JPART 15:113–131
Management Tenure and Network Time:
How Experience Affects Bureaucratic
Dynamics
Eric Gonzalez Juenke
University of Colorado at Boulder
ABSTRACT
Meier and O’Toole have developed an empirical model that allows scholars to test for the
impact of managers on a system and its outputs. In this article I attempt to add to
management theory and analysis by examining the impact of time in the system and
management tenure. I use ordinary least squares to replicate and expand upon Meier and
O’Toole’s results, using school superintendent survey responses along with outcome
measures from school districts in Texas. The most interesting results suggest that (1)
networking has a much larger impact when one controls for experience with the system; (2)
experience with the system has independent effects on outcomes; (3) management tenure
interacts with networking, resulting in greater outcomes; and (4) new managers may find
alternative (possibly deceitful) ways of affecting outcomes other than working their networks.
The public management field is in the midst of a theoretical and empirical upheaval
concerning the role played by networks in the delivery of public services. The rise of
public/private cooperation in the public sphere has cast doubt on the picture of the modern
bureaucracy as a hierarchical system of inefficiency. I continue the process of examining
management effects through public/private networks by exploring the frequently discussed
but infrequently tested idea of time within a network. Much of the management literature
treats networking as a one-shot phenomenon, ignoring ‘‘managerial experience’’ differences across organizations, but this study treats the relationships formed over time as a
critical element of network management success.
The article is fairly straightforward in that it adds a number of new components to a
previously developed model of management, to look separately at ‘‘new’’ and ‘‘established’’ managers in their respective networks. There are two overriding themes: First,
what effects do new managers have in their networks? Do they find it easy to operate, or do
they need time to develop relationships and build trust? The converse is asked about
established managers: Do they make outcome production inefficient because they are
Thanks go to Kenneth J. Meier, Laurence J. O’Toole Jr., Sean Nicholson-Crotty, Nick Theobald, and four
anonymous reviewers for help with earlier versions of the manuscript. Address correspondence to the author at
[email protected].
doi:10.1093/jopart/mui006
Journal of Public Administration Research and Theory, Vol. 15, no. 1
ª 2005 Journal of Public Administration Research and Theory, Inc.; all rights reserved.
114
Journal of Public Administration Research and Theory
apprehensive of change, or do they find it easier to maneuver between network contacts in
order to exploit resources?
The second theme involves the networks themselves. Does the tenure of the top
manager in a bureaucratic organization tap into an underlying concept regarding the
network? If the top manager has endured over time, what does this say about the network’s
ability to produce positive outcomes? This view speaks to the other side of ‘‘network
stability.’’ How much does the job security of the leader of a system reveal about the
stability of other actors in the system? Presumably, constant turnover in other nodes should
not reflect well on the manager and does not bode well for her or his time in office,
especially if networking matters. This analysis tests whether networking matters for public
education outcomes; more specifically, it provides evidence to address ‘‘under what
circumstances’’ networking matters.
The article is laid out into three main sections. The first section covers the theoretical
reasoning behind the analysis; it provides a number of possible motives for why ‘‘network
tenure’’ should contribute to better outcomes. The section also presents alternative beliefs
concerning time spent in the network and how it may diminish innovation and decrease the
benefits of network activity. The second section describes the management model used to
perform the analysis and explains the data and methods. The third section contains the
results, as well as an interpretation of what they mean for bureaucracies in general and
what they mean for scholars interested in the dynamics of network management. It also
presents some surprising findings and presents possible explanations for deviant
management behavior in certain environments.
RELATIONSHIP BUILDING IN NETWORKS: HOW TIME AFFECTS ACTIVITY
Management scholars define policy networks in similar ways. Kickert, Klijn, and
Koppenjan define them as ‘‘patterns of relations between independent actors, involved in
processes of public policy making. Interdependency is the key word in the network
approach’’ (1997, 6). O’Toole describes networks as ‘‘structures of interdependence
involving multiple organizations, where one unit is not just the formal subunit or
subordinate of the other in some larger hierarchal arrangement’’ (1997, 117). Alter and
Hage (1993) fill out the definition by specifying the joint production capabilities of
networks in the public setting. Last, networks can involve both public and private actors, an
important definitional detail for this particular study. These characteristics define
‘‘networks’’; they are patterns of communication that lie outside of the traditional policy
structure but do not exclude traditional components of the hierarchy.
The model developed and tested by Meier and O’Toole (2001) formalizes a theory of
management and examines the effects of network activity on a system and its outputs.1 The
model is described in section two, but it is necessary here to clarify the authors’ theoretical
framework. The model focuses on system outputs (and/or outcomes) and the amount of
influence system components have over these outputs. This framework is important
because it allows us to test whether systems are inertial (autoregressive), influenced mostly
by environmental factors, influenced by organizational factors, or influenced by aspects
under the manager’s control. It is this last feature that management scholars are most
1
I would like to thank Kenneth J. Meier and Laurence J. O’Toole Jr. for the use of their data. I have made
minor adjustments and additions to their original data set, but for the most part, it remains intact from their earlier work.
I accept full responsibility for errors in, or misinterpretation of, the data and analyses.
Juenke
Management Tenure and Network Time
interested in. Put simply, Do managers matter? And if so, when and how do managers
matter?
More specifically, my interest concerns management effects through network activity.
When leaders interact with others outside of the hierarchy, are they producing benefits for
the organization, or are they just wasting time? Michael Howlett (2002) uses evidence from
four Canadian policy sectors to show that the type of networks involved in policy change
has a large influence on outcomes. Provan and Milward (1995) similarly demonstrate that
the type of network structure has significant effects on outcomes in the health service
industry. These are valuable pieces of research that support the study of policy networks,
but they do not address the frequency of network activity and its possible consequences for
outcomes.
Meier and O’Toole (2001) specifically test whether the frequency of contact with
‘‘network nodes’’ has an impact on outcomes (see also O’Toole and Meier 2003a). Using
survey data collected in 1999 from the top managers of public school districts in Texas,
Meier and O’Toole (2001) find that the amount of time spent interacting with both public
and private network nodes has a significant and positive relationship with the district pass
rate for a high-stakes exam. The more time superintendents in these districts allocated to
communicating with local business leaders, other superintendents, state legislators, school
board members, and the state education agency, the greater was the district pass rate. Meier
and O’Toole’s initial study is important because it was the first large-N study of its kind,
and it specifically speaks to the frequency of network contact rather than the type. Also, the
authors combine survey data with other sources to show how this type of research should
progress in the future. The results from their analysis are suggestive, but they do not tell the
entire story.
