.3. thug-«N11 4MP).
L at... 1.

u; . . . , v
. .. 5
.111! (I,
:1:.:;. 119;};
.r f. 1.. £1.75; . p
V... 21.71

 

 

 

. z _.
v .9194
,

xi. «

 
 

4 ..
i 5.5 :5
I} ,r

, a
I L . z... 1.
tr!) .2. f, r . 7 f!.

7(3. , .J:
711.312

5.; 9:13... 4
1 ..l5}?/,(
.1: ...lr...y:;
gift!!!

aaifiz... .

 

 

. itfnwfy
‘1’: I

 

 

 

..,. Z .
..,.p.s..rc.
....,4{.44

       

 

 

    

 

 

 

 

 

. 4. fl?!
é.,.rt(.)! \.3r1
11731.

2w 1 . I.
4 by rank"? ; .

 

 

 

THESIS

Iii—'1”

7 if,

This is to certify that the

dissertation entitled

The Processing and Aggregation of Information

Within Organizational Structures

presented by

San gmook Kim

has been accepted towards fulfillment
of the requirements for

. . . ience
Ph D. degreein Polltlcal Sc

 

 

__IhmasJJammm__

Major professor

Date I" ’7‘ an"

MSU is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

.n—‘An

r a
LIBRARY
“hum State
University J

————_

v—v—

 

 

L

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return. on or before date due.

 

 

DATE DUE DATE DUE DATE DUE
esp 216 :99.-

its?)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Affirmative Action/Equal Opportunity Institution
c:\clrc\datedm.pm3-p.1

 

 

 

 

 

 

 

 

 

 

THE PROCESSING AND AGGREGATION OF INFORMATION
WITHIN ORGANIZATIONAL STRUCTURES
By

Sangmook Kim

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department Of Political Science

1992

 

 

 

ABSTRACT

THE PROCESSING AND AGGREGATION OF INFORMATION
WITHIN ORGANIZATIONAL STRUCTURES

By

Sangmook Kim

The main theme of this research is to analyze the processing and aggregation of
informatiOn within organizational structures using a mathematical approach with a formal
theoretical method and Bayesian statistical decision theory.

The first research question speaks to the problem of organizational structure: how
does organizational structure affect the information transmitted through it? By examining
how individuals process information, and an organization aggregates it, this study will
show that organizational structure determines the information which managers receive.
Different reports might be given to the top manager given a different structure.

The second research question stems directly from the first: what kind of
organizational structure will be most appropriate in information processing? When the
function of an organization is to predict a certain event from the environment, the top
manager would want to spend his time and energy as little as possible in information
processing as well as to predict the event as accurately as possible. The key question of
structural design involves the question of how to reduce information cost, and how to
increase event predictability. By discussing criteria suCh as information cost and event
predictability, and by examining information processing in four basic types of
organizational structures, this Study will determine which type is most appropriate for

information processing.

 

 

 

The third question focuses on the problem of strategic behavior in information
processing: what kind of incentive mechanism can be used by the top manager to ensure
that subordinates report information honestly in a particular kind of organizational
structure? By assuming that field agents are influence maximizers and that there is a
feedback process, this study will prove that the relative weight rule will be the Bayesian
incentive compatible mechanism. Under this mechanism, the dominant strategy of the
field agent is to report honestly; and that of the manager is to use the relative weight rule
appropriately. Given the existence of the incentive mechanism, we can compare
information cost and event predictability in organizational structures.

Finally it discusses the relation between organizational sn'ucture and environment.
This study will Show that different principles of information processing will be
emphasized when facing different characteristics of the environment, and that different

structures will work better in different environments.

 

 

 

Copyright by
Sangmook Kim
1992

 

To my family

 

ACKNOWLEDGENIENTS

I wish to express my sincere thanks to Professor Thomas H. Hammond for his
great personality, continuous guidance and encouragement.

I would like to thank Professors Jeffrey Hill, Scott Gates, Kenneth Williams, and
Larry Heimann, for serving on my dissertation committee. Also my special thanks go to
Professor Brian Humes of the University of Nebraska, Lincoln, Renee Smith of the
University of Rochester, Marcus Cheatham, Jonathan Monroe, and Tammy Essig, for
reviewing the earlier drafts. The assistance of the Political Science Department staff is
greatly appreciated.

Special thanks to the VanAlstine family--Aaron, Jack, and Margie--in Marquette,
Michigan, for their deep friendship.

Finally I wish to give special thanks to my wife, Young-Ac, who has prepared
luncheon boxes over four years, and my daughter, Iris Jeehyun, who has said "Please

don’t go to school. Play with me!" with a wonderful kiss every morning.

 

List of Tables

Table of Contents

List of Figures

1. Introduction

2. Literature Review

3. Dimensions in Organizational Research
3.1 A hypothetical setting

3.2 Information aggregation and feedback
3.3 Principles in information processing
3.4 An information processing game
4. Information Processing in Organizational Structures
4.1 Four models of organizational structures
4.2 Flat one-to-one structure
4.3 Flat one-to-many structure
4.4 Tall one-to-one structure
4.5 Tall one-to-many structure
4.6 Comparative analysis of organizational structures
5. An Example
6. Organizational Structure and Environment
7. Conclusion
BIBLIOGRAPHY

vii

P a g e
viii

ix

21
21
32

40

53
53
57
62
67
73
78
94
109
114

121

 

 

 

 

List of Tables

Table 1: Information Processing and Aggregation
Table 2: Four Models of Organizational Structure
Table 3: Characteristics of Organizational Structures

Table 4: Optimal Organizational Structure in Different Environments

viii

Page
38
57
85

113

 

 

 

 

Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
Figure 7:

Figure 8:

List of Figures

An example of flat one-to-one structure
Information processing in flat one-to-one structure
An example of flat one—to-many structure
Information processing in flat one-tovmany structure
An example of tall one-to-one structure

Information processing in tall one-to-one structure
An Example of tall one—to-many structure

Information processing in tall one-tO-many structure

ix

Page
86
87
88
89
9O
91
92

93

 

 

 

 

 

1. Introduction

The literature on organizational design has long studied the properties of
organizational structures. However, it is still not clear how much structure matters, and
what structure should lie chosen under what circumstances.

One important aspect of an organizational structure involves its impact on
organizational information processing. However, current studies of organizational design
have not attempted to analyze information processing directly nor to link organizational
structure to information processing. Organizational structure may affect what kind of
information the organization gathers, how information is processed and aggregated, how
efficient and reliable the organizational information processing will be, and what the
organizational decision will be. The purpose of this dissertation is to discover how much
organizational structure affects information processing, and what kind of organizational
structure will be most appropriate for its organizational environment.

I will address three questions about organizational structure and information
processing. The first research question speaks to the problem of organizational structure:
how does organizational structure affect the information transmitted through it? Current
studies Of organizational design argue that organizational structure exerts a strong
influence on information flows in organizations, but these studies do not show precisely
how organizational structure influences information flows. By examining how individuals

process information, and how an organization aggregates it, this study will Show that

organizational structure determines both the scope of information and the set of

 

 

 

 

 

2

information aggregation rules. Organizational structures are about aggregation procedures
that transform the individual reports into a collective one. Hence to design an
organizational structure is to choose a certain kind of aggregation procedure. Different
reports might be given to the top manager given a different structure.

The second research question stems directly from the first: what kind of
organizational structure will be most appropriate in information processing? Current
studies argue that different organizational structures are appropriate in different situations.
However, they do not demonstrate how different structures produce differences in
information processing nor do they show how to evaluate information processing in
different organizational structures. They have only formulated abstract hypotheses. When
the function of an organization is to predict a certain event from the environment, the top
manager would want to predict the event as accurately as possible but also to spend as
little of his time and energy as possible in information processing. The key question of
structural design involves the question of how to reduce information cost, and how to
increase event predictability. By discussing criteria such as information cost and event
predictability, and by examining information processing in four basic types of
organizational structures, this study will determine which type is most appropriate for
information processing.

The third question focuses on the problem of strategic behavior in information
processing: what kind of incentive mechanism can be used by the top manager to ensure
that subordinates report information honestly in a particular kind of organizational

structure? If it is useful to deal with the effect of organizational structure on information

 

 

 

 

3

processing, it is equally important to consider the role of strategic behavior in information
processing. Thus both the organizational structure and the incentive mechanism used to
ensure sincere information processing should be linked in the attempt to develop effective
organizational responses to the environment. Even a properly structured organization does
not ensure sincere information transmission. If the field agent is an expected utility
maximizer, he would use a successful strategy and avoid an unsuccessful one. If a
strategy is not successful, he will search for alternatives. By assuming that field agents
are influence maximizers and that there is a feedback process, this study will prove that
the relative weight rule will be a Bayesian incentive compatible mechanism. Under this
mechanism, the dominant strategy of the field agents is to report honestly. This
mechanism will insure honest reporting given the organizational structure.

This study will focus on how structure affects information processing in the
organization. Other upward flows such as policy alternatives or conflict resolution remain
to be considered.‘ For the purpose of showing clearly the effects of structure on
information flow, all other factors influencing information processing except structure are
assumed to be constant. However, after discussing and comparing organizational
structures, the assumption will be modified. This study will analyze information
processing in four models of organizational structures using a mathemathical approach

with a formal theoretical approach and Bayesian statistical decision theory. A

 

Sah and Stiglitz (1985) studies the aggregation of individual errors under different organizational structures.
Calvert (1985) argues that the rational decision maker prefers a biased information over a neutral one.
Hammond (1986) researches the aggregation of policy options and the resolution of conflicts under different
organizational structures.

 

 

 

 

4

formalization of organizational structure is useful since it shows why an organizational
structure matters, and it is general enough that it can be made more specific with addition
of several different variations.

The concept of information processing provides a useful method to explain the
characteristics of organizational structures. In this study, an organization is viewed as an
information processing system (Marschak and Radner 1972, Galbraith 1969, 1977,
Tushman and Nadler 1978, Egelhoff 1982). Information processing refers to the
gathering, the transmission, and the aggregation of information in organizational decision
making procedures. It is a vital aspect of organizational decision making since
organizational decisions depend on information that is imperfect in a number of obvious
ways (Feldman and March 1981). Organizations process information for effective
management of both "uncertainty," referring to "the difference between the amount of
information required and the amount of information possessed by the organization," and
"equivocality," that is "the existence of multiple and conflicting interpretations about an
organizational situation" (Daft and Lengel 1986, p.556).

Organizational structure exerts a strong influence on information flows in
organizations (Aguilar 1967, Galbraith 1977, Egelhoff 1982). Organizational structure
and internal processing systems determine the scope of information provided to decision
makers (Daft and Lengel 1986). An organization is "(1) composed of people and groups
of peOple (2) in order to achieve some shared purpose (3) through a division of labor (4)
integrated by information-based decision processes (5) continuously through time"

(Galbraith 1977, p.3).

 

 

 

 

 

5

Following this definition of organization, this study is based on several
assumptions about organizations. A basic assumption is that an organization is an open
social system which is affected by, and dependent upon, its environment (Lawrence and
Lorsch 1967, Thompson 1967 , Pfeffer and Salancik 1978). Since there are several
sources of uncertainty to which the organization should respond, it should be able to cope
with environmental uncertainty. As Otley (1988, p.87) put it: "Both organizational
structure and the mechanisms used to establish internal control are linked in the attempt
to develop effective organizational responses to environmental conditions."

The second assumption is that an organization is an information processing system.
The organization’s structure functions to process and aggregate information through
hierarchical routes (Galbraith 1969, 1977, Tushman and N adler 1978, Knight and
McDaniel 1979, Kmetz 1984). To anticipate environmental uncertainty, an organization
observes the external environment, gathers data, and transmits information. "Information
processing comprises (1) the assembly of raw data and information, i.e., the collection of
inputs, and (2) modification of the inputs, which include transformation, analysis, or
synthesis, into different forms" (Kmetz 1984, p.276). Organizational decision making
relies on the information transmitted through the organizational structure.

The third assumption is that an organization is a composition of individuals and
divisions based on the principle of the division of labor (Gulick 1937, Mintzberg 1979).
Following the principle of "grouping" which is "a fundamental means to coordinate work
in the organization" (Mintzberg 1979, p.106), an organization tries to group together

subordinates between which information aggregations are important. Each individual or

 

 

 

 

6

subunit utilizes a specific method and technology to observe the environment and gather
information which differs from the others. Relying on these general assumptions, this
study views an organization as an information processing system. In the following
section the procedures how to analyze information processing in organizational structures
will be introduced.

This study will be organized as follows. The next chapter will review current
studies on organizations. This literature review will confirm how important is this
research, and show that organizational structure itself has not been thoroughly
investigated.

Chapter 3 will describe a hypothetical setting of this research and the problems
of information processing. The goal of an organization in this study is to predict the
probability of a particular recurring event. The field agents gather information from the
external information source (the environment) and transmit the reports on the probability
of the occurrence of the event in the near future to the manager. The tOp manager, who
represents the organization, gathers the information and makes the organizational
prediction of the event. The individual in the organization is assumed to be a Bayesian
player: he/she uses the Bayesian rule in making the report, and follows the principle of
expected utility maximization. Secondly, this study will discuss information aggregation
and feedback. The organizational structure is about information aggregation. The role
of managers is to aggregate the reports from the subordinates and to simplify the
information. The relative weight rule will be developed as an appropriate aggregation

rule. Thirdly, information cost and event predictability will be discussed as the principles

 

 

 

 

 

7

of information processing. The information cost of an organization is the sum of the
managers ’ information processing costs and the field agents’ observation costs. But much
more important will be the costs of information overload on higher—level managers. Event
predictability refers to how accurate is the organizational prediction of the event. For
event predictability, the more information an organization generates, the more accurately
the organization predicts the event, and the more agreements about the probability an
organization produces, the more reliable and objective is that probability. The design of
a certain organizational structure means how to give emphasis to these principles. Finally
it will attack the third research question by demonstrating that the problem of
misrepresentation can be solved by applying the relative weight rule as an incentive
mechanism. Under the assumptions that all individuals are Bayesian players and that
there exist feedback processes, the relative weight rule is a Bayesian incentive compatible
mechanism. In an information processing game the dominant strategy of the field agent
is to report honestly and that of the manager is to use the relative weight rule
appropriately.

Based on two dimensions important in organizational design (specialization and
coordination), Chapter 4 will generate four basic models of organizational structures: Flat
one-to-one structure, Deep one-tO-one structure, Flat one-to-many structure, and Deep
one-to-many structure. It will discuss the characteristics of, and the advantages and
disadvantages of, each kind of organizational structure. Different organizational structures
transmit different information to the top manager. The information cost of tall structures

is less than that of flat structures for the top manager as well as the whole organization.

 

 

 

 

 

8

The flat one-to-one structure generates greater information predictability than the other
structural alternatives, given the existence of the relative weight rule. When considering
both information cost and information predictability, the tall one-to-one structure is better
than the other structures when the number of subordinates is not too small.

Chapter 5 will illustrate the models with a numerical example. This example
demonstrates how organizational structures affect the processing and aggregation of
information in the organization.

Chapter 6 will analyze the relation between organizational structure and
environment. When categorizing the environment with two dimensions dealing with
degree of information homogeneity and degree of information stability, this chapter will
Show that different principles of information processing will be emphasized when facing
different characteristics of the environment, and that different structures will work better
in different environments.

The concluding chapter will summarize the results of the study, discuss the issues

of coalition and risk preference and suggest some directions for further research.

 

 

 

 

2. Literature Review

Contingency theory, the studies of structure and strategy, transaction cost theory,
and principal-agent theory are the bodies of literature most relevant to this research.
Contingency theory emphasizes a fit between organizational environment and
organizational structure. The studies of structure and strategy focus on the causal
relationship between strategy and structure. Transaction cost theory tries to compare
different governance structures, mainly focusing on market structure and hierarchical
structure. Principal-agent theory asks how to design an incentive system which secures
sincere behavior in organizations.

Contingency theorists argue that the structure of an organization should match or
fit together internal and external organizational variables (Thompson 1967, Lawrence and
Lorsch 1967). The central arguments of contingency theory can be expressed in two
equivalent ways:

1. [An organization]’s efficiency is dependent on the relation between the
state of the environment and the form of the organization; or

2. Under conditions of efficiency, organizational form is correlated with
the state of environment (Grandori 1987, p.1).

Burns and Stalker (1961) suggested that the mechanistic form -- formal and
bureaucratic -— is effective in a stable environment, while the organic form -- less formal
and more flexible -- is effective in a rapidly changing environment. Different
organizational principles are thus appropriate in different organizational environments

(Woodward 1965). If a firm is to be successful, its design should be contingent upon the

 

 

 

 

10

characteristics of the environment in which it operates. Lawrence and Lorsch (1967) and
Lorsch and Allen (1973) observed that the achievement of a degree of differentiation
consistent with the requirements of the environment and the achievement of a degree of
integration consistent with the required interdependence of these parts was related to
higher organizational performance. "The best way to organize depends on the nature of
the environment to which the organization must relate" (Scott 1981, p.114).

Earlier findings on the relation between organizational environment and structure
were conflicting. Several researchers Observed that an increase in environmental
uncertainty was associated with a decentralized organizational structure (Lawrence and
Lorsch 1967, Duncan 1973). On the other hand, other studies found that an increase in
organizational uncertainty was associated with a centralized organizational structure
(Huber et a1. 1975, Bourgeois et a1. 1978). In addition, some studies showed that there
was no evidence for the pr0position that organic or decentralized organizations perform
better than centralized organizations in uncertain environments (Mohr 1971, Pennings
1975). Recent empirical studies have also shown mixed results.1

Contingency theory is criticized for several flaws. Lawrence (1981) summarized
the major problems:

1. It is only static, not dynamic.

 

Koberg and Ungson (1987) observed that fit between environment and organizational structure does not explain
performance. Dastrnalchian and Boag (1990) studied the relationships between marketing department structure
and organizations’ dependencies on markets and customers. The result showed that both specialization and the
degree of integration of the marketing department were positively affected by market dependency. Centralization
of decision making was positively affected by customer dependency. The results were more evident in
successful firms. Departmental formalization was also positively affected by market dependency in successful
firms. Randolph et. a1. (1991) analyzed the effects of the fit between technological innovation and organizational
structure on small businesses. They found that fit has a significant positive relationship with financial
performance for low growth companies, but it is unrelated to performance in high growth companies.

 

 

 

 

11

2. It did not treat the concept of environmental uncertainty with conceptual clarity.

3. It started a trend that has gone too far in seeking a never-ending stream of

contingent variables to account for organizational features.

4. It did not offer an explanation for the observed uniformities (Lawrence 1981).
However, more important is that contingency theory does not explain the core: why and
how different organizational structures are appropriate in different organizational
environments. Since it does not analyze the characteristics or roles of organizational
structure itself, it fails to relate the description of strucrure with that of organizational
functions. Nevertheless, it indicates that organizational structures need to be compared
with regard to the characteristics of organizational environment.

The studies on organizational Structure and strategy have debated two contradictory
arguments: structure follows strategy (Chandler 1962, 1977, Rumelt 1974), and strategy
follows structure (Bower 1970, Hedberg et al. 1976, Bobbitt and Ford 1980). The
doctrine that structure follows strategy was initiated by Chandler (1962), who
demonstrated a clear historical connection between the firm’s internal structure and the
sc0pe of its activities. He argued that structural choice follows from strategic choice.
Superior performance is argued to be the product of an appropriate fit between strategy
and structure. He found that the most fundamental change in large American corporations
was a move from centralized, functionally departmentalized structures (the U Form) to
multi—divisional ones (the M Form). Galbraith and Nathanson (1978, 1979) concluded
that the organization must achieve a fit between its strategy, its structure, and its
processes. But the concept of "fit" is still unclear: "Although the concept of fit is a useful

one, it lacks the precise definition needed to test it and to recognize whether an

organization has it or not" (Galbraith and Nathanson 1979, p.266).

 

 

 

 

12

On the other hand, Bower (1970) argued against the doctrine that structure
follows strategy; instead, the organizational structure affects the kind of information which
the top manager will receive from subordinates: "When management chooses a particular
organization form, it is providing not only a framework for current operations but also
the channels along with which strategic information will flow..." (Bower 1970, p.287).
Pettigrew (1973) explained strategic action as the result of an internal political process
which is affected by the division of labor. Fredrickson (1986) reviewed the
strategy/structure debate and concluded that the organizational structure has important
deterministic effects on the strategic decision process and its outcome. Hammond (1986)
explained how organizational structures might act in ways similar to legislative agendas
and thereby affect policy outcomes. Hammond (1991b) concluded that strategy follows
structure since the organizational structure affects what information the t0p manager
receives, and what alternatives are made available to him. Bourgeois and Astley (1979)
suggested that the relationship between strategy and structure must be reciprocal.
Depending upon what part of the strategic process is observed, both "structure follows
strategy" and "strategy follows structure" can be correct propositions (Burgelman 1983).
Recently, Chandler (1991) has said that structure has as much impact on strategy as
strategy has on structure. Boschken (1990) divided structure into two variables, micro-
structure and macro-structure. Micro-structure means "a set of coordinated subunits
assigned the critical tasks of designing for the whole organization and creating appropriate
implementation policies and changes in operational structure," while macro—structure

contains all organizational activities (Boschken 1990, p.136). His case study showed that

 

 

 

 

13

the micro-structure provides a means to determine strategy, while strategy determines the
design of a macro—structure. The earlier studies showed that there is no strong proof that
a fit between strategy and structure leads to effective performance (Galbraith and
Nathanson 1978). Results from recent empirical studies remain inconclusive.2

These studies provoke the basic question of this research: why does a
organizational structure matter? These studies indicate that an organizational Structure has
its own effects on organizational functions, whether strategy formulation or strategy
implementation or both. But still these studies do not explain what are the characteristics
of different organizational structures nor do they prove how different structures affect
differently.

Transaction cost theory views the organization as a "stable pattern of transactions"
(Ouchi 1980, p.140). "A transaction occurs when a good or service is transferred across
a technologically separable interface" (Williamson 1985, p.1). Transaction costs mean
"the negotiating, monitoring, and enforcement costs that have to be borne to allow an
exchange between two parties to take place" (Jones and Hill 1988, p.160). Transaction
costs can be divided into the two types: ex ante and ex post costs (Williamson 1985).
The first are the costs of drafting, negotiating, and safeguarding an agreement. The

second include the setup and operating costs of the governance structure. Both the ex

 

Bart (1986) observed that the relationship between strategy and structure is severely constrained. In the study
of small and medium sized organizations, Miller (1987) found that the way a business is organized affects its
strategic planning and strategy-making procedures. Provan (1989) examined the influence of internal
organizational power on strategy and concluded that department power has a strong influence on the planning
and implementation of strategy in organizations. In a study of the five largest Canadian banks, Murray and
J avidan (1987) conclude that it is misleading to look for the exact casual relationship between strategy and
structure, because the two are inextricably bound.

 

 

 

 

14

ante and ex post costs of transaction are interdependent. The sources of these costs are
the transaction difficulties that may be present in the exchange process. The factors
producing transaction difficulties are: bounded rationality, opportunism, uncertainty and
complexity, small numbers, information impactedness, and asset Specificity (Williamson
1975, 1985, Jones and Hill 1988). Transaction cost theory argues that "the costs of
economic activities are not technologically-determined but dependent on the form of
organization under which the activities are conducted" (Yarbrouch and Yarbrouch 1988,
p.7). The core methodological properties of transaction cost theory are:

l. The transaction is the basic unit of analysis.

2. Human agents are subject to bounded rationality and self-interest.

3. The critical dimensions for describing transactions are frequency,

uncertainty, and transaction-specific investments.

4. Economizing on transaction costs is the principal factor that explains

viable modes of contracting; it is the main issue with which organizational

design ought to be concerned.

5. Assessing transaction cost differences is a comparative institutional

exercise (Williamson and Ouchi 1981, p.367).

Coase (1937) used the concept of transaction costs to explain the emergence of
the firms. Coase argued that firms only emerge when an exchange in the market place
may be more costly than the same exchange in a hierarchical organization. The firm can
reduce the transaction costs by lessening information costs, lowering the number of
necessary contracts, and lengthening the terms of some contracts. Alchian and Demsetz
(1972) argued that when it is difficult to monitor or meter the individual contributions of
each member in a team production, and each member has an incentive to shirk, hierarchy

has definite monitoring and enforcement advantages for limiting opportunism.

Williamson (1975) argued for market trading and internal organization as competing

 

 

 

15

modes of handling transactions.3 Under conditions of market failure, hierarchy is more
efficient than the market mechanism in executing transactions since the contracting parties
have greater control and surveillance over uncertainties associated with a transaction and
over the personal opportunism of the parties involved. The multi-divisional structure is
a more effective means of allocating capital than the other structures.

Transaction costs are economized by assigning transactions to governance
structures in a discriminating way (Williamson 1985). Kleig, Crawford and Alchian
(1978) focused on vertical integration of production. They pointed out that whenever
asymmetric transaction specific investments exist, dependence exists; whenever
dependence exists, there exists the potential for opportunistic exploitation of those who
are dependent; and, some remedies to the problems associated with dependence are
implicit contracts and vertical integration. Ouchi (1980) suggested that the clan form of
governance of transactions has some efficiency and advantages. A clan can be seen as
an alternative mode of organizing, in addition to markets and hierarchies. A clan can be
seen as a set of control mechanisms that are legitimated by "a high degree of goal
congruence, typically through relatively complete socialization brought about through high
inclusion" (Ouchi 1980, p.136).

Jones (1983) analyzed how organizational culture emerges from the institutional
arrangements developed to regulate the transactions between members of an organization.

Transaction cost minimization determines the efficient property rights structure that

 

Williamson (1975, 1985) and Chandler (1962, 1977) argued that hierarchies replaced markets because they were
more efficient. However, Williamson focuses on transaction costs, while Chandler on administrative
coordination.

