PreprintPDF Available

Computational Policy Process Studies

Authors:
  • Simon Institute for Longterm Governance
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

While policy process theory has converged on the view that policymaking can be studied as a complex system, the literature has only minimally used the methodological complement to the theory - experiments performed with computational models. Implementations are rare, mainly pushed by computer scientists in trans-disciplinary work and often so detached from mainstream theory that they form a separate line of research instead of testing theories from the social sciences. This paper builds on the theory of policy processes and computational sciences to advance the computational turn of policy process studies. We examine how and why complexity science lends itself to study policymaking, propose a workflow to guide the creation of computational policy process models, describe the contours of a computational approach to policy process modeling and define goals for future research that follow from this computational turn. Overall, we aim to promote a computational turn of policy process studies that is empirical and hypothesis-driven.
Content may be subject to copyright.
Computational Policy Process Studies
Maxime Stauffer1,2, Konrad Seifert2, Isaak Mengesha3, Igor Krawczuk4, Jens
Fischer5,6, and Giovanna Di Marzo Serugendo1
1University of Geneva
2Simon Institute for Longterm Governance
3Complexity Science Hub in Vienna
4Ecole Polytechnique F´ed´erale de Lausanne
5Universit´e de Toulouse III - Paul Sabatier
6University of Potsdam
April 2021
8100 words
Abstract
While policy process theory has converged on the view that policymaking can be studied
as a complex system, the literature has only minimally used the methodological complement
to the theory - experiments performed with computational models. Implementations are rare,
mainly pushed by computer scientists in trans-disciplinary work and often so detached from
mainstream theory that they form a separate line of research instead of testing theories from
the social sciences. This paper builds on the theory of policy processes and computational
sciences to advance the computational turn of policy process studies. We examine how and why
complexity science lends itself to study policymaking, propose a workflow to guide the creation
of computational policy process models, describe the contours of a computational approach to
policy process modeling and define goals for future research that follow from this computational
turn. Overall, we aim to promote a computational turn of policy process studies that is empirical
and hypothesis-driven.
1
Introduction
An important development has marked peace and conflict studies, policy analysis and migration
studies in recent years: the use of computational methods to collect, analyse and generate data to
explore social phenomena, often summarized as ‘computational social science’ CSS (Conte et al.,
2012; Lazer et al., 2020). This paper contributes to such a computational turn in policy process
studies. Policy process studies (PPS) refers to the scholarship that examines the cognitive, behav-
ioral and social processes that underpin public policymaking within governments and international
organizations (Howlett, 2020). While often neglected at the benefit of policy analysis (e.g. the
evaluation of policies’ effectiveness) (Cairney & Weible, 2017), PPS sheds light on the determinants
of (un)successful collective action.
In the 21st century alone, humanity has witnessed a vast range of challenges: a global economic
crisis, a global pandemic, accelerating climate change and accelerating technological progress. To
make sure technology is beneficial and shocks can be coped with, it is essential to better understand
how societies can achieve the governance of such transgenerational global public goods (Bostrom,
2013).
The computational turn that we articulate in this paper finds its anchoring in the convergence
of policy process theory over the past decade. Explicitly or implicitly, various theories of the policy
process unite in the view that policy processes satisfy the hallmarks of complex systems (Cairney,
2012; Geyer & Rihani, 2010; Mor¸ol, 2013; Teisman & Klijn, 2008). Complex systems consist of many
heterogeneous parts whose interactions in networks lead to the emergence of macro-level outcomes,
often in a nonlinear fashion (Geyer & Rihani, 2010; B. D. Jones & Baumgartner, 2005b). In the
realm of policymaking, this means individuals and groups interact dynamically to create policies.
Complex systems can be described by complexity theory which has its origins in physics before
being successfully applied to chemical, biological (Mitchell, 2011) and social systems (Helbing et al.,
2015), e.g., in the study of conflict (c.f. Stauffer, 2021). Looking at the erratic pattern of public
budgets (B. D. Jones et al., 2009) and the size and complexity of policy networks (Koliba, Meek, Zia,
& Mills, 2018; Van Waarden, 1992), it becomes apparent that complexity theory is an appropriate
lens for policy process research.
With applications of complexity theory usually come computational methods that enable the
formalization of system mechanisms and, through algorithmic processes, the generation of emergent
patterns at the system level (Epstein, 2006; Page, 2018). Together, complexity theory and com-
putational methods form the field of complexity science which aims to identify laws and systemic
signatures across complex systems (Thurner, Hanel, & Klimek, 2018). Even though the theory of
complexity has been extensively applied to policy process studies, the field lacks computational ap-
plications. Only then will the actual science of complexity be used to study collective action. The
purpose of this paper is to facilitate this transition.
Facilitating the application of complexity science to policy process studies promises exciting sci-
entific progress by furthering Herbert Simon’s vision of policy process studies. Simon contributed
pioneering work on the fundamentals of bounded rationality (Simon, 1957), its applications to poli-
cymaking (Simon & March, 1976), the architecture of complex systems (Simon, 1991), the mecha-
nism of preferential attachment in networks leading to power-law distributions (Simon, 1955), and
computational methods (Simon, 1969). In this, Simon’s view was hypothesis-driven, based on his
empirical observations and relied on the formalization of complex systems and the study of their un-
predictability (B. D. Jones, 2003; Mintrom, 2015; Moynihan, 2018). With the exception of a handful
of scientists (e.g. B. D. Jones and Baumgartner (2005b); Rhodes and Marsh (1992); Thomas (2017)),
2
scholarship in policy process studies has delved instead more into critical, narrative and qualitative
analyses (Beardsworth, 2020) or purely rational, simplistic approaches (Enserink, Koppenjan, &
Mayer, 2013), thus moving away from Simon’s approach to social sciences. We here advocate to
re-engage with the early work of one of the fathers of the field to help enrich the study of policy
processes moving forward.
In other domains, such as conflict and migration, the computational turn is developing in paral-
lel to the mainstream (Lazer et al., 2020). Mathematicians, computer scientists and physicists are
attracted by the challenge of understanding social phenomena, the potential social impact thereof,
and the new challenge of applying their methods to social systems (Watts, 2016). While this trans-
disciplinary interest is laudable, it often is highly detached from decades of literature tackling the
same questions, albeit in a less technical manner. In this paper, we argue for a computational turn
in its truest sense, i.e. building upon the iterative theory-building that has characterised policy
process studies for almost a century. It includes the work by, for instance, Sabatier and Weible
(2018), Jones & Baumgartner (2009), Geyer & Rihani (2010), Mor¸ol (2013), Cairney (2012), Geyer
& Cairney (2015), and Teisman & Klijn (2008). In recent literature, authors have started to portray
this computational turn and made theoretical strides in shaping it in the most valuable way possible
(Johnson, 2015). Our paper builds on them to pave the way for this advancement.
A computational turn of policy process studies suggests either a change in the education of social
scientists (e.g. by providing computational method courses to undergraduate and graduate students)
or a change in the formation of research teams. The authors of this paper are bringing together
their backgrounds in political science, behavioral science, mathematics, physics, computer science
and machine learning, exemplifying such teams. We hope that this paper speaks to the reality,
interests and sensitivities of the different fields that need to contribute to achieve this advancement
of policy process studies.
This paper is divided into four sections. First, we survey the literature that portrays policy pro-
cesses as complex systems in examining complexity, their added-value and shortcomings. Second,
we provide a short introduction to computational models and why they are particularly well-suited
to study social systems. Third, we outline the contours of a computational approach by linking ex-
amples of computational methods to phenomena of common interest in the policy process literature.
Fourth, we sketch an agenda for further research, including general guidance for the research process
as well as key research questions.
1 Complexity in policy processes and methodological fron-
tiers
Policy processes consist of the interplay between the various forces that characterise policymaking
systems. They notably include a vast variety of policy actors and their surrounding environment
that nudges or incentives them to behave in a particular way and make certain choices. Similarly to
conflicts, these dynamics have aroused interest among social scientists because they are characteristic
of heterogeneous individuals aligning to take collective action. Where conflicts are characterized by
individuals who may outright kill each other, the stakes of policymaking are less salient. The process
of figuring out how best to use pooled resources is vital to understand, as they define the visions
of our constitutions, societal narratives, common goals, specify the instruments for collective action
and whether their outputs are beneficial to society (Bednar, 2012; Gintis, Doebeli, & Flack, 2012).