The present study attempts to add to both their theory and their analysis by examining
the impact of time in the system and management tenure. There are many ways in which
time in the network can affect the outcomes of such activity. Three theoretical reasons lead
us to expect that management tenure and time in the network should have positive
outcomes for the organization: the first is manager centered and has to do with getting to
know the network; the second is network centered and deals with stability and network
evolution; the third is the interaction of the previous components and concerns trust
building in network environments. The article covers each component briefly before
presenting alternative hypotheses regarding network activity outcomes.
Before continuing, it is important to qualify the extent to which we can draw causal
links between networking theory and the measure used here. The strong and suggestive
results reported in the article must be tempered using similar language from Meier and
O’Toole: ‘‘Clearly, this measure is simplified. It ignores all aspects of networking aside
from frequency and direction—for instance, skill, reputation, and a number of strategic
considerations’’ (2001, 281). But initial measurements of complex relationships like
networking activity are always going to be rudimentary. Meier and O’Toole continue,
‘‘Still, the measure taps the efforts managers choose to put into managing externally, in the
network’’ (2001, 281). The type of networking activity scholars discuss, the cooperative
dynamics, and the group-produced outputs are not directly measured here. This is an
important qualification but should not detract from the information contained in the
indirect behavioral measurement used to operationalize the model in this and other
research. Not only are these indirect measures theoretically consistent, they have, here and
elsewhere, been empirically validated.
115
116
Journal of Public Administration Research and Theory
Learning about the Network
Individuals who are new to a system take some time to develop relationships with interests
inside and outside the organization. Meryl Reies Louis (2001) presents a model of
‘‘newcomer experience’’ that formulates a number of hypotheses about the behavior of
new actors. While the model does not speak directly to management-level individuals, it
does not exclude them. Reies Louis (2001, 378) argues that newcomers face information
deficiencies, exclusion from informal and influential networks, and a peripheral position of
control. Michael Howlett (2002) similarly argues that new actors change the entire
dynamics of a network. He (2002, 246) attributes policy change to the combination of new
actors with new ideas into the system; the absence or presence of each relative to the other
has manifest consequences for outcomes. This is a self-evident set of assumptions;
newcomers in most settings are at a disadvantage relative to those in the organization, even
if the newcomers are in a position of greater formal control.
This argument demonstrates a big difference between newcomers to hierarchal
organizations (inside the system) and newcomers to networks (involving those outside the
system). Agranoff and McGuire (1999b, 21) suggest that managing networks and
managing hierarchies are two different activities. They explain, ‘‘The primary activities of
the network manager involve selecting the appropriate actors and resources, shaping the
operating context of the network, and developing ways to cope with strategic and
operational complexity’’ (1999a, 1999b). Klijn and Teisman characterize networks as
games and describe the manager’s less-than-optimal position by stating, ‘‘An essential
feature of network management is the assumption that all the actors involved possess a
margin of liberty which they will use in the game’’ (1997, 105). Clearly, new managers
suffer relative to more mature actors in the networks.
This is even more problematic in the school district setting, where interests are
entrenched and issue salience is high. New superintendents may quickly become frustrated
trying to deal with combative school board members or an overwhelmed and
uncooperative state education bureaucracy. Conversely, more experienced superintendents
have closer ties with many of their network nodes, understanding which ones need to be
cajoled and which ones need to be avoided altogether.
This relationship building goes beyond management tenure, however. Time spent
within the system in another capacity (as a teacher, principal, or assistant superintendent)
should also serve as a comparative advantage in affecting system outcomes. Many
managers have been building network relationships (as well as professional reputations)
for many years in the hierarchy before even being considered superintendent candidates.
These superintendents should realize the greater benefits of such (preemptive) networking
once given the chance to lead. O’Toole and Meier (2003a) find that the amount of time a
superintendent has spent in the district in any capacity has a significantly strong and
positive relationship with outcomes, demonstrating the connection between network
experience and student performance.
Network Evolution
The second theoretical link between tenure and networking outcomes has to do with the
development of the network over time (Termeer and Koppenjan 1997). Hoffman (1999)
explains that issue change causes an organizational field to form around particular topics
and that these organizational fields change over time relative to the changing environment.
Juenke
Management Tenure and Network Time
Using content analysis of chemical industry journals in the United States, he shows that
issue variation begins a process of organizational field (network) change to cope with new
problems. The important insight Hoffman provides is that measuring networks at one time
may not tell the entire story because the ‘‘field’’ may be at one particular stage of
development. Alter and Hage (1993) use similar language to describe how the networks
they observe have evolved over time.
This second theoretical argument is different from the one described in the previous
section because it reasons that the tenure of the top manager is an indicator of network age.
If the superintendent of a school district has been at the job for a long time (over ten years,
for example), then we can assume that the district is reasonably pleased with the job she or
he is doing. Of course, this does not hold in all cases, but on average, bad districts will not
keep their top manager on the job for too long. The short tenure of new managers may give
us extra information about the development of the network in that district. Alter and Hage
(1993, 252–254) show that network age may have a deleterious effect on outcomes because
rigid structures begin to replace the open exchange system that characterizes young
systems. This counterhypothesis will be examined in the last part of this section, but it is
important to note the development of a language that incorporates the ‘‘age’’ of a network
irrespective of the experience of the manager (Alter and Hage 1993, 253).
These ideas are all tied to network stability. If networks are like games, as Klijn and
Teisman (1997) suggest, then the players should learn about one another over time. The
trust that develops during repeated iterations is discussed below, but cooperation can take
place even in the absence of trust. Milward and Provan (2000, 247) clarify this by bringing
in the development of ‘‘reputations’’ and social structures that characterize the repeated
interactions of established networks. They argue, ‘‘At some point, an unstable network
becomes stable enough to perform reasonably well while still remaining flexible and
adaptive’’ (2000, 253). Like Alter and Hage, the authors warn that flexibility can become
rigid over time, suggesting an alternate hypothesis, but the point remains that networks in
the same policy area may be at different levels of development, and we need to account for
this in our measures. In another study, Milward and Provan (1998, 217) find that network
stability increased network effectiveness, even though it was not a sufficient condition for
effectiveness. I argue that the manager’s tenure is a good indicator of this network stability.
