The Micro-Dynamics of Network Leverage: Implications for
Change Agents External to an Organization
Ken Frank! and Ran Xu", William R. Penuel?
1: College of Education, Michigan State University, MI
2: College of Engineering, Virginia Tech, VA
3: College of Education, University of Colorado, CO
*; Correspondence to ranxu@vt.edu
Abstract
Much of the impact of a policy depends on how it is implemented, especially as mediated by
organizations such as schools, hospitals, or law enforcement agencies. Furthermore,
implementation depends on each organization’s capacity to absorb innovations based on its
culture, routines, and leadership. Here we extend the concept of absorptive capacity to include
the intra-organizational social dynamics that occur during the diffusion or implementation of an
innovation. In particular, we attend to the potential for intra-organizational polarization along
pre-existing lines. We then use agent based models to examine the interplay of intra-
organizational social dynamics and the external change agent who seeks to direct the
organization by introducing venues which contain information encouraging specific behaviors.
We find that when organizational members are salient to one another, external change agents
who attempt to direct organizations by introducing strongly oriented venues (e.g., professional
development emphasizing specific practices) may unintentionally accentuate existing cleavages
in the organizational network, inhibiting full implementation of the immediate policy as well as
reducing organizational capacity to implement future innovations. Thus the external change
agent should consider the short term interaction with the intra-organizational social dynamics as
well as the organization’s longer term absorptive capacity.
Keywords: Agent-based Model, Network Leverage, Influence, Selection, Extemal Change,
Venues.
Page 1
1. Introduction
Most public policies are implemented by organizations which develop expertise and allocate
internal resources to deliver services or programs (DeCarolis & Deeds, 1999; Kilduff & Tsai,
2003; Wemer, 2004; Kilduff et al., 2006; Scott, 2008; Weiss, Bloom and Brock, 2014). For
example, educational policies are implemented by schools which change their instructional staff
or curricular materials as teachers ultimately deliver instruction to students (Bidwell & Kasarda,
1985; DeCarolis & Deeds, 1999; Weiss, Bloom and Brock, 2014). Similarly, health policies are
implemented by hospitals, insurers, and associations (Barley, 1990; Poon et al., 2004; Watt et al.,
2005); welfare-to-work programs are implemented by states and pro offices (Weiss, Bloom
Ae Soe and immigration policy is implemented through law enforcement agencies
(Ridgley, a
An organization’s specific ability to implement a policy or innovation can be described in
terms of its absorptive capacity, defined as “an ability to recognize the value of new information,
assimilate it, and apply it to commercial ends.” (page 128, Cohen and Levinthal, 1990).
Absorptive capacity typically includes elements of the organization’s communication with the
extemal environment (such as mediated by leadership), existing expertise that relates to an
innovation, and the character and distribution of expertise within the organization. But here we
emphasize that each of these elements can be enacted and enhanced through intra- organizational
networks. That is, formal leaders can more effectively guide the organization to new behaviors
when they are well-established in the informal network as well (Moolenaar, Daly, and Sleegers,
2010; Moolenar and Sleegers, 2015; Hopkins et al., 2013). Similarly, expertise can be effectively
cultivated and distributed when conveyed through informal networks. In fact, an organization’s
culture can be characterized partly in terms of its capacity to distribute relevant resources
through networks (Frank et al., 2015).
In this study we emphasize that the networks through which absorptive capacity is manifest
are themselves dynamic. Networks not only convey information and noms, but networks
themselves are modified by diffusion of information and implementation processes (Xu & Frank,
2016). In this sense, an organization’s capacity is partly a function of the resilience of its network
to sustain information flows before, during, and after the diffusion of new information or
practices within the organization. Thus we attend to the intra-organizational social dynamics that
affect each member’s response to an externally generated policy shock.
In an example of the social dynamics of Sbonmptive capacity, Frank et al. (2013) found that
the pressures and institutions associated with No Child Left behind (NCLB) contributed to
polarization in instructional practices among teachers within schools. This occurred as teachers
were initially affiliated with cohesive subgroups, or cliques, within schools that featured different
orientations and receptivity to NCLB related practices. Under the pressure of NCLB some
subgroups became more aligned with instructional practices affiliated with NCLB and others less
so as they lacked expertise associated with and orientation to NCLB practices. Thus schools
became more polarized; ironically, the national policy intended to equalize opportunity across
children contributed to the unintended consequence of uneven instruction. Such unevenness can
ultimately create organizational challenges of coordination and collaboration beyond the focuse
of an intervention associated with NCLB (Woodward, 1965; Bidwell, 1965; Thompson, 1967),
broadly contributing to inequitable opportunities for students within schools (Frank et al,
forthcoming), as well as between schools which experienced differing levels of dysfunction
(Frank et al., 2015).
In this study, our ultimate goal is to inform the action of change agents by revealing how the
effects of their actions depend on the intra-organizational micro-dynamics of the organization
they seek to change. In particular, we show how a forceful change agent can exacerbate pre-
existing intra- organizational rift lines. This can polarize the organization, ultimately inhibiting
the implementation of the change agent’s intended policy.
Page 2
In the next Section, we describe the general scenario we consider and then present
hypotheses based on conventional thinking about change agents. We then develop agent-based
models expressing the extemally generated policy or incentive in terms of information to which
organizational members are exposed and then explore the internal dynamics in terms of the
salience of the organization to its members. A fter that, we experiment with our system in terms
of the actions of extemal change agents seeking to direct the organization to particular behaviors.
2. Theoretical Framework
2.1 General Scenario
We consider a set of actors in an organization; each actor engages in certain behaviors, such
that the behaviors or beliefs contribute to the organization outputs. The actors start with different
behaviors but have common utility functions determining how they are influenced by network
partners and how they choose network partners. Given this baseline we will show that
polarization occurs when the organization is only moderately salient in the sense of actors being
influenced by other members and exhibiting a preference for like-minded others.