Examining policymaking processes may also increase legitimacy by fostering transparency and
3
spotting where they may be broken (Cairney, 2019; Dye, 1992). As policy processes often involve
conflicts of subjective value judgements, objective analyses of underlying processes can pin-point
reasons for disagreements, and subsequently reduce or outright collusion. Scholars of policy pro-
cesses have advanced many theories, case studies and empirical analyses to describe such convoluted
phenomena (Weible & Sabatier, 2018). In this section, we survey why those processes satisfy the hall-
marks of complex systems and sketch current methodological frontiers in addressing the complexity
of policymaking.
1.1 Theoretical convergence: policy processes as complex systems
Theories of policy processes converge on their conceptualization as complex systems. They, however,
do not always do so explicitly (Cairney, 2012). That is, they sometimes propose other frameworks
that broadly rely on the same ideas but without using the language of complex systems. In what
follows, we first introduce the hallmarks of complex systems before describing how it is used - often
implicitly - in policy processes studies (summarised in Table 2).
Complex systems satisfy two key properties: (1) many, heterogeneous, interacting and adaptive
parts that lead to (2) the emergence of system-level dynamics and properties (Mitchell, 2011; Pines,
2018; Thurner et al., 2018). Explanations of such systems need to be reduced by identifying the key
mechanisms that govern such dynamics, instead of isolating each moving parts or system levels (e.g.
micro versus macro). As such, complexity science can describe otherwise inexplainable phenomena
in physical, chemical, biological and social systems.
First, complex systems have many constituent parts. In chemical systems, such parts are
molecules while in social systems they are individuals or groups. These parts can be heteroge-
neous; have sets of characteristics that vary from part to part. In chemical systems, for example,
characteristics of molecules are atomic composition and ionization while, in social systems, charac-
teristics of individuals include beliefs, interests, roles and more. Those components interact with
one another. In chemical systems, for example, molecules of hydrogen and oxygen interact in a
glass of water. In social systems, individuals may interact spatially (e.g. neighbours) or through
network structures. Lastly, such parts adapt their behavior as a function of their interactions and
environment. In chemical systems, molecules exchange atoms and electrons while, in social systems,
individuals may change their mind or their behavior as a function of social cues. The described
processes characterise the micro-level dynamics of complex systems.
Second, these micro-level dynamics lead to the emergence of macro-level dynamics and properties.
Those dynamics and properties are defined as ‘emergent’ because they cannot be traced back to the
behavior of the system’s individual constituents (Mitchell, 2011; Thurner et al., 2018). A glass of
water, for instance, cannot be understood by simply looking at isolated molecules of hydrogen and
oxygen, but needs additional information about the ratio of molecules and environmental parameters
like pressure and temperature. Similarly, in social systems, a conflict cannot be traced back to the
position of single individuals but is the product of their interactions and relative positions. Moreover,
those dynamics may be nonlinear, that is, a small change at the micro-level may have large effect
at the macro-level and vice-versa. At sea level, for instance, warming up water from 10°to 80°
Celsius does not change its chemical state, while going from 99°to 100°Celsius leads to a phase
shift from liquid to gas. Similarly, in social systems, conflicts or opinion dynamics are nonlinearities
due to negative or positive feedback loops that absorb or amplify information cascades and can lead
to conflict outbreak (Lee, Daniels, Myers, Krakauer, & Flack, 2020) or shifts of popular opinion
(Bassett, Alderson, & Carlson, 2012; Castellano, Fortunato, & Loreto, 2009; McBurnett, 1996).
4
Properties of complex systems Equivalents in policy processes
Many components Many actors and events interact in policy processes
Heterogeneity Policy actors have different motivations, expertise, roles, resources, etc.
Interaction in networks Policy actors can form alliances and interact in policy arenas
Environmental constraints Policy actors’ behavior is shaped by environmental events, institutional structures and social cues
Coevolution Policy actors update their focus and beliefs as a function of their interactions with other policy actors
Emergence Agendas and policies are macro-level outputs that cannot directly be linked to micro-level configurations
Nonlinear dynamics Opinion dynamics or the behavior of public budgets are results of positive and negative feedback loops
Table 1: The hallmarks of complex systems in policy processes
These properties have been widely used in policy process theories. First, a group of scholars has
explored the application of complexity theory to policy processes (Cairney, 2012; Geyer & Rihani,
2010; Klijn, 2008; Mor¸ol, 2013; Teisman & Klijn, 2008). They define policymaking processes as
networks of policy actors who have different personal characteristics and participate in an informal
and formal process of collective decision-making. This process depicts erratic features which they link
to the ideas of emergence and nonlinearity. The emergent properties are either coalitions, agendas,
policies, or decisions. Second, there is an implicit convergence among the theories of the policy
process as reported in the work of Sabatier & Weible (2007; 2018). The following are non-exhaustive
examples of this convergence.
The advocacy coalition framework proposes a framework to define the co-evolution of coalitions
as a function of their interactions and the resulting adaption of individual’s beliefs (Pierce, Giordono,
Peterson, & Hicks, 2019). The multiple streams framework characterises policymaking as consisting
of different flows of activity that, when coupled, can lead to drastic, sudden changes at the system
level. This coupling is often due to the behavior of a small group of actors and to changes in policy
environments (M. D. Jones et al., 2016). The policy feedback theory conceptualises feedback loops
from policies to policy processes, thus theorising amplification or absorption dynamics between the
products of policies processes and their inner workings (Mettler & Sorelle, 2014). Policy learning
defines the network-based dynamics of continuous adaptation of policy actors as a function of new
information, thus portraying the adaptive capacity of the participants of policy processes (Moyson,
Scholten, & Weible, 2017). Last but not least, punctuated equilibrium theory empirically identifies
power-laws in budget changes and theorises mechanisms of inertia and sudden change based on
positive and negative feedback loops within micro-level processes (Baumgartner, Jones, & Mortensen,
2018).
All of these theories converge on the view that policymaking consists of bounded-rational,
attention-limited (thus adaptive) actors whose interactions lead to system-level outcomes such as
policies. The presented theories have been applied to numerous case-studies and seem applicable
in democratic and authoritarian political contexts alike (Chan & Zhao, 2016). Comparative studies
in punctuated equilibrium theory, for instance, have found empirical regularities across almost all
democracies as well as China and international organizations (B. D. Jones et al., 2009; Lam & Chan,
2015; Lundgren, Squatrito, & Tallberg, 2018). Table 1 summarises the properties of complex systems
and their equivalents in policy processes. Overall, the literature has progressed on understanding
both micro- and macro-level dynamics (Moynihan, 2018).
All in all, scholars of policy processes have been increasingly conceptualizing policy processes in
a way that exhibits the hallmarks of complex systems (Geyer & Cairney, 2015). Whether they will
explicitly use complexity theory as their handle is a question related to disciplinary tradition and
attachment to the particularities of certain theoretical frameworks. As for this paper, the diagnosis
is clear: the analysis of policymaking processes would highly benefit from methodological advances
in complexity science. As it is already used at least implicitly, all it may require is a methodological
shift, which we address in the next section.
5
1.2 Methodological frontiers in examining complexity
While the application of complexity theory - even if implicit - to understand policy processes is
increasingly accepted, the employed methods fall short of providing system explanations that are in
line with complexity science. In this section, we present the main methodological approaches used
in the current policy process literature, discuss their limitations and outline recent methodological
progress that suggests a promising but rocky path for a computational turn of policy process studies.
Table 2 summarize the key references that contribute to the study of policymaking as a complex
system.
The works that explicitly use complexity theory to describe policy processes is mostly theoretical.
They focus on refuting or validating the use of complexity theory in policy process studies and
make recommendations for future scholarship, such as the unification of theories, moving away from
simplistic approaches, or recognising unpredictability (Geyer & Cairney, 2015). In contrast, the
scholarship applying the theories implicitly is either purely theoretical, purely qualitative or purely
quantitative.
The different theories are subject to debate. Scholars unpack criticisms and refine theoretical
mechanisms, often as a result of empirical insights and key events that shed light on policy processes
(e.g. 2008 economic crisis, Fukushima’s tsunami or Brexit (Giger & Kl¨uver, 2012)).