O’Toole and Meier (2003a) come closest to testing this hypothesis by using ‘‘time in the
district’’ in their stability model, but they are primarily concerned with the stability of the
network nodes in their study and do not use superintendent tenure to adequately measure
the top-level manager’s role in developing those networks.
Building Trust
The third theoretical setting used to derive hypotheses for this study comes from the game
theoretic perspective. If network interaction over time involves repeated games with
players who have their own interests, then it is imperative for managers to develop trust
with the respective nodes. If other network actors do not trust that the manger will fulfill
her or his obligations, then the effects of network activity will be zero (or negative).
O’Toole reveals that this is especially critical in network situations because ‘‘there are
fewer inducements toward cooperation and greater impediments across the multiactor
array’’ (1997, 131). While trust is important in all interactions, it takes on special
significance in a network.
117
118
Journal of Public Administration Research and Theory
La Porte and Metlay (1996, 342) lay a good foundation for understanding trust in an
unknown environment. They define trust as the ‘‘belief that those with whom you interact
will take your interests into account, even in situations where you are not in a position to
recognize, evaluate, and/or thwart a potentially negative course of action by ’those
trusted’’’ (1996, 342; see also Barber 1983 for a similar definition). This is comparable to
the type of environment experienced by actors in networks. The ‘‘transaction costs of
recovering trust’’ are high in these environments, which promotes a cooperative
atmosphere in network settings. La Porte and Metlay (1996, 345) conclude by suggesting
that organizational constancy (stability) is the most important factor in promoting trust
among actors.
Agranoff and McGuire make an explicit case for trust in network environments
because of the relatively new interest in private/public networks: ‘‘The ability of organizational actors to operate together within a single network is thus less dependent upon
a shared belief system, ideology, and common world view than it is a shared rationale for
organizing embodied in the project or program itself’’ (1999b, 29). Clearly, time in the
network is critical to building trust. This goes beyond simple time in the district; rather, it
speaks to the specific position of the top manager and her or his ability to engender trust
among network nodes. Not only must they trust the top manager, but also she or he must
provide a means for building trust among the other actors so that the benefits of group
activity will be realized.
Measuring trust in a network is difficult, so instead, this line of reasoning is included
with the other two to explain why we should expect management tenure to indicate a level
of trust that may not be developed between new managers and their networks. If network
activity is about building outside relationships that are beneficial to the system (whether or
not they are buffering or exploitative), and these relationships involve trust building and
iterative games of strategic behavior, then the more interaction a manager has with the
network nodes, the more developed and beneficial these relationships will be. Thus, system
experience is expected to weight network interactions in favor of managers with longer
tenures, and lack of experience, to hinder new managers’ interactive efforts.
An Alternative Theory
As specified above, some network scholars theorize that too much time in the network may
produce bureaucratic inefficiency and ineffectiveness. Milward and Provan (2000) and
Alter and Hage (1993) both support the notion that network evolution can go too far. Over
time, previously useful and cooperative networks may become rigid and unproductive. In
fact, this is the type of structure-building and uncooperative behavior that bureaucracies
often exhibit, which drew the ire of public administration scholars and contributed to the
recent explosion of private/public networking (Barzelay 1992; Osborne and Gaebler 1992).
In order to gain control over a network, managers may institute formal measures, which
can kill innovation and risk taking. If so, we should see no effects from networking in more
mature systems because the managers (or some other network node) have taken over and
reduced the group’s capacity to accomplish goals. The present study argues that more
experienced managers who find themselves in this situation (a network where activity is
fruitless) will decrease their involvement and produce small (or no) outcomes through
networking. These ideas suggest testable hypotheses; using the model and data set
developed by Meier and O’Toole (2001) we can produce evidence that supports or refutes
Juenke
Management Tenure and Network Time
the theory. The tested hypotheses are listed below. They serve as a good summary of the
theory thus far, although hypothesis 4 will be explained in the next section.
Hypotheses
H1 Network management effects will increase after controlling for time spent in the system,
and system experience will have independent, positive effects on outcomes.
H2 Management tenure (time spent in the lead position within the system) will have a
positive independent effect on outcomes.
H3 Management tenure will positively interact with network management to boost the effects
of network activity on outcomes.
H4 Because new managers’ network activity will have less impact on outcomes, the effects
of alternative (non-networking) outcome-oriented strategies will be greater under new
managers (the converse is true for more experienced managers).
THE DATA, MODEL, AND METHODS
The hypotheses are tested in a policy area uniquely suited for this topic: public education.
This policy space is highly salient and divisive, with numerous interests vying for control.
The data set combines survey data from public school districts in Texas with district
information provided by the Texas Education Agency (TEA) for the years 1995 through
1999 (see Meier and O’Toole 2001 for a complete description of the data set). In the U.S.
education system, superintendents are the top bureaucrats at the district level (the most
visual and relevant education unit in the system) and thus the most important actors in the
network of public and private individuals producing district education outputs; as the
linchpin between the internal and external environment, they are the network managers.
Five hundred seven superintendents returned surveys that asked a number of questions
regarding tenure and time in the district, as well as frequency of contact with network
actors. The network array involves both public and private actors (school board members,
local business leaders, the TEA, other superintendents, and state legislators). It is the
variance in the frequency of interaction with this network that the measurement taps into,
and it is this variance that provides a critical indirect measure of network management.2
The dependent variable is the district pass rate on the Texas Assessment of Academic
Skills (TAAS) test, a high-stakes exam that was used by the state as a performance measure
to rate all schools and districts in Texas.3 This variable is chosen because it is a good measure
of system outcomes. Also, it is the top priority of almost all of the managers in the data set
according to their responses to questions asking them to rank the goals of the district.
Meier and O’Toole explain that management is measured in two ways: management
quality and network management. Management quality is captured as the residual from an
estimation of managerial salary and has been validated in earlier work (Meier and O’Toole
2
The ‘‘management’’ aspect of the variable comes from the superintendent’s position as the top-level bureaucrat
in charge of bringing in new resources from the external environment. The actors are interdependent, but in this
particular context, the manager is accountable for the outputs produced by the network, and it is her or his interaction
with the network (or lack thereof) that I am interested in.
3
The TAAS is no longer used in Texas; it has been replaced by a similar high-stakes exam.