Next, we consider agents external to the organization who seek to change the organizational
outputs. For example, a change agent might seek to increase general endorsement of the teaching
practices consistent with a particular educational policy. By definition, actors extemal to an
organization cannot change organizational outputs by changing their own behaviors.
Furthermore, we assume external agents cannot influence behavior through extensive direct
interactions with members of the organization, which would effectively bring the external agent
inside the organizational boundary (Williamson, 1981). Thus the external agents must create
mechanisms for introducing information or influencing behaviors within the organization. We
think of these as sustained shocks, or venues. For example, public school districts may try to
influence teachers’ practices by creating sustained professional development (Garet et al., 2001;
Desimone et al., 2002), the AHA (American Hospital Association) may provide seminars and
leadership development programs for hospitals to improve the implementation of health care
reform (AHA, 2010), or jobtyists may provide venues for interaction among members of the US
senate (Fiorina & Abrams, 2008). Through these venues extemal agents can expose
organizational members to information intended to affect the behavior of the members of the
organization.
Critically, members of the organization must choose to participate in the venue (attend to the
event or visit the website) based on the attributes of the venue. In this sense the venues create
social spaces which increase the probability that any two actors participating in the venue will
interact (Feld, 1981). As such participants in the same venues have increased probability of
sharing information or capacity to impose a norm on one another.
Our research question then concerns what happens when an extemal change agent
introduces a venue containing information Supponing a policy goal into the social dynamics of
an organization? For what internal dynamics is the change agent able to direct the whole
organization in the desired direction of the venue’s orientation, and under what conditions do the
micro social dynamics generate unintended consequences?
2.2. Hypotheses based on C onventional Beliefs of External C hange Agents
Before we investigate how the intemal social dynamics of organizations affect their systemic
response to exogenous shocks, we consider conventional thinking about how to use exogenous
shocks to shape an organization. To begin, there is high face validity for attempts to change
Page 3
organizations by introducing shocks which have valences in the desired direction. For example,
if one wants to teachers to teach in a different way one introduces professional development (the
exogenous shock) that trains teachers in the desired practices (Garet et al., 2001; Desimone et al.,
2002; Weiss, Bloom and Brock, 0214). The same holds for any form of professional
development. Similarly, political parties hold rallies and create media events (the exogenous
shocks) to push the electorate towards the beliefs of the party, with the party’s ultimate goal of
moving the electorate enough to gain political power (Heaney & Rojas, 2015). This approach to
systemic change can be summarized in a baseline hypothesis:
Hi: Change agents can direct an organization to a desired policy goal by introducing a venue
conveying information supporting that goal
Note that although extemal agents may intuitively seek to efficiently allocate their resources, the
hypothesis is specified independent of the intemal dynamics of the social system. Thus we have
the corollary that the effect of an exogenous shock will be more dependent on the strength or
valence of that shock than on the internal social dynamics of the system.
Our theoretical development suggests a second hypothesis inhering in the intra-
organizational social dynamics. A bsorptive capacity rightly attends to the capacity of the
organization’s existing internal structures to diffuse information and practices. But an
organization’s capacity to absorb an innovation also depends on the internal dynamics manifest
during absorption; during implementation, networks, status, and the distribution of information
are likely to change in ways that can support or impede an organization’s capacity to absorb an
external innovation. For example, teachers with specific expertise in whole senguage instruction
may gain status if their school adopts a whole language pedagogy. If some resent that elevation
because of personality conflicts with the specific teachers or the general differentiation of status,
then the school may polarize during implementation (Glidewell et al., 1983; DePaulo et al.,
1983). Such dynamic polarization would limit the extent of implementation beyond what inhered
in the static characteristics of the school at the time of implementation. Thus our second
hypothesis is:
Ho: The capacity of the change agent to direct the organization to a desired goal depends on the
intra- organizational social dynamics which can contribute to polarization.
The key is that the intra-organizational social dynamics can affect how shocks are distributed
throughout a system, ultimately affecting the eae response to the shock (Frank & Fahrbach,
1999; Xu and Frank, 2016). Furthermore, the shock itself can accentuate extant pattems of
interaction which then shapes how the system responds to that shock as well as its future
capacity to distribute information. In the next Section we tum to more formal models of intra-
organizational social dynamics so that we may explore how these dynamics affect systemic
responses to exogenous shocks.
3. Models
In this section we illustrate the basic models we use for the agent-based simulations
(Wilensky & Rand, 2015). The key point is that to study the organizational response we must
examine how it emerges out of individual behavior. In particular, we consider how actors in the
organization deliberately choose their behavior (influence process) and with whom they interact
(selection process) to maximize utilities which reflect their desire to reduce transaction costs to
access new information under different levels of the salience of the organization. We then initiate
Page 4
our simulations with two subgroups with different behaviors (representing baseline
differentiation within most organizations), and we experiment to learn which types of strategies
(such as created by change agents) exert the most leverage on the organization given the intra-
organizational social dynamics.
3.1 Theoretical Basis for the Model
Drawing on economic literature, policies are implemented by rm incentives for
individuals or organizations (Schneider & Ingram, 1990; Gneezy et al., 2011), and thus changing
the behavior of organizations as corporate actors or as a collective of taeda Our
assumption is that these changes in incentives induced by policy are usually complex and context
specific, and thus they are not immediately comprieasible to all of the members within the
organization (Williamson, 1981). For example, teachers are uncertain about the implications of
incentives associated with the Common Core for their choices of curriculum and instructional
practice (Cobum et al., 2016). With high uncertainty/complexity i in the environment, there are
high transaction costs/risk to access info rmation. This increases organizational salience, as actors
rely on their organizations/immediate subgroups to reduce the transaction cost/risk to access new
ormation. Furthermore, actors will also align their behavior with their immediate subgroup
members to maintain the organization/group membership which can be protective in uncertain
conditions (Lin et al., 2001). In contrast, with low uncertainty/complexity in the environment,
transaction cost/risk is relatively low, and the salience of the organization is low, actors will
directly seek sources for new information re incentives of the policy, and change their
behavior based on this new information (Fi ae aa, 2011). Thus actors balance their networks
ee those who engage in similar rhe and those “tip possess non-redundant
information.