 

 

 

 

16

defines the organizational culture. Jones (1983) argued: "the form of culture that emerges
in an organization may be seen as a consequence of the way in which the firm attempts
to economize on the transaction costs associated with its production function. The origin
of an organizational culture is the system of property rights that is used to structure the
exchange relationships between organization members" (p.465). Recently Williamson
(1991) identified three generic forms of economic organization —- market, hybrid, and
hierarchy -- and explained that three generic forms are distinguished by different
coordinating and control mechanisms and by different abilities to adapt to disturbances.

Empirical studies have provided mixed evidence on the major arguments of
transaction cost theory. Some studies showed that there is a positive relationship between
the multi-divisional innovation and overall economic performance, while others
demonstrated negative or non—significant relations (see Cable 1988). Fligstein and Dauber
(1989) reviewed empirical studies of structural change in corporations and concluded that
the transaction cost perspective is not supported empirically. One interesting study
suggested that the economic approach to organizational structure is valid but insufficient,
and needs to be combined with the ecological and political approach (Palmer et al. 1987).

The criticisms of transaction cost theory are mainly focused on two questions:
what are transaction costs?; and how do hierarchical employment relations reduce these
costs? Robins (1987) stated that: "while transaction-cost analysis offers considerable
potential as a prescriptive approach to the problems of business strategy, it has far less
promise as a means of dealing with the larger issues associated with the evolution of

organizational form" (p.82). Dow (1987) said that transaction cost analysis will not

 

 

 

17
provide any causal explanation for the origin or persistence of these structures. Demsetz
(1988) argued that the transaction cost is only one element of the cost of purchasing from
others, and that there are a variety of others. Dow (1987) and Hammond (1991b)
indicated the difficulty of comparing transaction costs. Hammond (1991b) explained:

Transaction cost analysis would lead us to expect that changing a structure

will change the set of transactions that occur because the costs of some

transactions will now be greater than their benefits, while the costs of other

transactions will now be less than their benefits. But since the two
structures are now engaging in somewhat different transactions, there is no
common basis for comparison; they are simply two different organizations

doing different things. On what basis can it then be argued that one

su'ucture is more efficient than the other? (p.13).

The usefulness of transaction cost theory is to provide an analytical framework for
comparing the relative costs demanded by different organizational structures. But since
the precise nature of transaction cost is unclear, and since without common basis it is
difficult to compare transaction costs in different structures with different conditions, it
is still questionable how one structure can be compared with another.

Principal-agent theory views an organization as "a cascade of principal-agent
relationships," each superior acting as a principal in relation to his subordinates, and as
an agent in relation to his own superior (Radner 1991, p.218). Principal-agent
relationships arise whenever the principal cannot perfectly and costlessly monitor the
agent’s action and information (Spence and Zeckhauser 1971, Jensen and Mecking 1976,
Holmstrom 1979, 1982, Fama and Jensen 1983, Pratt and Zeckhauser 1985, Hart 1990).
Principal-agent theory assumes:

l. The principal is in a position to design the monitoring and incentive

mechanism.
2. All the benefits from improvements in performance go to the principal.

 

 

 

18

Principal-agent theory starts by making general assumptions about the preference structure
of the actors, the nature of uncertainty present, and the information distribution between
actors (Jensen 1983). This model incorporates two basic features of organizations:
asymmetric information and goal conflict among organization members. The principal-
agent theory is concerned with how the principal can design an incentive system which '
motivates his agent to act in the principal’s benefit. An equilibrium of the principal-agent
model is:

1) Given the announced compensation-pair, the agent chooses his action so as to

maximize his own expected utility.

2) Given the optimizing behavior of the agent described in 1), the principal

chooses a compensation-pair that maximizes his own expected utility (Radner

1991, p.237).

Alchian and Demsetz (1972) suggested that monitoring by hierarchical superiors
is necessary where team production makes it impossible to prevent shirking. Jensen and
Meckling (1976) argued the problem of the principal-agent relationship exists in all
organizations and in all cooperative efforts. Jensen (1983) defines an organization as a
legal entity that serves as a nexus for a complex set of contracts among disparate
members. Thus the behavior of the organization is the equilibrium behavior of a complex
contractual system made up of agents with diverse and conflicting objectives.

The principal-agent problems of asymmetric information can be divided into five
categories:

...consider the problem of an employer (the principal) hiring a worker (the

agent).... If the employer knows the worker’s ability, but not his effort

level, the problem is moral hazard with hidden actions. If neither player

initially knows the worker’s ability, but after the worker accepts a contract

he discovers it, the problem is moral hazard with hidden information. If
the worker knows his ability from the start, but the employer does not, the

 

 

 

 

 

 

19

problem is adverse selection. If in addition to the worker knowing his

ability from the start, he can acquire education observable by the employer

before they make a contract, the problem is signalling. If the worker

acquires his education in response to the contract offered by the employer,

the problem is screening (Rasmusen 1989, p.134).

Holmstrom (1982) enumerated the three desired characteristics of an incentive
system: Nash equilibrium, budget balancing, and Pareto efficiency. Holmstrom examined
the problem of free-riding and the role of competition in the case of a principal
monitoring many agents. He showed that a problem of moral hazard with hidden actions
may occur when there is no uncertainty about output. To solve the shirking problem the
principal needs to either enforce the penalties or to finance the bonuses.4 Groves (1973,
Groves and Ledyard 1977, Groves and Loeb 1979) exarrrined the entire class of incentive
mechanisms in the context of general team decision models, and discovered a class of
truth-inducing mechanisms which are immune to misrepresentation. But Miller (1992,
p.154) responded that: "There do exist incentive mechanisms that can induce efficient
levels of effort (Holmstrom’s joint forcing contract) or accurate revelation of information
(Groves-Loeb mechanisms), but they all violate budget balancing." Eisenhardt (1989)
summarized the contents of principal-agent theory and then concluded that agency theory
provides a unique, realistic, and empirically testable perspective on the problems of
c00perative effort (see also Moe 1984, Levinthal 1988). However, principal-agent theory
tells us about Optimal incentive schemes but not directly about organizational structure

(Hart 1990).

Principal-agent theory indicates that the problem of the principal-agent relationship

 

Brehm and Gates (1991) reviewed several models of hierarchical supervision and control.

 

 

 

20

exists in all organizations, and tell us about incentive schemes. But still unclear is how
to design an Optimal incentive mechanism for solving this principal-agent problem in
organizational contexts. Also it does not tell directly about organizational structure. Thus
it is necessary to combine the studies on incentive mechanisms with the studies on
organizational structure.

This literature review indicates that an organizational structure has its own impacts
on organizational functions, and that organizational structures need to be compared with
regard to the environmental characteristics, and the relative costs. It also indicates that
the problem of strategic behavior in organizational contexts should be solved. However,
this literature review shows that organizational structure itself has not yet been thoroughly
investigated, and that the basis of comparing organizational structures has not yet been
developed. Still unanswered but important are the issues: what are the characteristics Of
different organizational structures, and how do different structures affect differently; why
and how are different organizational structures appropriate in different organizational
environments; and how can the organization design an Optimal incentive mechanism for
ensuring the sincere behavior? These issues will be thoroughly discussed and solved
through the following chapters. The problem Of incentive mechanism design, and the
basis of comparing organizational structures in information processing will be analyzed
in Chapter 3. The characteristics of organizational structures will be discussed and
compared in Chapters 4 and 5. The relation between organizational structure and

environment will be explained in Chapter 6.

 

 

 

3. Dimensions in Organizational Research

3.1 A hypothetical setting

Assume that the goal of an organization is to predict the possibility of a certain
recurring event A--for example, military coups, battles, international conflicts, and riots--
which has a set of dimensions, A = (A ,,...,A,,). Assume that the environment E has a set
Of independent sectors E,, E = (E,,...,E,,), (E, n E!) = O, E, U EJ- = E,- + E}. The
environment is the part Of the external information flow to which the organization
responds through Observation, belief, and action. E can be understood as the information
source through which the organization can get information on A. The event A can be
predicted by knowing the environment E.

The responsibility of subordinates or field agents is to observe the environment
and report the probability of event A in their area. The field agent would gather some
number of samples which he would evaluate as tO whether event A will occur or not.
Each field agent is independent from the Others: it is assumed that there is neither
horizontal Opinion sharing nor peer reviews, and each agent has his own method of
analyzing the data. Thus, even though several agents may Observe the same sample (i.e.,
from the same Ei), their conclusions may be different. The field agent’s report does not
depend on the Others’ reports nor does he/she know how the others make their
observations. The top manager who represents the organization gathers reports from all
subordinates, whether directly or indirectly, and formulates an organizational conclusion

from these reports. At each time, the organization predicts the probability whether A will

21

 

 

 

22

occur or not at a certain point of time in the future.

A1. Individuals as Bayesian players

This study assumes that the individuals in an organization maximize expected
utility and act as Bayesian players when they acquire information, revise their
probabilities, and act upon this new information. A Bayesian decision model of
organizational behavior provides a useful framework for structuring information
processing. Individual strategies for reporting an estimate of the probability of event A
are governed by the expected utility of alternative reports, that is, by the utility attached
to each outcome weighted by the assessed probability of its occurrence. The individual
would transmit the report he believes will maximize his expected utility. From the
Bayesian point of view, the individual has both his own subjective probability, which
represents his knowledge and beliefs, and his own utility, which represents his tastes and
preferences.

Individual behavior is empirically quite consistent with a Bayesian approach.
Experimental research shows that individuals and groups within an organization act as
Bayesian players (Viscusi and O’Connor 1984, Viscusi 1985, Viscusi and Magat 1987).
For example, after analyzing an experiment on job risks with chemical workers, Viscusi
(1985, p.384) concludes:

Overall, individuals do not possess perfect information about the risks they

face, but they do have Opportunities to revise beliefs based on their

experiences. The Observed behavior patterns are consistent with the

principal predictions of a Bayesian learning process and subsequent
adaptive behavior.

 

 

 

23

The individual can generate a unique probability distribution over the states of
event A. In Bayesian perspective, any individual has a subjective probability distribution
over the possible values of any parameter that he does not know (Savage 1954, Raiffa
1968). Each individual has past experiences and beliefs that can influence probability
estimation. Before Observing the environment, he already has some prior probability of
event A. This probability value is a subjective probability in the sense that it derives
totally from the individual’s personal information about event A. According to Savage
(1962, p.163), the subjective probability is "a certain kind Of numerical measure of the
Opinions of somebody about something"; see also French (1982). All probabilities are
subjective; some are more "Objective" than others only in that a larger group Of field
agents would assign the same values for these probabilities based on their own
information sets (Cyert and DeGrOOt 1987, ch.2).

The theory of subjective probability is concerned with quantifying judgments of
likelihood (Fishbum 1986, Kreps 1988). Let us assume that there is a Boolean algebra
B of subsets X1, X2,... of a universal set S. Q is the empty set and Q g X, g S for every
X,- in B. A probability measure on X,- is a real valued function p: X,- —> [0, 1] satisfying
p(S) = l and p(X1 u X2) --= p(Xl) + p(X7) if X1 0 X2 = Q. Then let us read X1 > X2 as
"X1 is more likely than X2". This binary relation > is "a qualitative probability" (Savage
1954). The basic axiom of subjective probability shows that:

X1 > X2 iff p(Xl) > p(XZ).

This requires that (i) > is asymmetric and transitive, (ii) 8 > G (nontriviality), (iii) XI 2

Q (nonnegativity), and (iv) if X1 0 X3 = X2 0 X3 = Q, Xl > X2 iff X1 U X, > X2 U X3

 

 

 

 

24
(additivity).

This subjective information is used in conjunction with any available objective
information. After collecting a sample Of information from the environment, he will use
the data to form a likelihood function Of event A and then form a posterior probability by
multiplying his prior probability with that likelihood. In continuously iterating
Observations through time, subordinates process information at time I, each using his past
knowledge on event A at time t - 1 as his prior probability. At time t, he collects some
additional data. Then he will transmit the revised report based on his prior probability
and the additional data. Each time a sample Of data is observed, the prior distribution on
event A is updated. By means of this updating, the unknown parameter is determined
asymptotically as the variance of its distribution decreases with observations.

In the simplest form, Bayes theorem (Raiffa 1968, Kmenta 1986, ch.6-4, Smith
1988, Lindley 1990) involves two factors: a specific value of the parameter of interest,

say, 0, and the sample Observation, x. The theorem states that:

P(Glx) = P(G) P(xlO)
P(x)

For any value of 0, P(Glx) represents the updated, posterior probability of that
value of 6, and P(G) is the prior probability. P(xIO) is the probability of observing the
sample given 0, which can be readily identified as the likelihood function. This function
summarizes the information about 0 as provided by the observed evidence in the
environment. P(x) represents the probability of observing the given sample by whatever
the value of 6 is. It is the marginal probability of sample Observations Obtained by

adding probabilities over all values Of 0. Therefore, this quantity does not vary with 0;

 

 

 

25

its role is that of a normalizing constant, which ensures that the probabilities P(le) Of all
possible values of 9 add up tO unity. For this reason, P(x) is frequently ignored. Thus
Bayes theorem for data x and parameter 0 is:

P(le) cc P(x|9)P(9) or _

Posterior distribution cc prior distribution x likelihood function,

where cc means "is proportional to".

To construct a posterior distribution we need to formulate the likelihood function
and the prior distribution. If no sample evidence is available, decisions are based solely
on the prior information. But Otherwise, if the prior information is noninformative or
diffuse, the posterior probability is based only on the observed data. The formation of
the prior distribution is unique to the Bayesian approach. If the prior distribution depends
on a parameter 9, it is convenient tO have a standard family Of prior distributions that can
be used to represent individuals’ prior information. If P(le) and P(G) belong to the same
class of distributions, they are called "natural conjugates". This concept of natural
conjugate distributions is very useful in Bayesian inference. One convenient property
of such a family is that if the prior distribution is provided in the. form of a formula, and
if this formula, combined with the likelihood function, leads to the same functional form,
it is called a conjugate family Of prior distributions (Winkler 1968, DeGroot 1970, ch.9).

The most commonly used conjugate family of distributions for 6 is the family
Of beta distributions (Cyert & DeGroot 1987, ch.2, Hastings and Peacock 1974, pp.30-33).
An organization needs to know whether event A will occur or not, and each individual

is assumed to report the possibility Of event A based upon his Observation of the

environment. Event A is a binary variable that has only two states, occurrence (A = 1)

 

 

26

or non-occurrence (A = 0). The probability of occurrence is denoted as 6, while that of

non-occurrence as (1 - 0). Observing one piece of evidence of the occurrence of event

A in the environment is treated as 91, and Observing evidence of the non-occurrence of

event A as (l - 9)‘. A sequence of random variables A1,...,A,, are independent and

identically distributed. The parameter 9 is said to have a beta distribution with

hyperparameters 0t and B (or > 0 and B > 0) if the probability distribution function Of 0

is:

13(9): P(a +J3)9a-:(1 -9)B'1 = ea'1(1_elB-l
1‘ (001“(13) BULB) 0 s 6 $1

0, Otherwise.

where B(Ot,B) is the beta function with parameters or, B,
given by B(a,[3) = ,1 pa' 1(1 - p)fl'ldp.

A beta distribution’s mean, E(9), is 0t and variance is OtB

a + B (0.+B)2((1+B+1).

If the prior distribution cc 9“(1 - 0)’5 and the likelihood oc 97(1 - (9)5, then the

posterior cc 0“”(1 - (3)5”. If or > B, the prior information favors values of 9 greater than

.5, and if 0t < B, the prior information favors values of 0 less than .5 (Hastings & Peacock

1974, p.31, Figure 5.1). Larger values Of (or + B) imply greater certainty in the expected

value of 6.

In this study it is assumed that individuals in an organization are Bayesian players.

The individual i has his own prior information on the possibility Of event A, gi(A), and

the likelihood function of event A based on his Observations (0,) on the environment E,

1,-(Ei). Thus the individual’s posterior probability distribution function Of event A is:

FAA) = HEX/4): l.(E;)].

 

 

27

All of the probability distributions--p,-(A), g,(A) and 1,-(E,))--belong to the same class of
beta distribution. For example, imagine a Bayesian player, a,, wants to estimate the
probability of event A. In the past a1 observed that A occurred 4 of 7 times, and then al’s
prior information, g,(A), can be written B(4,3).1 He then observes an additional 8
samples, 0,- = 8, and has the likelihood, 1,- == B(5,3) which shows A occurred 5 times out
Of 8. Then by following the simple Bayes rule, the posterior probability, p,(A), will be
[(4+5)/(4+3+3+5)] = [9/ 15] = .6. But at certain points, for simplicity, it is assumed that
each has a common prior probability or a diffuse prior probability on event A. The
individual’s reporting strategies will be discussed at the section of information processing

game.

A2. Individuals as expected utility maximizers

The Bayesian expected utility framework is of value not only for its normative
significance in suggesting how individuals should process information but also for its
predictive power in information processing.

It is assumed that the individual in an organization is under uncertainty, since each
field agent can observe only a portion Of the environment, and cannot get the all of the
information which is essential for a perfect prediction Of event A, and since each manager

cannot directly Observe any portion of the environment but relies upon only the

 

 

B(X,y) = 0"";1- 02"“
B(x,y)

The mean of B(x,y) = x
x + y

 

 

 

28

transmitted reports.

Individuals are assumed to follow the principle of maximizing expected utility:
that is, in a given decision situation the individual should choose the alternative with
maximal expected utility (Gardenfors and Sahlin 1988).2 Let us denote the payoff, wij,
as the result of choosing the alternative, r, when the state Of the world turns out to be S].

Then the assumptions of expected utility maximization are:

1. Values Of payoffs: wil- > wk, iff U(w,.j) > U(w,,,).

2. Values of alternatives: r,- > r. iff U(w(r,~)) > U(w(rj)).

J

3. Information about states: P(sj) 2 0, Z P(sj) = 1.

4. Probability independence: P(sjlri) = P(sj), for all Si and all r,—.
Then the expected utility of r, is EU(r,-) = Z P(sj)U(w,-J(r,)).

The goal of the individual i is: max EU,-.

5. Sure-thing principle (Savage 1954):

If the person would not prefer f to g, either knowing that the event B
Obtained, or knowing that the event ~B Obtained, then he does not prefer
f to g. Moreover (provided he does not regard B as virtually impossible)
if he would definitely prefer g to f, knowing that B Obtained, and, if he
would not prefer f to g, knowing that B did not obtain, then he definitely
prefers g to f (pp.21-2).

There exists a probability measure and a utility measure such that for all
alternatives r, and rj,

Lindley (1990) argues that in Bayesian perspecrive every utility is an expected utility:

You build for Yourself a small world including some data, excluding Others, and within that small
world construct a model for Your beliefs. Now such a world is part Of a larger world and what You
say about the small world may appear incoherent when viewed in the larger perspective....Utility in that
small world is an expectation over a larger world that contains it (p.50 and p.54, italics in original).

 

 

 

 

29
,- > rj iff EU(r,-) > EU(r,-).

For any information processing behavior that satisfies these assumptions, there exists a
utility function that assigns values to different outcomes in such a way that the chosen
alternative always has the highest expected utility.3

The field agent is assumed to maximize his expected utility, which is assumed to
be a function Of his influence on information aggregation; that is, his utility is determined
basically by whether his report is treated as important or not. The field agents want to
influence the organizational decision. The goal Of the organization is to predict event A,
and the responsibility of field agents is to Observe the environment and report the
probability Of the event in their area. For making the organizational prediction, the
manager needs to aggregate the reports from the field agents. When aggregating them,
the manager may treat some reports more importantly and the Others less importantly.
A field agent whose report is treated more importantly exerts more influence on the
organizational prediction. Since information processing is the only way to influence the
organizational decision, each field agent tries to get more weight in information

aggregation.

 

 

The principle and the assumptions of maximizing expected utility have been criticized from several
viewpoints. Allais (1979, 1987) says that the utility Of a monetary prize may be affected by the
probability Of that prize. Allais argues that the probabilities and utilities may not be independent.
Kahneman and Tversky (1979) argues that what determines the value Of an alternative is its possible
changes in wealth, where changes are evaluated in relation to a reference level. Thus the choice of
alternatives is determined by both the utility Of the outcomes and a reference level. They demonstrate
several classes of decision situations in which actual preferences violate the axioms of expected utility
theory. Machina (1982, 1983) has developed a theory of expected utility without the sure-thing
principle. He discusses several types of violations Of the independence axiom which entails the sure-
thing principle and presents the generalized expected utility approach with the function U(x; F) where
x is a certain prize and F is a cumulative distribution. See Bernard (1984) for general discussions on
utility functions, and see Fishbum (1981) for a review of theories of subjective expected utility.

 

 

 

30

It is assumed that the field agent’s utility is a function of his influence on
information aggregation: he can maximize his expected utility only by maximizing his
relative influence on information aggregation. When assuming that each field agent
Observes the environment and processes the probability continuously through the time
dimension, a field agent’s influence on information aggregation is based on his reputation
in the organization." The report Of a field agent with a good reputation can get more
weight because, through the continuous reporting of information to the manager, it has
been shown that his reports are valuable and correct. A field agent with a bad reputation
would receive little or no weight. ' It is assumed that the utility Of each field agent is a
function Of a relative weight, w,, which measures his influence in information aggregation.
Field agents maximize expected utility by maximizing the relative weight, w,, with the
appropriate use of report r,. The utility function (see Bayarri and DeGrOOt 1988) is

assumed to be:

Ur =f(Wi) = 10g__w_.- _
2w,- , i¢j,w,-+2wj=1,i+2i=n.

EU.- = P[U(W.-(r.-, A = 1))l + (1 - P)[U(W;(rp A = 0))1, A 6 {0, 1}.
max EU,-

where r, is ai’s report and p is a,’s p,(A) that event A will occur at t + 1.

That is, the field agents care only about maximizing their relative weight, anddo not care

about anything else. The incentive compatibility and principal-agent literature put

 

 

For a multi-time information processing, a field agent can calculate his expected utility differently when
considering discount factors. With the variations of the discount factors, a field agent will be patient or impatient
in reaping the weight in the future. Issues on time dimension and discount factors remain to be considered.

 

 

31

emphases on shirking, monetary payments (such as salaries and bonuses), profit
distributions and so forth. But this study is not working in that particular tradition but
focusing only on the relative influence on organizational prediction.

The manager is assumed to maximize his expected utility which is a function of
information cost and event predictability. Managers have personality types placed on a
dimension between the extreme Of carefully considering information cost, but not
worrying about event predictability, and the extreme Of carefully considering event
predictability but not thinking about information cost. Information cost for each
individual (z'c) is defined as the cost of analyzing the reports, r, which he receives and
making the report, rm, which he announces. Event predictability (ep) refers to how
closely the announced expectation on event A approaches the true event. Information cost
of each manager would vary with organizational structure, which defines the number of
levels in the hierarchy and the span of control of managers. The information cost can be
estimated by the manager’s position in the organization and the number of reports given
to that manager. If the information cost would be fixed, the manager’s utility is only a
function of event predictability.5

The goal Of the manager m is to maximize his expected utility by maximizing

event predictability of his report rm with the given information cost:

Um = f(ic, ep).

Um = f(ep), if ic is constant,

 

 

lnforrnation cost and event predictability will be more thoroughly discussed in the section dealing with the
principles in information processing.

 

 

 

 

32
max EU". = P[U(ep(rm, A = 1))l + (1 - p)[U(ep(rm, A = 0))1. A e {0, 1}.
Since Um = flep), m would maximize the expected value of ep, B(ep).
max B(ep) == plep(rm, A = 1)] + (1 - P)[ep(rm. A = 0)]

where rm is m’s report and p is pm(A) on that event A will occur at t + l.

3.2 Information aggregation and feedback

The problem Of the Optimal design of organizational structures is viewed in terms
of the aggregation Of information in the organization. Organizations differ not only in
what kinds of information individuals generate, but also in how the organizations
aggregate information. The basic function Of aggregation is to simplify the volume Of
information which flows through the organizational structure and to reduce information
processing cost (Keren and Levhari 1989). In information processing, each level of the
organization aggregates upward-directed information and disaggregates downward-directed
feedbacks. The aggregation problem is the problem Of aggregating or combining different
estimates from the field agents into a corresponding combined quantity. It is to transform
multidimensional input into unidimentional output.

Simpson’s paradox demonstrates that different aggregation methods can affect the
information that the top manager would use for organizational decisions (Saari 1987,
Haunsperger and Saari 1991). Simpson’s paradox states that "the conclusion of the
aggregated data differs from a common conclusion Of the subpopulations" (Haunsperger
and Saari 1991, p.252). Let us examine an example Of Simpson’s paradox: '

Assume each individual has a beta probability distribution function and

reports B(x,y) where x is the number of good results and y is that of bad

 

 

ll

 

 

 

 

33

results. Thus the mean value of the probability distribution is calculated
as [x/(x+y)]. Suppose a certain drug is tested in Michigan State
University. First, the data are gathered by two subordinates and the
aggregated results are transmitted to the manager. The duty Of a, in the
Department of Agriculture is to test the new drug, (1,, and the standard
treatment, d2, and to report which one is better in one village, for example,
Cherry Lane Village. That Of a2 is to do the same in another village,
Spartan Village. The manager would gather the reports from a, and a2 and
make a decision whether d, or d2 is accepted for public use. a, Observes
that in 9 out of 24 cases d, cures the sick, while in 2 out of 6 cases (12 is
successful. Thus a, reports that d, is better than (12 based on the
probability calculations, p(d,) = .375 > .333 = p(dz). a2 Observes that 3 Out
Of 6 cases a, is successful, and 11 out of 24 cases d2 is good. SO a2
reports that d, is better than (1; with p(d, == .500 > .458 = p(dz). The
manager receives the same report from both a, and a2 and with no
hesitation he pronounces that the Department of Agriculture approves d,

for public use based on the experimental result.