119
120
Journal of Public Administration Research and Theory
2002). Managers who are paid above and beyond what is expected using a mix of
reasonable predictors (education, age, race, gender, and a number of district characteristics) are considered to be of higher quality than those who are paid much less than is
forecast by the model.4 To tap into networking behavior, the superintendents were asked
how frequently they interact with the five network nodes, using a six-point scale ranging
from ‘‘daily’’ to ‘‘never.’’ The factor scores for the five components of the network array
capture ‘‘network management style’’; all five items loaded positively on the first factor
(O’Toole and Meier 2003b, 19).
It should be reiterated that there is reason to be skeptical about the operationalization
of the ‘‘network’’ captured here. Rather than a direct measure of network development or
interaction, Meier and O’Toole provide an approximation of the amount of contact the
manager has with her or his networking partners. A better measure of network interaction
would involve tapping all of the nodes and observing the dynamics of the group (not an
individual, as is the case here), but such a measure awaits future scholars who have access
to a large number of organizations and their respective nodes. On the other hand, we are not
uninterested in the frequency of contacts the superintendent has with her or his partners,
because the superintendent is the manager of the network that produces district outputs. As
a networking indicator, the measure taps into the right information from the most important
individual in this context; any type of network interaction presupposes contact. It is not a
perfect networking measure, but it is full of relevant information for networking scholars
interested in empirical tests.
The time dynamic of the aggregate data set creates a time-series cross section that
enlarges the number of observations available for analysis. The shallow pool will not give
us much leverage on questions of causality, but it will provide a more complete picture of
‘‘tenure’’ and ‘‘time in district’’ networking effects over a simple cross-section design.
Ordinary least squares regression with fixed effects for years is used to estimate the
parameters.
THE MODEL
The O’Toole and Meier management model is discussed in greater detail in a number of
earlier studies (Meier and O’Toole 2001; O’Toole and Meier 2003a, 2003b). It is uniquely
suited for this analysis because it allows us to test competing factors affecting system
outcomes. Briefly, the model differentiates between environmental and managerial effects
on outcomes. Environmental variation, like clientele characteristics and organizational
constraints, are controlled for across units, allowing researchers to empirically separate the
effects of managerial quality and behavior. ‘‘M2’’ in the model is a measurement of the
manager’s efforts to both buffer external shocks to the system and exploit available
resources through contact and cooperation with internal and external actors (Meier and
O’Toole 2001, 9). If managers who work more with their network produce outcomes that
are significantly better than those who do not, after controlling for the environmental
features that help explain performance, then we can successfully implicate managerial
behavior in the causal process.
4
Again, this is an indirect indicator of a difficult concept to measure (‘‘managerial quality’’), but it is
theoretically consistent and performs well in empirical tests. School boards assess the performance of superintendents
on a yearly basis and set their salary accordingly: ‘‘In that determination, we think that management quality plays a
role—not an exclusive role, but a role nonetheless’’ (Meier and O’Toole 2002, 632).
Juenke
Management Tenure and Network Time
In the U.S. education setting, the superintendent is accountable to elected school
board members and most often bears the responsibility for performance failure. Successful
managers learn how to work with the school board to find solutions to district failures, take
advantage of resources from local businesses to supplement public funds, and
communicate with other superintendents to learn about successful programs in other
districts. Managers who are never in touch with these important network actors are likely to
produce different outcomes than those who report that they are in touch with these
copartners on a daily basis.
The empirical tests are just one tool, certainly not the only tool, that provides access to
an extremely complicated set of phenomena. If managerial activity does not matter, or if
the measure is faulty, we should find no significant effect when it is included in the tests.
The null hypothesis for each test is that managerial quality and frequency of network
contact will have no influence on system outcomes. The hypotheses listed above are tested
using the same basic model from Meier and O’Toole (2001), examining only the additive
form (with the nonlinear effects of tenure) in this article, with modifications for time
suggested above. Below are the two formal models tested:
Ot ¼ b1 Ot1 þ b2 Xt þ b3 M2 þ b4 MQ þ b5 Tenuret þ b6 Timet þ et
½1
Ot ¼ b1 Xt Tenuret þ b2 M2 Tenuret þ b3 MQ Tenuret þ b4 Timet Tenuret
þ b5 Exemptt Tenure þ b6 Xt þ b7 M2 þ b7 MQ þ b8 Timet þ b9 Exemptt þ et ½2
In these equations, Ot 5 outcomes at time t. These are the district’s pass rate on the state’s
high-stakes test,
Ot–1 5 last year’s pass rate,
M2 5 network management measure,
MQ 5 management quality measure,
Xt 5 a vector of environmental and organizational constraints (Meier and O’Toole 2001),
Tenure 5 experience within the district, as a superintendent, at time t,
Time 5 time spent in the district by the superintendent in any capacity,
Exempt 5 % of students (in specific category) exempted from taking the TAAS test
et 5 an error term
The same control variables from the original authors’ analysis are used except for one
additional component. ‘‘Exempt’’ is the percentage of students who have been exempted
from taking the TAAS test for each district for any reason. There is reason to believe that
this designation is sometimes used as an alternative method of bringing up the district pass
rate (Bohte and Meier 2000; Meier and Bohte n.d.). Underperforming students who are not
legally exempt from taking the test may be asked to stay home on test day or may be given
a ‘‘special ed.’’ or ‘‘limited English’’ exemption.