3.2 Formal Specification of the model
In this section, we formally define our agent based models. Specifically, there are three
interdependent processes involved, namely information seeking, behavior change, and network
change:
1. Actors will seek information from other actors to gain a better understanding of the
incentives of the policy;
2. Actors will change their behaviors according to information they access, as well as the
behaviors of those from whom they seek information;
3. Actors will change their network relations in order to access non-redundant information.
Information Seeking Process. Each actor has an information list, which consists of unique
pieces of information they possess. Each piece of information makes a unique contribution to the
actor’s understanding of the incentives of the policy, which could be either consistent or
inconsistent with the intended direction of the policy. In each round, each actor is considered as
an ego who seeks information from those in their networks (alters), who will randomly provide
one piece of information in their possession to the ego. If the information is new to the ego, then
ego will add this piece of information to its own information list; if the information is redundant,
then it will not go into ego’s information list.
Influence Process. Each round, actors will choose their behavior according to their previous
behavior, new information they receive, as well as the behaviors of their network members.
Page 5
Specifically, we choose a variation of Friedkin and Johnsen’s (1990) influence model and
specify our model as (see Frank & Fahrbach, 1999):
Wad
¥,=C-@)y, 1,4 fo uaa (1)
Wort
Where yit represents the behavior of actor i at time t. yit-1 represents actor i’s behavior at time t-1,
and lit-1 represents the effect of information on behavior. We define positive information as
information that is consistent with the policy intent; and we define negative information as the
information that is inconsistent with the policy intent. The second term represents the mean
behavior of actor i’s network members at time t-1, where wt 1=1 if actor] is a member of the
network of i at time t-1, 0 otherwise.
Given this model, a represents the salience of the organization on the actor’s changes in
behaviors. For a high value of a egos respond strongly to the behaviors of their networl
members. This is above and beyond the information the ego obtained as a result of interacting
with organization members — large a represents a normative effect of others in the organization.
This normative effect can be due to ego’s identification with others in the organization (Frank,
2009), because of a shared sense of fate (Portes and Sensenbrenner, 1993) or a sense of shared
mission in the organization (Williamson, 1981). Consider teachers in the NCLB example, where
uncertainty about the implications of NCLB for a given school could be high (Penuel et al.,
2009). In this case the school as an organization creates a strong filter of the effect of outside
institutions (Frank et al., 2013). Therefore, teachers are inclined to conform to the norms in their
intra-organizational networks with whom they share a common fate, as well as local conditions.
When a is low, ego’s behavior is primarily a function of ego’s prior behavior (yit-1) modified
by new information (1) to which ego is exposed. Importantly, this information can be conveyed
by members of the organization (see below) but in this capacity other members of the
organization carry no more weight than any other in ego’s network — the organization member is
simply a vector for conveying the information independent of the organizational context. For
example, when there is considerable tumover among school faculty, teachers loose allegiance to
the school (Ingersoll, 2001; Bryk and Schneider, 300), As aresult they act more atomistically,
responding to more to individual incentives conveyed through information provided by
individuals within or outside the school.
Selection Process. Each round, agents decide with whom to establish a network tie, assuming an
actor’s out-degree is constant. In this case, the salience of the organization as represented by a is
associated with the standard homophily term in a network model |yit-1 — yjt-1|. That is, actors
prefer to interact with others of similar behavior when they have a strong identification or
affiliation with others in the organization. When they do not, they seek merely to interact with
others who can instrumentally shorten their path lengths to potentially new information (Frank
and Fahrbach, 1999). This reduction in path lengths is represented by (mpjjt-1 -1) which occurs as
a result of an ego’s choice of with whom to interact given Ee" s current network. For example, if
the network distance between agent i and agent j is 3 (mpj=3), and network distance between
agent i and agent k is 5 (ipi=5) then agent i will gain more utility by connecting to k instead of
j, because k is more likely to have new information regarding the incentives of the policy fori.
Uj. (Mp Vie ja) =e) (mp; 1) ~a Vir Vita | (2)
Page 6
With high transaction costs to access information, the salience of organization a is high, and
actors will prefer similar others to reduce the transaction costs/risks. When transaction costs/risks
is low, the salience of organization a is low, and actors will prefer distant others who have
potential for new information regarding the incentives of the policy. Here again if we consider
the NCLB example, the uncertainty in the environment is high, so teachers will gain more utility
by selecting like-minded others, as a result teachers are more likely to select others with similar
teaching practices (Penuel et al., 2009). Combining with influence process above they eventually
form subgroups with homogeneous practices within the subgroup and heterogeneous practices
between subgroups.
After actors make initial selections, in each round they also decide whether they want to
maintain the network ties they have made before. As actors are motivated to seek new
information, we specify that when actors access redundant information from a network member,
there is higher probability that the actor will re-evaluate and dissolve the network tie. This
decision process is a function of how many consecutive times an actor is exposed to redundant
information from the network member:
x
P.a4 (3)
Where Pit is the probability that actor i will maintain the network tie with j at time t, Aisa
constant between 0 and 1, and x is a integer between 0 and +c, which indicates how many
consecutive times actor i is exposed to redundant information from j. So if actor i accesses new
information from j, x will be 0 and the probability that i will maintain a network tie with j at time
tis 1. The first time actor i is exposed to redundant information from j, x will be 1 and the
probability to maintain the tie at time t becomes A. If actor i continues to be exposed to
information from j, x will increase by 1 each time until the tie is discontinued or actorj provides
new information to actor i. Note that even if the connection is discontinued, it may be resumed if
a is high and actor i andj already share similar behaviors because of previous interactions.