When used as lenses to understand specific cases, those theories are usually accompanied by
qualitative methods, ranging from ethnographic research, to interviews and social network analy-
ses (Keman & Woldendorp, 2016). Those approaches often remedy the lack of granular data on
the micro-level dynamics of policymaking processes. Therefore, scholars need to conduct (semi-
)structured interviews with policy actors to access information about beliefs, behaviors, network
structures and power dynamics. These research results are often used for thematic analyses or
critical analyses that deconstruct social realities (Levy, 2007).
Scholars also employ quantitative methods to validate or refute theories. Surveys, Open Govern-
ment Data, media data, or the Comparative Agendas Project (Baumgartner, Breunig, & Grossman,
2019) provide sources of micro-level and macro-level data on policy processes, allowing the analysis
of the relationship between changes at the micro-level and their impact on changes at the macro-
level. The methods employed are either descriptive statistics or statistical models such as linear
regressions (Curini & Franzese, 2020).
The current methods - theoretical, qualitative or quantitative - constrain the potential of policy
process studies unless complemented. First, the literature that ties together complexity theory and
policy processes mostly focuses on the meaning of complexity as a concept applied to policymaking.
It is about building a narrative and a set of mental models to understand complex policymaking
systems. While we think building this narrative is important to change paradigms in social science,
it often falls short of identifying empirical facts and laws of policymaking systems. Therefore, it is
insufficient to describe empirical insights about the complexity of policymaking systems.
Second, building a complexity narrative of policy processes also opens the door to relativism.
Especially when associated with concepts from chaos theory, the complexity narrative attracts pro-
ponents of postmodernism, interpretivism and more (Morcol, 2001). While these critical schools of
thought have led to important reflections about the social sciences and the role of the limitations
and power of researchers in attempting to understand social phenomena, they tend to suggest that
knowledge exists only if highly nuanced (Healy, 2017). This assumption, often implicit, has fos-
tered doubts about the value of abstraction, formalization, quantitative and computational methods
6
(Drucker, 2016), which has slowed down the adoption of computational methods in policy process
studies.
Third, policy process studies are divided into roughly two clusters of scholars. On the one hand,
scholars push very reductionist approaches and focus on e.g. linear relationships between micro-level
and macro-level data through black-box, statistical models. On the other hand, scholars embrace
all-encompassing approaches and resort to long essays or text-books to describe complexity. These
two groups seem to position themselves differently with respect to modeling social phenomena. The
first group focuses on models that do not account for the characteristics of complex systems while,
the second group, does not seem to engage with any form of formal modeling. The computational
turn is a way to reconcile these clusters, by putting together scientists who formalize policymaking
systems and others who have developed detailed theories thereof. By facilitating this turn, we want
to motivate the search for more complex but humble models that do not try to hide their assumptions
and simplifications but try to test specific theoretical propositions as best as possible.
However, it is important to highlight relevant quantitative advances in policy process studies.
First, statistical analyses applied to punctuated equilibrium theory using leptokurtic distributions
have been widely tested and generated empirical regularities across countries (Chan & Zhao, 2016;
Seb´ok & Berki, 2017; Sharp, 2019). They notably show a power-law distribution of changes in public
budgets. These robust findings are one of the first results that look like the ones complexity science
generates on other systems: macro-level patterns that are found across systems, which suggest
that systems behave the same way and thus lead to the formulation of hypotheses for micro-level
dynamics. In fact, scholars have found similar findings in conflict dynamics, where the frequency
and severity of conflict also follow a robust a power-law (Clauset, 2018; Richardson, 1960).
The second advancement refers to the development of network or agent-based models to formalize
the micro-level dynamics of policy processes and generate macro-level phenomena (Cioffi-Revilla &
Rouleau, 2010; Klein, 2017; Kollman & Page, 2006; Richards & Doyle, 2000; Thomas, 2017), follow-
ing what complexity scientists have been applying to other areas of social science (Axelrod, 1997;
Bhavnani, Donnay, Miodownik, Mor, & Helbing, 2014; Guerrero & Axtell, 2013). This progression
is promising as, so far, quantitative models were statistical (i.e. modeling of data) while in network
and agent-based models, the idea is to model not just observations but the generating mechanisms
of systems themselves. A third advancement consists of deep learning models using decision trees
and forest fire dynamics to replicate power-laws in budget changes and link them with micro-level
processes (Hegelich, 2016, 2017). While in their infancy, these models – network, agent-based and
deep learning – are increasingly connected to policy process theory, and thus illustrate the compu-
tational turn that goes hand in hand with parallel scholarship. However, to date, these models have
not generated particularly useful insights beyond unpacking the barriers to computational modeling
of policy processes. They are also published in journals poorly known to policy process scholars. In
the rest of this paper, we propose avenues to further progress.
2 Computational models to understand policymaking
The advent of computers was essential for the study of complex systems. They enable the formal
and algorithmic study of systems in a way that can be anchored in and validated by empirical
data. Computers are a powerful tool to study systems whose behaviors are too complex for human
intuition to grasp and cannot be reduced to linear relationships. Computational methods include
any computational tool that serves to collect and model data as well as model systems (Lazer et
al., 2020). In this paper, we focus on computational models of systems and this section briefly
7
Reference title Author(s) & year Implicit or explicit
use of complex system Type of contribution Contribution Formal modeling
Complexity in Public Management Teisman and Klijn (2008) Explicit Theory Mapping of properties of complex systems
on the characteristics of public policymaking processes NA
Complexity and Public Policy: A New Approach
to Twenty-First Century Politics Geyer and Rihani (2010) Explicit Theory
Mapping of properties of complex systems on the
characteristics of public policymaking processes;
and discussions of implications
NA
Complexity Theory in Political Science and
Public Policy Cairney (2012) Explicit Theory Discussion of the use of complexity to understand
policymaking processes and implications for the field NA
A Complexity Theory for Public Policy Mor¸ol (2013) Explicit Theory Mapping of properties of complex systems on the
characteristics of public policymaking processes NA
Handbook of Complexity and Public Policy Geyer and Cairney (2015) Explicit Theory; review
Collection of chapters on the use of complexity theory
to explain policymaking processes; and description on
the use of agent-based models in policy process studies
Steps for agent-based
modeling
Common Approaches for Studying advocacy:
Review of Methods and Model Practices of the
Advocacy Coalition Framework
Pierce et al. (2019) Implicit Theory; review
Description of the use of co-evolutionary coalitions to
explain policymaking processes; report that 70 to
100% analyses are qualitative
NA
A River Runs Through it: A Multiple Streams
Meta-review M. D. Jones et al. (2016) Implicit Theory; review Description of interplay of actor and environment to
explain policymaking; report 88% of studies are qualitative NA
Policy Feedback Theory Mettler and Sorelle (2014) Implicit Theory Describe feedback loop between policy outcome
and policy processes NA
Policy Learning and Policy Change: Theorizing
their Relations from Different Perspectives Moyson et al. (2017) Implicit Theory; review Describe adaptive behaviour of individuals and groups NA
Punctuated Equilibrium Theory: Explaining
Stability and Change in Public Policymaking Baumgartner et al. (2018) Implicit Theory Describe power-law in budget changes and theorise
positive and negative feedback loops at the micro-level Statistical modeling
Political Complexity: Nonlinear Models of Politics Richards and Doyle (2000) Explicit Review of models Collection of models of politics, with unclear link
to policy process theories Various
Computational Methods and Models of Politics Kollman and Page (2006) Explicit Review of mo dels Review of models of politics, mostly fo cusing on
electoral systems and institutional design Agent-based modeling
MASON RebeLand: An Agent-based Model of
Politics, Environment, and Insurgency Cioffi-Revilla and Rouleau (2010) Explicit Exploratory mo deling Formalization of a large-scale agent-based model
with unclear link to policy process theories Agent-based modeling
Policy Emergence: an agent-based approach Klein (2017) Explicit Exploratory modeling Formalization of an agent-based model based on
combination of policy process theories Agent-based modeling
Modeling Contagion in Policy Systems Thomas (2017) Explicit Exploratory mo deling Modeling of attention contagion in policy networks,
with clear link to policy process theories Agent-based modeling
Deep Learning and Punctuated Equilibrium Theory Hegelich (2017) Explicit Exploratory modeling Modeling of patterns of attention Deep neural networks
Table 2: Sixteen key references that make contributions to studying policymaking as a complex
system
presents what computational models are and why they are useful, before outlining a process of
evidence-driven computational modeling for policy processes.