This type of organizational ‘‘cheating’’ may not be systematic, but there is a large
amount of discretion at each level of the system. Superintendents who do not tolerate the
use of inappropriate exemptions will find ways to curb ‘‘cheating,’’ while others may leave
the discretion in the hands of street-level bureaucrats as a way to allow the practice to
121
122
Journal of Public Administration Research and Theory
Table 1
Management and Performance: Replication and the Adjusted Sample
Dependent Variable 5 Student Exam Pass Rates
Independent Variables
Base Models
Lagged Pass Rate
Network Management
Teacher’s Salaries (k)
Class Size
Teacher Experience
Noncertified Teachers
% State Aid
% Black Students
% Latino Students
% Low-Income Students
Constant
R2
F
N
—
0.7035**
(4.60)
0.4665**
(4.31)
0.3117**
(4.72)
0.1943*
(1.90)
0.1873**
(5.30)
0.0173**
(2.09)
0.2167**
(13.49)
0.1091**
(10.39)
0.1670**
(11.16)
75.66**
(25.32)
0.59
187.8
2,534
0.7206**
(56.84)
0.1842*
(1.81)
0.2133**
(2.97)
0.0502
(1.14)
0.0753
(1.17)
0.0927**
(3.95)
0.0004
(0.07)
0.0548**
(4.98)
0.0352**
(4.98)
0.0191*
(1.87)
21.67**
(9.88)
0.82
708.68
2,534
New Sample
—
0.6244**
(3.45)
0.6187**
(4.89)
0.4848**
(5.38)
0.0250
(0.22)
0.2070**
(4.69)
0.0007
(0.08)
0.2049**
(10.64)
0.1120**
(8.78)
0.1574**
(8.49)
75.51**
(21.33)
0.57
122.43
1,681
0.7264**
(48.15)
0.2602*
(2.22)
0.2707**
(3.29)
0.0731
(1.24)
0.2109**
(2.84)
0.1365**
(4.78)
0.0000
(0.00)
0.0548**
(4.26)
0.0418**
(4.98)
0.0029
(0.23)
21.07**
(8.25)
0.82
495.18
1,681
Note: Dummy variables for individual years not reported. Absolute value of t-scores in parentheses.
**p , .05; *p , .10.
continue. In order to have an impact on outcomes (immediate or prolonged) new
superintendents might seek out alternative solutions to networking in order to achieve
quick performance results. This variable is included in each of the analyses to indicate the
level of ‘‘alternative outcome strategies’’ employed by the district as a whole.
RESULTS AND ANALYSIS
The first part of the analysis involves a replication of Meier and O’Toole’s (2001, 285)
results; these are presented in the first two columns of table 1. My results are identical
except for the ‘‘Teacher Experience’’ coefficient, which is marginally higher in their table.
Table 1 also presents a corrected sample size, one that includes only performance and
environmental observations that are appropriate for the time frame of the managerial
survey responses. If a respondent had been the superintendent for less than five years (the
time frame of the performance data), then the data from any of the years they had not been
in that position were dropped (for example, ‘‘three year tenure’’ 5 first two years for that
individual were dropped). The ‘‘Network Management’’ coefficient moves from 0.184 to
0.260 from their sample to mine when the model includes the lagged pass rate. This is
Juenke
Management Tenure and Network Time
Table 2
Management and Performance: The Effects of Time, Tenure, and Quality
Dependent Variable 5 Student Exam Pass Rates
Independent Variables
Lagged Pass Rate
Network Management
Time in District
Management Tenure
Management Quality
Teacher’s Salaries (k)
Class Size
Teacher Experience
Noncertified Teachers
% State Aid
% Black Students
% Latino Students
% Low-Income Students
R2
F
N
—
0.7457**
(4.14)
0.0908**
(4.06)
0.0077
(0.18)
0.8659**
(4.22)
0.5608**
(4.37)
0.5801**
(6.35)
0.0870
(0.75)
0.2127**
(4.83)
0.0052
(0.52)
0.2121**
(11.01)
0.1031**
(8.12)
0.1669**
(9.01)
0.59
99.06
1,663
0.7161**
(46.94)
0.2828**
(2.39)
0.0282*
(1.93)
0.0084
(0.30)
0.1641
(1.22)
0.2771**
(3.29)
0.1259**
(2.08)
0.1909**
(2.51)
0.1423**
(4.93)
0.0017
(0.26)
0.0557**
(4.27)
0.0380**
(4.52)
0.0123
(0.98)
0.82
383.30
1,663
Note: Dummy variables for individual years not reported. Absolute value of t-scores in parentheses.
**p , .05; *p , .10.
encouraging because it validates Meier and O’Toole’s original findings; within the
corrected sample, network management has larger effects. The new sample is used
throughout the analyses.
In table 2 the results of the test on hypothesis 1 are included. Two measures of
network experience as well as Meier and O’Toole’s measure of managerial quality are
included. District experience is operationalized with the number of years the respondent
has been in the district in any capacity, and the number of years the respondent has been the
superintendent in the district is used to measure management tenure. Both of these
variables were adjusted for time in the data set so that a superintendent who had been in the
district five years in 1999 was counted as being in the district ‘‘four years’’ in 1998, ‘‘three
years’’ in 1997, and so on.
As the results in table 2 indicate, the number of years a respondent has been in the
district in any capacity has an immediate effect on both the network management variable
123
124
Journal of Public Administration Research and Theory
Table 3
Do Managers with Longer Tenure Network More?
Dependent Variable 5 Networking Measure
Independent Variables
Coefficient
t-score
Management Quality
Time in District
Management Tenure
R2
F statistic
N
0.0594**
0.0111**
0.0073
0.04
7.16
1,663
2.11
3.64
1.25
Note: Dummy variables for individual years not reported. Nine controls not presented.
**p , .05 (two-tailed test).
and the system outcomes.5 In this model, the management coefficient moves from 0.624
(results from table 1) to 0.746 and from 0.260 to 0.283 in the models controlling for last
year’s TAAS pass rate. Independent of its impact on management, the amount of time a
superintendent has spent in the district has significant effects on outcomes as well.
Management tenure appears to have no independent association with outcomes, but
this is not a good test of hypothesis 2; its interactive relationship with management will be
examined later. Management quality, however, has a strong and significant influence on
test outcomes. While this relationship disappears when the lagged pass rate is included, it
reappears in an interesting and consistent manner in later results.
Before presenting the full model, it would be beneficial to discuss the relationship
among network management, quality, time, tenure, and the use of exemptions as an
alternative tool. We wish to know if longer-tenured superintendents and higher-quality
managers will network and use exemptions more often than newer superintendents.
Bivariate correlation results (Appendix) indicate that more experienced and higher-quality
managers will network less than younger, lower-quality managers. This relationship is
again evidenced in table 3, where networking is regressed on the three other measures (as
well as nine environmental controls). All three relationships are negative (though they all
have very small effects on networking). This is interesting but still consistent with the
theory, namely, that it is the quality of contact, not necessarily the frequency, that makes a
difference for more tenured superintendents in their networks. Also of note is the tiny
amount of variance explained by the twelve variables, a finding consistent with Meier and
O’Toole’s results and one that suggests that network management is highly contingent on
individual discretion.
Table 4 reports the outputs of a similar model, but one that tries to explain test
exemptions rather than networking. Low-performing students can be designated as
‘‘learning disabled’’ or ‘‘language disabled,’’ or some students may be asked to refrain
from coming to school on the TAAS test day. These low performers are then excluded from
the district’s overall pass rate. In the high-stakes atmosphere of state test performance,
managers may find these alternative means of producing results more attractive than other
long-term approaches.