Second, the manager directly gathers data from both Cherry Lane Village
and Spartan Village. For d, among 30 cases only 12 samples show that
d, is successful, [(9+3)/(24+6)], while for (12 13 out of 30 demonstrates d2

is good, [(2+11)/(6+24)]. Thus p(d,) = .400 < .433 = p(dz). When

 

 

 

 

34

aggregated, the data show d2 is better than (1,. It shows that the way to

aggregate information is an important determinant in information

processing. (From a talk by DC. Saari at Washington University on March

1, 1988, and Saari 1987, p.2).

Simpson’s paradox shows that different aggregation methods may make different
conclusions from the same data. If a design of organizational structure means to choose
a certain kind of aggregation procedures, different organizational structures may generate
different decisions from the same volume of information as suggested in Simpson’s
paradox.

The role Of managers is to gather the reports from his subordinates and to
aggregate them into his own likelihood function. For managers, the question is how to
aggregate information. The manager is assumed to formulate his own aggregation rule,
am, as a Bayesian player. An aggregation rule transforms any set of individual reports on
event A into an organizational (or a divisional) report. Formally, an aggregation rule (1,,
is a correspondence from the n—fold Cartesian product Of reports on p(A) into p(A), dm:
[0, 1]’’ ——> [0, l]. The basic rule in information aggregation would be the relative weight
rule: d,,, = (w,,...,w,,) (Morris 1983, French 1985, Genest and Zidek 1986, Gradstein &
Nitzan 1988, Bayarri and DeGrOOt 1988, DeGrOOt and Mortera 1991). The aggregation
rule depends only on the manager’s assessment Of the field agent’s relative weight. The
aggregation rule involves the application Of Bayes rule to formally revise the weight after
receiving each report and Observing the occurrence Of event A.

OSw,Sl, Zwi=1.

 

35

When aggregating reports, the manager will assign a relative weight to each report based
on his evaluation of each subordinate." The report of each field agent receives a certain
prior weight in information aggregation, and the weight is then updated by the observed
state of event A -- whether it has occurred or not. I assume that the manager uses the
relative weight rule as his aggregation rule because the manager can easily use it in a real
world. Since each field agent has a beta probability distribution function on event A, each
time the field agent reports r, as B,(a, b) it shows that the expected value Of 6 is [a/(a +
b)]. Only the individual posterior probability distributions can be reported by r,, whereas
the individually Observed data sets and prior distributions cannot be communicated. The
manager m compares r, and assigns a relative weight, w,, to r,- based on his evaluation on
a,. For example, m estimates r, at time t based on his experience with a, until I - 1. If
m thinks that r, at time t - 1 was accurate and valuable, then he will be more likely to
consider r, at t to be valuable. If m determines that r,- is more valuable than r,, i at j, he

will give a relatively heavier weight to r, as w, > W], i as j; however, w,- = W]. if the

 

 

Bayesian studies of combining probability distributions have developed a number of ways to aggregate
subjective probability distributions (see French 1985, Genest and Zidek 1986). Let us denote it as an
aggregated information by u = f(r,, d...) and r.- is 053 report.

 

Linear opinion pool: p. = X w,r,-, 0 S w,- S 1, Z w,- = 1.
Generalized linear opinion pool: t1 = 2 win, -1 s w,- 5 1, Z w,- = 1.
Logarithmic opinion pool: r1 = [I r,."_"
[I r,“ (111.
Generalized logarithmic opinion pool: p. = 2 H {'5
is H r.” dtL g = 13(9)-

Log—Odds Opinion pool: ti = Z w,log(o,/o,), 0,- = Pg/(l - Pi). 0., = m’s prior.

 

 

 

 

 

 

36

importance of r,- and r]. is the same. The weight for r,- depends not only on m’s assessment
of a,’s relative trustworthiness or reputation but on those of the other field agents as well.
Therefore, the weight is relative. The weights are variable in the sense that as m learns
more about a,- and r,- through time, w,- must be updated (Morris 1983). The manager
decides ai’s relative weight based on his prior assignment and the present report with
regard to the other reports:

Wt = f(th'1’ War]: ri r_,-),

where r_, = (r,,...,r,-,,, r,+,,...,r,,) and w,‘ " means the relative weight to a, at time t-l.
The assigned weight to each field agent directly decides his utility. If the manager has
no prior information about his subordinates, then he will use an equal weight with w,- =
(1/n). With r,, the aggregated information, am, is a function of the basic aggregation rule
am:

it”. =f(r, cl...) = W’r = Z W.»
where w’ is the transpose of the matrix w.

rm = f(gm(A), PM).

The manager’s likelihood function is a summary of a set of reports that is a
collection of probability functions. The manager constructs his posterior report, rm, based
on his prior information on event A and its likelihood. When the manager’s prior is
uniform, that is, gm(A) oc constant, the manager should adopt the aggregated information
as his own report (Morris 1977). However, in aggregating reports, there is a special
principle. If all subordinates report the same information, the manager would accept it
as the final report no matter what his prior information is, because one of the purposes

of aggregation is to save the time and energy of the managers. If there is a consensus

 

 

 

37

among subordinates, the manager would not invest his own time to make the report. This
principle is stronger than the "Management By Exception" rule.7 The unanimity
principle in information processing is:

mmmityprinciple: rm denotes m’s final decision on p(A). r,,, is the same as k

if and only if all subordinates report k, that is,

r,,, = k, if and only if V i, r,- = k, i = 1,...,n. (Morris 1983, Clemen and Winkler,

1990)

Feedback is a vital element in information processing systems. Kaufman (1973)
defined "administrative feedback" as "all the processes by which the bureau leaders--the
whole headquarters—-are apprised of subordinate behavior down to the lowest
organizational level" (p.1). Feedback from a higher level of a hierarchy can serve two
major functions for an organization member.8 Feedback can provide subordinates with
an estimate as to what reported information is correct or valuable, or wrong or not
valuable in making organizational decisions. It can provide reinforcement, rewarding a
correct report by increasing its weight, and punishing a wrong one by decreasing its
weight. Rewards and punishments for subordinates are confined entirely just to
information processing issues. Since subordinates care only about maximizing their

relative weights, and do not care about anything else, rewards and punishments are

completed by changing the relative weight in information aggregation. Both functions can

 

 

If at least k of the n field agents agree on the probability of event A, the manager will accept that probability
as his report; if fewer than k field agents agree on that probability, the manager will invest his own infortnation
cost to make the report himself. It is assumed that k > n/2 (see Hammond and Horn 1985, p.52).

In the field study, Becker and Klimoski (1989) showed that, while holding the other feedback variables constant,
feedback from supervisory and organizational sources was related to reported job performance while feedback
from peers and self was not. Higher performers received more positive feedback.

 

38

influence each subordinates’ subjective probability and Observational accuracy. Feedback
shows whether the manager is consistent in applying the relative weights (Te’eni 1991).
Feedback can affect information processing by motivating the managers to choOse
appropriate weights and by instructing the field agents to choose successful reporting
strategies. Through the process of feedback the manager can update his aggregation rule

and the field agent can revise his prior probability and strategy.

Table 1: Information Processing and Aggregation

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time & step Field agents a, Manager at
t, 1 has prior g,(A) has prior g,,,(A)

2 observes E
3 reports r, assigns prior weight w,
4 aggregates r
5 reports rm

t+ 1, 1 event A event A
2 updates g,(A) updates gm(A)
3 observes E revises w,-
4 reports r, assigned the updated weight
5 aggregates r
6 reports rm

1+ 2, 1 event A event A

 

 

 

 

I assume that observation and information processing is continuous through time
dimension with feedback. The process of feedback and updating the weight could be
carried out repeatedly as t ——> oo. Since the revised weight is going to be used at the next

time, it is natural for the field agents to try to maximize their own weights by

 

39

appropriately choosing the reporting strategies that they report to the manager. Based
on this feedback each report can be evaluated in the near future. The time interval
between t and t + 1 is assumed to be as short as possible. Thus feedback provides the
learning mechanism through which the organization can generate stable patterns of
organizational behavior over time. Table 1 summarizes the procedures the field agents
and the manager use to process information.

At time t, m receives r = (r,,...,r,) and aggregates it to n," by using (1,, Then m
reports rm based on gm(A) and ,um. At t + 1, m observes event A and then updates d",
based on the state, A = 0 or 1, and r. w, would be revised by the degree how r, is
approached to the true value of event A, with regard to r,-, j at i. If a,- reports r,- = 1, and
A = 1, a,- can increase w,- at t + 1 with w,-’+ ’ .>_ w,‘. Since w, is decided with regards to rj,
even though a,- reports r, = 1 (or 0) when A = 1 (or 0), if all of the other field agents
report the same, then w,' + ’ is the same as w,‘. The general principle is that the more
accurate r,, the heavier the weight that will be assigned to r,. The updating rule for

weights (see Bayarri and DeGroot 1988) is that:

 

 

If A = 1, W; J" = w.‘r.’
2 witri‘ 9
ifA = 0, WiH-I = W;‘(1 'fl
2 Wit(1 " ri‘) , O S W; S 1, i: 1,...,n.

For example, let us assume that there are 3 field agents, (a,, a2, 03). At time t, m
assigns an equal weight to each agent, w,- = 1/3. Manager m receives r = (n, r2, r3) =

(B1(9.1). 1320.3). [33(4,6)). SO by using d,,,, m gets pm = 1/3(9/10) + 1/3(7/10) + 1/3(4/10)

 

 

 

 

40

= .67. If g,,,(A) is diffuse, then m reports that the probability of the occurrence of event
A at t + 1 is .67. If at t + 1 event A has actually occurred, then m updates am such as w,‘
“'1 = [1/3(9/10) -:- 20/30] = .45, wz” ’ = .35, W3” ’ = .2. When m aggregates the reports
at t + 1, he would use the updated weights, w = (.45, .35, .2). Since r, is more accurate

than the Others, w, is heavier than the others, while r3 receives less weight.

3.3 Principles in information processing
We can consider the following criteria in information processing for comparing

different models of organizational structure.

A1. Inforlnation cost

Information cost, IC, means the overall cost to the organization Of information
processing. This involves the cost of Observing the environment, analyzing reports, and
making an organizational report. The information cost Of an organization is the sum Of
the managers’ information processing costs and of the field agents’ observation costs. It
is assumed that the unit cost of Observing a sample from the environment, (0), is the same
for different organizational structures. It is also assumed that the greater the structural
level Of a manager, the greater is the unit cost of analyzing a report.9 For example, the
top manager’s unit cost Of analyzing a report is greater than a division manager’s. The

information cost for the manager (ion) is a function of the structural level of that manager,

 

Starbuck (1971) explained the reason why the span Of control near the top of the hierarchy is supposed to be
smaller than near the bottom is there is greater need for coordination near the top.

 

 

41

Lm, and the number of reports to that position, N(r).lo It will be an exponential function
because the information cost will be geometrically increased by receiving more reports.
As receiving one more report, the manager needs not only to analyze it but also to
compare it with the others. The field agent’s information cost (ici) is the simple product
of the number of observations, 0,, times the unit cost of observation, c. Therefore, the

information cost of an organization is:

IC = 2 ic,,, + 2 ic, = ZEN” + 2 c '0,-
where c is constant, c > 0, m = 1,...,r, i = 1,...,n.

This functional form for IC represents that the total information COSt of an organizational
structure is the sum of information costs of the managers and the field agents. For
example, if a division manager receives 3 reports from his subordinates, then his
information cost is 23 = 8. If a field agent observes 12 samples from the environment,
his information cost is 120. If all the organizational structures have the same n numbers
Of field agents who generate the same k number of observations, then 2 ic, = c: n 'k is
the same for all the organizational structures.

Let us call IC(') the information cost in an organization. Then 12 is preferred to
g iff the information cost of h is less than that of g:

h P g iff IC(h) < IC(g).

 

 

Williamson (1970) considered two variables to represent organization costs: the span of control, and the number
of hierarchical levels.

 

 

 

42

A_2. Event predictability

Event predictability is the fundamental criterion for comparing organizational
structures in information processing. It refers to how accurate the organizational
prediction of event A is. It is defined as the absolute value Of the difference, or deviation,
between the organizational prediction on event A, re, and the actual state of event A.
However, it cannot be known until the organization observes the actual state of event A.
Furthermore, it can vary at each time when the organization knows the state of the event.
It can be estimated by considering the variables which are closely related with event
predictability. For estimating event predictability, the basic condition is that the
information received from field agents is accurate and unbiased. Managers should use
a certain kind of incentive mechanism to insure honest information transmission from the
subordinates. Under the condition where an appropriate incentive mechanism is used,
information predictability is estimated as a function of both information efficiency and

information reliability.

A2. 1. Information efficiermv

Information efficiency refers to the size of the sample an organization uses to
estimate A from the environment. According to the probability theory, if the observations
themselves are accurate, the more observations an organization generates, the more
accurately the organization predicts the event. If one has detailed information about his
jurisdiction, the associated probability interval would be narrow. Let us call two

different organizational structures h and g, and the number of Observations which an

 

 

 

43

organization generates as N ( ). Then h is preferred tO g iff the number of observations
in h is greater than that in g:

h P g iff N(h) > N(g).

A2.2. Information reliabilitL

Information reliability refers to how much agreement there is within an
organization about the probability of event A. The more agreement about this probability,
the more reliable the organization believes the prediction to be, and the more objective
is the probability. Let us call AN(') the number of similar reports in an organization.
h is preferred to g iff the number of similar reports in h is greater than that in g:

h P g iff AN(h) > AN(g).

The more information efficiency, the more event predictability, and that the more
information reliability, the more event predictability. Therefore, event predictability is
proportional to both information efficiency and information reliability. I assume that
event predictability, EP cc NAN. This functional form for EP represents that EP can be
estimated with the number of observations and the number of major agreements on the
probability. It shows that EP will increase when increasing the number of Observations,
and when achieving more agreements about the probability. Let us call event
predictability in an organization EP(-). h is preferred to g iff the information
predictability of h is greater than that of g:

EP==1-lA-rol oc f(N, AN), A={0,1}.

EP oc NAN.

 

 

h P g iff EP(h) > EP(g).

A3. The tradeoff between information cost and event predictability

The goal of the top manager is to design an organizational structure which can
minimize information cost and maximize event predictability. However, if it is
impossible to satisfy these two principles simultaneously, he would make a trade-off
between these principles. If information cost is more critical than event predictability, he
would choose the organizational structure which can reduce information cost. On the
other hand, if event predictability is more critical, he would design the organizational
structure which would maximize event predictability even while paying more information
cost. The tOp manager’s choice (ch) Of an organizational structure is based on the overall
information cost and on the event predictability Of an organization.

ch = f(lC, EP) = _l;13__

IC

This functional form for ch represents that the value of ch will vary directly with EP and
inversely with IC. The value of ch will be increased when EP is increased, and

proportionally decreased when IC is increased.

3.4 An information processing game
Information processing in an organization is subject to the strategic behavior of
subordinates (Simon 1976, Cyert and March 1963). Information is the only factor that

determines the individual’s relative weight and utility. "In reporting at every level,

 

 

45
hierarchy is conducive to concealment and misrepresentation" (Wilensky 1967, p.43).
Cyert and March (1963) stated:

Where different parts Of the organization have responsibility for different

pieces of information relevant to a decision, we would expect some bias

in information transmitted due to perceptual differences among the

subunits and some attempts to manipulate information as a device for

manipulating the decision (p.67).

O’Reilly and Roberts (1974) investigated the directionality of information flows
in organizations and the impact of trust and influence on these flows. The study showed
that more favorable information is passed upward, while unfavorable but important
information is passed laterally. But they also found that upward information filtration is
heavily influenced by u'ust between superior and subordinate. Thus to solve the problem
of information distortion and misrepresentation it is necessary to employ a mechanism or
relationship that would ensure honest reporting. In the case of public good provision,
Groves and Loeb (1975) studied the mechanisms which provide an incentive for each
individual to send truthful information, so that an optimal quantity of the public input will
be provided by the coordinator. Under this mechanism each individual will be charged
the difference between the budget and the reported benefits of the other individuals.
Since misrepresentation cannot give any advantage, truth telling will be the dominant
strategy.11 Groves and Ledyard (1977) explained:

Even though consumers are completely free to misrepresent their demands

for public goods, the tax and allocation rules are structured in such a way
that in equilibrium it is in each consumer’s individual self-interest to reveal

 

 

In a Groves mechanism the dominant strategy of the subordinate is truth telling, but that of the manager will be
misrepresentation since it cannot consider the designer’s strategic behavior (Hammond and Miller 1989, Miller
1992). Miller (1992) suggested that it is necessary that the manager credibly commits himself to these efficient
incentive schemes.

 

 

 

46

his true demand or valuation of the public good (p.783).

But the Groves mechanism cannot balance the budget and so reaches nonoptimal
outcomes. d’Aspremont and Gerard-Varet (1979a) indicated, using Bayesian approach
to incomplete information, that if a compatibility condition is imposed on individual
beliefs, and if a Bayesian solution is given to the incentive problem, then one may find
mechanisms to simultaneously ensure incentive comparability and budget balancing.
This section will search for an incentive mechanism through which information processing
is immune to the strategic use of information.

An organizational structure can be thought of as a vector of equilibrium strategies
in a non—COOperative game (see Dow 1990). A noncooperative game is a game in which
the individual’s Optimal decisions depend on his beliefs about the play of his opponents.
The concept of a Nash equilibrium plays a central role in a noncooperative game. The
way to approach Nash equilibrium is to propose a strategy combination and test whether
each player’s strategy is a best response to the other’s strategies.

Information processing in the organization can be analyzed by an extensive form
game since it contains the following information:

(1) the set of players

(2) the order of moves--i.e., who moves when

(3) the players’ payoffs as a function of the moves that were made

(4) what the players’ choices are when they move

(5) what each player knows when he makes his choices

(6) the probability distributions over any exogenous events (Fudenberg and

Tirole 1991a, p.77).

For extensive form games, the concept of Nash equilibrium is not sufficient since:

(1) Nash criterion does not require players to make nondominated choices in equilibrium;

 

 

47

(2) there is nothing strictly irrational about choosing weakly dominated strategies
(Bernheim 1984, Pearce 1984); and (3) Nash equilibrium may involve unreasonable
behavior at unreached parts of the extensive form (Van Damme 1983, Fudenberg and
Tirole 1991a, ch.3). Thus much effort has been focused upon refining the concept of
Nash equilibrium. One of the equilibrium refinements is the concept of perfect Bayesian
equilibrium. Perfect Bayesian equilibrium results from combining the ideas of subgame
perfection, Bayesian equilibrium and bayesian inference. The information processing
game would be analyzed by applying the concept of perfect Bayesian equilibrium
(Fudenberg and Tirole 1991b).

Information processing in an organization should be sequentially iterated through
a time dimension. I assume that information processing is continuous through time
dimension as t -—> co, and the time interval between t and t + 1 is as short as possible.12
The game at time t would be affected by the previous game at time t - 1, and would
influence to the next game at time t + 1. An information processing game will be defined
as following.

Let N = {a,,...,a,,, m} be the set of players, who are n field agents, and a manager.
For deep structures, an organization game would be analyzed as a two stage game -- with
a,- and dm, and with dm, and tm. In the game G, let us assume:

1. There is a given organizational structure ‘1’ from time t = 0.

2. At each time t, a,- has a subjective probability distribution function over the

 

 

In this kind of multi-time information processing game, we can also consider the other dimensions such as
discount factors and risk preferences. When including these dimensions into the game, the reporting strategies
of field agents may be more specified with different types of field agents. But these issues remain to be
considered.

 

 

 

*7—

48

states of event A. It is independent with each other. a,- would have his posterior
probability distribution, p,(A), based on his prior, g,(A), and his likelihood function,
l,(o,). The prior is continuously revised by the posterior and the state of event A.
a, would update his probability distribution function through the time dimension
by applying Bayes rule:

13.04) = far-(A), 4(0)), 8.04)”! = f(p.(A)‘, A).

3. At each time t, a,- would choose a reporting strategy function, r,, based on p,(A)
and his belief, b,, on the manager’s information aggregation strategies and the
other agents’ reporting strategies. Basically the agent’s possible strategies are to
report his own posterior probability distribution honestly, r, = p,(A), and to
misrepresent his posterior probability distribution, r, at p,(A). Before reporting r,,
a, knows p,(A) not pj(A) nor rj of the other agents, j ¢ i.

ri: pi(A) —> [Or 1]

4. For both a,- and m, the belief b should be consistent with Bayes’ rule by
depending on the previous history of the information processing and aggregation
until I - 1, h"’, and the given strategies:

b = f( Ihbl, 1361,“).

5. a,’s utility U,(w£) at t depends on the previous weight w,’ ’ ’, r, and r_,-. At each
time t, r, should maximize his expected utility, EU,(w,), before knowing the true
state of event A.

6. At each time t, m would use a certain aggregation method for aggregating r and
update it by which w,‘ + ’ would be revised. m would choose an aggregation
strategy which should maximize his expected utility EUm. m has his own belief
on r.

7. It is common knowledge that every player is a Bayesian player. Everybody
knows that each player chooses a strategy that maximizes his expected utility
given his belief, and each player updates his belief and probability distribution
function.

 

8. The information processing game G is a multi—time game with r and dm, where
each a, reports simultaneously in each time, and r, at each time is transmitted
before event A is actually occurred or not.

9. The strategies yield a Bayesian Nash equilibrium, if possible, not only for the
whole game G, but also for any subgame g starting in each time t.

 

 

 

*7—

49

The states Of event A are represented by the two values: s(A) = {0, 1}. a,- chooses a
strategy (report) function r,(-):p,- —> [0, 1] from his strategy space r,- = {r,.(-)}. m
chooses an aggregation rule, d(°):[0, 1]’l —> [0, 1] from his aggregation strategy d =--
{a’(:)}. Let us define a,’s utility as U,- = f(w,-) = f(r,-, r_,-, w"’). m’s utility is defined as
U,,, = f(d,,,, ic, ep).
For all a,, given aggregation rule d,,, and given organizational structure ‘1‘,
max EU,(r,-, r,,-, w‘ ‘1 I am, II’)
= max 2 P(s)U,-(r,-, r,,-, w‘ '1 1 dm, ‘1’).
For m, given the set of reports r and given organizational structure ‘I’,
max EU,,,(d,,,, ic, ep l r, ‘1’)

= max 2 p(s)U,,,(d, ic, ep I r, ‘1’).

In a Bayesian game, the criterion for the best strategy is that it should give a
player the highest expected utility. Expected utility is computed by multiplying utilities
times probabilities and then adding them all together (Myerson 1985). The Bayesian
equilibrium is defined as the combination of strategies and beliefs where neither field
agents nor the manager can gain more by playing differently, given an organizational

S II'LICIUI'C.

Definition 1. Perfect Bayesian equilibrium in the information processing game

 

A perfect Bayesian equilibrium (PBE) of a multi-time information processing game is an
assessment (r*, dm’l‘, b) such that at any time t and under l1’,

1. (r*, d,,,* , b) is reasonable by satisfying the assumptions,
2. for each time t and the history If ’ ,the strategies for the remainder of the game
are a Bayesian equilibrium given the beliefs.

 

 

 

 

 

50

V (1,, EU,(r*,-, r*,,- | b,, (1*, ‘1') 2 EU,(r,., r*,,. l b,, (1*, ‘1’).

V m, EU,,,(d,,,* l b,,,, r*, ‘1’) 2 EU,,,(d,,, l bm, r"‘, ‘1’).

For the manager to maximize his expected utility by improving information
predictability, he should give his subordinates the correct incentives to reveal their
posterior probability distribution honestly. To the extent that his private information on
event A affects the decision of the manager, any field agent may find it advantageous to
distort the information he transmits since he cannot be compelled to reveal it honestly
without the correct incentives (d’Aspremont and Ge’rard-Varet 1979b).

Let us examine whether the relative weight rule, the aggregation rule of the
manager, can give each field agent the correct incentives. If, under the relative weight
rule, by reporting their posterior probability distributions honestly, all field agents can get
more expected utility than by the other reporting strategies, the relative weight rule can

compel honest reports. If so, we can call it a Bayesian incentive compatible mechanism.

Definition 2. Bayesian incentive compatible mechanism

A mechanism 3 will describe the relative weight aggregation rule of the manager, d,,,*,
and the vector of the agents’ relative weight, w, that is, E = <d,,,*, w> (see d’Aspremont
and Ge’rard-Varet 1979b, 1982, Myerson 1985, d’Aspremont, Cre’mer and Gérard-Varet
1990). We denote by r*,. the honest strategy for field agent a,, such that r*,. = p,(A) for
all p,- 6 p,. A mechanism E is Bayesian incentive compatible iff it is a Bayesian
equilibrium for all field agents to report their posterior probability distributions honestly
such that

V at: Wt("*p r*-i ' dm*) 2 WM, r*-£ | d*m):

where r*,. = p,(A), r, ¢ p,(A), and r*,,- = p,,-(A).

 

 

 

¥

13

 

51

Theorem 1: The mechanism 3 is Bayesian incentive compatible with the given utility

function of the field agents.

Proof. For a,, the goal is to maximize the expected utility by appropriately choosing r,-

at any time I (see Bayarri and DeGrOOt 1988). For notational simplicity, p = p,(A)
at t.

max EU,-

max E(log w,- )
ZWJ‘

max [Pu-(Win, A = 1)) + (1 - P)U.-(W.-(rs, A = CD] ,A = {0, 1}.

m assigns w,- at I based on w,- at t - 1 and on r at t such that”:

 

 

IfA = 1, W; : will?

ij"’rj‘ , i at j, w,- + 2w}- : 1,
ifA = 0, w; = wild-r11

z Wj".1(1 " rjr).

E(log w,-‘ ) = plog w,-"’r,-‘ + (1 — p)log w-L’U - ff)
2 wit 2 wit-1r; 2 ij1(1 _ rjr)

 

= plogr,‘ + (1 - p)log(1 - r,’) + (terms not involving r,)
To maximize the expected utility, let us get the first order condition.

