same options?
©??The answer is most likely no? (or at least we cannot assume it
is yes)
©??Why don't they consider all the same options?
©??If it is cultural then you have a bias associated with examining
'similarly situated' firms (e.g., all Swedish); i.e., you would have no
real variance and would inflate the role of minor factors
©??They are engaged in a 'location' game
©??If it is bounded rationality what are the boundedness conditions?
©??What is the decision rule? Firms? Boards? Managers?
©??What would they do if they had to consider the same options?
©??Extract out the bias from the fundamental determinants
What Does This Imply (about FDI Models)?
©??At its simplest, we are not measuring any real decision
model
©??Which begs the question of exactly what we are measuring
©??Panel data may not be giving us anything like a natural
experiment (in fact it is very 'unnatural')
©??In other words, our data do not reflect the process of the
decisions being made (we can't extract the rule)
©??Qualitative case analysis is not picking up the
complexity in a form that is relevant
©??What is the fundamental problem?
©??We are assuming a process rather than determining the rules
that drive the data
©??We assume that managers have access to all options and
simply select amongst the plethora available
©??We assume that singular models are the best characterization
8
20/10/12
The Importance of Differences:
Modeling Heterogeneity Preemptively
Thinking About Heterogeneity
©??Heterogeneity is at the heart of many of our theories
©??Cultural differences
©??Institutional differences
©??Institutional distance
©??Path dependence
©??Firm differences
©??RBV/KBV
©??Dynamic Capabilities
9
20/10/12
Traditional Thinking
©??Observed Heterogeneity
©??Individual Level
©??Clusters of individuals displaying specific "cultural" traits in
some pattern that can be represented
©??Country based or "representational" (e.g., power-distance)
©??Organizational Level
©??Characteristic effects that can be accounted for through
observable fixed or characterizable random effects
©??Firm Size, Industry, Firm specific dummies, etc.
©??Unobserved heterogeneity assumed to fall into the error
term, with the DV characterizable by one "dominant"
model
Standard Modelling
Y
i
= o + |
v
j
X
v
j
+ I
v
k
Z
v
k
+ c
i
Where some Z
k
are "controls" for
"observed" heterogeneity
c
i
includes "actual" variance and
"unobserved" heterogeneity
Independent Variable j
10
D
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
20/10/12
A Different Characterization
©??There may not be one dominant model but many
possible models
©??Each individual/firm is a "composite" or "mixture" of
these many possible models
©??Heterogeneity is now
©??Direct and Observed
©??Indirect and Observed
©??Unobserved (some of which we will be able to make Direct and
Observed via modeling)
Modeling Heterogeneity (one example)
M1: Y
1
= o
1
+ |
1
v
j
X
v
j
+ c
1
Each i is characterized as:
e
1i
M1 + e
2i
M2 + ?
i
M2: Y
2
= o
2
+ |
2
v
j
X
v
j
+
c
2
Independent Variable j
11
D
e
p
e
n
d
e
n
t
V
a
r
i
a
b
l
e
20/10/12
A "Better" Characterization
©?? The standard model is simply a special case of the mixture
model with e
1i
=1 for all firms
©?? We can distinguish between controls that need to be in the
model (e.g., as part of the X
v
j
rather than as outside the
model)
©?? We must theoretically know what is a X
v
j
and what is a Z
v
k
©?? Unobserved heterogeneity is revealed through the mixture
equation (e
1i
M1 + e
2i
M2 + ?
i
)
©?? Note that this separates the model variance (c
1
and c
2
) from
true heterogeneity variance ?
i
(hence the individual models
are "better" fitted)
©?? Observed heterogeneity is modeled directly; e.g., [either
separately or more efficiently within the mixture equation]
as e = , + ì
v
k
Z
v
k
+ u
Why Do We Need to Do This
©??We do not model heterogeneity as it appears in reality
but put very hard restrictions on our models that we do
not readily acknowledge
©??Our theories have a very limited and potentially incorrect
characterization of differences
©??Bayesian modeling approaches are increasingly de
rigueur in many fields
©??Classical approaches can be considered special cases
©??IB is a field where "heterogeneity" matters
significantly and our contribution can be in helping to
integrate this further into thinking, theory and modeling
12
20/10/12
What Does This Imply?
©?? Econometric models assume a very strict structure without
recognizing the implications of that structure
©?? Decision = f(Criteria) + c
©??All of the excluded process is assumed into the error
©?? They do not account for individual managerial/firm level
heterogeneity. For example, assume that managers have
different levels of experience. What might that imply?
©?? Decision = f(Criteria) + c
©??All of the heterogeneity is assumed to be in the error
13
doc_309647095.docx