The rates of influence — k. Note that in the influence process and selection process, we use the
same parameter a to represent the salience of the organization that can affect both actors’
influence and selection processes. However, the relative rates at which influence and selection
occur can vary. For example, the rate of influence would be high relative to that of selection if
actors may rapidly adopt the behaviors of those in their network but are slow to change network
ties based on homophily. Therefore we express the rate of influence relative to that of selection
as k (O<k<1), and incorporate k into the influence process in (1):
raj
Ww
=k 1. +k
Yq =U kay, ly tha 4)
jt
Generally, when k > 0 actors retain only their previous behaviors — influence occurs slowly
relative to selection; as k increases the process of influence occurs faster relative to selection, and
when k > 1 influence occurs at the same rate as selection.! If k is small, then actors maintain
balance by focusing more on selecting members who engage in similar behaviors than
conforming in behavior to those of network members. Thus our models allow us to express the
* Tf selection does not occur and therefore the network does not change, the system converges to the same end point
for 0<k<1 (Frank and Fahrbach 1999).
Page 7
system in terms of the interplay of between influence and selection using two parameters:
salience of the organization («) and the rate of influence relative to the rate of selection (k).
4. Simulation Methods
We then initiate our simulations with two subgroups with different behaviors (representing
baseline differentiation within most organizations), and we experiment to leam which types of
implementation strategies created by change agents exert the most structural leverage on the
system. We also examine how effects of these strategies interact with the salience of the
organization and the rates of influence relative to selection.
Our primary focus is on efforts of agents external to the organization to change the behavior
of organization members or the organization as whole. Interpreting the change agent in our
models, the external agent does not exert leverage on the organization by changing its own
behavior or network. Instead, the external agent is limited to creating venues or events that
express a particular orientation or behavior to which members of the organization can be
attracted and exposed. For example, national policymakers might seek to introduce professional
development programs into schools and districts. The providers of these programs do not enter
the schools as full agents of the schools, seeking to establish network ties and change their own
behaviors. Instead the sponsors seek to change behaviors by providing information or
representing national norms, Furthermore, they provide opportunities for subsets of teachers to
coven. become exposed to one another, and perhaps create new network ties (Spillane et al.,
Given our discussion above, the choice for the external change agent concems how strongly
to express a position in the venue the agent creates. A strong position may represent the policy
well, but may create unintended consequences in terms of network dynamics and ultimately the
adoption rate in an organization. Therefore, the change agent must choose the venue with an eye
yard the attendant network consequences as well as the direct intended consequences for
ehavior.
We express the position of the change agent in terms of the valence of the events the agent
introduces into the system. We describe two types of venues, one containing information
strongly supporting a policy which we call positive venues (the effects of negative venues can be
understood by symmetric arguments conceming information that supports an alternative policy
and behaviors), and the other containing an almost equal balance of positive and negative
information supporting a policy.
4.1 Simulation Process
We perform agent based simulations in Netlogo 5.2.0 (Wilensky, 1999). Specifically, in each
round: (i) Each actor randomly seeks one piece of information from each of his/her network
members; (ii) Each actor decides whether to end the current tie and to start new tie based on the
probability equation in [3]; (iii) If new ties are to be established, an actor calculates the utility of
establishing a tie with each of other actor based on the selection equation [2]; (iv) Each actor
then establishes ties with other actors with highest utilities, holding the out-degree (number of
others identified as network ties) for each actor constant;” (v) As actors select with whom to
form network ties, they are influenced by the new information they receive as well as the mean
? For example, if an actor is initialized with 3 out-going ties, then in each round it retains the number of out-going
ties as 3.
Page 8
behavior of their network members, and adjust their behaviors based on the influence model in
[4]. Actors only update their behavior when they receive new information from their alters,
otherwise actors will retain their prior behaviors. The information I is set to be 1.05 for positive
information (consistent with policy effort) and 0.95 for the negative information (inconsistent
with policy effort).
For each experiment described below we stop the simulation when (1) every actor obtains all
pieces of information in the system, (2) or after 600 iterations? We set 2 to be 0.8, and we vary
the uncertainty salience of the organization (a) from 0.3 to 1 by intervals of 0.05, and chose the
relative rate of influence (k) to be 0.1 or 0.5. In each configuration we simulated 200 rounds,
with a total of 200*15*2=6000 simulations.
4.2 Experiment Condition
4.2.1 Baseline Condition
In the baseline condition we initialize each network as follows: (1) we create two subgroups,
each consists of 10 actors, one we called a positive group and one we called negative subgroup;
(2) we create relatively dense networks within subgroups and sparse networks between
subgroup, which results in a density of 0.2 and clustering- coefficient around 0.4; (3) behaviors
within aroun follow a normal distribution with standard deviation 1; for the positive
subgroup the mean behavior is 12, and for the negative subgroup the mean behavior is 8; (4) for
actors in the positive subgroup, each actor starts with 3 random pieces of positive information
and 2 random pieces of negative information. While for actors in the negative subgroup, each
actor has 2 random pieces of positive information and 3 random pieces of negative information.
In this way information is set up to be aligned with the behavior of the actor. The unique pieces
of information are randomly drawn from a total of 15 pieces of negative information and 15
pieces of positive information.
4.2.2 Effects of Venues Created by Extemal Change A gents
We initiate each simulation with 30 rounds given the baseline dynamics established by equations
[2] through [4]. We then introduce shocks as venues with a particular set of information whose
sum we refer to as a valence. Aligning with our assumptions about extemal change agents, these
venues do not have capacity to change their behaviors or network ties. Actors in the organization
can select these venues to be part of their networks based on the selection process as in equation
[2], and they can be influenced by venues based on the influence process as in equation [4]. The
information contained in the positive valenced venue is 13, one within-group standard deviation
higher than the initial mean behavior of the positive group. The positive valenced venue also
contains 3 pieces of positive information that are new to the organization, representing the
external information exerted by change agents that supports the policy goal.