2.1 What are computational models?
Before the advent of personal computers in the 1990s, the formalization of social systems fell short of
replicating complex dynamics. They were able to generate either deterministic dynamics (Richard-
son, 1960) or chaotic dynamics (Saperstein, 1984). Later on in the 1990s, networks of computers
could facilitate purely theoretical computational research in e.g. international relations and the val-
idation or refutation of traditional theories (Duffy, 1992). As of the 2000s, computational models
became more precise and increasingly empirically validated (Clauset & Gleditsch, 2012; Weidmann
& Salehyan, 2013). In sixty years, the field of social system formalization moved from systems of
equations to computational toy models and on to empirically-validated computational models. It is
this latter, empirically-validated computational type we want to encourage in policy process studies.
Computational models can be referred to as dynamic formalizms because they combine two core
components. First, models are implemented as systems of equations, computer code or a mix of both
(Thurner et al., 2018). This process of formalization aims to reduce the description of system parts
and mechanisms to a minimum of components in a way that is theoretically and/or empirically valid.
To model complex systems, formalizations often include update rules or feedback loops, which are
iterative mechanisms that lead to adaptation, absorption or amplification. Second, computational
power is used to run the formalization and activate models’ iterative processes. The iterations then
generate system dynamics.
To examine social systems, a prominent class of computational models are agent-based models
(Page, 2008). These models formalize the characteristics and adaptive behaviors of agents (e.g.
policy actors), their mode of interactions and network structures (e.g. topology of policy networks)
and the macro-level properties (coalitions, attention dynamics, collective decisions, .. . ) that result
from iterative processes of agent-level adaptation (e.g. opinion changes, decisions, etc.). Agent-
8
based models are increasingly popular in political science, especially in the study of intra-state
conflict (Neumann & Lorenz, 2019).
“Is it possible to model human behavior?” is a question that is repeatedly raised in political
science departments and conferences. This paper builds on the insight that, in fact, all descriptions
of human behavior are models. Verbal descriptions may seem more faithful because of their fuzziness
and room for interpretation, allowing people to project their own models on what is meant. Yet,
current verbal descriptions of policy processes based on complexity theory are difficult to test em-
pirically because they are often too general. Therefore, a minimum level of formalization is needed.
Similarly, mathematical models without implementation fall short of eliciting complex dynamics.
It is the coupling of mathematical formalization and computational power that can usefully model
human behavior and explain social dynamics (Helbing et al., 2015). All models are wrong, some
models are useful (Epstein, 2008). Any modeling effort needs to state its limitations clearly, but
models do not need to be perfect. They need to fit the theory and the data and be tested against
reality.
2.2 The value of computational models
Computational models are particularly useful because they formalize and generate dynamics of
iterative, stateful, path-dependent processes. This is important because social systems feature plenty
of iterative processes. Conflicts, opinion dynamics or economic supply and demand relationships, for
example, are all iterative processes that boil down to how agents update as a function of the behavior
of other agents. Modeling such iterative processes enabled the study of emergent cooperation among
heterogeneous agents, for instance (Axelrod, 1997).
Computational models also enable the study of the dynamics of systems of equations that are
not analytically solvable. Differential equations of complex systems rapidly become too complex to
be solved by hand and most are non-deterministic (Thurner et al., 2018). Therefore, implementing
those equations as computational models enables researchers to generate dynamics and then analyse
them statistically. Relatedly, while it is extremely difficult to make point-predictions about complex
systems, computational models can make accurate predictions of outcome distributions by running
thousands of simulations (Lazer et al., 2020; Thurner et al., 2018).
Computational models enable to generate data on either micro-level or macro-level dynamics
when no real data is available (Epstein, 2006). Therefore, models may be built on theory and then
used to generate data with which statistical analyses are possible. This is particularly useful to
examine complex systems for which data collection is very difficult (secret processes, or dangerous
locations) and to analyse counterfactuals in possible future outcomes, assuming the model has been
sufficiently validated (Page, 2018).
Computational models also allow researchers to reconcile the study of micro-level and macro-level
dynamics. As the mainstream literature tends to dissociate both (e.g. micro- versus macroeconomics,
or policy analysis versus behavioral public policy), the use of computational models can link macro-
level dynamics to the behavior of their constituent parts. They directly integrate the concept of
emergence and can produce the nonlinear dynamics associated with emergent phenomena.
In social systems, there is often one history that cannot be repeated, which prevents the rigorous
understanding of what-if scenarios. Computational models allow researchers to repeat artificial
scenarios and compare them to each other as statistical equivalents (Thurner et al., 2018).
9
It is, however, very challenging to reliably attain the added-value we proposed. The next section
presents a guide to maximise the success of computational modeling efforts.
2.3 From model conceptualization to the identification of signatures of
policymaking systems
Recent literature on computational social science has aggregated pieces of advice for creating valuable
models. Bhavnani and colleagues (2020) summarise this advice in a process to develop robust
evidence-driven computational models. We adapt their process to policy process studies. The
advantage of this process is that it places computational modeling within a larger scientific process,
thus linking different stages of research design.
Figure 1 illustrates the resulting five step process: (1) model conceptualization; (2) model im-
plementation; (3) data construction; (4) model validation and refinement; and (5) counterfactual
analysis. As such, it guides researchers in the identification of the signatures of complex policymak-
ing systems. While the figure depicts a linear process, it is inherently recursive. In the rest of this
section, we describe each step and link it to the policy process literature.
Model
conceptualization Model
implementation Data construction Model validation
and refinement Counterfactual
analyses
Process
Aim Theoretical and
empirical anchorage Formal model of
decision dynamics Collect data on micro
and macro dynamics Validate and improve
model empirically Explore system
configurations
Figure 1: A process for computational modeling in policy process studies (adapted from Bhavnani
et al., 2020)
Model conceptualization: the first step of computational modeling is the definition of parts, mech-
anisms and outcomes. Model conceptualization is, in a way, a synthesis of many different information
sources. To conceptualize a model of policymaking, it is possible to draw from existing theories of
policy processes (Weible & Sabatier, 2018), existing case-studies, expert opinion, interviews and sur-
veys. Altogether, these sources of information enable the specification of system parts, the selection
of specific mechanisms and link micro-level with macro-level dynamics. A crude, conceptual model
of policymaking, for instance, can be deduced from punctuated equilibrium theory as shown in Box
1.
Box 1. A simple verbal model of policy processes based on punctuated equilib-
rium theory
Policy actors have two adaptive characteristics which is their current attention and opinion,
they interact within a decentralized network structure, and update their attention and opinion
10
as a function of their neighbours’ characteristics. Actors pursue two strategies: trying to reach
better positions in the network and forming coalition with value-aligned actors. Iteratively, this
process leads to aggregate patterns of attention such as periods of inertia and periods of change,
and contagion of opinion. (B. D. Jones & Baumgartner, 2005a, 2005b).
Model implementation: once the model is conceptualised and matches descriptions and data from
available sources of information, one can implement it. Implementation includes formalization of
the model as systems of equation and computer code. The formalization specifies the relationship
between variables and their associated probabilities, how the model aggregates observables (like
patterns of attention), and how it reports results. Formalization often is a crucial step to test the
robustness of the conceptual model because it forces researchers to spell out modeling assumptions
and will thus challenge fuzzy definitions. Relevant mathematical and software programming tools
must be employed and their use justified (see next section). Model implementation also involves
many checks to test whether the implementation produces the intended dynamics. Sensitivity anal-
yses are useful to check whether the model behaves in a realistic manner. In policy process studies,
the implementation of the conceptual model could draw from differential equations that help formal-
ize co-evolving networks (Thurner et al., 2018). See the next section for an approach to formalize
policy processes.
Data construction: once the model is implemented, it is called a toy model, as its results will
not be empirically valid and can only serve exploratory purposes. To produce empirically valid
results, the third step is to identify, collect and prepare data to seed and validate the model (see
next paragraph). Data is required to specify the initial conditions of the model, such as the number
of policy actors, their initial focus, the network topology and the frequency and size of exogenous
signals, such as media coverage. To validate the simulated results - such as aggregate patterns of
attention, the rate of new decisions, and the evolution of network topologies - more data is required.
Finding useful data is a challenge. Data on micro-level processes is scarce and often qualitative.