BE (log w!) = p - (l-p)
Br} 2 wj‘ r,‘ (1 - ri‘)

When [BE/Bri’] = 0, r,’ may maximize or minimize the expected utility.
[P/rr'] - [(1 - p)/(1 - r0] = 0
P - pr,’ + pr,’ 1 r; = 0

1:?-

t or t - 1 is n_ot_ a multiplier but means a certain point in time dimension.

 

 

 

 

52

If the second derivative of r,‘ is less than zero, that is, ODE/arid < 0, than we can
say that r,‘ would maximize the expected utility when r,‘ = p.

a[8E/Br,-‘] = (-p + p - 1) < 0.

Thus when r,’ = p, a,- can maximize the expected utility. Since p = p,(A) at t, r,’
should be a,’s honest posterior probability distribution on event A.

r*, = p,(A) and the mechanism 3 = <d*,,,, w> is Bayesian incentive compatible.

QED.

In the perfect Bayesian equilibrium of the information processing game, the
dominant strategy of the field agent is to report honestly. That of the manager is to use
the relative weight rule appropriately.” The existence of the relative weight rule as a
Bayesian incentive compatible mechanism provides a common basis when comparing
different organizational structures since it makes us expect that all subordinates will
behave sincerely.

The problem of organizational design is viewed in terms of information
aggregation. To design an organizational structure is to determine a set of information
aggregation rules in consideration of the two principles, information cost and information

predictability.

 

Theoretical equilibrium is never attained in reality, but the analysis of equilibrium conditions helps specify the
behavioral tendencies in organizations. These behavioral tendencies can be used to make predictions (Barney
1990).

 

 

 

 

4. Information Processing in Organizational Structures

4.1 Four models Of organizational structures

This chapter will analyze and compare the processing and aggregation of
information within different Organizational structures. It will generate four simple models
of organizational structures, and discuss the characteristics of, and the advantages and
disadvantages of, each kind of organizational structures in information processing. Then
these four structures will be compared with each other.

Organizational structure defines the lines of authority and communication between
individuals and divisions, as well as information transmission through these lines of
communication and authority (see Gulick 1937, Chandler 1962, Galbraith 1977,
Mintzberg 1979, Child 1984, Hammond 1986). Organizational structure serves to produce
organizational outputs, to regulate the influence of individual variations, and to ensure that
individuals conform to the requirements of the organization (Hall 1982). An
organizational structure has essentially two Objectives: it facilitates the flow of
information within the organization in order to reduce the uncertainty in decision making;
and it achieves effective coordination-integration (Duncan 1979). In this study,
organizational structure is defined as the hierarchical rules governing how information
processing and aggregation must take place. The design of an organizational structure
determines a certain set of information processing and aggregation routes.

The focus here will be on two dimensions of organizational structure which

affecting information processing and aggregation: specialization and coordination (Gulick

53

 

 

 

54
1937, Chandler 1962, Galbraith and Nathanson 1978).l Specialization is dividing the

work of the organization into divisions or individuals for efficient performance.
Coordination is integrating the activities of specialized divisions or individuals in order
to achieve organizational objectives: "Coordination involves fitting together the activities
Of organization members, and the need for it arises from the interdependent nature Of the
activities that organization members perform." (Argote 1982, p.423). Both dimensions
are essential to effective information processing.

Four models of organizational structures will be created in order to test the effects
of changes in su'ucture on information processing. Based on the coordination principle,
we can consider two kinds of organizational structures: "flat" and "tall" structures. As
Child (1984, p.59) Observes, "Tall and flat structures are usually identified by the number
of hierarchical levels there are in an organization relative to its total size. A tall structure
is one that has many levels in relation to total numbers employed, while a flat structure
is one that has few levels relative to total employees." Based on the specialization
principle, we can consider two kinds of matching problems: "one-tO-one" and "one-to
many" matching. This refers to the choice of a particular form of the organization’s
interaction with its environment (Baligh and Burton 1980). As organizations become
more specialized, they have to achieve more coordination. Balancing specialization and
coordination is the key to designing effective organizational structures.

One-to-one matching assigns a separate jurisdiction to each field agent. It

 

Similar principles in organizational design are "differentiation" and "integration" (Lawrence and Lorsch
1967).

 

 

 

55

promotes organizational efficiency in the sense that the more observations the
organization generates with the same information cost, the more efficient it is.
Specialization increases as jobs are broken down into more simple components (Dewar
and Simet 1981). If the environment can be decomposed into many subsets, E ,,...,E,,, the
organization can greatly speed up its information generation by assigning each field agent,
a,, to each subenvironment, E,. Specialization by one-tO-One matching enables field
agents to become experts in their jurisdictions. They are relying less on individual biases
and more on empirical Observations. The negative aspect of specialization is that it gives
each field agent informational monopolistic power on each jurisdiction since the
organization can get access to the information on E,- only through (2,.

One-to-many matching assigns a common jurisdiction to all field agents. The
purpose is organizational reliability (Landau 1969, Bendor 1985, Lerner 1986, Heimann
1992). Landau (1969) argues that we can decrease the probability of organizational
failure and increase organizational reliability by using redundant factors (see Bendor 1985,
Heimann 1992). Redundancy in information processing occurs when some portion of
information is produced from a common jurisdiction. The other aspect of redundancy is
that it makes field agents compete with each other, since the information from the
environment is pooled, as they try to prepare the best possible estimate of event A.

The structural arrangements that produce redundancy are grouped into two basic
categories, "duplication" and "overlap":

Duplication provides redundancy by emphasizing parallelism. Overlap

provides redundancy by emphasizing ambiguity. The ambiguity is in the

jurisdictions Of the overlapping units. Define jurisdictions as
organizationally mandated functions in designated task areas assigned to

 

 

 

56

a unit. In overlap some of the functions assigned to A are also assigned

to B. Thus overlap ought really to be called "partial overlap". By contrast

in duplication all the assigned functions of A and B are identical; A and

B are completely interchangeable (Lerner 1986, p.336, italics in original).
Duplication is the simplest form of redundancy (1991). It makes the' comparison of
reports from a common jurisdiction easier, while with an overlapping jurisdiction it is
difficult to figure out what portion Of reports are produced from a common area. Thus
in this study duplication is adopted as the structural arrangement of redundancy. In one-
to-one matching each field agent has his own jurisdictional area, while one-to-many
matching is a duplication Of jurisdiction in which all field agents have the same
jurisdiction.

The taller an organizational structure, the more emphasis on intermediate
coordination processes, while the flatter an organizational su'ucture, the lower the degree
of information distortion (Worthy 1950, Melzer and Salter 1962, Wilensky 1967, Carzo
and Yanouzas 1969). A tall structure means that there exist intermediate aggregation
procedures between the field agents and the top manager. There is a delegation of some
information aggregation tasks to lower levels of the organizational structure. This reduces
information costs for the tOp manager but risks distortion in information processing. A
flat structure exists when there are direct information routes between the field agents and
the top manager. There is no interruption or deterioration in information processing, but
it imposes more information costs on the top manager when analyzing the reports.

Distinguishing between tall and flat structures, and between one-to-one matching

and one-tO-many matching yields four different kinds of organizational structures for

information processing. The characteristics of each will be analyzed.

 

 

 

 

57

Table 2: Four Models of Organizational Structure

 

 

 

Flat Structure Tall Structure
One-to-one matching Flat one-to-one structure Tall one-to-one structure
One-tO-many matching Flat one-to-many structure Tall one-to-many structure

 

 

 

 

 

4.2 Flat one-to-One structure

A flat one-to-one structure consists of one top manager, tm, and a number of field
agents, a,, i = 1,...,n. Each a,- has his own jurisdiction and observes a sector E,- from
which he can get information on A,. Based on his observation of E,, each a, produces a
report r,, and send it to rm, who aggregates r,- and makes an organizational report.

The environment E would be segmented into separated jurisdictions, E,. It
facilitates increased specialization (Duncan 1979). Thus the field agent a,- generates
information based on the Observations in his jurisdiction Ei. As a Bayesian player, the
field agent has his own prior estimate of the probability of event A, g,(A), and the
likelihood function of event A based on his information gathering. The field agent’s
posterior probability is based on his prior and on his observations of his jurisdiction.
Whenever he receives information, he evaluates whether it indicates the occurrence of
event A. Each a,- constructs his own beta probability distribution function of event A, i.e.,
Pi(A) =f(gi(A): “Er”

Each a,- observes his jurisdiction continuously, and gathers a certain amount of
information at each time. Since he has a restricted jurisdiction, he can cumulate his
knowledge on his jurisdiction through continuous observations so that he can analyze data

more precisely. He gathers more samples from his jurisdiction, and he is gradually

 

 

 

 

W

 

 

 

58

relying more on the objective observations on E,. If he accumulates detailed information
about the same E,, its associated probability interval about event A would be narrow: a
larger sample contains more information about the jurisdiction and hence allows more
precise estimation. Since a,- can have a narrow focus on E,- and can draw a large sample
from E,, it is expected that a,’s posterior probability distribution function on A, will be
narrow and sharply peaked, and thus have lower variance. Based on his posterior
probability, the field agent would report r,- to the tOp manager.

Since each field agent has his own jurisdiction, each report is based on
observations from different segments of the environment. Each field agent makes 0,-
observations of data from E,. So the total numbers of Observations on E in this structure
would be 20,-, i = 1,...,n. If all the field agents observe the same k number of data from
each jurisdiction Eb then the total number of observations in this structure would be nk.
This structure thus transmits n reports to the top manager.

A field agent may behave strategically because a field agent with an independent
jurisdiction has some power from his information monopoly. This is because each field
agent can easily distort information without arousing the suspicion of the other field
agents. Even though the report may be doubted by the tOp manager, the field agent can
persuade the tOp manager by demonstrating his expertise on that jurisdiction. This power,
based on information monopoly and expertise on his jurisdiction, might allow all field
agents to behave strategically, and each field agent might use his own power to maximize
his relative weight on information aggregation. If the field agent a,- expects that a

strategic report (r,- at p,(A)) will provide him more expected utility than a sincere report

 

 

 

59
(r,- = p,(A)), he will not tell truth; that is, r,- 9* p,(A) if w,(r,- at p,(A)) > wjr, = p,(A)), and
r,- = p,(A) if w,(r,- = p,(A)) 2 w,(r,- :t p, A )). The top manager needs to give field agents an
incentive for honest reporting. Therefore, the top manager will apply the relative weight
rule as an Bayesian incentive compatible mechanism.
[Figure 1 about here]

The top manager is the only one responsible for coordination in this flat one-to-
one structure. The top manager takes responsibility for the synthesis of all specialized
inputs. The major role of the top manager in this structure is to decide how to assign the
relative weights to the reports. The top manager receives the reports from all the field
agents and aggregates them into a likelihood estimate of event A. Since there are the
direct reporting routes between the n field agents and the top manager, the top manager
must analyze and compare all n reports. When he applies the relative weight rule, he
needs to allocate a weight to each report. He will incur great information costs if the
number Of field agents is large. The information tm receives is Zr, = Zflg,(A), l,(E,-)) and
he assigns w,- = f(w,-"', wf’, r,, rj), i ¢ j. By using the reports from the field agents and

his aggregation method, the top manager constructs his likelihood, that is, ,um, = Xw,-r,-.2

Then the top manager makes the organizational report on event A, based on his prior and

 

 

The relative importance of jurisdiction can be considered in information aggregation. Sometimes not all
jurisdictions are equally important in the final evaluation of the event. In this case, the top manager needs to
modify the relative weight rule to include a factor on jurisdictional importance. But in this study I assume that
all jurisdictions have equal importance since for analyzing how organizational structures affect information
processing and aggregation it is necessary to fix the other factors constant. The other purpose of the relative
weight rule is to ensure honest reporting that is the basic condition of comparing organizational structures.
However, I expect that even with somewhat different form combining jurisdictional importance, the relative
weight rule is still useful for both aggregating information and ensuring sincere reporting because jurisdictional
importance is not controlled by field agents but determined by the top manager. It will be discussed further in

the conclusion chapter.

 

 

 

60
this likelihood. The process of making the organizational report is: Zf(g,(A), l,(E,)) —-> it,“

with g,,,,(A) —> r,,,, = f(g,,,,(A), pm) = f(g,,,,(A), l,,,,(E)). At this point, the probability reports
from the different jurisdictions can be combined into a single probability based on the
whole environment: [0, 1]" ——> [0, 1].

Feedback functions to enforce the relative weight rule to work effectively. The
field agents will report truthful estimates because the Bayesian incentive compatible
mechanism ensures w,( r, = p,(A)) 2 w,( r, ¢ p,(A)), as shown by Theorem 1. The only way
to maximize the expected utility is to report honestly as r, = p,(A). Based on the deviation
between a field agent’s estimate of the probability of event A and the actual state of event
A, the top manager would revise the relative weight of the field agents. Both the top
manager and the field agents may update their prior beliefs through feedback. They may
compare their estimates and the actual state of the event, and adjust their prior beliefs.
Based on the change in his relative weight, the field agent can revise his information
evaluation method by examining whether he interpreted the data precisely. The
information processing in flat one-tO—one structure would be summarized as Figure 2.

[Figure 2 about here]

When the top manager evaluates this structure and compares it with the other
alternatives, he would focus on the two principles-—information cost and event
predictability. The overall information cost of this organizational structure is the sum of
the top manager’s information processing costs and of the field agents’ observation costs.
The unit cost of analyzing intelligence from the environment is assumed to be constant.

The field agent’s information cost (ic,) is the product of the number of Observations, 0,,

 

 

61

and the unit cost of an observation, 0. Thus, ic, = c '0,. For n field agents, the sum of
observation costs is (2 c - 0,, i = 1,...,n). The top manager needs to analyze and
compare all the reports from the field agents. The first determinant of the top manager’s
information cost is the number of reports, which is the same as the number of the field
agents in this structure. Since this structure generates n reports at each time and it has
only two levels of hierarchy, the top manager’s information processing cost is 2”. Thus
the total information cost of this structure is (2’l + 020,, i = 1,...,n). If all field agents
observe the same k number of data from the jurisdictions, and c = 1, the information cost
would be (2" + nk).
Given the existence of the Bayesian incentive mechanism, each field agent would
report r, = p,(A). Then event predictability can be estimated by the function of both
information efficiency and information reliability. This structure emphasizes information
efficiency since it is increased by a specialization of the jurisdiction of a field agent.
Information efficiency is based on the number of observations, N, which this organization
generates. It is (N = 20,) when 0, is the number of observations of 0,. Since event
predictability cannot be known until Observing the state of event A at the next time, and
each time it may be varied, it cannot be directly estimated. However, it can be evaluated
by calculating its expected value. The expected value of event predictability is
proportional to the function of information efficiency and information reliability, as EP
<><= NAN, as defined in Chapter 3. The expected value Of EP of this structure is supposed
to be proportional to a certain point on the interval [20,, (20,)"]. If all the field agents

generate the same k data, then the interval would be [nk, (nk)"].

 

 

 

62

The advantages of this flat one-to-one structure are: (1) information gathering can
be more efficiently performed if the environment of-the—whole is decomposed into a
certain number of separated jurisdictions for matching each jurisdiction with each field
agent; that is, E, with a,, and if the field agent is allowed to specialize in each jurisdiction;
and (2) there is less procedural error since the top manager can get access to the
information from the field agents without any filtration or distortion.

The top manager of this flat one-to-one structure has a more deterministic role in
information processing than the other models since (1) there is only one position to
aggregate the reports and only the top manager can manage information aggregation; and
(2) the top manager is the only one who has a broad vieWpoint on the whole environment
E. The top manager has a greater breadth of information about the environment than the

field agents, while the latter have more detailed information about their own jurisdictions.

4.3. Flat one-to—many structure

A flat one-tO-many structure consists of one tOp manager, 1m, and a number of
field agents, 0,, i = 1,...,n. All a, observe the same portion of all sectors of E, that is, e,
= (e,,...,e,,), e, e E,, e, e E, e = e, = e,, i ¢j.

Each portion of environment e would be observed by all field agents. Each field
agent a, generates information based on his observations of the common jurisdiction.
Thus the jurisdiction would be scrutinized from n different perspectives. Even though all
observe the same jurisdiction, the estimates might be different because each field agent

has his own method of Observation and analysis. For example, a, and 0,, i ¢ j, observe

 

 

 

 

63

the same intelligence. One possible result might be thatga, evaluates it as evidence for
the possible occurrence of the event, while a, decides that it indicates the event will not
happen. Thus even with e, l,(e) might be different from l,(e). On the other hand, a, and
0, might draw the same conclusion from e.

As a Bayesian player, the field agent has his own prior belief on the probability
of event A. He then gathers some information from the environment, and constructs his
posterior probability distribution function of event A as p,(A) = f(g,(A), l,(e)). Since a,
makes only a small number of Observations on each e,, a,’s probability distribution
function on A is broad--it comes all of E—-but shallow. Based on his posterior probability,
the field agent would report r, to the top manager.

Since all the field agents have a common jurisdiction, all reports are based on the
same observations with different estimation methods. Thus if r,(A) is the same as r,(A),
information is more reliable since there is a consensus between the two different
estimations. 2r,(A) would be the sum of independent evaluations on the same data. In
this structure, each field agent observes 0 number of data, 0 = 0, = 0,, i ¢ j, from the
environment e, and thus the total number of observations of this structure is the same as
the number of an individual’s observations 0 because of the duplication effect. If a,
gathers k sample data from e, the total number of Observations in this structure would be
k since all field agents get the same data, but each a, may interpret the data differently.
The reduction in the effective sample size, compared to one—tO—one structure, results in
more weight being given to the prior probability in the computation of the posterior

probability (Winkler 1985). On the whole, the set of the reports from the field agents

 

 

 

 

 

64

would cover only the shallow and broad Observations on the environment. Also for E,,
it gathers only k/n data.
[Figure 3 about here]

The top manager’s role in this structure is to synthesize the different points of
view. Since the field agent gets information on all dimensions of the event, the
differences in the reports from the field agents indicate the differences in their subjective
interpretations of the event. Thus, information aggregation by the top manager is focused
on the coordination of the different evaluations. If there is consensus among the field
agents’ reports, there is no need to analyze and compare the reports by the unanimity
principle. Only the top manager needs to announce the agreed probability as the
organizational report. But if there is no consensus, the top manager will incur great
information costs since he receives n number of reports. tm receives 2r, = 2f(g,(A), l,( e) )

l t

and assigns w, = f(w," , w-

] ", r,, r,), i at j. By using the reports and his aggregation rule,

the top manager arrives at his likelihood, um, = 2w,r,. Then the top manager makes the
organizational report on event A, by combining his prior belief and this likelihood. The
process of making the organizational report is: 2f(g,(A), l,(e)) —> am, with g,,,,(A) —> r,,,, =
f(gm(A), Mm) == f(g.m(A), 1.48)).

Feedback reinforces the Bayesian incentive compatible mechanism, and functions
as the process to adjust the subjective perspectives of the field agents and the top manager
on the event. Through the adjustment process the top manager may update the relative
weights of field agents and his prior belief, and each field agent would revise his prior

belief and estimation method. The information processing in flat one-to-many structure

 

 

 

 

 

65

would be summarized as in Figure 4.
[Figure 4 about here]

The information cost of this organizational structure is the sum of the top
manager’s information processing costs and the field agents’ observation costs. The unit
cost of observing evidence from the environment is assumed to be constant. The field
agent’s information cost (ic,) is the product of the number of observations, 0, which is the
same for all a, and the unit cost of an observation, c. Because all field agents observe
the same data from the duplicated jurisdiction, each must analyze them independently, and
so each incurs an information cost. Thus, ic, = ic = c - o. For n field agents, the sum
of observation costs is n ' c '0. The top manager needs to analyze and compare all the
reports from the field agents. The top manager’s information cost is determined by the
number of reports, which is the same as the number of the field agents in this structure.
Since this structure generates n reports at each time and it has only two levels of
hierarchy, the tOp manager’s information processing cost is 2”. For comparability, if all
field agents observe the same k data and c = 1, the total information cost of this structure
is (2" + n ' k).

Event predictability is proportional to the function of both information efficiency
and information reliability. This structure emphasizes information reliability since it is
increased by the duplication of the jurisdiction for all field agents. Information reliability
is based on the number of similar reports in an organization. The more consensus on the
state of the event the organization achieves, the more reliable the prediction is. The

expected value of event predictability of this structure is proportional to NAN, where N is

 

 

 

66

the number of observations and AN is the number of same reports on the estimation.
Thus, event predictability is proportional to a certain point on the range [0, 0”]. For
comparability, if all the field agents generate k data, then the range would be [k, k”].

One aspect Of this redundant suucture is that with duplicating jurisdictions, the
field agents compete against each other over maximizing their relative weights, and
information is scrutinized from different perspectives. Thus, it benefits from diverse
evaluation methods. This leads to more reliable reports since the field agents make the
best estimates in competition and there may be more of a chance to achieve consensus
on the probability of event A among field agents.

The other aspect of this redundancy is that this structure permits the field agents
to predict the event by constructing posterior probabilities based on the environment e.
The function of information aggregation is basically distributed to the field agents. Thus
there might be tensions among competing field agents or inner-bargainin g among colluded
field agents since their reports can be easily compared with each other.

The advantages of this flat one—to-many structure will be: (1) the reports from the
field agents would be more reliable since there is the common jurisdiction and the field
agents may try to make the best estimate of event A; (2) the field agents as well as the
top manager are able to have a broader perspective on the environment and event A, and
it offers the field agents greater scope and responsibility in their prediction; and (3) as in
flat one-to-one structure, there is less procedural error in information processing since
there is no filtration or distortion between the field agents and the top manager. By using

the common jurisdiction, this structure seeks to benefit from the diverse views and

 

 

 

67

judgments of field agents in predicting event A and provides means of identifying and
correcting errors that would be undetectable in one-to-one structures.

The top manager has a less deterministic role in information processing since (1)
all field agents as well as the top manager have a broader understanding on the
environment and event A; and (2) the field agents share the role of information

aggregation with the top manager.

4.4. Tall one-to-one structure

A tall one-to-one structure consists of one top manager, tm, several division
managers, 0172,, j = 1,...,v, v < n, and a number of field agents, 0,, i = 1,...,(n - v). The
total number of the subordinates of the top manager is fixed as n as in the other
organizational structures. The number of divisions is less than or equal to the number of
field agents, v S (n - v). Each division has (n-v)/v field agents.3 Each division has its
own divisional environment, DE,, in which each field agent in this division has his own
jurisdiction, E,. 0, observes E, and processes r, to dm, who aggregates the reports and
transmits a divisional report, dr,, to tm. The tm then aggregates these reports and make
a final evaluation on the occurrence of event A.

The information processing in this structure involves three factors: how the field
agents generate information; how the division managers aggregate the reports; and how

the top manager analyzes the divisional reports. The environment E would be segmented

&

For simplicity, it is assumed that each division has an equal number of field agents. However, unequal
allocations of the field agents (e. g., Division 1 has 3 field agents, and Division 2 has 5) produce the same results.
When (n-v)/v is not integer, it means there is unequal allocations of field agents.

 

 

 

68

into separate jurisdictions, E,, i = 1,...,(n - v). An E, might be larger than that in a flat
one-to-one structure since (n - v) < n. The characteristics of information generation by
the field agents are similar to those in a flat one—to-one structure. The field agent 0,
generates information based on the observations on his jurisdiction E,. As a Bayesian
player, the field agent has his own prior information on the probability of event A, g,(A),
and the likelihood function of event A based on his information gathering. Whenever
he gets information, he evaluates whether it indicates the occurrence of event A. Since
each field agent has his own prior belief on event A and his own method of evaluating
information, he subjectively estimates the information from the jurisdiction. Each 0,
constructs his own beta probability distribution function of event A, i.e., p,(A) = f(g,(A),
l,(E,)). If 0, has detailed information about an area, its associated probability interval
would be narrow. A larger sample contains more information about the jurisdiction and
hence allows more precise estimation. Since a, can have a narrow but deep focus on E,
but with a large sample from his jurisdiction, it is expected that 0,’s posterior probability
distribution function on A, is narrow and sharply peaked and thus has the low rates of
variance. Based on his posterior probability, the field agent would report r, to the
division manager.
[Figure 5 about here]

Since each field agent has his own jurisdiction, each report is based on the
observations on a different segment of the environment. Thus even though r,(A) is the
same as r,(A), each probability is based on a different jurisdiction, E, at E,, and a different

evaluation method. Each field agent observes 0, number of data from E, and then the

 

 

 

69

total information generations in this structure would be 20,, i = 1,...,(n - v) on E. Thus
the estimations on 20, would be included in 2r, reports. If all the field agents observe
the same k number of data from each jurisdiction, then the total number of observations
in this structure would be (n - v)k. This structure transmits (n - v) reports to the division
managers.

Each division manager receives (n - v)/v reports from the field agents in his
division. A major function of division managers is the filtering and editing of
information. The role of the division manager is to compare and analyze the reports
based on his own perspective and to transmit the divisional report to the top manager.
Each division manager has a broader viewpoint of the divisional environment, while the
field agents have more detailed knowledge of their own jurisdictions. dm receives 2r, =
2f(g,(A), l,(E,)) and assigns w, =f(w,"’, w,“’, r,, r,), i ¢ j. By combining the reports and
his own aggregation rule, the division manager aggregates the reports, ,udm = 2w,r,, i =
1,...,(n - v)/v. Then the division manager makes the divisional report on event A, based
on his prior and this likelihood which is based on the divisional environment. The
process of making the divisional report is: 2f(g,(A), l,(E,)) —> ad," with gdm(A) —) drj -_-
flganM), 14m) = f(g,,m(A), ldm(DE,)). At this stage, the probabilities based on the field
agents’ jurisdictions can be combined into the condensed probabilities based on the
divisional environments: [0, 1]("“’) —> [0, 11”-

The top manager receives the reports from only the division managers. Since

there are indirect reporting routes between the field agents and the top manager, the tOp

manager can get access to only filtered and prescreened information. But this reduces the

 

 

 

 

70

top manager’s information processing cost since v < n. When the top manager applies
the relative weight rule, he needs to allocate the weights to the divisions. Thus it might
be easy to update the relative weights at each time since the number of the divisions is
smaller than that of the field agents.