The baseline behavior of the near-neutral venue is fixed at the mean behavior of all the actors in
the system at time 30. The near-neutral venue also contains 10 equal pieces of positive
information and negative information that are already in the system. We then move the venue
slightly away from neutral by including 3 pieces of positive information that are new to the
organization.
4.2.3 Key Outcome measures
3 We stop at 600 iterations because in a baseline experiment where we start from random networks, in most
simulations actors obtain all pieces of information in the system within 500 iterations.
Page 9
We are interested in two outcome measures. The first measure is the probability of full
information diffusion (Rogers, 2010). It is calculated as the percentage of the total simulations in
which all actors obtain all pieces of information in the organization. This represents the extent to
which actors have acquired all the information to evaluate the incentives of the policy. The
second measure is the mean behavior of members of the organization. It is calculated as the
mean behavior of the actors as simulation ends. This represents the extent to which actors have
adopted behaviors consistent with the policy.
5. Results
Probability of Full Information Diffusion (k=0.1) Probability of Full Information Diffusion (k=0.5)
Probability
a o
ry
Probability
\
\ .
\ _
2° bi e ae
selec) snene()
: :
Figure 1. Simulations initiated with two subgroups for different rates of interpersonal influence (k).
Probability of full information diffusion decreases with salience (a), with more dramatic decrease when a
positive venue is introduced. (A) Probability of full information diffusion vs salience when rates of
influence is low (k=0.1); (B) Probability of full information diffusion vs salience when rates of influence
is high (k=0.5).
5.1 Diffusion of information
In Figure 1 the black lines represent conditions in which no extemal venues are
introduced establishing a baseline against which to compare scenarios in which venues are
introduced into the system. The baseline condition shows that the probability of full information
diffusion (all the actors obtain all unique pieces of information in the system) decreases as
organization salience (a) increases. For low values of « actors establish an integrated network of
interaction with others of similar or different behaviors, thus allowing the diffusion of
information across the organization. On the other hand, for high values of a the network becomes
factionalized, inhibiting the diffusion of information between factions. These trends apply
regardless of the relative rate of influence (k), although when influence is larger as on the right,
the tendency for factions is mitigated for .8< a <.9 and therefore more information is diffused.
Page
10
Tuming to the effects of venues, as a increases, the red lines show a significantly sharper
decrease (relative to baseline black) in the probability of full information diffusion when a
positive venue is introduced (for a > .45), with the probability decreasing to near zero when
a> .8. This is because the positive venue attracts actors of like behaviors (the yellow dots),
accentuating their predispositions, and thereby distancing them from those of the initially
opposite behavior. On the other hand, the system is more able to diffuse full information when a
near-neutral venue is introduced (for a > .9 the equilibria for the near-neutral event are
comparable to those for the baseline because the near-neutral event cannot compensate for the
effects of homophily for een high organizational salience). Thus is because the near-
neutral venue attracts actors of either orientation, providing opportunities for them to continue to
exchange information and influence one another.
5.2 Change in Behavior
Mean Behavior for Each Subgroup (k=0.1) Mean Behavior for Each Subgroup (k=0.5)
zs 2 {= ovina
near-neural verue
— positive verve
ee ae
nik a ",
Behavior
L
03 04 oS 06 O07 0&8 09 10
Salience (a)
A
Salience()
8
Figure 2. Simulations initiated with two subgroups for different rates of interpersonal influence (k).
Divergence of behavior between subgroups increases with salience (a), more dramatic divergence
emerges when a positive venue is introduced. (A) Mean behavior of each subgroup vs salience when rate
of influence is low (k=0.1); (B) Mean behavior of each subgroup vs salience when rates of influence is
high (k=0.5).
Figure 1 shows that the trends for diffusion are not dramatically altered when influence is
presence (k=.5). But Figure 2 shows a more complex interaction between influence and the
distribution of behavior between subgroups. In the baseline condition (represented by the black
lines) the subgroups (x for the positive subgroup, o for the negative subgroup) maintain
separation across values of a when influence is low (k=.1). But when influence is high (k=.5) the
subgroups are more similar to each other for moderate levels of a. This is because members of
different subgroups are able to influence one another and maintain similarity for low to moderate
organizational salience (a<.5). It is only for high salience (a>.5) that actors are drawn to
interactions within their subgroups to generate polarization in behavior.
Page
11
The effects of the venues on behavior are shown with the colored lined in Figure 2. When
influence is low, the pattern of separation between the subgroups for a near-neutral venue is
similar to that for baseline, except all behaviors are moderately elevated because of the 3 pieces
of positive information embedded in the near-neutral venue). The red lines show stronger
separation and at lower values of a as members of the positive subgroup are attracted to the
positive venue, and as a result become more extreme in their behaviors. This also creates a social
distance between the positive and negative subgroups, and as a result there is little
counterbalance to the normative pressures within the negative subgroup, making it more
extreme.
When influence is stronger, as on the right, the subgroups maintain integration for low
salience (a < .5) in baseline and for positive or near-neutral venues. Thus influence can
compensate for the tendency for homophily to drive subgroups apart provided salience is low to
moderate. But the mitigating effects of influence on polarization is diminished for high salience
(a alpha > .5) in which case polarization occurs regardless of the presence or strength of valence
of a venue (although the separation between the subgroups is smaller across conditions for high
influence than forlow).
Across our results the effect of the event depends on the salience of the organization.
When salience is low (a < .5) an external change agent introducing a positive venue can shift the
mean behavior of the organization without inducing polarization. On the other hand, when
salience is high (a > .5) a positive event accentuates polarization, ultimately inhibiting the
diffusion of information and constraining the overall change in behavior (increases in the
behavior of one subgroup are offset by decreases in the behavior of the other). Finally, the
relative rate of influence (k) amplifies the distinction between high and low salience.