There is some progress in collecting such data under the label ‘big data’ by relying on email data to
identify network topology and social media data to identify attention patterns and political behavior
(Donnay, 2017). Surveys and behavioral experiments can also provide complementary information,
especially on behavioral profiles (Moser & Kalton, 2017). Data on macro-level processes, such as
agendas, attention or decisions, is easier to find in a quantitative format. The Comparative Agendas
Project centralises data on public budgets (Baumgartner et al., 2019) and the Open Government
Data group reports decisions and bills issued by national governments (Ubaldi, 2013). In most
cases, the identified data will not be directly usable to seed and validate the model. It will require
“construction” of datasets that can be imported into the model by, for instance, matching dataset
variables with model parameters. Or by drawing probability distributions from datasets to define
update rules, frequency of exogenous signals, etc.
Model validation and refinement: once data is prepared, researchers can use it to seed a model’s
initial conditions and validate simulated outcomes. Multiple methods can be used to measure the
fit of simulated results with empirical data, including the root-mean-square-deviation, but also
precision, recall and F1 scores. F1 scores are particularly useful as they provide information on how
many observations the model gets right or wrong, instead of a simple overall correlation or deviation
(Goutte & Gaussier, 2005). Data can be used to test internal validity (using in-sample data from a
specific case) and then external validity (using out-of-sample data from other cases). Here again, the
insights from punctuated equilibrium theory are useful. The power-law of public budgets has been
empirically validated across contexts. It would be interesting to build a model of the micro-level
11
dynamics that lead to its emergence and then explore whether its mechanisms are valid in various
contexts.
Counterfactual analysis: once models are empirically validated, at least internally, it is possible
to use them as artificial laboratories to explore their signatures. Conducting counterfactual analyses
allows to understand how the system behaves when different initial conditions are chosen. For
instance, one can explore different levels of network centrality, different attention probability updates,
different frequencies and sizes of exogenous signals, and explore how the system behaves as a function
of those changes. In complexity science, it is essential to explore critical points - the equivalent of
the 99 to 100 degree shift in water temperature that leads to a phase transition. For instance,
does increasing network density from one specific value to the next lead to radical changes at the
macro-level? Those analyses could allow the robust identification of the drivers of policymaking. As
such, they can then be used to generate recommendations on how to design policymaking processes
(see section 4).
This five-step process allows researchers to move from the conceptual models, often described in
policy process studies, to the exploration of common signatures of complex policymaking systems.
It is important to iterate on these five steps, instead of trying to perfect each of them only once. The
questions in the following two sections relate to modeling particular phenomena in policy process
studies and avenues to be explored with this process.
3 The contours of a computational approach to model policy
processes
Manoeuvring from model conceptualization to model implementation provides two key challenges.
Firstly, it raises the question of finding an accurate formalization and, secondly, it requires adequate
tools to implement the dynamics of said formalization.
This section reviews successful modeling approaches and sketches the contours of their imple-
mentation. Subsequently, we apply our approach to policy processes outlined in Box 1. While this
section is far from exhaustive, it aims to exemplify the transition from the conceptualization to
the implementation of modeling policy processes. By specifying the strategy of our approach we
illustrate potential to better understand policy processes.
3.1 A case study in opinion dynamics
Our goal in this section is to retrace what makes for a successful formal model of social dynamics.
We aim to highlight advantages and challenges on the path to attain a causal model and with it a
mechanistic explanation of policy processes.
It is in principle hard to make a precise verbal explanation such that it uniquely defines a system
in its micro- and macro-states. Or even to prove a causal relationship between them. The immediate
payoff of any formal model of the policy process is a proof-of-concept for the internal consistency
of stated axioms. Enabling us to explore scenarios and generate hypotheses. A prominent example
being Axelrod’s computer tournaments to find optimal strategies for the iterated prisoner’s dilemma
(Axelrod & Hamilton, 1981). After finding the winning strategy (“tit-for-tat”), Axelrod et al.
could provide an explanation of why such behavior is common in any species with similar selection
12
pressure. However, the process of policymaking constitutes an open system, meaning that it is highly
interconnected on many different levels of abstraction with some larger system, outside the scope
of any modeling attempt. Consequently, models of such systems will most likely be unsuitable for
forecasting and restricted to improve the qualitative understanding of each constituent’s contribution
to the overall dynamics.
For example, R. Hegselman and U. Krause followed these considerations when developing their
model of opinion dynamics (Hegselmann, Krause, et al., 2002). The goal of their model was specified
as providing mechanistic explanations for the process of opinion fragmentation. What are the con-
ditions for consensus or polarization to emerge? The first step was to formulate axiomatic rules that
they believed to be the essential rules of the process. The second step was to define a correspond-
ing formal mathematical framework that, besides allowing for further extensions, most importantly
answered the following question: To what extent do the system dynamics follow from stated as-
sumptions? In what follows, we retrace their approach and formalization and, thus, illustrate the
process of formalizing social phenomena.
3.1.1 The minimal model
With nas the number of agents contained in the considered population, some discrete-time T=
1,2, ... and some continuous opinion xi(t) with tTfor each agent ion the interval [0,1] the
fundamental building-blocks for an opinion dynamics model are defined. We then want to define
the most basic updating mechanism for each actor, namely taking the average over the opinion of
neighbouring agents and the agent’s own opinion. However, actors do not weigh everyone’s opinion
equally. Therefore we define weights for the all connections in the undirected network, resulting in
the adjacency matrix A:= (aij) of the network. As such, the updating mechanism can be compactly
written as:
x(t+ 1) = A(t, x(t)) ·x(t) for tT(1)
Formalizing the general research question to: how does a given initial opinion profile of the
entire network influence the final opinion distribution? Equipped with this formalization we can now
investigate the most basic case with Aas a stochastic matrix. This allows estimating convergence
time to the steady-state depending on the structure of A. We identify the network structure and
opinion updating rule of each actor as the key elements that dictate the dynamics. Hence, making
them the focus of further analysis (Hegselmann et al., 2002).
3.1.2 Extensions of the minimal model
Building on this minimal model, Hegselman and Krause introduced many interesting extensions,
such as the property of susceptibility for each actor or “hardening of positions” where actors weigh
their own opinion more strongly over time. Moreover, they analyzed non-linear extensions, which
were introduced by defining conditions under which weights can become zero - meaning actors are
ignoring each other. One possible implementation is to define a confidence level εito each agent
similar to the Deffuant-Weisbuch model (Deffuant, Neau, Amblard, & Weisbuch, 2000). Such models
are called bounded confidence models, where the confidence level acts as threshold parameter that
defines the level of discrepancy of opinions beyond which actors ignore each other. Further research
investigated the implications of asymmetric confidence intervals between the agents and of the
13
underlying network structure. From this we learn that choosing the right level of abstraction (i.e.
minimal model) is essential and allow for useful extensions at a later point. Ideally, we do not
arrive at a single, large and specific self-consistent model, but rather at a cluster of different models
exploring implications of different assumptions.
3.2 A simple model of policy processes
Drawing from the outlined considerations, we now propose a minimal model for policy processes
based on the verbal model describe in Box 1.. We first make our assumptions for the system explicit
and subsequently propose a formalization of them. There are three components to our model: (1)
the internal state of actors, (2) the network topology and (3) the updating mechanism.
3.2.1 Components of the model
(1) The internal states are characterized by the actors attention and their deeper beliefs/opinions.
While both change over time, we assume that they happen at different timescales, with attention
changing in much shorter intervals (B. D. Jones & Baumgartner, 2005a). Additionally, we believe
that so-called “focusing events” shape attention dynamics but are negligible in opinion dynamics.
Furthermore, we incorporate bounded rationality by limiting the attention that can be distributed
among different topics for each actor. Lastly we equip every actor with an idiosyncratic strategy
preference, which will be specified in (3). This results in the following: Actor iwill have an opinion
vector x0
i(t)[0,1]nand a corresponding attention vector xA
i(t)[0,1]nthat, together with the
strategy si[0,1] will define the internal state vector Xi(t)[0,1]2n+1.