5
Summary statistics and correlations between all new variables used in this article are found in the Appendix.
Juenke
Management Tenure and Network Time
Table 4
Do Managers with Longer Tenure Allow More Exemptions?
Dependent Variable 5 Percent Students Exempted
Independent Variables
% All Exempt
% Low Income
Lagged Pass Rate
Network Management
Management Quality
Time in District
Management Tenure
R2
F
N
0.0401**
(2.02)
0.1707
(1.18)
0.1459
(0.89)
0.0094
(0.52)
0.0373
(1.08)
0.27
28.54
1,187
0.1006**
(3.35)
0.1382
(0.63)
0.1552
(0.63)
0.0091
(0.33)
0.0723
(1.38)
0.15
11.45
1,186
% Special Ed.
% LEP
0.0229
(1.38)
0.2849**
(2.37)
0.0585
(0.43)
0.0082
(0.54)
0.0267
(0.92)
0.25
23.02
1,187
0.0294**
(3.32)
0.0906
(1.41)
0.1591**
(2.18)
0.0010
(0.12)
0.0070
(0.45)
0.25
30.46
1,187
Note: Control variables not presented. Absolute value of t-scores in parentheses. % LEP 5 % Limited English Proficient.
**p , .05 (two-tailed test).
I regress four different (but highly correlated; see Appendix) test exemption measures
on to the same management variables and nine environmental controls.6 The results are
consistent with Bohte and Meier (2000), who find a large amount of leftover ‘‘exemption
variance’’ unexplained by legitimate environmental causes. In fact, the results indicate that
much of the discretion to ‘‘cheat’’ may come from lower-level sources (principals and
teachers, unmeasured here), because the management variables have little significant
impact. The more important conclusion drawn from these results is that superintendents
who have longer tenures are not necessarily associated with higher levels of test
exemptions when controlling for all other influences. This finding will become important
later.
In order to test for interactive effects between superintendent tenure and network
management, I split the sample into three groups. There are a number of ways this can be
done; in a previous version of this article I split the sample by ‘‘five years’ experience as
superintendent in this district’’ because it is theoretically reasonable to assert that it takes
about this much time to establish the kind of nodal relationships discussed at the beginning
of the article. Given that the average superintendent has been managing her or his district
for about five years, this split proved fruitful.7 In this version of the article, however, I
wanted to examine finer interactions between tenure and networking, so I split the sample
into thirds using the 25 percent and 75 percent tenure frequencies to motivate the test.8 This
produces three samples bounded at the bottom by two years’ experience and by seven
6
The test exemption data for this test, as well as the results presented in tables 5 and 6, are for the years
1997–1999. Because of this, the sample size for these tests was reduced by about two-fifths. The results are very
similar to previous findings, however (an earlier version of this article).
7
Joint F-tests on the two samples indicate that they are indeed two separate groups in the population.
8
At first sight, this does not look like an even three-way split. By including the managers who had been in control
for the years on the edge of the split (managers with tenures of two and seven years) in the bottom and top groups,
respectively, I am able to create a well-balanced set of subsamples. Table 5 gives a better picture of what I am doing.
125
126
Journal of Public Administration Research and Theory
Table 5
Cheating on the TAAS? Tenure Effects on Networking (Using ‘‘% Students Exempt’’)
Dependent Variable 5 Student Exam Pass Rates
Independent Variables
2 Years or Less
Network Management
0.4849
(1.48)
0.4547
(1.43)
0.0951**
(2.11)
0.1198
(1.81)
78.91**
(10.91)
0.49
26.04
399
Management Quality
Time in District
% Students Exempt
Constant
R2
F
N
Management Tenure Grouping
More than 2 Years/Less than 7
0.7525**
(2.34)
0.5729
(1.33)
0.0558
(1.37)
0.2055**
(3.06)
53.68**
(6.33)
0.48
30.63
445
7 Years or More
1.124**
(2.88)
1.441**
(3.06)
0.0981**
(2.32)
0.1401
(1.82)
75.23**
(8.06)
0.44
19.50
343
Note: Control variables not presented. Absolute value of t-scores in parentheses.
**p , .05 (two-tailed test).
years’ experience at the top (see the Appendix for a detailed summary of the tenure
variable). The same results are produced whether one splits these groups into four samples
as well. It is not necessarily the chosen years of the split that matter; rather, it appears that
something very interesting happens from the earliest, to the average, to the latest
management tenures that is causing the findings given in tables 5 and 6.9
The first set of results uses the total percentage of students exempted from the test to
measure TAAS ‘‘cheating.’’ New managers with two years of experience or less (column
one) appear to be constrained by their environment. Not only do networking and quality
have no impact on pass rates, but also the constant is very high and the environmental
measures have significant and strong effects (results not shown). The one managerial
characteristic that does distinguish these administrators, however, is the amount of time
they spent in the district prior to becoming superintendent. The real impact of a standard
deviation change in this variable (nine years) leads to a change in the TAAS pass rate of
almost one percentage point when controlling for all other effects. The percentage of
students exempted from the test is significant (one-tailed test), and its impact is large (one
standard deviation change leads to about a 1 percent score increase).
In the second group, more mature managers find that networking pays off, as does
‘‘cheating.’’ More interesting here is that the constant drops from seventy-eight in the
first group to fifty-three in the second group. This suggests that the samples are very different, something I cannot fully explain (the finding is not the result of outliers, nor does
it go away when the sample is split into smaller groups). It also backs up the previous
9
Joint F-tests were performed on these samples as well. The top two categories (tenure . two years) were
significantly different from the other two groups; however, the newest managers were not found to be statistically
different from the other two categories. Given the results, we should be cautious about interpreting the coefficients
for the newest managers, but I have a great deal of confidence in differences between the top two groups. The
interpretation of the results for the young managers is meant to be suggestive, not conclusive.