6. Discussion
Our context applies to the micro level action of the agent who seeks to change the behavior
of an organization. This might apply to a national policymaker who seeks to influence schooling
by changing the instructional practices of teachers (DeCarolis & Deeds, 1999; Porter et al., 2011;
Weiss, Bloom and Brock, 2014). We make what we believe to be an authentic definition of an
agent who truly extemal to the — anization, and therefore must exert leverage by creating events
(e.g., professional development) that will contain information supporting the intended change
(Garet et al. 2001; Desimon et al., 2002). It is then incumbent upon the members of the
organization to attend the event, absorb its information and distribute it throughout the
organization. Given this context, our agent-based models reveal equilibria in terms of the
distribution of information and attendant behaviors within the organization.
We start by assuming that there already exist at least modest divisions within the
organization such as by formal departments or informal cliques (we discuss this in our
assumptions checks below). Given the existence of these divisions, our dynamic analysis shows
there is a tendency for polarization in the system when actors are able to influence one another or
select others based on similarity of behavior. Thus it is these tendencies for polarization that the
change agent encounters in trying to influence organizational behavior. In this sense our baseline
finding adds a dynamic element to the literature on absorptive capacity which typically focuses
on the ability of an organization’s static structures to facilitate communication and coordination.
But our findings are more specific than just that internal dynamics matter. In particular,
when social salience is high the change agent can more effectively influence the organization by
Page
12
creating a near-neutral venue which will mitigate against the polarization of the organizational
members. For example, if the employees of a school have a strong affiliation with the school
then external change agents might paradoxically exert the most leverage by creating professional
development that exhibits an even-handed, or near-neutral orientation to a particular set of
teaching practices. In contrast, when social salience is low, change agents can more directly
influence organizational members by creating a venue which strongly represents the orientation
of the agent. This might obtain when the members of a school have weak affiliation with the
school, such as when turnover rates are high (Ingersoll, 2001). Thus when the organization is
weak the standard economic emphasis on changing incentives can change organizational
behavior. The dependence of the optimal action of the change agent on the level of social
salience confirms our second hypothesis, that the effect of the change agent is dependent of the
internal social dynamics of the organization.
Our support for hypothesis 2 informs our assessment of hypothesis 1, that change agents can
alter an organization by introducing a venue with an orientation in the desired direction. We now
understand that how a change agent exerts leverage depends on the internal social dynamics of
the organization. We emphasize that the actions of change agents are not merely sensitive to
organizational network structure or the distribution of behavior, but to the intemal dynamics
which depend on the location of behavior in the network as well as the processes of influence
and selection through which behaviors change and networks are modified. It is because of these
dynamics that the effects of external change agents on an organization may go well beyond their
intended actions.
Our second key result is that the process of implementation itself can change the intra-
organizational dynamics. In particular, a change agent who introduces a strong venue into an
organization may accentuate existing cleavages in the system. In the extreme, this can create
polarization or factions, limiting the organization’s capacity for coordination and to diffuse
future innovations. Because changes in network structure have implications for the diffusion of
any innovation, and because polarization may endure and be difficult to overcome, the effects of
change agents may go extensively beyond their immediately intended goal.
Methodologically, we emphasize our parsimonious formalization of the intemal dynamics of
organizations (Chang & Harrington, 2006). Through the formalization we are able to express the
dynamics of diffusion in terms of the processes of influence through and selection of network
members. Thus although we recognize the formal organization as defining the broad conditions
of diffusion, the action of the individual actors are derived from network dynamics of the
individuals within organizations. The parsimony of our models is itself a theoretical contribution,
as we describe the essential internal dynamics in terms of two parameters, organizational
salience (a) and the rate of influence relative to that of selection (k). While no doubt other factors
affect influence (level of expertise, trust, etc.) and selection (proximity), our models allow the
theoretical exploration of two key elements of intemal dynamics that affect consequences of the
actions of agents extemal to the system.
7. Conclusion
The intent of any policy is to change experiences of end users. Much of that experience is
shaped by action within organizational boundaries. But organizations are not monolithic. In
particular, organizations typically feature formal divisions or pt eee in which informal
interactions are concentrated. These subgroups define the lines of potential polarization when
extemal shocks are introduced into the organization. Ignoring this potential can generate serious
Page
13
unintended consequences that can undermine the immediate intent of action as well as the
organization’s capacity to learn, coordinate, and adapt to future innovations. Thus we urge
change agents to attend to the network dynamics intemal to the organizations responsible for
implementing their innovations.
Reference
American Hospital Association, (2010). Implementing Health Care Reform. Retrieved from
http://www.aha.org/content/00-10/100722-message-card-health-reform.pdf
Barley, S. R. (1990). The Alignment of Technology and Structure through Roles and Networks. Administrative
Science Quarterly, Vol. 35, No. 1, Special Issue: Technology, Organizations, and Innovation. (Mar., 1990),
pp. 61-103.
Bidwell, C. E. (1965). The school as a formal organization In JG March (Ed.), Handbook of organizations
(pp. 972-1022).
Bidwell, C. E., & Kasarda, J. D. (1985). The Organization and its Ecosystem: A Theory of structuring in
Organizations (Vol. 2). JAI press.Birkinshaw, J., Hamel, G. and Mol, MJ., 2008. Management
innovation. Academy of management Review, 33(4), pp.825-845.
Brass, D. J., Galaskiewicz, J., Greve, H. R., & Tsai, W. (2004). Taking stock of networks and
organizations: A multilevel perspective. The Academy of Management Journal, 47(6), 795-817.
Bryk, A., & Schneider, B. (2002). Trust in schools: A core resource for improvement. Russell Sage
Foundation.
Chang, M. H., & Harrington, J. E. (2006). Agent-based models of organizations. Handbook of
computational economics, 2, 1273-1337.