The tm receives 2dr,- = Zflgde), flan.) = 2178mm), 14,,(DE,)). BY using the
reports from the division managers and his aggregation method, the top manager
constructs his likelihood, that is, um, == 2w,dr,. Then the top manager makes the
organizational report on event A, based on his prior belief and this likelihood. The
process of making the organizational report is: 2f(g,(A), l,(E,)) ——> 2dr, = f(gd,,,(A),
1d,,(DE,)) —> it," with g,,,,(A) -—-> r,,,, = f(g,,,,(A), am) = f(g,,,,(A), l,,,,(E)). At this point, the
probabilities based on the different divisional jurisdictions can be combined into the single
probability based on the whole environment: [0, 1](""’) —> [0, 1]" —> [0, 1].

Feedback in tall one-to-one su'uctures has these three-stage procedures: (1)
feedback from the actual state of the event; (2) feedback from the top manager to the
division managers; and (3) feedback from the division manager to the field agents. By
comparing his prediction with the actual state of the event, each actor might revise his
prior belief and his estimation method. After observing the state of the event, the tOp
manager would update the relative weights on the division managers and inform them of
the revised ones. Based on the feedback from the state of the event and from the top
manager, the division managers would adjust their perspectives on the divisional
jurisdictions and also revise the relative weights of the field agents. Given the feedback

from the division manager and the state of the event, the field agents would update their

 

 

 

71

prior beliefs and information estimation methods. Information processing in tall one-to-
one structure is summarized in Figure 6.
[Figure 6 about here]

The information cost of this organizational structure is the sum of the managers’
information processing costs and of the field agents’ observation costs. As before, the
unit cost of observing evidence from the environment is assumed to be constant. The
field agent’s information cost (ic,) is the product of the number of Observations, 0,, and
the unit cost of an Observation, 0. Thus, ic, = c ' 0,. For (n - v) field agents, the sum
of observation costs is (2 c ' 0,, i = 1,...,(n - v)). In this structure with 3—1evel
hierarchies, the managers’ information processing cost will be estimated as the sum of the
information cost of the tOp manager and of the division managers. Each division manager
needs to analyze and compare (n - v)/v reports from each division’s field agents. Thus
the information processing cost of each division manager is 2‘" ' V)". The sum of the
information cost of all division managers is 22"’ ' "”" = 12(2)"I ‘ ””‘1 The top manager
receives only v number of reports from the division managers. His information cost is
3". The managers’ information processing cost would be (v(2)"' "’”" + 3”). Then the total
information cost of this structure is (020, + v(2)(""’”" + 3", i=l,...,(n - v), v < n).

This information cost will vary with the number of divisions. If the top manager
wants to decrease his own information processing costs, he will reduce the number of
divisions so that each division manager needs to analyze more reports and to pay more
of the information cost. If the field agent observes the same k number of data from each

jurisdiction and c = 1, then the total information cost of this structure would be (k( n - v)

 

 

 

72

+ v(2)"""”" + 3", v < n).

This structure emphasizes information efficiency, instead of information reliability
since it is increased by the specialization of the jurisdiction of a field agent. Information
efficiency is based on the number of observations, N, which this organization generates.
It is (N = 20,, i = 1,...,(n - v)) when 0, is the number of observations of 0,. Since event
predictability is proportional to a function of information efficiency and information
reliability, such as EP cc NAN. EP of this structure is pr0portional to a certain point on
the interval [20,, (20,)" ' "], i =1,...,(n ~v). If all the field agents generate the same k data,
then the interval would be [(n - v)k, ((n - v)k)""’].

The advantages of this tall one-tO-one structure are: (1) this structure would greatly
reduce the burdens of the top manager to analyze and compare many reports by
introducing middle-level managers whose task is to analyze the reports from the field
agents, and transmit the condensed information to the top manager, while keeping the
total number of his subordinates fixed; (2) as in flat one-tO-one structure, information
gathering can be more efficiently performed if the environment is decomposed into a
certain number of separated jurisdictions; and (3) this structure provides more evaluation
procedures which make the organizational report more accurate.

There is a conventional wisdom that the quality of information finally received by
the top manager will probably be very different from the original reports of the field
agents (Downs 1967). Wilensky (1967) postulated three structural conditions which
increase the possibility of information distortion: (1) hierarchy, which restricts free

information flows; (2) specialization, which reduces horizontal communications between

 

 

 

73

the field agents; and (3) centralization, by which the top manager is too far removed from
the actual observations. Wilensky (1967) argued: "The greater the number of ranks and
the greater the number of organizational units involved in a decision process, the more
the distorting influence of rank and jurisdiction and, consequently, the greater the chance
of an intelligence failure" (p.57). However, this research indicates that information
distortion can be removed by providing appropriate feedback.

The top manager of this tall one-to-one structure has moderately deterministic role
in information processing because he shares the function of information aggregation with
the division managers but he is the only one who can have information about the whole

environment.

4.5. Tall one-tO-many structure

A tall one-to-many structure consists of one top manager, 1m, several division
managers, dm,, j = 1,...,v, v < n, and a number of field agents, 0,, i = 1,...,(n - v). Each
division has (n - v)/v field agents. In each division all field agents have the same
jurisdiction. Hence all 0, in a division observe the same portion of all E,, that is, e, =
(81,.-.,e,,,,,), e, e E,, e, e E, e = e, = e,, i ¢j.

Field agent 0, generates information based on his observations of the common
jurisdiction. Information generation by the field agents is similar with that in the flat one-
to-many structure. The jurisdiction would be scrutinized from (n - v) different
perspectives. Even though all observe the same jurisdiction, each makes an estimate

which might be different because each field agent has his own method of analysis. As

 

 

 

 

74

a Bayesian player, the field agent has his own prior belief on the probability of event A,
and he can continuously update it through feedback processes. He gathers a certain
amount of information from the environment and makes a likelihood function. Then he
consu'ucts his posterior probability distribution function of event A as p,(A) = f(g,(A),
l,( e) ). Based on his posterior probability, the field agent would report r, to the division
manager.

[Figure 7 about here]

Since all the field agents share a common jurisdiction, all reports are based on the
same observations with different estimation methods. Thus if r,(A) is the same as r,(A),
information is more reliable since there is an agreement between the two different
estimations. 2r,(A) would be the set of independent evaluations on the event A. In this
structure, each field agent observes 0 number of data, 0 = 0, = 0,, i at j, from the
environment e and the total information generation would be the same as 0 because of
duplication effect. If 0, gathers k sample data from e, the total number of observations
in this structure would be k since all field agents get the same sample data, but each 0,
may interpret the data differently. Thus 2r, is more affected by the field agents’
subjective beliefs.

Each division manager receives (n - v)/v reports from the field agents in his
division. The role of the division manager is to compare and analyze the reports based
on his own perspective and to transmit the divisional report to the top manager. The
reports r, = f(g,(A), l,(e)) would be re-evaluated from the view point of the division

manager. 0’m receives 2r, = 2178101), l,(e)) and 38813118 W,- =f(W,-"’, Wf'l, 1“,, rj): i #11

 

 

 

 

 

75

Each division manager aggregates only the reports from his own subordinates. By
combining the reports and his own aggregation rule, the division manager aggregates the
reports, pd," = 2w,r,, i = 1,...,(n - v)/v. Then the division manager makes the divisional
report on event A, based on his prior and this likelihood which is based on the common
jurisdiction. The process of making the divisional report is: 2f(g,(A), l,(e)) -—> pd", with
gd,,,(A) —-> dr, = f(g,,,,,(A), pm) = f(g,,,,,(A), l,,,,(e)). At this stage, the probabilities based on
the different estimation methods can be combined into the probabilities based on the
divisional aggregations: [0, 1](”"’) —> [0, 1]”-

The top manager receives the reports from only the division managers. Since
there are indirect reporting routes between the field agents and the top manager, the top
manager can get access only to filtered information. But this reduces the tOp manager’s
information processing cost since v < n. When the top manager applies the relative
weight rule, he needs to allocate the weights only to the divisions. Thus it might be easy
to update the relative weights at each time. tm receives 2dr, = 2f(gd,,,(A), pm) =
2f(gd,,,(A), ldm( e) ). By using the reports from the division managers and his aggregation
method, the top manager constructs his likelihood, that is, 14",, = 2w,dr,, j = 1,...,v. Then
the top manager makes the organizational report on event A, based on his prior belief and
this likelihood. The process of making the organizational report is: 2f(g,(A), l,( e) ) —-> 202‘,
= f( 81mm), 1146)) -r Hm with gal A) -> rm. = f(gm(A), Hm.) = f(gtm(A), 1,m(e))- At this Point,
the probabilities can be combined into the single probability: [0, 1](”“’) —> [0, 1]" —>
[0, 1].

Feedback in this flat one-to—many structure functions as the way to reinforce the

 

 

 

 

76

Bayesian incentive compatible mechanism, and as the process to adjust the subjective
perspectives of the field agents and the top manager on the event. After observing the
state of the event, the tOp manager would update the relative weights on the division
managers. Based on the feedback from the state of the event and from the tOp manager,
the division managers would revise the relative weights on the field agents. By the
feedback from the division manager and the state of the event, the field agents would
update their prior beliefs and information estimation methods. The information processing
in tall one-to-many structure would be summarized in Figure 8.
[Figure 8 about here]

The information cost of this organizational structure is the sum of the managers’
information processing costs and of the field agents’ observation costs. As before, the
unit cost of observing evidence from the environment is assumed to be constant. The
field agent’s information cost (ic,) is the product of the number of observations, 0, which
is the same for all 0,, and the unit cost of an observation, c. Even though all field agents
observe the same data from the duplicated jurisdiction, and so each needs to analyze them
independently, each needs to pay an information cost. Thus, ic, = ic = c ' 0. For (n -
v) field agents, the sum of Observation costs is (n - v) ' c - o. The managers’
information processing cost will be the sum of the information cost of the division
managers and the top manager. Each division manager needs to analyze and compare (n -
v)/v reports from the field agents. Thus the information cost of all division managers is
22‘“ ‘ ‘0’" = v(2)"‘ ' W". The tOp manager receives v number of reports from the division

managers. His information cost will be 3". The total information cost of this structure

 

 

 

77

is ((n - v)c . o + v(2)(" ' "V" + 3", v < n). This information cost will vary by the number
Of divisions. If all field agents observe the same k number of data, and the unit cost of
observation is 1, then the total information cost of this structure is (( n - v)k + v(2)"""”"
+ 3", v < n).

Event predictability is proportional to the function of both information efficiency
and information reliability. This structure emphasizes information reliability since it is
increased by the duplication of the jurisdiction for all field agents and since this structure
provides the three steps to estimate the event with the same e. Information reliability is
based on the number of similar reports in an organization. The more consensus on the
state of the event the organization achieves, the more reliable the prediction is. Event
predictability, EP, of this structure is proportional to N“, where N is the number of
information generations and AN is the number of major agreement on the estimation.
Thus EP is proportional to a certain point on the interval [0, 0‘" "9]. For comparability,
if all the field agents generate k data, then the interval would be [k, k‘” ' ”’].

This tall one-to-many structure, by using the common jurisdiction, seeks to benefit
from the diverse views and judgments of field agents in predicting event A. The
advantages of this structure are: (l) as in flat one-to-many structure, the reports from the
field agents would be more reliable; (2) the field agents as well as the managers are able
to have a broader perspective on the environment and event A, and it offers the field
agents greater scope and responsibility in their prediction; and (3) as in tall one-to-one
structure, by introducing middle-level managers the top manager can save his information

processing cost, and the means for more evaluation is provided.

 

 

 

78

The top manager has a much less deterministic role in information processing

because he shares the information aggregation function with the division managers as well

as the field agents.

4.6. Comparative analysis of organizational structures

The previous sections have studied information processing in different

organizational structures with the fixed n number of subordinates. This section

summarizes the results by demonstrating that different organizational structures transmit
different information to the top manager and by demonstrating, with regard to information

cost and event predictability, what kind of organizational structure is more attractive to

a top manager who designs an organization.

Proposition 1 : Each organizational structure can transmit different information to the
top manager, with the same number n of subordinates.

Proof: Organizational structure determines the way to aggregate reports into a collective
one as explained in the previous sections. Since different organizational structures

operate different aggregation methods, different organizational structures process
different information to the top manager. The following table shows how different

reports are given to the top manager.

]

 

! tm receives:

(1) 2r, = 2f(g,(A), l,(E,)), i = 1,...,n.

(2) 27', = 2f(g,(A), l,(e)), i = 1,...,n.

(3) 2dr,- = 2778mm), ldm(DEj))Ij = 1r°°°rV.
(4) 202‘, = Zf(gd...(A). lane». 1‘ = 1,...,v.

 

Flat one-to-one structure

 

Flat one-tO-many structure

 

Tall one-tO-one structure

 

 

 

 

 

1 Tall one-to-many structure

 

(1) ¢ (2) a (3) r (4).
QED.

 

Even with the same n number of subordinates and the same environment E,

 

79

 

different organizational structures generate different reports. The goal of a tOp manager,
or an organizational designer who designs an organization, is to minimize information
cost and to maximize event predictability. If the top manager decides that information
cost is more critical than event predictability, he would choose the organizational structure
which can minimize information cosr regardless of event predictability. The following

proposition shows what kind of organizational su'ucture a manager needs so that he pays

less information cost with the same n number of subordinates. As defined in Chapter

3.3, the information cost of an organization is the sum of the field agents’ observation
costs and of the managers’ information processing costs. That is, IC = 2 c ' 0, + 2
LmNm. Let us assume that all the field agents generate the same k number of observations
and the unit observation cost of the field agents is fixed as c = 1. The number of the
field agents is greater than or equal to the number of the divisions in tall structures, (n -

v) 2 v.
The total information, cost of tall su'uctures is less than that of flat

Proposition 2:
structures, if n > 2.

Proof: The information costs of organizational structures are:

 

,[ Information cost

 

nk+ 2’I

” Flat structures
(n - v)k + v(2)"""”" + 3" ~

 

 

 

 

j Tall structures

 

where n, v, k are all integers, n 2 2v, v 21, n > 2, k 2 1.

If both (1) nk > (n - v)k and (2) 2" > v(2)"""”" + 3” are satisfied,
then (3) nk + 2” > (n - v)k + v(2)(" “’)"’ + 3" is also true.

(1) Since n > v and k 2 1, the inequality (1) is proved.

 

 

80

 

If (2) is verified with the minimum numbers of n and v, then the inequality is
proved with all numbers. The minimum numbers satisfying n > 2, v 2 1, and n

22varen=3andv=1.

Withn=3andv= 1, 2"—23=8>7=22+3l =v(2)‘"“')"+3’.
(ii) 2’I > v(2)(""')"’ + 3", n > 2, n 2 2v, v 21.

By (i) and (ii), the inequality (3) is proved.4
QED.
Proposition 2.1 : The top manager’s information processing cost is less in tall structures
than in flat structures, if n > 2.

Proof: By the proof of Proposition 2, the managers’ information processing cost of tall
structures is less than that of flat structures.

2’I > v(2)(’""”’ + 3", n > 2, n 2 2v, v 2 1.

v(2)(n - v)/v > 0

 

tm’s information cost
2n
3V

 

Flat structures

 

 

Tall su'uctures

 

 

 

Therefore, 2" > 3".
QED.
Tall structures are more attractive to the top manager who seeks to reduce

information cost.’ On the other hand, if the top manager emphasizes event predictability

 

The result is also true even with unequal sizes of divisions in tall structures. The minimum numbers of n and

v which make it possible to have the unequal sizes are n = 5 and v = 2 -- e.g., Division 1 has 2 field agents and
Division 2 has 1 field agent. The information cost of this structure will be 3k + 22 + 2‘ + 32 = 3k + 15, while
that of the flat structure with n = 5 is 5k + 2’ = 5k + 32. The inequality (3) is proved since 5k + 32 > 3k + 15,

for k 2 1.
When criticizing transaction cost theory, Hammond (1991b) indicates that we need to differentiate the top
manager’s costs from those of his subordinates:
It is more accurate to say that each structure reduces some actors’ transaction costs while
increasing the transaction costs of other actors. It follows that structural design hinges on the
question of whether the chief executive’s transaction costs are more important than those of

the subordinates, or less important (p.63, fn.21).

 

 

 

81

more than information cost, he would choose the organizational structure which can
maximize event predictability, with a given n number of subordinates, regardless of
information cost. The general axioms of event predictability (EP) are that the more
information efficiency, the more event predictability, and that the more information
reliability, the more event predictability. Given the existence of the Bayesian incentive
compatible mechanism, event predictability is proportional to the function of information
efficiency and information reliability, as EP oc NAN, where N is the number of
observations which an organization generates, and AN is the number of major agreement
on the estimation. EP is proportional to a certain point on the interval of NA”.
Proposition 3: The flat one-to-one structure generates greater event predictability than

the other structural alternatives, given the existence of the relative

weight rule as the Bayesian incentive compatible mechanism.

Proof: Let us assume that all the field agents observe the same k number of data, k 2 1.

 

Interval of NAN

 

Flat one-to-one structure [nk, (nk)"]

 

Flat one-to-many structure [k, k"]

 

Tall one-to-one structure [(11 - v)k, ((n - v)k)"""’]

 

Tall one-to-many structure [k, k(""’)]

 

 

 

 

Using a minimax strategy, the top manager will choose the organizational structure
which guarantees the maximum among the minimums of those intervals. Since
nk > (n - v)k > k, n > 2, k 2 1, n > v 2 1, the flat one-to-one structure ensures the

greatest information predictability among four structural alternatives.
QED.

The top manager would choose the tall structure in order to reduce information

costs. For increasing event predictability, the top manager would design a flat one-to-one

 

 

82

structure with using a Bayesian incentive mechanism. But the top manager would want
to design an organizational structure which can minimize information cost and maximize
event predictability simultaneously. Then the top manager’s choice (ch) of an
organizational structure is based on the function of both information cost, IC, and event
predictability, EP:

ch = f(IC, EP) = EP
IC

An organizational structure which ensures greater value in the choice function is better
for the top manager. The following theorem demonstrates what kind of organizational
structure is better choice for the tOp manager.

Theorem 2: The tall one—to-one structure is better than the other structural alternatives
for the top manager, if n > 4.

Proof: Let us assume that all the field agents generate the same k observations, and the
unit observation cost is constant as c = 1, and n 2 2v, v 2 l, k 2 1. The tOp
manager is supposed to design an organization based on the choice function.

 

ch = EP , and if EP oc the minimum
IC of the interval of NAN

 

 

 

Flat one-to-one structure (1) nk
nk + 2"
Flat one-to-many structure (2) k
nk + 2"
Tall one-to-one structure (3) (n - v)k

 

(n - v)k + v2(”"’)’" + 3"

 

Tall one-to-many structure (4) k
(n - v)k + v2‘”"’”" + 3"

 

 

 

 

 

If the value of an organizational structure in the choice function is greater than the
others, it is the better for the top manager.

If the information cost is same, the greater the information predictability, the

 

83

greater the value of the choice function.

(1)>(2), nk>k,k21.
(3)>(4), (n-v)k>k,n_>.2v,v21.

Thus the choice is reduced to (1) and (3). Now let us suppose:
(i) (1) < (3).
If the inequality (i) is satisfied with the minimum numbers of n and v, then the

result is also true with any numbers of n and v. The minimum numbers satisfying
n > 4, v 2 1, n 2 2v are n = 5 and v = 1. With these minimum numbers:

(1) = nk = 5k = 5k
nk+2" 5k+25 5k+32
(3) = (n - Vlk = __4k__ = 4k

 

(n-v)k+v2"'“’”"+3" 4k+24+3 4k+19
Regardless of k, when n > 4, the inequality (i) is satisfied.6 For example, if k =
1, (1) = .135 < .174 = (3). If k = 100, (1) = .940 < .955 = (3). Thus the
inequality (i) is verified.
QED.

When considering both information cost and event predictability, the tall one-to-

one structure is better than the other organizational structures, with n > 4.7 At some

 

 

Even though there is an unequal allocation of the field agents satisfying the conditions It > 4, v 2 l, n 2 2v, such
as Division 1 has 2 field agents and Division 2 has only 1 field agent, the result is same. For instance, if n =
5, v = 2 and k = l, (1) = .135 < .167 = (3). Also when n > 4, any kind of structural variation with v 2 1 gives
the greater value in the choice function than the structure with v = 0.

Several Studies in the literature on structures are related to this result. Keren and Levhari (1979) suggested a
model of pure hierarchy which oversees a given number of productive units. The model permits the calculation
of an optimum formal structure of the hierarchy in terms of the span of control at different levels. When
considering planning time, the total planning time in tall structure will be less than in flat structure so long as
the number of productive units exceeds four. With the assumption that organization members are infallible in
the absence of overload, Drenick (1986) demonstrated mathematically that flat structures may not achieve the
objective of lightening the load on the top manager and tall structures are uniquely suited to the avoidance of
overload. Keren and Levhari (1989) analyzed the aggregation and error in the multi—divisional form of
organizational structure. The aggregation introduces error into decisions. They argued: "the degree of
aggregation has to be selected so as to attain the optimal trade-off between the total variance of this choice
process and administration costs" (p.217). Their model demonstrated that the span of control expands as one
goes down the tiers of the hierarchy, and aggregation is the most radical in the lower levels of the hierarchy.

 

 

 

84

point in the growth of an organization it is necessary to formalize tall structure and to
reduce information processing cost. While a flat one-to—one structure is able, by
generating more information, to deal effectively with greater amounts of uncertainty than
the tall one-to-one structure, there are greater information costs with this increased
information generation capacity. With the great number of subordinates the disadvantage
of information cost in the flat structure offsets the advantage of information predictability.
However, if the top manager is willing to suffer the greater information processing costs,
the flat one-to-one structure is more attractive since the top manager can allocate all the
subordinates as the field agents, who do the actual observations, and the top manager can
get all the reports.

Faced with an important event to estimate, the top manager would emphasize
event predictability more. If it is necessary to analyze a huge amount of intelligence in
order to predict the event, a flat one-to-one structure is preferable. When there is only
a limited information, a flat one-to-many structure would be better since the information
can be scrutinized from diverse perspectives. Table 3 summarizes the characteristics of
each organizational structure in information processing.

The shape of the organizational structure -- not only the number of ranks but also
the number of personnel at each level -- conditions the upward flow of information
(Wilensky 1967, Hammond 1986, 1990b, 1991b). Structural arrangements in
organizations determine the information processing patterns which, in turn, affect
organizational outcomes. Organizational structures can also determine the specific role

that the top manager will play in making organizational reports (Graber 1992).

 

85

Table 3: Characteristics of Organizational Structures

 

Flat one-to-one

Tall one-to-one

Flat one-to-many

Tall one-to-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

structure structure structure many structure
consists of: top manager, tm top manager, tm top manager, tm top manager, tm
v division, dmj v division, dml-
n field agents, a,- n field agents, a, 72 field agents, a, n field agents, a.-
a,’s E,. E,. e e
jurisdiction
0; reportsr r; = f(g.-(A). IKE.» r.- = f(gt(A). l.-(E.-)) n = 1782(4), l,(e)) rt = f(8.(A). l,(e))‘
generates: nk data (n - v)k data k data k data
dmj receives: (n - v)/v reports (n - v)/v reports
tm receives: It reports v reports It reports v reports
tm receives: Zn: Zf(g,-(A), I,(E,-) 2drj=2f(gd,.(A), 140,.) 27.: 2f(g.~(A), l.-( e) Eda-=ZflgmM).
to...)
14.2»- = ZWi’i Kim = Zwiri
dmj uses: Wt =f(Wr‘-lv ri) W; = f(wl!-!) rt)
tm uses: w, = f(w,-"’, r,-) wj = f(w"", drj) wi = f(w,"’, r,-) w]. = f(wj"1, drj)
For a,- specialization & specialization & redundancy & redundancy &
monopoly monopoly competition competition
tm & a,- direct interaction indirect interaction direct interaction indirect interaction
on has: more deterministic moderate deterministic less deterministic much less
role role role deterministic role
Information relatively high relatively low relatively high relatively low
cost
Event more efficient more efficient more reliable more reliable
predictability

 

 

is generally:

 

more reliance on
objective observations

 

more reliance on
objective observations

 

more reliance on
prior beliefs

 

more reliance on
prior beliefs

 

 

 

 

 

86

 

Figure 1: An example of flat one—to-one structure

 

 

 

Ei

 

 

 

 

Pdor

 

 

 

 

 

 

Pfior

 

 

 

Field agent ai ‘

 

 

n- = f(gi(A). Ii(Ei))

 

 

 

 

 

 

87
WE)
9/04
gtmtA) _
feedback

 

feedback

 

 

Top manager tm

 

 

rtm = f(gtm(A), /tm(E))

 

Event A

Figure 2: Information processing in flat one—to-one structure

 

88

 

 

 

 

 

 

 

Figure 3: An example of flat one-to-manv structure

 

 

tm

 

 

 

 

 

 

 

Pfior

 

 

 

 

”(9)

W

 

 

Pnor

 

 

 

89

 

Field agent ai

 

‘—

 

 

 

i

ri = f(gi(A). Ii?» .

 

tm A
Top manager tm

 

feedback

 

 

 

 

 

feedb k /
ac E

 

vent A

Figure 4: Information processing in flat one-to-manv structure

rtm = f(gtm(a), Hm?»

 

 

Ei

 

 

 

 

 

£2

 

 

 

 

E3

 

 

 

 

E4

 

 

 

9O

 

 

 

 

dm1

 

 

 

 

 

 

tm

 

 

 

 

 

 

dm2

 

 

 

 

Figure 5: An example of tall one-to-one structure

 

 

 

 

 

 

 

 

 

\fiki.