Juenke
Management Tenure and Network Time
Table 6
Cheating on the TAAS? Tenure Effects on Networking (Using ‘‘% Special Ed. Exempt’’)
Dependent Variable 5 Student Exam Pass Rates
Independent Variables
2 Years or Less
Network Management
0.5238
(1.61)
0.4524
(1.43)
0.0934**
(2.09)
0.2268**
(2.95)
79.69**
(11.09)
0.49
26.85
399
Management Quality
Time in District
% Special Ed. Exempt
Constant
R2
F
N
Management Tenure Grouping
More than 2 Years/Less than 7
0.8068**
(2.50)
0.6887
(1.60)
0.0610
(1.50)
0.2489**
(2.94)
52.14**
(6.17)
0.44
25.96
445
7 Years or More
1.133**
(2.88)
1.490**
(3.16)
0.1056**
(2.50)
0.0671
(0.75)
75.57**
(8.06)
0.43
19.11
343
Note: Control variables not presented. Absolute value of t-scores in parentheses.
**p , .05 (two-tailed test).
conjecture that results in districts with newer managers are being driven by inertia, while
more mature managers have a higher variance of impacts on their districts by utilizing
networking, experience, and test exemptions. Evidence from the third sample strengthens
this hypothesis.
The third column completes the picture by showing the substantial effects of quality
networking and good management on pass rates. More mature managers (keeping in mind
that they network less) appear to have developed more beneficial (or more efficient)
relationships with their networks than younger ones and continue to reap the benefits of
‘‘cheating.’’ A standard deviation change in networking is associated with a one percentage
point increase in test outcomes, as is a standard deviation change in quality when
controlling for all other factors. Here, ‘‘time in the district’’ regains its significance and
power.
Table 6 presents the results of the same model except that the dependent variable is
now the percentage of students in the district with a special education test exemption.10
The results are consistent and even more vivid than those from the previous analysis. At
each stage of tenure, network management increases its influence on outcomes, and young
managers find that it does little to change test results. Instead, these superintendents find
that cheating has a greater influence on scores, as the coefficient is significant and
produces a one point increase in pass rates when an average change in exemption rates
occurs.
10
Results for the other two ‘‘cheating’’ measures (% Limited English Proficiency [LEP] and % low income
exempted) are not reproduced in this article, but the low income measure results are consistent with those shown here,
while those for LEP appear to be operating without any influence from either management or cheating variables
altogether. This is consistent with the high correlation between ‘‘low income’’ cheating and the ‘‘special ed.’’ and ‘‘total
exemption’’ measures, as well as the lower correlations between ‘‘% LEP exempt’’ and the other three variables
(highlighted in the Appendix).
127
128
Journal of Public Administration Research and Theory
A similar pattern of inertia and discretion is seen as we move from one sample to the
next. In the middle group, network management begins to generate effects, while the
constant drops to 52 percent passing and some of the environmental variables begin to
show weaker effects (results not shown). As we move to the third sample, the results show
consistency with the previous table as networking and quality impacts grow in size
and significance, but here the effect of cheating becomes indistinguishable from zero
(results consistent with the previous version of this article using a two subsample split).
Again, it is clear that we are dealing with three different samples as the constant shoots
back up to 75 percent passing. I offer some possible reasons for this in the concluding
remarks.
SUMMARY CONCLUSIONS
Meier and O’Toole’s management model is validated and strengthened in every test
presented here. Results in table 1 show that networking has strong and independent effects
on test outcomes, even within a corrected (and reduced) sample. Results in table 2
demonstrate that when ‘‘time in the network,’’ ‘‘management tenure,’’ and ‘‘management
quality’’ are included in the models the effects of networking are larger than when they are
excluded. Second, management quality and time in the network have independent and
positive relationships with policy outcomes. Even though higher-quality managers with
longer tenure work the network less (table 3), superintendents who have been in control of
the bureaucracy longer gain more positive outcomes through networking than those who
are newer to their networks (tables 5 and 4). Networking and tenure interact to weight the
amount of leverage a manager has on her or his environment. Interaction with the network
nodes over time increases the quality of those relationships, thus producing better results
(with less interaction).
There are two surprising but provocative findings as well. The first involves a measure
that was suggested by the (small) literature on alternative output strategies and was found to
be significantly linked to superintendent tenure. Even though superintendents with longer
tenures may ‘‘cheat’’ (or allow cheating) as frequently as others do, they do not reap the
benefits of this behavior as well as those newer to the school district. Thus, newer managers
get more returns for non-networking behavior, possibly because they have not established a
relationship with their networks and get no investment returns on working the contacts,
even though they may do it more often. Other scholars should pursue tests for alternative
strategies among newer managers in networks, but presently the results speak volumes as to
what can be uncovered with different measures and methods of investigation.
Second, there is something very important happening in the split samples presented
here (and they appear to be truly representative of an underlying phenomenon as opposed
to being an aberration of the split samples). There is a reason new managers are new to
their district. Either they are new because the previous manager was doing a poor job and
was fired, thus placing them in a poorly performing district that moves on inertia and is
heavily reliant on bureaucratic and clientele resources (suggesting an undeveloped
network), or they are new to the district because the previous manager was doing well and
was promoted, producing possibly contradictory circumstances. Clearly we know the latter
scenario occurs, but generally speaking it is not driving the results reported here. Thus the
data speak to the frequency of the former circumstance and suggest that the factors driving
performance vary over the course of a superintendent’s tenure.
Juenke
Management Tenure and Network Time
A new manager in a district has a small initial effects on performance because the
momentum of the organization overwhelms her or his networking efforts. But over time,
longer-tenured managers (more than seven years here) exert their influence over the
organization, and the momentum shifts to managerial inertia. There is a reason that these
managers still have their jobs, and on average it is because they have learned how to tap
their network resources and continue to produce quality outcomes. The evidence is
suggestive rather than conclusive in regard to these strong statements, but it appears
consistently in the results.
The surprising confirmation of the idea that system development can be distinguished
by managerial tenure endorses the proposal of network evolution and suggests that these
networks might be at different stages of progress. The thought that top-level management
tenure serves as an indicator of network development is stimulating because it provides
information about the network itself, not simply the actors working in it.11 Like the
‘‘cheating’’ finding, this is a surprising result and requires further investigation, but the
current forum is insufficient. Clearly, though, network actors who have been working
together for a longer period of time have overcome much of the environmental inertia that
plagues new managers in new networks.
Last, the evidence soundly rejected the argument that mature networks were
functioning at diseconomies of scale. More experienced managers appear to have built
relationships on performance and reputation and, by definition, contribute heavily to the
stability of the system while retaining the benefits of group behavior. The ‘‘time in district’’
and ‘‘tenure’’ of the top-level managers appear to be rich with information; future studies
should exploit these seemingly innocuous indicators and test for weighted network effects
to further explore the dynamics of cooperative behavior in public organizations.