Cobum, C. E., Hill, H. C., & Spillane, J. P. (2016). Alignment and Accountability in Policy Design
and Implementation The Common Core State Standards and Implementation Research. Educational
Researcher, 45(4), 243-251.
Coburn, C. E., & Russell, J. L. (2008). District policy and teachers’ social networks. Educational
Evaluation and Policy Analysis, 30(3), 203-235.
Dahl, G., Loken, K., & Mogstad, M. (2014). Peer effects in program participation. American Economic
Review, 104, 2049-2074
Damschroder, L.J., Aron, D. C., Keith, R. E., Kirsh, S. R., Alexander, J. A., & Lowery, J. C. (2009).
Fostering implementation of health services research ‘findings into practice: a consolidated
framework for advancing imy yn science. Impl tation science, 4(1), 50.
DeCarvlis, D. M., & Deeds, D. L. (1999). The impact of stocks and flows of organizational knowledge on
firm performance: An empirical investigation of the biotechnology industry. Strategic Management
Journal, 20, 953-968.
DePaulo, B. M., Nadler, A., & Fisher, J. D. (1983). New directions in helping: Help seeking (Vol. 2).
Desimone, L. M., Porter, A. C., Garet, M. S., Yoon, K. S., & Birman, B. F. (2002). Effects of professional
development on teachers’ instruction: Results from a three-year longitudinal study. Educational
evaluation and policy analysis, 24(2), 81-112.
Page
14
Durkheim, E., & Swain, J. W. (2008). The elementary forms of the religious life. Courier
Corporation.
Feld, S. L. (1981). The focused organization of social ties. American journal of sociology, 86(5), 1015-
1035.
Figlio, D. N., Hamersma, S., & Roth, J. (2011). Information shocks and social networks (No. w16930).
National Bureau of Economic Research.
Fiorina, M. P., & Abrams, S. J. (2008). Political polarization in the A merican public. Annu. Rev. Polit.
Sci., 11, 563-588.
Frank, K. A. (1995). Identifying cohesive subgroups. Social networks, 17(1), 27-56.
Frank, K. A. (2009). Quasi-ties directing resources to members of a collective. American Behavioral
Scientist, 52(12), 1613-1645.
Frank, K. A., & Fahrbach, K. (1999). Organization culture as a complex system: Balance and information
in models of influence and selection. Organization Science, 10(3), 253-277.
Frank, K. A., Muller, C., Schiller, K. S., Riegle-Crumb, C., Mueller, A. S., Crosnoe, R., & Pearson,
J. (2008). The social dynamics of mathematics coursetaking in high school 1. American Journal of
Sociology, 113(6), 1645-1696.
Frank, K. A., Penuel, W. R., & Krause, A. (2015). What Is A “Good” Social Network For Policy
Implementation? The Flow Of Know-How For Organizational Change. Journal of Policy Analysis
and Management, 34(2), 378-402.
Frank, K. A., Zhao, Y., & Borman, K. (2004). Social capital and the diffusion of innovations within
organizations: The case of computer technology in schools. Sociology of Education, 77(2), 148-171.
Frank, K. A., & Zhao, Y . (2005). Subgroups as meso-level entities in the social organization of
schools. Social organization of schooling. New Y ork, NY : Sage Publications.
Frank, K. A., Zhao, Y., Penuel, W. R., Ellefson, N. C., & Porter, S. (2011). Focus, fiddle and friends:
Sources of knowledge to perform the complex task of teaching. Sociology of Education, 84(2), 137-
156.
Friedkin, N. E., & Johnsen, E. C. (1990). Social influence and opinions. Journal of Mathematical
Sociology, 15(3-4), 193-206.
Garet, M. S., Porter, A. C., Desimone, L., Birman, B. F., & Yoon, K. S. (2001). What makes professional
development effective? Results from a national sample of teachers. American educational research
journal, 38(4), 915-945.
Gellner, E. (2015). Muslim Nations. Nationalism, Language, and Muslim Exceptionalism, 17.
Glidewell, J. C., Tucker, S., Todt, M., & Cox, S. (1983). Professional support systems: The teaching
profession. New directions in helping, 3, 189-212.
Gneezy, U., Meier, S., & Rey-Biel, P. (2011). When and why incentives (don't) work to modify
behavior. The Journal of Economic Perspectives, 25(4), 191-209.
Page
15
Gupta, A. K., & Govindarajan, V. (2000). Knowledge flows within multinational corporations. Strategic
Management Journal, 21, 473-496.
Hatch, M. J., & Schultz, M. (200s ane the strategic stars aligned for your corporate brand. Harvard
business review, 79(2), 128-1
Heaney, M. T., & Rojas, F. (2015). Party in the Street: The Antiwar Movement and the Democratic Party
after 9/11. Cambridge University Press.
Hopkins, M., Spillane, J. P., Jakopovic, P., and Heaton, R. M. (2013). Infrastructure redesign and
instructional reform in mathematics: Formal structure and teacher leadership. Elementary School
Journal. 114(2): 200-224.
Ingersoll, R. M. (2001). Teacher tumover and teacher shortages: An organizational analysis. American
educational research journal, 38(3), 499-534.
Kilduff, M., & Tsai, W. (2003). Social networks and organizations. Thousand Oaks, CA: Sage.
Kilduff, M., Tsai, W., & Hanke, R. (2006). A paradigm too far? A dynamic stability reconsideration of
the social network research paradigm. Academy of Management Review, 31(4), 1031-1048.
Langevoort, D. C. (2002). The Organizational Psychology of Hyper-Competition: Corporate
Irresponsibility and the Lessons of Enron. Geo. Wash. L. Rev., 70, 968.
Lin, N., Cook, K. S., & Burt, R. S. (Eds.). (2001). Social capital: Theory and research. Transaction
Publishers.
Lipsky, M. (1980). Street level bureaucrats. Nova Y ork: Russel Sage.