 

 

 

 

 

Ei
gi(A
Pfior
—> Prior
Pnor

 

 

 

 

 

 

91

 

Field agent ai

 

‘——~

 

 

ri = f(gi(A), ”(E/7)

 

 

 

 

 

 

 

 

 

 

 

 

feedback
dmA . . . .
DIVISIOI‘I manager dm1 +__.
drj-_-f(gdm(A), ldm(DEj))
tm A feedback
Top manager tm
rtm = f(gtm(A), Itm(E))
feedback

 

Event A

figure 6: Information processing in tall one-to-one structure

 

 

 

92

 

 

 

dm1

 

 

 

 

 

 

 

 

 

 

 

 

 

dm2

 

 

 

 

 

 

 

 

 

Figure 7: An example of tall one-to-manv structure

 

 

 

 

 

93

 

 

 

 

 

Field agent 31' ‘—

/gi(A/'
Prior ’1 = ”91.04): ”(9)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

feedback
+ prior dm A Division manager dmj ‘_.
drj -_— f(gdm(A), /dm(e)
tm A
Prior Top manager tm feedback
I rtm = (gtm(A). Itme)
feedback

 

 

 

 

Event A

figure 8: Information processing in tall one-to-mjmv Structure

 

 

 

 

 

5. An Example

The findings of this research will be illustrated through a numerical example. This
example will demonstrate that different organizational structures transmit different
information to the top manager, and that different organizational structures generate
different information costs and event predictabilities. Let us consider a top manager, rm,
who has six subordinates, (1,, i = 1,...,6. The responsibility of subordinates is to analyze
information, and report the probability of event A in their jurisdictions. Event A is a
binary variable that it will occur (A = 1) or not (A = O). The probability of occurrence
of the event is denoted as 9, while that of non-occurrence as (1 - 6). Depending upon the
reports from the subordinates, the top manager periodically makes predictions of the
probability of the occurrence of event A. The top manager wants to know how each
organizational structure will affect the reports given to him, and which will be most
appropriate in information processing.

Four models will be discussed, each with seven members--the top manager and
six subordinates organized in four different ways. The first one is a flat one—to-one
structure in which tm receives 6 reports from the field agents, and each a,- has his own
jurisdiction. The second model is a flat one-to—many structure in which rm receives 6
reports from the field agents who share a common jurisdiction. The third model is a tall
one-to-one structure in which 1m receives 2 reports from division managers, dm, and dm2.
Each division consists of a division manager and 2 fieldagents. Each field agent has his

own jurisdiction. Finally, there is a tall one-to-many structure in which zm receives 2

94

 

95

 

reports from division managers. All field agents have a common jurisdiction.
Information processing in this example will follow the procedures developed in
Chapters 3 and 4. All individuals in the organization are Bayesian players. Each
subordinate is independent from the others, and each has his own method of analyzing
the data. At each time, there is an information set which consists of 36 data each of
which may indicate whether the event will occur or not. Each field agent can evaluate
only 6 pieces of data. All have the beta probability distribution function, B(a, b). The
expected value of the probability of occurrence of the event, E(O), in the beta distribution
B(a, b), is a/(a+b). Observing evidence of the occurrence of event A is treated as adding
1 to a in B(a, b), and observing evidence of the non-occurrence of event A as adding 1
to b in B(a, b). Assume also that all individuals have a common prior probability, [3( 1,1)
that is, E(9) = 1/2, on event A, and that the managers will give an equal weight to each
report at time t. Now at time t the information set is given to the organization. The
following table of information set shows that each E,, i =1,...,6, consists of 6 pieces of

data, and e also consists of 6 pieces of data each of which belongs to each E,.

Information set at t

 

 

 

 

 

96

Since each evaluates data with his own criteria, even though several agents may
observe the same evidence, their conclusions may be different. For example, when (11 and
a2 observe a piece of data "f" from the environment, al may estimate that it is evidence
of the likely non-occurrence of event A at t + 1, while a, may treat it as an evidence of
the likely occurrence of the event. The evaluation criteria of subordinates are suggested
as follows. For example, when a, observes one of the elements {a, b, c, d, e, f} from E,,
he will interpret it as an evidence of the likely non-occurrence of event A at t + 1, and
he will add 1 to b in B(a, b). On the other hand, when he observes one of the elements
{g, h, i, j} from E,., he will interpret it as an evidence of the likely occurrence of event

A at t + 1, and he will add 1 to a in B(a, b).

For a1, {a, b, c, d, e, f} = 0; {g, h, i, j} = 1.

For a2, {a, b, c, d, e, g} = 0; {f, h, i, j} =1.

For a3, {a, b, c, d} = 0; {e, f, g, h, i, j} = 1.
For a4, {a, b, c, d, g} = 0; {e, f, h, i, j} =1.
For as, {a, b, c, e, f} = 0; {d, g, h, i, j} =1.
For a6, {a, b, d, g} = 0; {e, e, f, h, i, j} = 1.

Information processing in the flat one-to-one structure at t

The information set is divided into 6 subsets, E,., i = 1,...,6. Each (11 needs to
analyze each E,., and report r,- to tm. Each time, this structure observes all 36 cases.
(1) a1 observes E1 which consists of 6 pieces of data, and estimates them with his own

criteria. E1 = {e, j, h, g, i, g}. Among 6 observations, 5 cases indicate the occurrence

 

 

 

97

of event A at t + 1; only one shows the non-occurrence. Then a1 constructs his likelihood
function as 11(El) = B(5,1). Since al’s prior probability g1(A) is [3(1, 1), (21 ’s posterior
probability on event A is p,(A) = g,(A)ll(E1) = B(5+1, 1+1) = [3(6, 2). Under the incentive
mechanism, (11 behaves sincerely and reports rl as same as pl(A). Thus (11 reports a
probability, [3(6, 2), to rm at t.

(2) a2 observes 6 pieces of from E2, and finds that 3 cases indicate the likely
occurrence of the event at t + 1, while the other 3 indicate the likely non-occurrence of
the event. Then a, formulates his posterior probability, and reports r2 = B(3+1, 3+1) =
[3(4, 4).

(3) a3 observes 6 pieces of from E3, and finds that 5 cases indicate the likely
occurrence of the event at t + 1, while the other 1 indicates the likely non-occurrence of
the event. Then a, formulates his posterior probability, and reports r3 = [3(5+1, 1+1) =
[3(6, 2).

(4) a4 observes 6 pieces of from E,, and finds that 4 cases indicate the likely
occurrence of the event at t + 1, while the other 2 indicate the likely non-occurrence of
the event. Then (14 formulates his posterior probability, and reports r4 = B(4+1, 2+1) =
[5(5, 3).

(5) a5 observes 6 pieces of from E5, and finds that 5 cases indicate the likely
occurrence of the event at t + 1, while the other 1 indicates the likely non-occurrence of
the event. Then as formulates his posterior probability, and reports r5 = [3(5+1, 1+1) =
[3(6, 2).

(6) a6 observes 6 pieces of from E6, and finds that 5 cases indicate the likely

 

 

 

98

occurrence of the event at t + 1, while the other 1 indicates the likely non—occurrence of
the event. Then (16 formulates his posterior probability, and reports r6 = [3(5+1, 1+1) =
[5(6, 2).

(7) tm receives r,-, i = 1,...,6, and aggregates them using the relative weight rule. At
time t, 1772 will use an equal weight, w, = 1/6. Then tm formulates his likelihood function
by aggregating the reports. 14",, = 2 win, 2' = 1,...,6.

,um = 1/6[B(6, 2) + [3(4, 4) + [3(6, 2) + [3(5, 3) + [3(6, 2) + [3(6, 2)] = [3(33, 15).

Since tm’s prior is also B(1, 1), rm = B(33+1, 15+1) = [3(34, 16). Thus the top manager

will predict that the probability of event A at t + 1 is 34/50 = .68.

Information processing in the flat one-to-m_any sti'uciugre at I

All subordinates need to observe a common jurisdiction, which consists of 6 cases,
e = { g, d, h, c, j, i}. The jurisdiction would be scrutinized from 6 different perspectives.
(1) al observes e, and estimates each evidence with his own criteria. Among 6 pieces
of data, 4 cases indicate the occurrence of event A at t + 1. a1 constructs his likelihood
function as l,(e) = [3(4, 2). Since a1’s prior probability is assumed to be B(1, 1), a,’s
posterior probability on event A is p1(A) = B(4+1, 2+1) = [3(5. 3)- Under the incentive
mechanism, al reports r1 = p1(A). Thus a, reports r1 = [3(5, 3) to tm at t.
(2) a2 observes the jurisdiction, and finds that 3 cases indicate the likely occurrence
of the event at t + 1, while the other 3 indicate the likely non-occurrence of the event.
Then a, formulates his posterior probability, and reports r2 = B(3+1, 3+1) = B(4, 4).

(3) a3 observes the jurisdiction, and finds that 4 cases indicate the likely occurrence

 

 

 

 

 

99

of the event at t + 1, while the other 2 indicate the likely non-occurrence of the event.
Then a3 formulates his posterior probability, and reports r3 = B(4+1, 2+1) = [3(5, 3).

(4) a4 observes the jurisdiction, and finds that 3 cases indicate the likely occurrence
of the event at t + 1, while the other 3 indicate the likely non-occurrence of the event.
Then 04 formulates his posterior probability, and reports r4 = [3(3+1, 3+1) = [5(4, 4).

(5) a5 observes the jurisdiction, and finds that 5 cases indicate the likely occurrence
of the event at t + 1, while the other 1 indicates the likely non-occurrence of the event.
Then as formulates his posterior probability, and reports r5 = B(5+1, 1+1) = [3(6, 2).

(6) a6 observes the jurisdiction, and finds that 3 cases indicate the likely occurrence
of the event at t + 1, while the other 3 indicate the likely non-occurrence of the event.
Then a.s formulates his posterior probability, and reports r6 = B(3+l, 3+1) = [3(4, 4).

(7) rm receives 6 reports, and aggregates them into his likelihood function with using
an equal weight 1/6:

,u", = 1/6[B(5, 3) + [3(4, 4) + B(5, 3) + [3(4, 4) + [3(6, 2) + [3(4, 4)] = [3(28, 20).

With his prior probability [3(1, 1), he formulates r,,,, = B(28+1, 20+1) = [3(29, 21). Thus

the top manager predicts that the probability of event A at t + 1 is .58.

Information processing in the tall one-to-one structure at I
There are 2 divisions each of which has 2 field agents. dml, acted by (13, receives

the reports from a1 and a2, and dm2, performed by a6, receives 2 reports from (14 and a5.

 

 

 

 

 

 

100

This structure observes 24 cases from E1, E,, E,, and Es.‘

(1) a1 observes E1, and estimates each case with using his own method. Among 6
observations, 5 cases support the occurrence of event A. (11 constructs his likelihood
function as l,(El) = [3(5, 1). Since al’s prior probability on the event is [3(1, 1), al’s
posterior probability on event A is p1(A) = B(5+1, 1+1) = [3(6, 2). (11 reports r1 = [3(6, 2)
to dm1.

(2) a2 observes 6 pieces of from E2, and finds that 3 cases indicate the likely
occurrence of the event at t + 1, while the other 3 indicate the likely non-occurrence of
the event. Then a, formulates his posterior probability, and reports r2 = [3(3+1, 3+1) =
[3(4, 4).

(3) a4 observes 6 pieces of from E4, and finds that 4 cases indicate the likely
occurrence of the event at t + 1, while the other 2 indicate the likely non-occurrence of
the event. Then (14 formulates his posterior probability, and reports r4 = [3(4+1, 2+1) =
[3(5, 3).

(4) a5 observes 6 pieces of from E5, and finds that 5 cases indicate the likely
occurrence of the event at t + 1, while the other 1 indicates the likely non—occurrence of
the event. Then as formulates his posterior probability, and reports r5 = B(5+1, 1+1) =
[3(6, 2).

(5) dm1 receives r1 and r2, and aggregates them using the relative weight rule. At t,

 

 

The top manager can also consider the different combinations of data That is, he can assign to a1 4 pieces of
data from E1 and 2 from E2, and to a2 2 pieces of data from E2 and 4 from E3, and so on. Then it can cover
all E,. with observing 4 pieces of data from each E,- (4 x 6 = 24). But it is nothing but the different assignment
of jurisdictions, which also generates different reports to the top manager.

 

 

 

 

 

101

dm1 will give an equal weight 1/2 to each report:

#4,.“ = 1/2[B(6, 2) + [3(4. 4)] = B(IO, 6).
With his prior probability function [30, 1), dm1 constructs his posterior probability, pm (A)
= B(10+1, 6+1) = [3(11, 7). Under the incentive mechanism dm1 behaves sincerely, and
thus he reports rd,"l = [301, 7) to tm.
(6) de receives r4 and rs, and formulates his likelihood function using the relative
weight rule:

Hana = 1/2[I3(5, 3) + [5(6, 2)] = [3(11. 5).
With his prior probability, dm2 reports rd,"2 = B(11+1, 5+1) = [3(12, 6) to 1m.
(7) Im receives rd”,1 and rdmz, and aggregates them into his likelihood function using
an equal weight 1/2:

[1,," = 1/2[B(11, 7) + [3(12, 6)] = [3(23, 13).
With his prior probability [3(1, 1), he will formulate r,,,, = B(23+1, 13+1) = [3(24, 14). The

top manager will predict that the probability of the occurrence of the event A is .632.

Information processing in the tall one-to-mjanv structure gt_t

There are 2 divisions. dm1, acted by a,, receives rl and r2, and dm2, performed by
(16, receives the reports from a4 and as. The division managers transmit the reports to 1722.
All field agentsual, a2, 0,, and a4--have the common jurisdiction e.
( 1) al observes e, and estimates each case with using his own criterion. Among 6
observations, 4 cases indicate the occurrence of event A at t + 1. al constructs his

likelihood function as l,(e) = B(4, 2). (11’s posterior probability on event A is p,(A) =

 

 

102

B(4+1, 2+1) = [3(5, 3). Under the incentive mechanism, (11 reports r1 = p,(A). Thus (11
reports r1 = [3(5, 3) to dm1.
(2) a2 observes the jurisdiction, and finds that 3 cases indicate the likely occurrence
of the event at t + 1, while the other 3 indicate the likely non-occurrence of the event.
Then a, formulates his posterior probability, and reports r2 = B(3+l, 3+1) = [3(4, 4).
(3) a4 observes the jurisdiction, and finds that 3 cases indicate the likely occurrence
of the event at t + 1, while the other 3 indicate the likely non-occurrence of the event.
Then (14 formulates his posterior probability, and reports r4 = B(3+1, 3+1) = [3(4, 4).
(4) a5 observes the jurisdiction, and finds that 5 cases indicate the likely occurrence
of the event at t + 1, while the other 1 indicates the likely non-occurrence of the event.
Then (15 formulates his posterior probability, and reports r5 = [3(5+1, 1+1) = [3(6, 2).
(5) dm1 receives r1 and r2. Using the aggregation rule, he will formulate his likelihood
function: A

at... = 1/2[[3(5, 3) + p(4, 4)] = [3(9, 7).
Using his prior probability and likelihood function, dm1 constructs his posterior
probability as pdml(A) = B(9+1, 7+1) = [3(10, 8). Under the incentive mechanism, he will
report rd”,1 = B(lO, 8).
(6) dm2 has r4 and rs, and aggregates them with an equal weight 1/2:

um = 1/2[t3(4, 4) + [3(6, 2)] = [3(10, 6).
Thus dm2 will report rdmz = [3(10+l, 6+1) = [3(11, 7) to rm.
(7) Im receives 2 reports, rd,"1 and rdmz, and formulates his likelihood function with an

equal weight:

 

103
Hm. = 1/2[l3(10, 8) + [3(11. 7)] = [3(21, 15)-

tm’s posterior probability will be rm = [3(21+1, 15+1) = [3(22, 16). The top manager

predicts that the probability of event A at t + 1 is .579.

Weight updating at t + 1

At t + 1, assume that event A has occurred, that is, A = 1. Then the managers
need to update the relative weights. The relative weight at t + 1 will be calculated by the
report at t and the weight at t. The updating rule when A = l is:

W-1 + I = w-‘r-‘

zwitri‘

 

(1) In the flat one-to-one structure, the top manager needs to update 6 relative weights
based on the previous equal weight and the reports at t since he receives 6 reports at each
timC- Xwir, = 33/48.

For (11, W1 at t + 1 will be 6/33. This updated weight is greater than 1/6.

For (2,, w2 at t + 1 will be 4/33. This updated weight is less than 1/6.

For 03, W3 at t + 1 will be 6/33. This updated weight is greater than 1/6.

For (14, W4 at t + 1 will be 5/33. This updated weight is less than 1/6.

For as, w5 at t + 1 will be 6/33. This updated weight is greater than 1/6.

For a6, w6 at t + 1 will be 6/33. This updated weight is greater than 1/6.

The sum of the updated relative weights is also 1. When the top manager
aggregates the reports at t + 1, he would use the updated weights.
(2) In the flat one-to-many structure, the top manager needs to update 6 relative

weights based on the previous equal weight and the reports at t since he receives 6 reports

 

 

 

104

at each time. Zwir, = 28/48.

For a1, wl at t + 1 will be 5/28. This updated weight is greater than 1/6.

For a2, W2 at t + 1 will be 4/28. This updated weight is less than 1/6.

For as, W3 at t + 1 will be 5/28. This updated weight is greater than 1/6.

For (2,, W4 at t + 1 will be 4/28. This updated weight is less than 1/6.

For as, ws at t + 1 will be 6/28. This updated weight is greater than 1/6.

For a6, W6 at t + 1 will be 4/28. This updated weight is less than 1/6.

The sum of the weights is 1. The tOp manager will use these updated weights to
aggregate the probabilities at t + 1.
(3) In the tall one-to—one structure, both the top manager and the division managers
need to update the relative weights. The top manager will update the relative weights of
the division managers, and each division manager will revise the weights of the field
agents. Zwm-‘rm‘ = 23/36. The updating rule for rd," when A = 1 is:

t + 1 t 1
Wm = Wan; ram;

demi'rdnu't
For dm1, wdml at t + 1 will be 11/23. This updated weight is less than 1/2.
For dmg, wdm2 at t + 1 will be 12/23. This updated weight is greater than 1/2.
In Division 1, Zwi‘r,‘ = 10/16, 1' = l, 2. In Division 2, Zwi‘ri’ = 11/16, i = 4, 5. The
updating rule for r, when A = 1 is:

W-‘ + l : w.‘r-‘

Zwttrt!

 

For all, w1 at t + 1 will be 6/10. This updated weight is greater than 1/2.

 

 

 

 

105

For a2, w2 at t + 1 will be 4/ 10. This updated weight is less than 1/2.

For a,,, w4 at t + 1 will be 5/11. This updated weight is less than 1/2.

For as, w5 at t + 1 will be 6/11. This updated weight is greater than 1/2.

The sum of the relative weights of division managers is 1. In each division, the
sum of the relative weights of field agents is also 1. When aggregating the reports at t
+ 1, each manager will assign the revised relative weight to each report.
(4) In the tall one-to-many structure, both the top manager and the division managers
need to update the relative weights as in the tall one-to—one structure. The top manager
will update the relative weights of the division managers, and each division manager will
revise the weights of the field agents in his division. Zwm‘rm‘ = 21/36. The updating

rule for rd”, when A = l is:

Zwm‘rdmt
For dm1, det at t + 1 will be 10/21. This updated weight is less than 1/2.
For dm2, WM at r + 1 will be 11/21. This updated weight is greater than 1/2.
In Division 1, Xwi‘ri’ = 9/16, i = l, 2. In Division 2, Zwi‘ri‘ = 10/16, i = 4, 5. The
updating rule for r,- when A = 1 is:

: w.‘ r .‘
Zwitrit

14).t+l

l

 

For all, w1 at t + 1 will be 5/9. This updated weight is greater than 1/2.
For a2, w2 at t + 1 will be 4/9. This updated weight is less than 1/2.
For a4, w4 at t + 1 will be 4/10. This updated weight is less than 1/2.

For as, w5 at t + 1 will be 6/10. This updated weight is greater than 1/2.

 

 

 

 

106

The sum of the relative weights of division managers is 1. In each division, the
sum of the relative weights of field agents is also 1. When aggregating the reports at t
+ 1, each manager will assign the revised relative weight to each report. Each time, the
managers need to revise the relative weights so that they can ensure honest reporting.

At time t + 1, a new information set will be given. The field agents in these four
models will also observe the information set, and process the probabilities on the
occurrence of event A at t + 2. Then the managers will aggregate the reports with using
the revised relative weights, w,‘+ 1. After observing the state of event A at t + 2, the
managers will update the relative weights again. Through these procedures, prior beliefs
and evaluation criteria will also be adjusted.

This example has demonstrated how different organizational structures affect the
processing and aggregation of information in the organization. First, it shows that each
different organizational structure transmits different reports to the top manager. Even
though the subordinates in these four structures observe the same information set, and all
individuals begin with the same prior probability, the reports given to the top managers
are different, as are the predictions by the top managers. For example, the aggregated
reports in the flat one-to-one structure is [3(33, 15), while that in the tall one-to-one
structure is [3(23, 13). Also when the top manager in flat one-to-one structure predicts
the probability of the occurrence of event A at t + 1 as .68, the top manager in the tall
one-to-many structure calculates that probability as .579.

Second, each organizational structure observes a different number of cases. For

instance, the flat one-to-one structure examines 36 cases, while the flat one-to-many

 

 

 

 

107

structure generates only 6 observations.

Third, different organizational structures need to pay different information cost.
The top manager’s information cost (fem) of flat structures is 26 but that of tall structures
is 32. The information cost (IC) of flat structures is 100 (= 36 + 26), whereas that of tall
structures is only 41 (= 24 + 2x22 + 32).

Finally, at t + 1, each organizational structure generates a different event
predictability. Event predictability is estimated as the function of both information
efficiency and information reliability. It is expected to be proportional to NAN, when N
is the number of observations, and AN is the number of major agreements on the
probability. The interval of NAN can be calculated. That of the flat one-to-one structure
is [36, 366] since this structure examines all 36 pieces of data, and the possible maximum
number of agreements is 6. That of the tall one-to—many structure is [6, 64] since this
structure examines only 6 pieces of data, and the possible maximum number of
agreements among the field agents is 4. From the reports at t, we can calculate the point
estimate of NAN in each structure. That of the flat one-to-one structure is 364 because 4
among 6 field agents agree on the probability of event A at t + 1, while that of tall one-
to-many structure is 62 because 2 among 4 field agents agree on the probability of the
event at t + 1. When assuming that event A has occurred at t + 1, event predictability can
be calculated by the deviation between the prediction and the actual state of the event,
that is, EP = 1 - |A - rml, A = {0, l}. The event predictability of the flat one-to-one
structure at this time is .68, while that of the flat one-to-many structure is .58. But EP

will vary at each time when observing the actual state of the event. Thus when designing

 

 

 

108

an organizational strucrure, the top manager needs to consider the interval of N"N as an

indicator of event predictability. The following table summarizes the results of this

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

example.
Flat one-to-one Flat one-to- Tall one-to-one Tall one-to—many
many

observes: 36 6 24 6 data
tm receives: 6 6 2 2 reports

ic,, [3(33, 15) [3(28, 20) [3(23, l3) B(21, 15)

run .68 .58 .632 .579

ic,, 64 64 9 9

IC 100 100 41 41
Interval of NAN [36, 36‘] [6, 6‘] [24, 24‘] [6, 6‘]
Point estimate 364 63 242 62
of NAN at t
EP, 1+1, A=1 .68 .58 .632 .579

 

 

 

 

 

 

6. Organizational Structure and Environment

Theorem 2 in Chapter 4 concludes that the tall one-to-one structure is in general
better than the other structural models, when assuming all factors except structure to be
constant. This result does not consider the characteristics of the environment even though
the organization gathers information from the environment. Now let us extend the
discussions of organizational structures to the relation between the environment and
organizational structure. The assumption is modified; all other factors except structure
and environment are constant. This study views an organization as open, information
processing system. Organizational information processing is affected by, and dependent
upon, its environment. The purpose of this chapter is to relate the structural models
deve10ped in Chapter 4 with the characteristics of the environment.

Most scholars have used two variables and a 2 x 2 matrix to classify the
environment. Thompson (1967) classified the environment as homogeneous-stable,
homogeneous-shifting, heterogeneous-stable, and heterogeneous-shifting. Lawrence and
Lorsch (1967) categorized it into low diversity and not dynamic, low diversity and highly
dynamic, high diversity and not dynamic, and high diversity and highly dynamic. Duncan
(1972) used simple-static, simple-dynamic, complex~static, and complex-dynamic. Let
us keep, for purposes of argument, the view of this study that assumes an organization
is an information processing system and the organizational environment is the source of
information perceived by the organization. The field agents would gather and analyze

information, and transmit the reports to the managers. Then finally the top manager

109

 

 

 

 

 

 

110

aggregates the reports and makes organizational announcements. When the tOp manager
designs an organizational structure, he will also consider the characteristics of the
environment as well as those of organizational structures. He will emphasize different
principles of information processing when facing different characteristics of the
environment.

Following Thompson (1967), we can characterize the environment with two
dimensions dealing with degree of information homogeneity and degree of information
stability. First, the environment is characterized as (relatively) homogeneous or
heterogeneous, indicating whether the information generated from the environment is
similar to one another. Second, the environment is classified as stable or shifting,
indicating whether the range of information variations is large. By using these two
dimensions, the environment would be classified as homogeneous-stable, homogeneous-
shifting, heterogeneous-stable, and heterogeneous-shifting.

With the homogeneous and stable environment, the organization can gather a
certain amount of similar information with small variations. Since information is
homogeneous, information reliability by achieving consensus in estimating information
is considered more important than information efficiency by increasing the number of
observations. For increasing information reliability, achieving consensus in estimating
information is more important than increasing the number of observations. It is
apprOpriate to assign a common jurisdiction to field agents since it seeks to benefit from
the diverse perspectives of field agents in predicting the event. Since information is

stable, the top manager himself does not need to review all reports, and then it is better

 

 

 

 

 

111

for him to reduce his information costs by introducing middle-level managers. With this
kind of environment, the top manager will give emphasis to increasing information
reliability and reducing information cost. Thus the appropriate structure with this
homogeneous and stable environment is a tall one-to-many structure that emphasizes
information reliability and smaller information cost.