APPENDIX
Describing the Data Set
Summary Statistics
Variable
Obs
Mean
Std. Dev.
Minimum
Maximum
Student Pass Rate
Networking Measure
Managerial Quality
Time in District
Superintendent Tenure
% Special Ed. Exemption
% Total Exemption
% Low Income Exempted
% LEP Exemption
4293
1682
4227
1791
1791
2781
2781
2769
2781
74.73729
0.0140538
11.60675
9.174539
5.246259
7.005645
9.433261
14.31083
1.278641
12.45746
1.000544
1.012744
9.209111
4.928431
4.977699
5.878383
8.202805
2.371469
14.3
2.29
0.037
0
0
0
0
0
0
100
2.92
17.14
41
38
32.5
40.8
51.2
26.5
11
Again, in the U.S. educational context, it is the superintendent’s tenure and networking activity that pertain to
educational performance (as the theory and evidence demonstrate). In other settings, it may be the tenure and
networking activity of that particular organization’s top-level manager that may indicate network development.
To be clear, the theory states that the networks help produce benefits; empirically, however, this is indicated by the
manager’s network activity.
129
130
Journal of Public Administration Research and Theory
Correlations between Key Variables
(obs 5 1187)
network
m qual.
time
tenure
% total
% sp. ed
% l.inc
% lep
network
m qual.
time
tenure
% total
% sp. ed
% l.inc
% lep
1.0000
0.0689
0.0973
0.0737
0.0324
0.0458
0.0202
0.0110
1.0000
0.0340
0.1495
0.0471
0.0212
0.0565
0.1199
1.0000
0.4840
0.0566
0.0703
0.0839
0.0061
1.0000
0.0322
0.0504
0.0840
0.0220
1.0000
0.8665
0.8837
0.5350
1.0000
0.7866
0.0661
1.0000
0.4532
1.0000
Detailed Summary: Management Tenure
Percentiles
1%
5%
10%
25%
50%
0.5
1
1
2
4
75% 7
90% 12
95% 15
99% 23
Smallest
0
0
0
0
Largest
35
36
37
38
Obs
Sum of Weight
Mean
1791
1791
5.246259
Std. Dev.
4.928431
Variance
Skewness
Kurtosis
24.28943
2.066745
9.104882
REFERENCES
Agranoff, Robert, and Michael McGuire. 1999a. Big questions in public network management research.
Paper prepared for the 5th National Public Management Research Conference, Texas A&M
University.
———. 1999b. Managing in network settings. Policy Studies Review 16:18–38.
Alter, Catherine, and Jerald Hage. 1993. Organizations working together. London: Sage.
Barber, Bernard. 1983. The logic and limits of trust. New Brunswick, NJ: Rutgers University Press.
Barzelay, Michael. 1992. Breaking through bureaucracy. Berkeley: University of California Press.
Bohte, John, and Kenneth Meier. 2000. Goal displacement: Assessing the motivation for organizational
cheating. Public Administration Review 60:173–82.
Hoffman, Andrew. 1999. Institutional evolution and change: Environmentalists and the U.S. chemical
industry. Academy of Management Journal 42:351–71.
Howlett, Michael. 2002. Do networks matter? Linking policy network structure to policy outcomes:
Evidence from four Canadian policy sectors 1990–2000. Canadian Journal of Political Science
35:235–67.
Kickert, Walter J. M., Erik-Hans Klijn, and Joop F. M. Koppenjan. 1997. Introduction: A management
perspective on policy networks. In Managing complex networks, ed. Walter J. M. Kickert,
Erik-Hans Klijn, and Joop F. M. Koppenjan, 1–13. London: Sage.
Klijn, Erik-Hans, and G. R. Teisman. 1997. Strategies and games in networks. In Managing complex
networks, ed. Walter J. M. Kickert, Erik-Hans Klijn, and Joop F. M. Koppenjan, 98–118. London: Sage.
Juenke
Management Tenure and Network Time
La Porte, Todd, and Daniel Metlay. 1996. Hazards and institutional trustworthiness: Facing a deficit of
trust. Public Administration Review 56:341–47.
Meier, Kenneth, and John Bohte. N.d. Inside the bureaucracy: Principles, agents and bureaucratic
strategy. Unpublished MS.
Meier, Kenneth J., and Laurence J. O’Toole Jr. 2001. Managerial strategy and behavior in networks: A
model with evidence from U.S. public education. Journal of Public Administration Research and
Theory 11:271–94.
———. 2002. Public management and organizational performance: The effect of managerial quality.
Journal of Policy Analysis and Management 21:629–44.
Milward, H. Brinton, and Keith G. Provan. 1998. Principles for controlling agents: The political economy
of network structure. Journal of Public Administration Research and Theory 8:203–21.
———. 2000. How networks are governed. In Governance and performance: New perspectives, ed.
Carolyn J. Heinrich and Laurence E. Lynn Jr., 238–62. Washington, DC: Georgetown University
Press.
Osborne, David, and Ted Gaebler. 1992. Reinventing government: How the entrepreneurial spirit is
transforming the public sector. Reading, MA: Addison-Wesley.
O’Toole, Laurence J., Jr. 1997. Implementing public innovations in network settings. Administration and
Society 29:115–38.
O’Toole, Laurence J., Jr., and Kenneth J. Meier. 2003a. Plus ça change: Public management, personal
stability and organizational performance. Unpublished MS.
———. 2003b. Public management in intergovernmental networks: Matching structural networks and
managerial networking. Unpublished MS.
Provan, Keith G., and H. Brinton Milward. 1995. A preliminary theory of interorganizational
effectiveness: A comparative study of four community mental health systems. Administrative
Science Quarterly 40:1–33.
Reies Louis, Meryl. 2001. Surprise and sense making: What newcomers experience in entering unfamiliar
organizational settings. In Classics of organization theory, 5th ed., ed. Jay Shafritz and J. Steven Ott,
337–90. London: Harcourt Publishers.
Termeer, C. J. A. M., and Joop F. M. Koppenjan. 1997. Managing perception in networks. In Managing
complex networks, ed. Walter J. M. Kickert, Erik-Hans Klijn, and Joop F. M. Koppenjan, 79–97.
London: Sage.
131