(2) A node in the network can have an unlimited number of edges. However, we take limited
processing capabilities into account by limiting the sum over all weights for all actors to a constant,
similar to the concept of Dunbar’s number (Dunbar et al., 2012). The strength of the connection
between two actors is then some function of the edge weight connecting them, alignment of their
internal states and common connections. The network is adaptive in nature, re-weighting and
rewiring its edges over time. This corresponds to fluctuations in the intensity and composition of
their relationships. We assume that the re-weighting process is more essential and therefore neglect
the rewiring process. Additionally rewiring often increases the computational cost for simulations
immensely. However, we will investigate the effect of introducing new and removing old nodes into
the network. As a consequence we have some stochastic matrix A(t)Rn×nthat defines the
network.
(3) Lastly the updating mechanism constitutes the core of our model of policymaking systems.
We interpret the underlying social networks as multi-layered co-evolving networks (Thurner et al.,
2018, p. 21-25). Similar to the model by Hegselman and Krause (2002), the evolution of every actor’s
state and the network is some function Fof its previous internal state and the weight matrix:
{A(t+ 1,x(t+ 1)),x(t+ 1)}=F[A(t, x(t)),x(t)] for tT(2)
This mutual dependence of opinion dynamics and network topology accounts for the co-evolutionary
properties of our model. Subsequently, we will postulate further assumptions on the function Fthat
are valid for policy processes in general.
14
Each actor has to update his internal state, namely his attention and opinions. Furthermore, ev-
ery actor is bounded-rational and has imperfect memory or learning capabilities. We can encapsulate
these dynamics in a “stress” function Heach actor tries to minimize:
min{H(A(t),x(t))}(3)
This approach closely resembles other successful model of opinion dynamics based on the Ising
model (Galam, 2008). Hshould take the states of the actors local/global neighbourhood into
account. Additionally, this function will have a stochastic/perturbative component that accounts
for the previously mentioned cognitive limitation. In the limit case of no cognitive ability, the actions
should appear to be random. The weights of the edges account for the influence that each neighbor
has on one particular actor. Optionally, this influence can be dependent on more parameters than
weights in A(t). The update of each actor’s opinion therefore depends on the network’s topology.
Furthermore, actors will have two strategies that guide their interactions. First they are driven
to increase their degree centrality (i.e. importance in the network) and second they want to form
coalitions with value- and goal-aligned actors. We denote the function of the gained utility of these
strategies with Cefor a centrality-driven and Cofor a coalition-driven strategy respectively. These
two strategies specify the preferred attachment-style of a certain agent (i.e. the connections they
want to establish or maintain). Therefore they constitute the driving mechanism of the evolution of
the topology. Each actor’s idiosyncratic strategy is determined by the previously defined constant
siin the following way:
max{siCe(A(t),xi(t)) + (1 si)Co(A(t),xi(t))}(4)
Note that this optimization only takes existing edges into account and reweighs them such that
the term is maximized for a certain strategy. The previously undefined function Fcan now be
identified as the application of these two optimization problems for a given actor iin the system.
3.2.2 Procedure and Extensions
We have outlined the fundamental mechanism of the model. Formulating these functions more ex-
plicitly will need empirical justification. Nevertheless, we can determine next steps of analysis. After
implementing the model we will investigate its properties through established schemes for sensitivity
analysis and Monte Carlo methods, answering questions such as: How does the system’s steady state
depend on its initial configuration? How sensitive is it to the actors rationality/randomness? How
does its rate of convergence towards the steady state depend on factors such as the dimensionality
of opinions or the capacity for attention? Do we already observe punctured equilibria and how can
they be explained without external shocks? Do certain strategy profiles correspond to specific known
groups such as “policy entrepreneurs” (Aviram, Cohen, & Beeri, 2020)? If so can we validate their
suspected influence on the system?
Additionally we want to make two implicit assumptions of our approach explicit: First, the
proposed model is time-discrete. Switching from discrete to continuous time models often changes
the dynamics profoundly. Consequently, we need to observe closely if this may distort the system
behavior we observe. We have chosen discrete time due to its computational advantages, but it is
not necessary for analysis in principle. Second, the model’s time evolution only depends on the state
of the previous time step, making it markovian. This means it does not remember its history. In
many complex systems this is an acceptable approximation and yields models with large explanatory
15
power regardless. Transferring this approach to a political system is not trivial and requires careful
consideration.
Exploring the proposed model would already yield insights on the behavior of policy processes.
In particular, it would provide a way to explore their dynamics in a way that reconciles both micro-
and macro-level dynamics. Such efforts can be expanded upon by adapting the model to allow for
exchanging actors over time, external events that have a global effect on attention (i.e. focusing
events). Once we have established an understanding of key properties, we can go on to introduce
more refined rewiring processes, more complex events and more heterogeneous properties for the
actors.
4 A research agenda for computational policy process studies
Taking stock, we delineate two main research avenues to achieve the computational turn of policy
process studies. First, we outline areas where progress is to be made - from formalizing policy
processes to generating empirical evidence about the drivers of policy change. Second, we articulate
three contributions that computational policy process studies can make in the near future.
First, the computational modeling process sketched in section 2.3. highlights the necessary steps
to move from a conceptual understanding of policy processes to the identification of their empirical
signatures. We see five areas of progress. First, the computational approach outlined in section 3.
must be expanded and refined. It is important to carefully select which concept of policy process
studies deserve formalization and then identify promising tools. Second, once the approach provides
isolated formal components of policy processes, one can select and assemble them to design toy
models. Toy models will be used to elicit dynamics that can be validated theoretically. The idea is
to check whether results come close to the expectations in the literature. Third, more effort must
be put into collecting data on the characteristics, behavior and network connection of policy actors.
Building datasets of micro-level determinants is crucial to advance empirical research and examine
how micro-level dynamics lead to macro-level outcomes. Fourth, toy models can be expanded and
seeded and validated with empirical data. Fifth, empirically-validated models can then be used to
explore counterfactual scenarios and identify critical points of policymaking systems.
These five steps can apply to single projects but also to the entire field of computational policy
process studies. As such, as computational projects multiply, they can feed into each other and
contribute to a better understanding of how to model policy processes, increased data validity, and
the identification of key critical points.
Second, if the proposed process is implemented and benefits from a strong community that
contributes to its development, we expect the field to progress towards three core advances. First, as
things stand, the literature on policy processes has identified a robust power-law of public budgets,
which is a decent proxy for policy change (B. D. Jones et al., 2009). The same literature has
identified drivers of political behavior and policy networks (Moynihan, 2018). Yet, both streams are
somewhat detached from one another (Thomas, 2017). Ontologically akin to Schelling’s work on
Micromotives and Macrobehaviors (Schelling, 2006), we expect the computational turn to reconcile
micro and macro level and provide empirical explanations of how policies change as a function of
small changes at the micro-level. While public budgets depict empirical regularity across contexts
(Baumgartner et al., 2018), the driving mechanisms may be contextually different (Chan & Zhao,
2016; Sharp, 2019). Moreover, the formalization of complex policymaking systems will also enable
the exploration of a large set of correlates and test the sensitivity of initial conditions (e.g. network
16
structures), the formation of coalitions and the effects of exogenous perturbations. As such, this
avenue looks very similar to the computational developments in the study of conflict. A power-law of
the frequency and severity of conflict was discovered in the middle of the twentieth century and has
been confirmed over time (Clauset, 2018; Richardson, 1960). In parallel, other scholars have theorized
mechanisms of inter-state and intra-state conflicts (Kalyvas, 2006). Only very recently, empirically
valid computational models formalize micro-level dynamics and generate validated power-laws of
conflicts (De Mesquita, 2006; Lee et al., 2020).
Second, reconciling micro-level dynamics and macro-level outcomes computationally and empir-
ically can, in turn, allow the exploration of levers (Meadows, 1997) for policy change. For instance,
how does shifting information provision (e.g. increasing information quantity in the system) com-
pare to shifting information processing (e.g. allowing actors to process more or less information)
versus changing network structures (e.g. clustering or declustering networks) versus changing rules
(e.g. incentives and reward structures) (B. D. Jones & Baumgartner, 2005a; van Veen, Kudesia, &
Heinimann, 2020)?
Third, the examination of those different levers within formal models can shed light on how to
intervene in policymaking systems. Therefore, the ultimate contribution of computational policy
process studies could be to generate evidence-based recommendations to reform policymaking such
that it better accounts for the bounded rationality of policy actors, the collective nature of its
processes, its changing and information-rich environments and its inherent ambiguity. For example,
which mechanisms can install the right feedback loops to foster an adequate division of power and
better participation in decision-making?