With the homogeneous and shifting environment, the organization can get a certain
amount of similar information with large variations. Since information is homogeneous,
the agreement on predicting the event is more important than the amount of information.
Assigning a common jurisdiction is better for getting consensus from diverse viewpoints
of field agents. Since the range of information variation is so large, the top manager
wants to analyze all reports with no filtration, and get a more direct role in aggregating
reports. Also for increasing event predictability, he will assign all subordinates as the
field agents to observe the environment. With this kind of environment, he will give
emphasis to increasing information reliability but have less concern for information costs.
The appropriate structure with this homogeneous and shifting environment is a flat one-to-
many structure that focuses on information reliability and the top manager’s direct role
in making the predictions.

With the heterogeneous and stable environment, the organization can gather a
certain amount of different information with small variatiOns. Since information is
heterogeneous, increasing information efficiency by increasing the number of observations
is considered more important than information reliability. For increasing information

efficiency, the organization needs to assign separate jurisdictions, each corresponding to

 

 

 

112

a relatively homogeneous segment of the environment. Thus information gathering can
be more efficiently performed with specialization. Since information is stable, the top
manager does not need to aggregate all reports with paying a great information cost. The
top manager can reduce his information costs by introducing middle-level managers
whose task is to transmit a condensed report. With this kind of environment, the top
manager will want to increase information efficiency and reduce information cost. The
appropriate structure with this heterogeneous and stable environment is a tall one-to—one
structure that emphasizes information efficiency and less information cost.

With the heterogeneous and shifting environment, the organization can get a
certain amount of different information with large variations. Since information is
heterogeneous, the organization needs to decompose the environment into a certain
number of separate jurisdictions, for analyzing more information with specialized
perspectives. Since the range of information variation is large, the top manager wants to
get access to the reports from the field agents without any filtration. For increasing event
predictability, he will assign all subordinates as the field agents to observe the
environment. With this kind of environment, he wants to increase information efficiency
but less concerns on information cost. The appropriate structure with this heterogeneous
and shifting environment is a flat one-to-one structure that emphasizes information
efficiency and the top manager’s deterministic role in the event prediction.

Different principles of information processing will be emphasized when facing
different characteristics of the environment. With the stable environment, the tall

structure is better than the flat structure since it can reduce information costs. With the

 

 

 

113

shifting environment, the flat structure is better than the tall structure since it can increase
event predictability. With the homogeneous environment, the one-to—many structure is
better than the one-to-one structure since it can increase information reliability. With the
heterogeneous environment, the one-to-one structure is better than the one-to-many
structure since it can increase information efficiency.

These results enrich the discussions of organizational structure. When the
environment is constant, the tall one-to-one structure is better than the others. When
considering other characteristics of the environment, different structures will work better

in different environments. The results are summarized in Table 4.

Table 4: Optimal Organizational Structure in Different Environments

 

Stable Shifting

Homogeneous Tall one-to-many structure Flat one-to-many structure

 

 

 

 

 

Heterogeneous Tall one-to-one structure Flat one-to-one structure

 

 

 

 

 

 

 

 

 

 

 

 

7. Conclusion

In this research I addressed three questions about organizational structure and
information processing. The first research question was "How does organizational
structure affect the information transmitted through it?" To answer it, I developed four
kinds of organizational structures focusing on specialization and coordination: flat one-to-
one, flat one-to-many, tall one-to-one, and tall one-to-many structures. Chapter 4 shows
that different organizational structures generate different information for the top manager
and that each organizational structure has its own unique advantages and disadvantages
in information processing. The discussions are summarized in Proposition 1 and Table
3. Chapter 5 illustrates the findings of Chapter 4 through a numerical example.

The second question was "What kind of organizational structure will be most
appropriate in information processing?" With regard to information cost and event
predictability, four kinds of organizational structures were compared. Propositions 2 and
2.1 demonstrate that the information cost of tall structures is less than that of flat
structures for the top manager as well as the whole organization. Proposition 3 indicates
that the flat one-to-one structure generates greater information predictability than the other
structural alternatives, given the existence of the relative weight rule. Based on
Propositions 2, 2.1, and 3, Theorem 2 demonstrates that when considering both
information cost and information predictability, the tall one-to-one structure is better than
the other organizational structures, with n > 4. When considering the relation between

organizational structure and environment, Chapter 6 shows that different principles of

 

 

 

 

 

115

information processing will be emphasized when facing different characteristics of the
environment, and different structures will work better in different environments.

The third question was "What kind of incentive mechanism can be used by the top
manager to ensure that subordinates report information honestly in a particular kind of
organizational structure?" Chapter 2 discusses information processing and aggregation,
and develops the relative weight rule as a Bayesian incentive compatible mechanism.
Theorem 1 demonstrates that under the assumptions that all individuals are Bayesian
players, and that there exist feedback processes, the relative weight rule is a Bayesian
incentive compatible mechanism. In the information processing game, the field agent will
report honestly and the manager will use the relative weight rule appropriately under a
Bayesian incentive compatible mechanism. With regard to this result, the issues of
collusion, and risk preference will be discussed in the following sections.

Organizational politics may play an important role in organizational information
processing. Within the information processing framework of this study, organizational
politics means the management of relative influence on the organizational predictions of
event A. The coalition among the subordinates is one of the most important players in
the political arena (March and Simon 1958, Cyert and March 1963). Stevenson et al.
(1985 ) define a coalition as "an interacting group of individuals, deliberately constructed,
independent of the formal structure, lacking its own internal formal structure, consisting
of mutually perceived membership, issue oriented, focused on a goal or goals external to
the coalition, and requiring concerted member action" (p.261). Following this definition,

a coalition in this study refers to a group of subordinates that work together to influence

 

 

 

 

 

116

the organizational predictions by reporting an agreed probability.

Under the existence of the relative weight rule as a Bayesian incentive compatible
mechanism, a coalition among the subordinates may not be developed for two reasons.
First, a coalition cannot escape from the fundamental problem of team theory--free-riding
(Alchian and Demsetz 1972, Marschak and Radner 1972). Under the coalition, each
coalition member still needs to observe the jurisdiction, and to make the posterior
probability. Then the coalition will reach an agreed probability through inner-bargaining
processes. In this situation no one wants to pay his own observation costs for making the
probability. Every member wants to just follow the other’s report with no effort to
evaluate the data. Thus every coalition member will become a free-rider. Second, the
coalition cannot ensure the maximum relative weight but an average weight. Since the
same report will be transmitted to the manager, each member can get only the average
weight equally. From the standpoint of maximizing the relative weight, it is not a good
strategy to join a coalition for sharing only an average weight. Thus when applying the
relative weight in information aggregation, the manager might not be manipulated by a
coalition among the subordinates.

The issue of risk preference may be important for understanding the individual
behavior in information processing. Now let us analyze whether'the degree of risk
aversion may be relevant to the information processing game. Theorem 1 shows that the
relative weight rule is a Bayesian incentive compatible mechanism. Under this incentive
mechanism, the dominant strategy of the field agents is to report honestly. Individual risk

preference is not considered in this game. But the degree of risk aversion may be

 

 

 

 

 

117

 

irrelevant since the Bayesian incentive compatible mechanism requires all field agents to
behave sincerely regardless of risk preference.

When all field agents are the relative weight maximizers, the risk-avoider (0,) may
care more about keeping his current weight, while the risk-taker ((1,) may care more about
increasing his weight in the future. The great merit of the relative weight rule is that the
relative weight of a field agent a, cannot be manipulated by himself but decided by both
r, and Er], j ¢ i. For the risk-avoider, keeping his current weight is dependent upon his
report and the others’ reports. Since the state of event A is under uncertainty, (11 may
report strategically r1 = .5 when p,(A) = .6 or p,(A) = .4. When (11 reports r1 = .5 but
p1(A) = .6, and the event occurs, if the others report rj > .5, w1 will be dramatically
decreased. When (11 reports r1 = .5 but p1(A) = .4, and the event does not occur, if r]. <
.5, w1 will also be reduced. Even though he behaves strategically, he cannot be sure
whether he can keep his current weight unless he knows all other reports. The dominant
strategy of the risk-avoider is to report sincerely.

For the risk-taker, increasing the weight is detemrined by the others’ reports as
well as his own. Since the state of event A is uncertain, (1.2 may report strategically; (12
will report r2 = 1 when p2(A) = .6 or r2 = 0 when p2(A) = .4. When (12 reports r2 = 1, and
the event occurs, if the others also report r,- = 1, w2 will be unchanged. When he reports
r2 = 0, and the event occurs, if rj > 0, W2 will be decreased. The strategic reporting
cannot ensure anything about the risk-taker’s relative weight unless he knows all other
reports. The dominant strategy of risk-taker is also to report sincerely. Since strategic

reporting cannot give any benefit to the field agents, they behave sincerely regardless of

 

 

 

118

risk preference.

The structure of an organization is not an immutable given, but rather a set of
principles over which the top manager exercises considerable choice. This study has
analyzed four simple models of organizational structures. The impact of organizational
structure is critical in information processing: it determines what kind of information the
organization can get; it decides what kind of aggregation procedures the organization
should follow; it affects how much cost the organization as well as the top manager need
to pay for evaluating information; it determines how efficient and reliable the
organizational information processing will be; and finally, different organizational
structures may result in different organizational predictions. The results of this research
give directions on how to understand the [effects of organizational structures on
information processing and how to design organizational structures with regard to
information cost and event predictability.

Several different directions for further research can be drawn from the implications
of this analysis. First, this study has assumed that the field agent’s utility is the function
of his relative influence on information aggregation. If the field agent’s utility is a
function of several variables, not one, such as income and leisure as in the principal-agent
framework, is it possible to design a Bayesian incentive compatible mechanism? How
can the relative weight rule be applied to ensure honest reporting? When the manager
considers both the relative weights of the reports and the relative importance of their
jurisdictions, how can the reports be aggregated? If the relative weight rule is still useful,

how can it be modified to include these two factors?

 

 

 

 

 

119

 

Second, since this study has assumed that there exists a direct feedback process
and that information processing is continuous through the time dimension, we can expect
that both the field agents and the top manager do not want to deviate from the dominant
strategies, and that honest reporting is guaranteed. But if there is no direct feedback or
if the event does not recur, how can the organization get honest reports from the field
agents? In multi-time information processing game, we need to consider the other
factors such as discount factors and risk preferences. When including these variables into
the game, the utility functions and reporting strategies of the field agents need to be more
specified. Also when the managers who want information to support their own bias, how
does this fact influence to information processing and aggregation? If there does not
exist an incentive mechanism, what kind of organizational structure is more appropriate
with regard to information cost and event predictability?

Finally, this study considers only the problem of aggregating probabilities in
organizational structures. Hammond (1986, 1991a, 1991b, Hammond and Thomas 1989)
argued that different organizational structures generate different policy outcomes. If
information is not probability but policy preferences, how can the top manager apply the
relative weight rule as the aggregation rule? Also how can compare organizational
structures with regard to information cost and information predictability?

Organizational structures affect the processing and aggregation of information
within organizations. Different organizational structures transmit different reports to the
top manager. To design an organizational structure is to determine a set of information

aggregation rules, and to give emphasis to different principles of information processing.

 

 

 

 

120

 

Each organizational structure has its own unique characteristics with its own information
costs and event predictability.

This research concentrates only on the processing and aggregation of probabilities
within organizational structures. For better understanding the workings and impact of
organizational structures, the other kinds of information processing and organizational

functions via organizational structures need to be studied.

 

 

 

 

 

BIBLIOGRAPHY

Aguilar, F. J. 1967. Scanning the Business Environment. New York: Macmillan.

Alchian, Armen A. 1979. Organizations and Environments. Englewood Cliffs: Prentice-
Hall.

, and Harold Demsetz. 1972. "Production, Information Costs, and Economic
Organization." American Economic Review 62: 777—795.

Aldrich, Howard E., and Jeffrey Pfeffer. 1976. "Environments of Organizations." Annual
Review of Sociology 2: 79-105.

Allais, Maurice. 1987. "Allais’s Paradox." The New Palgrave: A Dictionary of Economics,
1. New York: MacMillan: 80-82.

. 1979. "The So-Called Allais Paradox and Rational Decision under Uncertainty."
in M. Allais & O. Hagen, eds. Expected Utility and the Allais Paradox.
Dordrecht: Reidel: 437-663.

Allison, Graham T. 1971. The Essence of Decision. Boston: Little, Brown.

Aoki, Masahiko. 1984. The Co-operative Game Theory of the F irm. Oxford: Oxford Univ.
Press.

Argote, Linda. 1982. "Input Uncertainty and Organizational Coordination in Hospital
Emergency Units." Administrative Science Quarterly 27: 420-434.

Arrow, Kenneth J. 1985. "Information Structure of the Firm." American Economic Review
75: 303-307.

. 1974. The Limits of Organization. New York: Norton.

 

Astley, W. Graham, and Andrew H. Van de Ven. 1983. "Central Perspectives and Debates
in Organization Theory." Administrative Science Quarterly 28: 245-273.

Baligh, Helmy H., and Richard M. Burton. 1980. "Matching the Organization’s Structure
and Its Cooperative Market Relations." Theory and Decision 12: 311-324.

Barnard, Chester 1. 1938. The Functions of the Executive. Cambridge: Harvard Univ.
Press.

121

 

122

Barney, Jay B. 1990. "The Debate Between Traditional Management Theory and
Organizational Economics: Substantive Differences or Intergroup Conflict?"
Academy of Management Review 15: 382-393.

_, & W.G. Ouchi eds. 1986. Organizational Economics: Toward a New Paradigm
for Understanding and Studying Organizations. San Francisco: Jossey-Bass.

Bart, Christopher K. 1986. "Product Strategy and Formal Structure." Strategic
Management Journal 7: 293-312.

Bayarri, M.J. & M.H. DeGroot. 1988. "Gaining Weight: A Bayesian Approach." in J.M.
Bernardo, M.H. DeGroot, D.V. Lindley & A.F.M. Smith, eds. Bayesian Statistics
3. New York: Oxford Univ. Press: 25-44.

Becker, Thomas E., and Richard J. Klimoski. 1989. "A Field Study of the Relationship
between the Organizational Feedback Environment and Performance." Personnel
Psychology 42: 343-358.

Bendor, Jonathan B. 1985 . Parallel Systems: Redundancy in Government. Berkeley: Univ.
of California Press.

 

, and Thomas H. Hammond. 1992. "Rethinking Allison’s Models." American
Political Science Review 86: 301-322.

Bernard, Georges. 1984. "On Utility Functions. The Present State." Theory and Decision
17: 97-100.

Bemheim, B. Douglas. 1984. "Rationalizable Strategic Behavior." Econometrica 52: 1007-
1028.

Bettenhausen, Kenneth, and J. Keith Muminghan. 1985. "The Emergence of Norms in
Competitive Decision-Making Groups." Administrative Science Quarterly 30: 350—
372.

Bianco, William T. and Robert H. Bates. 1990. "Cooperation by Design: Leadership,
Structure, and Collective Dilemmas." American Political Science Review 84: 133-
147.

Blau, Peter M. 1974. On the Nature of Organizations. New York: John Wiley & Sons.

Bobbitt, HR, and JD. Ford. 1980. "Decision Maker Choice as a Determinant of
Organization Structure." Academy of Management Review 5: 13-24. '

Boschken, Herman L. 1990. "Strategy and Structure: Reconceiving the Relationship.“

 

 

123
Journal of Management 16: 135-150.

Bourgeois, L.J.,III, and W.Graham Astley. 1979. "A Strategic Model of Organizational
Conduct and Performance." International Studies of Management and Organization
9: 40-66.

, Daniel N. McAllister, and Terence R. Mitchell. 1978. "The Effects of Different
Organizational Environments upon Decisions about Organizational Structure. "
Academy of Management Journal 21: 508-514.

Bower, Joseph L. 1970. Managing the Resource Allocation Process: A Study of Corporate
Planning and Investment. Boston: Graduate School of Business Administration,
Harvard University.

Boyd, Brian. 1990. "Corporate Linkages and Organizational Environment:A Test of the
Resource Dependence Model." Strategic Management Journal 6: 419-430.

Brehm, John, and Scott Gates. 1991. "Donut Shops and Speed Traps: Evaluating Models
of Supervision on Police Behavior." Unpublished papers

Broome, John. 1985. "A Mistaken Argument against the Expected Utility Theory of
Rationality." Theory and Decision 18: 313-318.

, Bezalel Peleg, and Michael D. Whinston. 1987. "Coalition-Proof Nash Equilibria
1. Concepts." Journal of Economic Theory 42: 1-12.

Buck, James R. 1989. Economic Risk Decisions. Ames: Iowa State Univ. Press.

Buckland, Michael K. 1989. "Information Handling, Organizational Structure, and Power."
Journal of the American Society for Information Science 40: 329-333.

Buckley, Walter. 1967. Sociology and Modern Systems Theory. Englewood Cliffs:
Prentice-Hall.

Burgelman, Robert A. 1983. "A Model of the Interaction of Strategic Behavior, Corporate
Context, and the Concept of Strategy." Academy of Management Review 8: 61-70.

Burns, Tom, and GM. Stalker. 1961. The Management of Innovation. London: Tavistock
Publications.

Burton, Richard M., and BQrge Obel. 1984. Designing Efficient Organizations: Modelling
and Experimentation. Amsterdam: North-Holland.

Cable, John R. 1988. "Organizational Form and Economic Performance." In S. Thompson

 

 

124

and M. Wright, eds. Internal Organization, Efi‘iciency and Profit. Oxford: Philip
Allan: 12-37.

Calvert, Randhal L. 1985. "The Value of Biased Information: A Rational Choice Model
of Political Advice." Journal of Politics 47: 530-555.

Carzo, Rocco, Jr., and John N. Yunouzas. 1969. "Effects of Flat and Tall Organization
Structure." Administrative Science Quarterly 14: 178-191.

Chandler, Alfred D. 1991. "Corporate Strategy and Structure: Some Current
Considerations." Society 28(3): 35-38.

 

. 1977. The Visible Hand: The Managerial Revolution in American Business.
Cambridge: Harvard Univ. Press.

 

. 1962. Strategy and Structure. Cambridge: MIT Press.

Child, John. 1984. Organization, 2nd ed. London: Harper and Row.

 

. 1973. "Predicting and Understanding Organizational Structure." Administrative
Science Quarterly 18: 168-185.

 

 

. 1972. "Organization Structure, Environment and Performance: The Role of
Strategic Choice. " Sociology 6: 1-22.

Cho, In-Koo. 1987. "A Refinement of Sequential Equilibrium." Econometrica 55: 1367-
1389.

Clemen, Robert T. 1990. Making Hard Decision: An Introduction to Decision Analysis.
Boston: PWS-Kent Publishing.

 

. 1987. "Combinin g Overlapping Information." Management Science 33: 373-380.

 

. 1986. "Calibration and the Aggregation of Probabilities." Management Science
32: 312-314.

 

1984. Modeling Dependent Information: A Bayesian Approach. Ph.D.
dissertation, Indiana University.

a, and Robert L. Winkler. 1990. "Unanimity and Compromise Among Probability
Forecasters. " Management Science 36: 767- 779.

 

. 1985. "Limits for the Precision and Value of Information from Dependent
Sources. " Operations Research 33: 427-442.

 

 

 

125
Coase, Ronald. 1937. "The Nature of the Firm." Economica 4: 386-405.

Cooke, Roger M. 1991. Experts in Uncertainty. New York: Oxford Univ. Press.

Cremer, Jacques. 1980. "A Partial Theory of the Optimal Organization of a Bureaucracy.‘
Bell Journal of Economics 11: 683-693.

Crozier, Michael. 1964. The Bureaucratic Phenomenon. Chicago: Univ. of Chicago Press.

Cyert, Richard M. 1988. The Economic Theory of Organization and the Firm. New York:
Harvester Wheatsheaf.

, and Morris H. DeGroot. 1987. Bayesian Analysis and Uncertainty in Economic
Theory. Totowa: Rowman and Littlefield.

 

 

, and James G. March. 1963. A Behavioral Theory of the Firm. Englewood
Cliffs: Prentice-Hall.

Daft, Richard L., and Robert H. Lengel. 1986. "Organizational Information Requirements,
Media Richness and Structural Design." Management Science 32: 554-571.

 

 

d’Aspremont, Claude, & Louis-Andre Gerard-Varet. 1982. "Bayesian Incentive
Compatible Beliefs." Journal of Mathematical Economics 10: 83-103.

. 1979a. "Incentives and Incomplete Information." Journal of Public
Economics 11: 25-45.

 

. 1979b. "On Bayesian Incentive Compatible Mechanisms." In J.-J. laffont
ed. Aggregation and Revelation of Preferences. Amsterdam: N orth-Holland: 269-

288.

 

, Jacques Cremer, and Louis-Andre Gerard-Varet. 1990. "Incentives and Existence
of Pareto-Optimal Revelation Mechanism. " Journal of Economic Theory 51: 233-
254.

 

, A. Jacquemin, and J-F. Mertens. 1987. "A Measure of Aggregate Power in
Organizations." Journal of Economic Theory 43: 184-191.

 

Dastmalchian, Ali, & David A. Boag. 1990. "Environmental Dependence and
Departmental Structure: Case of the Marketing Function. " Human Relations 43:

1257-1276.

De Groot, Morris H. 1 970. Optimal Statistical Decisions. New York: McGraw-Hill.

 

126

, and Julia Mortera. 1991. "Optimal Linear Opinion Pools." Management Science
37: 546—558.

Demsetz, Harold. 1988. "The Theory of the Firm Revisited." Journal of Law, Economics,
and Organization 4: 141-161.

_. 1983. "The Structure of Ownership and Theory of the Firm." Journal of Law and
Economics 26: 375-390.

Dewar, Robert D., and Donald P. Simet. 1981. "A Level Specific Prediction of Spans of
Control Examining the Effects of Size, Technology, and Specialization." Academy
of Management Journal 24: 5-24.

Donaldson, Lex. 1990. "The Ethereal Hand:Organizational Economics and Management
Theory." Academy of Management Review 15: 369-381.

Dow, Gregory K. 1990. "The Organization as an Adaptive Network." Journal of
Economic Behavior and Organization 14: 159-185.

. 1987. "The Function of Authority in Transaction Cost Economics." Journal of
Economic Behavior and Organization 8: 13-38.

 

Downs, Anthony. 1967. Inside Bureaucracy. Boston: Little, Brown.
Drenick, RF. 1986. A Mathematical Organization Theory. New York: North-Holland.

Duncan, Robert. 1979. "What is the Right Organization Structure?" Organizational
Dynamics, Winter: 5 9-8 0.

. 1973. "Multiple Decision Making Structures in Adapting to Environmental
Uncertainty: The Impact on Organizational Effectiveness." Human Relations 26:
273-291.

 

1972. "Characteristics of Organizational Environments and Perceived
Environmental Uncertainty. " Administrative Science Quarterly 17: 313-327.

 

Egelhoff, William G. 1982. ”Strategy and Structure in Multinational Corporations: An
Information Processing Approach. " Administrative Science Quarterly 27: 435-458.

Eisenhardt, Kathleen M. 1989. "Agency Theory: An Assessment and Review. " Academy
of Management Review 1 4: 5 7— 74.

Etzioni, Amitai. 1964. Modern Organizations. Englewood Cliffs: Prentice—Hall.

 

 

 

 

127

 

Fahey, Liam. 1981. "On Strategic Management Decision Processes." Strategic
Management Journal 2: 43-60.

Fama, Eugene, and Michael C. Jensen. 1983. "Seperation of Ownership and Control."
Journal of Law and Economics 26: 301-326.

Feldman, Martha 8., and James G. March. 1981. "Information in Organizations as Signal
and Symbol." Administrative Science Quarterly 26: 171-186.

Fishbum, Peter C. 1986. "The Axioms of Subjective Probability." Statistical Science 1:
335-345.

. 1981. "Subjective Expected Utility: A Review of Normative Theories." Theory
and Decision 13: 139-199.

Fligstein, Neil, and Kenneth Dauber. 1989. "Structural Change in Corporate
Organization." Annual Review of Sociology 15: 73-96.

Fredrickson, James W. 1986. ”The Strategic Decision Process and Organizational
Structure. " Academy of Management Review 11: 280- 297.

French, Simon. 1986. "Calibration and the Expert Problem." Management Science 32:
315-321.

. 1985. "Group Consensus Probability Distributions: A Critical Survey." in J.M.
Bernardo, M.H. DeGroot, D.V. Lindley, and A.F.M. Smith, eds. Bayesian Statistics
2. New York: Elsevier Science Publishers: 183-202.

 

. 1982. "On the Axiomatisation of Subjective Probabilities." Theory and Decision
14: 19-33.

 

Friedkin, Noah E., and Michael J. Simpson. 1985. "Effects of Competition on Members’
Identification with Their Subunits. " Administrative Science Quarterly 30: 377-394.

Fudenberg, Drew, and Jean Tirole. 1991a. Game Theory. Cambridge: MIT Press.

. 1991 b. "Perfect Bayesian Equilibrium and Sequential Equilibrium."
Journal of Economic Theory 53: 23 6-260.

 

. 1989. "Noncooperative Game Theory for Industrial Organization: An
Introduction and Overview. " In R. Schmalensee & R. D. Willig, eds. Handbook
of Industrial Organization, vol. I. 261 -327.

 

Galbraith, Jay R. 1977. Organization Design. Reading: Addison-Wesley.

 

128

. 1973. Designing Complex Organization. Reading: Addison-Wesley.

 

. 1971. Matrix Organization: Design for High Technology. Cambridge: MIT Press.