Liu, C. C., & Srivastava, S. B. (2015). Pulling closer and moving apart: Interaction, identity, and
influence in the US Senate, 1973 to 2009. American Sociological Review, 80(1), 192-217.
Matsueda, R. L., & Anderson, K. (1998). The dynamics of delinquent peers and delinquent
behavior. Criminology, 36(2), 269-308.
McLean, B., & Elkind, P. (2013). The smartest guys in the room: The amazing rise and scandalous fall of
Enron. Penguin.
Moolenaar, N. M., Daly, A. J., and Sleegers, P. J. C. (2010). Occupying the principal position: Innovation
in schools. Educational A dministration Quarterly. 46: 623-670.
Moolenaar, N. M., and Sleegers, P. J. C. (2015). The networked principal: Examining principals' social
relationships and transformational leadership in school and district networks. Journal of Educational
Administration. 53(1): 8-39.
Ouchi, W. G. (1977). The relationship between organizational structure and organizational
control. Administrative science quarterly, 95-113.
Ouchi, W. G., & Wilkins, A. L. (1985). Organizational culture. Annual review of sociology, 11(1), 457-
483.
Penuel, W. R., Riel, M., Krause, A., & Frank, K. A. (2009). Analyzing teachers’ professional interactions
in aschool as social capital: A social network approach. Teachers College Record, 111(1), 124-163.
Page
16
Phan, T. Q., & Airoldi, E. M. (2015). A natural experiment of social network formation and
dynamics. Proceedings of the National Academy of Sciences, 112(21), 6595-6600.
Poon, E. G., Blumenthal, D., Jaggi, T., Honour, M. M., Bates, D. W., & Kaushal, R. (2004). Overcoming
barriers to adopting and implementing computerized physician order entry systems in US
hospitals. Health A ffairs, 23(4), 184-190.
Porter, A., McMaken, J., Hwang, J., & Yang, R. (2011). Common core standards the new US intended
curriculum. Educational Researcher, 40(3), 103-116.
Portes, A., & Sensenbrenner, J. (1993). Embeddedness and immigration: Notes on the social determinants
of economic action. American journal of sociology, 98(6), 1320-1350.
Portes, A., & Sensenbrenner, J. (1993). Embeddedness and immigration: Notes on the social determinants
of economic action. American journal of sociology, 98(6), 1320-1350.
Ridgley, J. (2008). Cities of refuge: Immigration enforcement, police, and the insurgent genealogies of
citizenship in US sanctuary cities. Urban Geography, 29(1), 53-77.
Rogers, E. M. (2010). Diffusion of innovations. Simon and Schuster.
Schneider, A., & Ingram, H. (1990). Behavioral assumptions of policy tools. The Journal of
Politics, 52(2), 510-529.
Scott, W. R. (2008). Institutions and organizations: Ideas and interests (4th ed.). Thousand Oaks, CA:
Sage.
Simon, H. A. (1962). The architecture of complexity. Proceedings of the A merican philosophical
society, 106(6), 467-482.
Simon, H. A. (1991). Bounded rationality and organizational learning. Organization science, 2(1), 125-
134.
Snijders, T. A., Pattison, P. E., Robins, G. L., & Handcock, M. S. (2006). New specifications for
exponential random graph models. Sociological methodology, 36(1), 99-153.
Snijders, T. A., Van de Bunt, G. G., & Steglich, C. E. (2010). Introduction to stochastic actor-based
models for network dynamics. Social networks, 32(1), 44-60.
Spillane, J. P., Kim, C. M., & Frank, K. A. (2012). Instructional advice and information providing and
receiving behavior in elementary schools: Exploring tie formation as a building block in social
capital development. American Educational Research Journal, 49(6), 1112-1145.
Stouffer, S. A., Lumsdaine, A. A., Lumsdaine, M. H., Williams Jr, R. M., Smith, M. B., Janis, I. L., .. &
Cottrell Jr, L. S. (1949). The American soldier: Combat and its aftermath.(Studies in social
psychology in World War II), Vol. 2.
Thomas, A. (2012). Three strategies to prevent unintended pregnancy. Journal of Policy Analysis and
Management, 31(2), 280-311.
Thompson, J. D. (1967). Organizations in action (New Y ork: McGrawHill, Inc).
Watt, S., Sword, W., & Krueger, P. (2005). Implementation of a health care policy: An analysis of
barriers and facilitators to practice change. BMC health services research, 5(1), 53.
Page
17
Weatherley, R., & Lipsky, M. (1977). Street-level bureaucrats and institutional innovation: Implementing
special-education reform. Harvard educational review, 47(2), 171-197.
Weber, M. (1946). Bureaucracy. From Max Weber: essays in sociology, 196, 232-235.
Weiss, M.J., Bloom, H. S., & Brock, T. (2014). A conceptual framework for studying the sources of
variation in program effects. Journal of Policy Analysis and Management, 33(3), 778-808.
Werner, A. (2004). A guide to implementation research. Washington, DC: Urban Institute Press.
Wilensky, U. (1999). NetLogo.
Wilensky, U., & Rand, W. (2015). An introduction to agent-based modeling: modeling natural, social,
and engineered complex systems with NetLogo. MIT Press.
Wilhelm, A. G., Chen, I. C., Smith, T. M., & Frank, K. A. (2016). Selecting expertise in context: Middle
school mathematics teachers’ selection of new sources of instructional advice. American Educational
Research Journal, 53(3), 456-491.
Williamson, O. E. (1981). The economics of organization: The transaction cost approach. American
Journal of Sociology, 87(3), 548-577.
Woodward, J. (1980). Industrial organization: Theory and practice. Oxford; New Y ork: Oxford
University Press.
Xu, R, and Frank, K. A. (2016). The Implications for Network Structure of Dynamic Feedback between
Influence and Selection. In International Conference on Social Computing, Behavioral-Cultural
Modeling, and Prediction. Springer International Publishing.
Page
18