The proposed research avenues characterise the computational turn of policy process studies.
However, this turn does not imply the dominance of mathematicians and physicists within the social
sciences. Historians, philosophers and political scientists do need to contribute their knowledge.
The idea is to combine their questions and insights with the methods and ways of thinking from
the computational sciences. It is a truly interdisciplinary project aiming to better understand the
dynamics of collective action. As such, it is itself a collective endeavour that must aggregate the
insights of various perspectives. Computational policy process scholars will have to drink their own
medicine.
Conclusion
The 21st century is characterised by many societal challenges that require global coordination. The
most impactful form of such explicit coordination is found in public policymaking. Understanding
policymaking systems is thus crucial to achieving action across societal strata. Policy process studies
has attempted to provide such explanations, ranging from the empirical law of budgets to the drivers
of political behavior. In particular, it has converged on an understanding of policymaking systems
that satisfies the hallmarks of complex systems. While the basic tenets of complexity theory are
widely used in policy process studies, the use of its methodological counterparts - computational
models - is rare and immature.
This methodological paper contributes to the computational turn of policy process studies. First,
we surveyed the literature on complexity and policy processes and identified the current methodolog-
ical frontiers in addressing complexity. Then, we provided definitions, added-values and a procedure
for computational modeling. In particular, we addressed why it lends itself well to the study of pol-
icy processes. Third, we sketched the contours of a computational approach to model policymaking
17
systems, by exemplifying the formalization of key features of policy processes, including social net-
works and adaptive behavior. Fourth, we presented two research avenues for computational policy
process studies: a research process going from formalization to the identification of the signatures of
complex policymaking systems; and three contributions that we expect computational policy process
studies to progress towards - empirical validation, identification of leverage points and mechanism
design.
All in all, this paper aims to appeal to both the social scientists that want to reform the field of
policy process studies and to the computational scientists who are attracted to the examination of
social phenomena. By combining theories of policy processes with recent advances in computational
modeling, this paper combines two fields that could contribute to a better understanding of collective
action.
References
Aviram, N. F., Cohen, N., & Beeri, I. (2020). Wind(ow) of Change: A Systematic Review of Policy
Entrepreneurship Characteristics and Strategies. Policy Studies Journal ,48 (3), 612–644. doi:
10.1111/psj.12339
Axelrod, R. (1997). The complexity of cooperation: Agent-based models of competition and col labo-
ration (Vol. 3). Princeton University Press.
Axelrod, R., & Hamilton, W. D. (1981). The Evolution of Cooperation. , 211 , 8.
Bassett, D. S., Alderson, D. L., & Carlson, J. M. (2012). Collective decision dynamics in the presence
of external drivers. Physical Review E ,86 (3), 036105. doi: 10.1103/PhysRevE.86.036105
Baumgartner, F. R., Breunig, C., & Grossman, E. (2019). Comparative Policy Agendas: Theory,
Tools, Data. Oxford University Press.
Baumgartner, F. R., Jones, B. D., & Mortensen, P. B. (2018). Punctuated Equilibrium Theory:
Explaining Stability and Change in Public Policymaking. In Theories of the Policy Process
(pp. 55–101). Routledge.
Beardsworth, R. (2020). Postmodernism Past, Present and Future. The SAGE Handbook of Political
Science, 203.
Bednar, J. (2012). Prosociality, Federalism, and Cultural Evolution. , 3, 14.
Bhavnani, R., Donnay, K., Miodownik, D., Mor, M., & Helbing, D. (2014). Group segregation and
urban violence. American Journal of Political Science,58 (1), 226–245.
Bhavnani, R., Donnay, K., Reul, M., & Backer, D. (2020). Notes and advice for computational
modelling projects. In Handbook of Research Methods in Political Science & International
Relations (Sage ed.). R. Franzese and L. Curini.
Bostrom, N. (2013). Existential risk prevention as global priority. Global Policy,4(1), 15–31.
Cairney, P. (2012). Complexity theory in political science and public policy. Political Studies Review,
10 (3), 346–358.
Cairney, P. (2019). Understanding public policy. Red Globe Press.
Cairney, P., & Weible, C. M. (2017). The new policy sciences: Combining the cognitive science of
choice, multiple theories of context, and basic and applied analysis. Policy Sciences,50 (4),
619–627.
Castellano, C., Fortunato, S., & Loreto, V. (2009). Statistical physics of social dynamics. Reviews
of Modern Physics,81 (2), 591–646. doi: 10.1103/RevModPhys.81.591
Chan, K. N., & Zhao, S. (2016). Punctuated Equilibrium and the Information Disadvantage of
Authoritarianism: Evidence from the P eople’s R epublic of C hina. Policy Studies Journal ,
44 (2), 134–155.
Cioffi-Revilla, C., & Rouleau, M. (2010). MASON RebeLand: An agent-based model of politics,
environment, and insurgency. International Studies Review,12 (1), 31–52.
18
Clauset, A. (2018). Trends and fluctuations in the severity of interstate wars. Science Advances,
4(2), eaao3580. doi: 10.1126/sciadv.aao3580
Clauset, A., & Gleditsch, K. S. (2012). The developmental dynamics of terrorist organizations. PloS
one,7(11), e48633.
Conte, R., Gilbert, N., Bonelli, G., Cioffi-Revilla, C., Deffuant, G., Kertesz, J., . . . others (2012).
Manifesto of computational social science. The European Physical Journal Special Topics,
214 (1), 325–346.
Curini, L., & Franzese, R. (2020). The SAGE Handbook of Research Methods in Political Science
and International Relations. SAGE.
Deffuant, G., Neau, D., Amblard, F., & Weisbuch, G. (2000). Mixing beliefs among interacting
agents. Advances in Complex Systems,03 (01n04), 87–98. doi: 10.1142/S0219525900000078
De Mesquita, B. B. (2006). Game theory, political economy, and the evolving study of war and
peace. American Political Science Review , 637–642.
Donnay, K. (2017). Big Data for Monitoring Political Instability. International Development Policy
— Revue internationale de politique de d´eveloppement,8(1). doi: 10.4000/poldev.2468
Drucker, J. (2016). At the intersection of computational methods and the traditional humanities. ,
43–68. doi: 10.25969/mediarep/11912
Duffy, G. (1992). Concurrent interstate conflict simulations: Testing the effects of the serial as-
sumption. Mathematical and Computer Modelling ,16 (8-9), 241–270.
Dunbar, R. I. M., Baumard, N., Hamilton, M. J., Hooper, P., Finkel, D. N., & Gintis, H. (2012).
Networking Past and Present. Cliodynamics,3(2). doi: 10.21237/C7clio3215774
Dye, T. R. (1992). Understanding public policy. Prentice Hall Englewood Cliffs, NJ.
Enserink, B., Koppenjan, J. F., & Mayer, I. S. (2013). A policy sciences view on policy analysis. In
Public policy analysis (pp. 11–40). Springer.
Epstein, J. M. (2006). Generative social science: Studies in agent-based computational modeling.
Princeton University Press.
Epstein, J. M. (2008). Why model? Journal of Artificial Societies and Social Simulation,11 (4),
12.
Galam, S. (2008). Sociophysics: A review of Galam models. International Journal of Modern
Physics C ,19 (03), 409–440. doi: 10.1142/S0129183108012297
Geyer, R., & Cairney, P. (2015). Handbook on complexity and public policy. Edward Elgar Publish-
ing.
Geyer, R., & Rihani, S. (2010). Complexity and Public Policy: A New Approach to Twenty-First
Century Politics. Policy and Society. Routledge.
Giger, N., & Kl¨uver, H. (2012). Focusing events and policy change: The aftermath of Fukushima. In
Proceedings of the European Political Science Association Conference, Berlin, Germany (pp.
21–23).
Gintis, H., Doebeli, M., & Flack, J. (2012). The evolution of human cooperation. Cliodynamics,
3(1).