. 1969. "Organization Design: An Information Processing View." Worldng paper
#425-69, MIT, Sloan School of Management.

, and Daniel A. Nathanson. 1979. "The Role of Organizational Structure and
Process in Strategy Implementation." In D.E. Schendel & C.W. Hofer, eds.
Strategic Management: A New View of Business Policy and Planning. Boston:
Little, Brown.

. 1978. Strategy Implementation: The Role of Structure and Process. St.
Paul: West Publishing.

Gardenfors, Peter, and Nils—Eric Sahlin. 1988. "Introduction: Bayesian Decision Theory -
- Foundations and Problems." In P. Gardenfors and N. Sahlin, eds. Decision,
Probability, and Utility:Selected Readings. Cambridge: Cambridge Univ. Press.

Genest, Christian, and James V. Zidek. 1986. “Combining Probability Distributions: A
Critique and an Annotated Bibliography." Statistical Science 1: 114-148.

Goodstein, Jerry, and Warren Boeker. 1991. "Turbulence at the Top: A New Perspective
on Governance Structure Changes and Strategic Choice." Academy of Management
Journal 34: 306-330.

Gortner, Harold F., Jalianne Mahler, and Jeanne Bell Nicholson. 1987. Organization
Theory: A Public Perspective. Chicago: Dorsey Press.

Graber, Doris A. 1992. Public Sector Communication. Washington, DC: Congressional
Quarterly.

Gradstein, Mark, and Shmuel Nitzan. 1988. “Participation, Decision Aggregation and
Internal Information Gathering in Organizational Decision Making." Journal of
Economic Behavior and Organization 10: 415-431.

Grandori, Anna. 1987. Perspectives on Organization Theory. Cambridge: Ballinger.
Groves, Theodore. 1985. “The Impossibility of Incentive-Compatible and Efficient Full
Cost Allocation Schemes." In HP. Young ed. Cost Allocation: Methods,

Principles, Applications. Amsterdam: North-Holland.

. 1973. “Incentives in Teams." Econometrica 41: 617-631.

 

 

129

, and John O. Ledyard. 1977. "Optimal Allocation of Public Goods:A Solution to
the ’Free Rider’ Problem." Econometrica 45: 783-809.

 

, and Martin Loeb. 1979. "Incentives and Divisionalized Firm." Management
Science 25: 221-230.

 

, Roy Radner, and Stanley Reiter, eds. 1987. Information, Incentives, and
Economic Mechanisms. Minneapolis: Univ. of Minnesota press.

 

Gulick, Luther. 1937. "Notes on the Theory of Organization." In Luther Gulick and
Lyndall Urwick eds. Papers on the Science of Administration. New York: Institute
of Public Administration.

Hage, Jerald, Michael Aiken and Cora B. Marrett. 1971. "Organization Structure and
Communications." American Sociological Review 36: 860-871.

Hall, David J., and Maurice A. Saias. 1980. "Strategy follows Structure!" Strategic
Management Journal 1: 149-163.

Hall, Richard H. 1991. Organizations:Strucrures, Processes, and Outcomes, 5th ed.
Englewood Cliffs: Prentice-Hall.

. 1982. Organizations: Structure and Process, 3rd ed. Englewood Cliffs: Prentice-
Hall.

Hammond, Thomas H. 1991a. "Toward a General Theory of Hierarchy: Books,
Bureaucrats, Basketball Tournaments, and the Administrative Structure of the
Nation-State." Unpublished paper.

. 1991b. "Strategy, Structure, and the Agenda of the Firm." Unpublished paper.

. 1990a. "In defence of Luther Gulick’s ’Notes on the theory of organization."
Public Administration 68: 143-173.

 

. 1990b. "The Agenda of the Firm: Structure and Incentive in Institutional Design."
Unpublished paper.

 

 

. 1986. "Agenda Control, Organizational Structure, and Bureaucratic Politics."
American Journal of Political Science 30: 379-420.

, and Jeffrey H. Horn. 1985. "’Putting One Over on the Boss’: the Political
Economy of Strategic Behavior in Organizations." Public Choice 45: 49-71.

 

, and Gary J. Miller. 1989. "Manipulating the Non-Manipulable: On Managing

 

 

130

Firms via Demand-Revelation Rules." Unpublished paper.

 

 

. 1985. "A Social Choice Perspective on Expertise and Authority in
Bureaucracy." American Journal of Political Science 29: 1-28.

 

, and Paul A. Thomas. 1990. "Invisible Decisive Coalitions in Large Hierarchies."
Public Choice 66: 101—116.

 

. 1989. "The Impossibility of a Neutral Hierarchy." Journal of law,
Economics, and Organization 5: 155-184.

 

Harman, Michael H., and John Freeman. 1977. "The Population Ecology of
Organizations." American Journal of Sociology 82:929-964.

Hart, Oliver. 1990. "An Economist’s Perspective on the Theory of the Firm." In O.E.
Williamson, ed. Organization Theory. New York: Oxford Univ. Press: 154-171.

Hastings, N. A. J., and J. B. Peacock. 1974. Statistical Distributions. New York: John
Wiley & Sons.

Haunsperger, Deanna B., and Donald G. Saari. 1991. "The Lack of Consistency for
Statistical Decision Procedures.” American Statistician 45: 252—255.

 

Hedberg, B.L.T., Paul C. Nystrom, and William H. Starbuck. 1976. "Camping on

Seasaws: Prescription for a Self-Designing Organization." Administrative Science
Quarterly 21: 41-65.

Heimann, C.F. Larry. 1992. "Administrative Errors and Bureaucratic Structure: Toward
a General Theory of Organizational Reliability." Unpublished paper presented at
1992 Midwest Political Science Association Meeting, April 9-11, Chicago.

Hellwig, Martin, and Wolfgang Leininger. 1987. "On the Existence of Subgame-Perfect

Equilibrium in Infinite-Action Games of Perfect Information." Journal of
Economic Theory 43: 55 - 75.

Hess, James D. 1983. The Economics of Organization. Amsterdam: N orth-Holland.

Hesterly, William 5., Julia Liebeskind, and Todd R. Zenger. 1990. "Organizational

Economics: An Impending Revolution in Organization Theory?" Academy of
Management Review 15 : 403-420.

Hoenack, Stephen A. 1983. Economic Behavior Within Organizations. Cambridge:
Cambridge Univ. Press.

131

Holmstrom, Bengt. 1982. "Moral Hazard in Teams." Bell Journal of Economics 13: 324-
340.

_. 1979. "Moral Hazard and Observability." Bell Journal of Economics 6: 74-91.

, and Jean Tirole. 1989. "The Theory of the Firm." In R. Schmalensee and RD.
Willig, eds. Handbook of Industrial Organization. Amsterdam: Elsevier Science
Publishers: 61-133.

Horowitz, Ira, ed. 1990. Organization and Decision Theory. Boston: Kluwer Academic.

Hrebiniak, Lawrence G., and William F. Joyce. 1985. "Organizational Adaptation:

Strategic Choice and Environmental Determinism." Administrative Science
Quarterly 28: 454-467.

Huber, George R, Michael O’Connell, and Larry L. Cummings. 1975. "Perceived
Environmental Uncertainty: Effects of Information and Structure." Academy of
Management Journal 18: 725-740.

Jensen, Michael C. 1983. "Organization Theory and Methodology." Accounting Review.
58: 319-339.

a, and W.H. Meckling. 1976. "Theory of the Firm: Managerial Behavior, Agency
Costs and Ownership Structure." Journal of Financial Economics 3: 305-360.

Jones, Gareth R. 1983. "Transaction Costs, Property Rights, and Organizational
Culture: An Exchange Perspective." Administrative Science Quarterly 28: 454:467.

 

, and Charles W.L. Hill. 1988. "Transaction Cost Analysis of Strategy-Structure
Choice." Strategic Management Journal 9: 159-172.

Kahneman, Daniel, and Amos Tversky. 1979. "Prospect Theory: An Analysis of Decision
under Risk." Econometrica 47: 263-291.

Kaufman, Herbert. 1973. Administrative Feedback: Monitoring Subordinates’ Behavior.
Washington, DC: Brookings Institution.

Keren, Michael, and David Levhari. 1989. "Decentralization, Aggregation, Control Loss
and Costs in a Hierarchical Model of the Firm." Journal of Economic Behavior
and Organization 11: 213-236.

 

. 1979. "The Optimal Span of Control in a Pure Hierarchy." Management
Science 25 : 1162-1172. .

 

 

132
King, Gary. 1989. Unifying Political Methodology. Cambridge: Cambridge Univ. Press.

Klein, Benjamin, Robert Crawford, and Armen Alchian. 1978. "Vertical Integration,
Appropriate Rents and the Competitive Contracting Process." Journal of Law and
Economics 21: 297-326.

Kmenta, Jan. 1986. Elements of Econometrics, 2nd ed. New York: Macmillan.

Kmetz, John L. 1984. "An Information Processing Study of a Complex Workflow in
Aircraft Electronics Repair." Administrative Science Review 29: 255-280.

Knapp, Tim. 1989. "Hierarchies and Controle New Interpretation and Revolution of
Oliver Williamson’s "Markets and Hierarchies" Story." Sociological Quarterly 0
425-440.

Knight, Kenneth E., and Reuben R. McDaniel, Jr. 1979. Organizations: An Information
Systems Perspective. Belmont: Wadsworth.

Koberg, Christine 8., and Gerardo R. Ungson. 1987. "The Effects of Environmental
Uncertainty and Dependence on Organizational Structure and Performance:A
Comparative Study." Journal of Management 13: 725-737.

Kreps, David M. 1990a. "Corporate Culture and Economic Theory" In J .E. Alt and K.A.
Shepsle, eds. Perspectives on Positive Political Economy. Cambridge: Cambridge
Univ. Press.

. 1990b. A Course in Microeconomic Theory. Princeton: Princeton Univ. Press.

 

 

. 1988. Notes on the Theory of Choice. Boulder: Westview Press.

, and Robert Wilson. 1982. "Sequential Equilibria." Econometrica 50: 863-894.

 

Laffont, Jean-Jacques. 1990a. The Economics of Uncertainty and Information.
Cambridge: MIT Press.

. 1990b. "Analysis of Hidden Gaming in a Three-Level Hierarchy." Journal of
Law, Economics, and Organization 6: 301-324.

 

 

. 1988. Fundamentals of Public Economics. Cambridge: MIT Press.

 

Landau, Martin. 1991. "On Multiorganizational Systems in Public Administration."
Journal of Public Administration Research and Theory 1: 5-18.

. 1969. "Redundancy, Rationality, and the Problem of Duplication and Overlap."

 

 

 

 

133
Public Administration Review 29: 346-358.

Lawrence, Paul R. 1981. "The Harvard Organization and Environment Research
Program.” In A.H. Van de Ven and W.F. Joyce, eds. Perspectives on Organization
Design and Behavior. New York: John Wiley: 311-337.

, and Jay W. Lorsch. 1969. Developing Organizations: Diagnosis and Action.
Reading: Addison Wesley.

 

. 1967. Organization and Environment: Managing Difi‘erentiation and
Integration. Cambridge: Harvard Univ. Press.

 

Lerner, Allan W. 1986. "There is more than one way to be redundant." Administration
and Society 18: 334-359.

Levinthal, Daniel. 1988. ”A Survey of Agency Models of Organizations." Journal of
Economic Behavior and Organization 9: 153-185.

Lewin, Arie Y., and George P. Huber. 1986. "Organization Design: Introduction to the
Focused Issue. ” Management Science 32: 513-538.

Lewis, Pamela S., and Patricia M. Fandt. 1989. "Organizational Design: Implications for
Managerial Decision-Making.” SAM Advanced Management Journal 54: 13-16.

Lindley, Dennis V. 1990. "The 1988 Wald Memorial Lectures: The Present Position in
Bayesian Statistics.” Statistical Science 5: 44-89.

 

. 1986. "Another Look at an Axiomatic Approach to Expert Resolution."
Management Science 32: 303-306.

Lorsch, Jay W. 1977. "Organization Design:A Situational perspective." Organizational
Dynamics 6: 2-14.

 

, and Stephen A. Allen 1H. 1973. Managing Diversity and Interdependence:An
Organizational Study of Multidivisional Forms. Boston: Harvard Univ. Press.

Luce, Duncan, and Howard Raiffa. 1957. Games and Decisions. New York: John Wiley
and Sons.

Machina, Mark J. 1983. "Generalized Expected Utility Analysis and the Nature of
Observed Violations of the Independence Axiom." In B.P. Stigum and F.
Wensth, eds. Foundations of Utility and Risk Theory with Applications.
Dordrecht: Reidel: 263-293.

 

134

 

. 1982. "Expected Utility: Analysis without the Independence Axiom."
Econometrica 50: 277-323.

 

Makowski, Louis, and Joseph M. Ostroy. 1987. "Vickrey-Clarke-Groves Mechanisms and
Perfect Competition." Journal of Economic Theory 42: 244-261.

March, James G., and Herbert A. Simon. 1958. Organizations. New York: Wiley.

Marschak, Thomas A. 1986. "Organization Design." In Kenneth J. Arrow and Michael
D. Intriligator eds. Handbook of Mathematical Economics. Amsterdam: North-
Holland.

, and Roy Radner. 1972. Economic Theory of Teams. New Haven: Yale Univ.
Press.

McGuire, CB, and Ray Radner, eds. 1986. Decision and Organization, 2nd ed.
Minneapolis: Univ. of Minnesota Press.

Meltzer, Leo, and James Salter. 1962. "Organizational Structure, and the Performance and
Job Satisfactions of Physiologists." American Sociological Review 27: 351-362.

Milgrom, Paul, and John Roberts. 1988. "Economic Theories of the Firm: Past, Present
and Future." Canadian Journal of Economics 21: 444-458.

. 1982. "Predation, Reputation, and Entry Deterrence." Journal of Economic
Theory 27: 443-460.

 

Miller, Danny. 1987. "Strategy Making and Structure: Analysis and Implications for
Performance." Academy of Management Journal 30: 7-32.

Miller, Gary J. 1992. Managerial Dilemmas: The Political Economy of Hierarchy.
Cambridge: Cambridge Univ. Press.

Mintzberg, Henry. 1983. Power In and Around Organizations. Englewood Cliffs:
Prentice-Hall.

. 1979. The Structuring of Organizations: A Synthesis of the Research. Englewood
Cliffs: Prentice-Hall.

 

. 1975 . Impediments to the Use of Management Information. New York: National
Association of Accountants.

 

Mirlees, James A. 1976. "The Optimal Structure of Incentives and Authority Within an
Organization." Bell Journal of economics 7: 105-131.

 

 

135

 

Moe, Terry M. 1984. "The New Economics of Organization." American Journal of
Political Science 78: 739-777.

Mohr, Lawrence B. 1971. "Organizational Technology and Organizational Structure."
Administrative Science Quarterly 16: 444—459.

Morris, Peter A. 1986. ”Observations on Expert Aggregation." Management Science 32:
321-328.

. 1983. "An Axiomatic Approach to Expert Resolution." Management Science 29:
24-32. '

. 1977. "Combining Expert Judgments: A Bayesian Approach." Management
Science 23: 679-693.

 

. 1974. "Decision Analysis Expert Use." Management Science 20: 1233-1241.

Muminghan. J. Keith. 1985. ”Coalitions in Decision-Making Groups: Organizational
Analogs.” Organizational Behavior and Human Decision Processes 35: 1-26.

Murray, Alan, and Mansour Javidan. 1987. "The Impact of Structure and managerial
Perceptions on the New Strategic Initiatives of Canadian Banks." International
Studies of Management and Organization 17: 5 8-67.

Myerson, Roger B. 1985. "Bayesian Equilibrium and Incentive-Comparability: an
Introduction. ” In L. Hurwicz, D. Schmeidler & H. Sonnenschein, eds. Social Goals
and Social Organization. Cambridge: Cambridge Univ. Press: 229-259.

Nash, J. F. 1951. ”Non-Cooperative Games. " Annals of Mathematics 54: 286-295.

O’Reilly, Charles A., HI. 1978. "The Intentional Distortion of Information in
Organizational Communication: A Laboratory and Field Investigation." Human
Relations 31: 173-193.

, and Karlene H. Roberts. 1974. "Information Filteration in Organizations:Three
Experiments." Organizational Behavior and Human Performance 11: 253-265.

Ordeshook, Peter C. 1986. Game Theory and Political Theory: An Introducrion.
Cambridge: Cambridge Univ. Press.

Otley, David. 1988. "The Contingency Theory of Organizational Control." In S.
Thompson and M. Wright, eds. Internal Organization, Efliciency and Profit.
Oxford: Philip Allan: 86-107.

 

 

 

 

Ouc

Pag

Pal

Fe

Fe

 

136

 

Ouchi, William G. 1980. "Markets, Bureaucracy, and Clans." Administrative Science
Quarterly 25: 129-141.

Page, Talbot. 1988. "Pivot Mechanisms as a Link between Probability and Preference
Revelation." Journal of Economic Theory 44: 43-62.

Palmer, Donald, Roger Friedland, P. Devereaux Jennings, and Melanie E. Powers. 1987.
"The Economics and Politics of Structure: The Multi-divisional Form and the
Large US. Corporation." Administrative Science Quarterly 32: 25-48.

Pearce, David G. 1984. "Rationalizable Strategic Behavior and the Problem of Perfection."
Econometrica 52: 1029- 1050.

Pennings, Johannes M. 1975. "The Relevance of the Structural-Contingency Model for
Organizational Effectiveness." Administrative Science Quarterly 20: 393-410.

Penrose, Edith T. 1959. The Theory of the Growth of the Firm. London: Basil Blackwell.
Pettigrew, A. 1973. The Politics of Organizational Decision Making. London: Tavistock.
Pfeffer, Jeffrey. 1981. Power in Organizations. Marshfield: Pitrnan.

, and Gerald Salancik. 197 8. The External Control of Organizations. New York:
Harper & Row.

Picard, Pierre. 1987. "On the Design of Incentive Schemes under Moral Hazard and
Adverse Selection." Journal of Public Economics 33: 305-331.

Pratt, J ., and R. Zeckhauser, eds. 1985 . Principals and Agents: The Structure of Business.
Boston: Harvard Business School Press.

Provan, Keith G. 1989. "Environment, Department Power, and Strategic Decision Making
in Organizations: A Proposed Integration." Journal of Management 15: 21-34.

Radner, Roy. 1991. "Dynamic Games in Organization Theory." Journal of Economic
Behavior and Organization 16: 217-260.

 

. 1986. "Review Essay: Information Pooling and Decentralized Decision Making."
in B. Grofman & G. Owen, eds. Information Pooling and Group Decision Making.
Greenwich: JAI Press: 195-217.

Raiffa, Howard. 1968. Decision Analysis: Introductory Lectures on Choices under
Uncertainty. New York: Random House.

 

 

R2

 

 

 

137

Randolph, W. Alan, Harry J. Sapienza, and Mary Anne Waston. 1991. "Technology-
Structure Fit and Performance in Small Businesses: An Examination of the ~
Moderating Effects of Organizational States." Enterpreneurship: Theory and
Practice 16: 27-42.

Rao, Ramomahan. 1989. Economic Efficiency of the Organizational Decisions of the
Firm. Berlin: Springer-Verlag.

Rasmusen, Eric. 1989. Games and Information. Oxford: Blackwell.

Riker, William H. 1962. The Theory of Political Coalitions. New Haven: Yale Univ.
Press.

Robins, James A. 1987. "Organizational Economics: Notes on the Use of Transaction-Cost
Theory in the Study of Organizations.” Administrative Science Quarterly 32: 68-
86.

Ross, Stephen A. 1973. "The Economic Theory of Agency: the Principal’s Problem."
American Economic Review 63: 134-139.

Roth, Alvin E., and Francoise Schoumaker. 1983. ”Expectations and Reputations in
Bargaining: An Experimental Study." American Economic Review 73: 362-372.

Rumelt, R. 1974. Strategy, Structure and Economic Performance. Cambridge: Harvard
Univ. Press.

Saari, Donald G. 1987. "The Source of Some Paradoxes from Social Choice and
Probability." Journal of Economic Theory 41: 1-22.

Sah, Raaj Kumar, and Joseph E. Stiglitz. 1986. "The Architecture of Economic
Systems: Hierarchies and Polyarchies." American Economic Review 76: 716-727.

. 1985. "Human Fallibility and Economic Organization." American
Economic Review 75 : 292-297.

 

Savage, Leonard J. 1962. "Bayesian Statistics." In R.E. Machol and P. Gray, eds. Recent

Developments in Information and Decision Processes. New York: Macmillan: 161-
194.

. 1954. The Foundations of Statistics. New York: John Wiley.

 

Schendel, Dan E., and Charles W. Hofer, eds. 1979. Strategic Management: A New View
of Business Policy and Planning. Boston: Little, Brown.

 

 

 

 

Sch

Sc:

Si]

(I)

 

 

 

138

Schervish, Mark J. 1986. "Comments on Some Axioms for Combining Expert
Judgemnts." Management Science 32: 306-312.

Scott, W. Richard. 1981. Organizations: Rational, Natural, and Open Systems. Englewood
Cliffs: Prentice-Hall.

Silberberg, Eugene. 1990. The Structure of Economics: A Mathematical Analysis, 2nd ed.
New York: McGraw-Hill.

Simon, Herbert A. 1976. Administrative Behavior, 3rd ed. New York: Macmillan.

Smith, J .Q.. 1988. Decision Analysis: A Bayesian Approach. New York: Chapman and
Hall.

Spence, Michael, and Richard Zeckhauser. 1971. "Insurance, Information, and Individual
Action." American Economic Review 61: 380-387.

Starbuck, William H. 1971. "Organization Growth and Development." In W.H. Starbuck,
ed. Organizational Growth and Development. Harrnonsworth: Penguin: 11-141.

Stevenson, William B., Jone L. Pearce, and Lyman W. Porter. 1985. "The concept of
"Coalition" in Organization Theory and Research." Academy of Management
Review 10: 256-268.

Te’eni, Dov. 1991. "Feedback in DSS as a Source of Control: Experiments with the
Timing of Feedback." Decision Sciences 22: 644-655.

Thompson, James D. 1967. Organizations in Action. New York: McGraw-Hill.

Tsui, Anne S. 1984. "A Role Set Analysis of Managerial Reputation." Organizational
Behavior and Human Performance 34: 64-96.

Tullock, Gordon. 1965. The Politics of Bureaucracy. Washington, DC: Public Affairs
Press.

Tushman, Michael L., and David A. Nadler. 1978. ”Information Processing as an
Integrating Concept in Organization Design.” Academy of Management Review 3:
613-24.

Van Damme, Eric. 1983. Refinements of the Nash Equilibrium Concept. Berlin: Springer-
Verlag.

Van de Ven, Andrew H., and W. Graham Astley. 1981. "Mapping the Field to Create a
Dynamic Perspective on Organization Design and Behavior." In A.H. Van de Ven

 

 

 

 

 

139

and W.F. Joyce, eds. Perspectives on Organization Design and Behavior. New
York: John Wiley.

Venkatraman, N., and John E. Prescott. 1990. "Environment-Strategy Coalignment: An
Empirical Test of its Performance Implications." Strategic Management Journal
11: 1-23.

Viscusi, W. Kip. 1985. "Are Individuals Bayesian Decision Maker?" American Economic
Review 75: 381—385.

, and Wesley A. Magat. 1987. Learning about Risk:Consumer and Worker
Responses to Hazard Information. Cambridge: Harvard Univ. Press.

, and Charles O’Connor. 1984. "Adaptive Responses to Chemical Labeling: Are
Workers Bayesian Decision Makers?" American Economic Review 74: 942-956.

Waddock, Sandra A., and Lynn A. Isabella. 1989. "Strategy, Beliefs about the
Environment, and Performance in a Banking Simulation." Journal of Management
15: 617-632.

Weber, Max. 1947. The Theory of Social and Economic Organization. trans. by AM.
Henderson and T. Parsons. New York: Free Press.

Weigelt, Keith, and Colin Camerer. 1988. "Reputation and Corporate Strategy: A Review
of Recent Theory and Applications." Strategic Management Journal. 19: 443-454.

Wilensky, Harold L. 1967. Organizational Intelligence. New York: Basic Books.

Williamson, Oliver E. 1991. "Comparative Economic Organization: The Analysis of
Discrete Structural Alternatives." Administrative Science Quarterly 36: 269-296.

. 1990. Organization Theory. New York: Oxford Univ. Press.

. 1985. The Economic Institutions of Capitalism. New York: Free Press.

. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. New York:
Free Press.

. 1970. Corporate Control and Business Behavior. Englewood Cliffs: Prentice-
Hall.

, and William G. Ouchi. 1981. "The Markets and Hierarchies Paradigm of

Research: Origins, Implications, Prospects." In A.H. Van de Ven and W.F. Joyce,
eds. Perspectives on Organization Design and Behavior. New York: John Wiley:

 

Win

 

 

 

 

 

 

 

 

140

347-370.

Winkler, Robert L. 1986. "Expert Resolution." Management Science 32: 298-303.
. 1985. "Information Loss in Noisy and Department Processes." In J .M.Bernardo,
M.H.DeGroot, and D.V.Lindley, eds. Bayesian Statistics 2. Amsterdam: Elsevier
Science Publishers: 559-570.

1981. "Combining Probability Distributions from Dependent Infonnation
Sources." Management Science 27: 479-488.

 

. 1968. "The Consensus of Subjective Probability Distributions." Management
Science 15:B-61-B-75.

 

Woodward, Joan. 1965. Industrial Organization: Theory and Practice. London: Oxford
Univ. Press.

Worthy, James C. 1950. "Organizational Structure and Employee Morale." American
Sociological Review 15: 169-179.

Yarbrouch, Beth V., and Robert M. Yarbrouch. 1988. "The Transactional Structure of the
Firm." Journal of Economic Behavior and Organization 10: 1-28.

 

 

 

 

 

 

   

lllll