Goutte, C., & Gaussier, E. (2005). A Probabilistic Interpretation of Precision, Recall and F-
Score, with Implication for Evaluation. In D. E. Losada & J. M. Fern´andez-Luna (Eds.),
Advances in Information Retrieval (pp. 345–359). Berlin, Heidelberg: Springer. doi: 10.1007/
978-3-540-31865-1 25
Guerrero, O. A., & Axtell, R. L. (2013). Employment Growth through Labor Flow Networks. PLOS
ONE ,8(5), e60808. doi: 10.1371/journal.pone.0060808
Healy, K. (2017). Fuck Nuance. Sociological Theory,35 (2), 118–127. doi: 10.1177/
0735275117709046
Hegelich, S. (2016). Decision Trees and Random Forests: Machine Learning Techniques to Classify
Rare Events. European Policy Analysis,2(1), 98–120. doi: 10.18278/epa.2.1.7
Hegelich, S. (2017). Deep learning and punctuated equilibrium theory. Cognitive Systems Research,
45 , 59–69. doi: 10.1016/j.cogsys.2017.02.006
19
Hegselmann, R., Krause, U., et al. (2002). Opinion dynamics and bounded confidence models,
analysis, and simulation. Journal of artificial societies and social simulation,5(3).
Helbing, D., Brockmann, D., Chadefaux, T., Donnay, K., Blanke, U., Woolley-Meza, O., . . . others
(2015). Saving human lives: What complexity science and information systems can contribute.
Journal of statistical physics,158 (3), 735–781.
Howlett, M. (2020). Policy instruments: Definitions and approaches. A Modern Guide to Public
Policy.
Johnson, L. (2015). Complexity modelling and application to policy research. Handbook on Com-
plexity and Public Policy.
Jones, B. D. (2003). Bounded Rationality and Political Science: Lessons from Public Administration
and Public Policy. Journal of Public Administration Research and Theory,13 (4), 395–412.
doi: 10.1093/jpart/mug028
Jones, B. D., & Baumgartner, F. R. (2005a). A Model of Choice for Public Policy. Journal of Public
Administration Research and Theory,15 (3), 325–351. doi: 10.1093/jopart/mui018
Jones, B. D., & Baumgartner, F. R. (2005b). The Politics of Attention: How Government Prioritizes
Problems. University of Chicago Press.
Jones, B. D., Baumgartner, F. R., Breunig, C., Wlezien, C., Soroka, S., Foucault, M., . . . Walgrave,
S. (2009). A General Empirical Law of Public Budgets: A Comparative Analysis. American
Journal of Political Science,53 (4), 855–873. doi: 10.1111/j.1540-5907.2009.00405.x
Jones, M. D., Peterson, H. L., Pierce, J. J., Herweg, N., Bernal, A., Lamberta Raney, H., & Za-
hariadis, N. (2016). A River Runs Through It: A Multiple Streams Meta-Review: A Multiple
Streams Meta-Review. Policy Studies Journal ,44 (1), 13–36. doi: 10.1111/psj.12115
Kalyvas, S. N. (2006). The Logic of Violence in Civil War. Cambridge University Press.
Keman, H., & Woldendorp, J. J. (2016). Handbook of Research Methods and Applications in Political
Science. Edward Elgar Publishing.
Klein, R. (2017). Policy Emergence: An agent-based approach (Unpublished doctoral dissertation).
TU Delft.
Klijn, E.-H. (2008). Complexity Theory and Public Administration: What’s New? Public Manage-
ment Review,10 (3), 299–317. doi: 10.1080/14719030802002675
Koliba, C. J., Meek, J. W., Zia, A., & Mills, R. W. (2018). Governance networks in public admin-
istration and public policy. Routledge.
Kollman, K., & Page, S. E. (2006). Computational Methods and Models of Politics. In L. Tesfatsion
& K. L. Judd (Eds.), Handbook of Computational Economics (Vol. 2, pp. 1433–1463). Elsevier.
doi: 10.1016/S1574-0021(05)02029-0
Lam, W. F., & Chan, K. N. (2015). How Authoritarianism Intensifies Punctuated Equilibrium:
The Dynamics of Policy Attention in Hong Kong. Governance,28 (4), 549–570. doi: 10.1111/
gove.12127
Lazer, D. M. J., Pentland, A., Watts, D. J., Aral, S., Athey, S., Contractor, N., . . . Wagner, C.
(2020). Computational social science: Obstacles and opportunities. Science,369 (6507), 1060–
1062. doi: 10.1126/science.aaz8170
Lee, E. D., Daniels, B. C., Myers, C. R., Krakauer, D. C., & Flack, J. C. (2020). Scaling theory
of armed-conflict avalanches. Physical Review E ,102 (4), 042312. doi: 10.1103/PhysRevE.102
.042312
Levy, J. S. (2007). Qualitative Methods and Cross-Method Dialogue in Political Science. Compar-
ative Political Studies,40 (2), 196–214. doi: 10.1177/0010414006296348
Lundgren, M., Squatrito, T., & Tallberg, J. (2018). Stability and change in international policy-
making: A punctuated equilibrium approach. The Review of International Organizations,
13 (4), 547–572. doi: 10.1007/s11558-017-9288-x
McBurnett, M. (1996). Complexity in the Evolution of Public Opinion. In L. D. Kiel & E. Elliott
(Eds.), Chaos Theory in the Social Sciences (pp. 165–196). University of Michigan Press.
Meadows, D. (1997). Places to Intervene in a System. Whole Earth ,91 (1), 78–84.
20
Mettler, S., & Sorelle, M. (2014). Policy feedback theory. Theories of the policy process,3, 151–181.
Mintrom, M. (2015). Herbert A. Simon, Administrative Behavior: A Study of Decision-Making
Processes in Administrative Organization. The Oxford Handbook of Classics in Public Policy
and Administration. doi: 10.1093/oxfordhb/9780199646135.013.22
Mitchell, M. (2011). Complexity: A Guided Tour (1. Edition ed.). Oxford: Oxford University Press.
Morcol, G. (2001). What Is Complexity Science? Postmodernist or Psotpositivist? Emergence,
3(1), 104–119. doi: 10.1207/S15327000EM0301 07
Mor¸ol, G. (2013). A complexity theory for public policy. Routledge.
Moser, C. A., & Kalton, G. (2017). Survey Methods in Social Investigation. Routledge.
Moynihan, D. P. (2018). A Great Schism Approaching? Towards a Micro and Macro Public Admin-
istration (SSRN Scholarly Paper No. ID 3481460). Rochester, NY: Social Science Research
Network.
Moyson, S., Scholten, P., & Weible, C. M. (2017). Policy learning and policy change: Theorizing
their relations from different perspectives. Policy and Society,36 (2), 161–177.
Neumann, M., & Lorenz, J. (2019). Agentenbasierte Simulation in der Politikwissenschaft. In
C. Wagemann, A. Goerres, & M. Siewert (Eds.), Handbuch Methoden der Politikwissenschaft
(pp. 1–24). Wiesbaden: Springer Fachmedien. doi: 10.1007/978-3-658-16937-4 34-1
Page, S. E. (2008). Agent-based models. The New Palgrave Dictionary of Economics: Volume 1–8,
47–52.
Page, S. E. (2018). The model thinker: What you need to know to make data work for you. Hachette
UK.
Pierce, J. J., Giordono, L. S., Peterson, H. L., & Hicks, K. C. (2019). Common approaches
for studying advocacy: Review of methods and model practices of the Advocacy Coalition
Framework. The Social Science Journal.
Pines, D. (2018). Emerging Syntheses In Science. CRC Press.
Rhodes, R. A., & Marsh, D. (1992). New directions in the study of policy networks. European
journal of political research,21 (1-2), 181–205.
Richards, D. E.-A., & Doyle, D. R. (2000). Political complexity: Nonlinear models of politics.
University of Michigan Press.
Richardson, L. F. (1960). Arms and insecurity: A mathematical study of the causes and origins of
war. Boxwood Press.
Sabatier, P. A. (Ed.). (2007). Theories of the policy process (2nd ed ed.). Boulder, Colo: Westview
Press.
Saperstein, A. M. (1984). Chaos—a model for the outbreak of war. Nature,309 (5966), 303–305.
Schelling, T. C. (2006). Micromotives and macrobehavior. WW Norton & Company.
Seb´ok, M., & Berki, T. (2017). Incrementalism and punctuated equilibrium in Hungarian budgeting
(1991-2013). Journal of Public Budgeting, Accounting & Financial Management,29 (2), 151–
180.
Sharp, T. (2019). Wars, presidents, and punctuated equilibriums in US defense spending. Policy
Sciences, 1–30.
Simon, H. A. (1955). On a Class of Skew Distributions Functions. Biometrika ,42 (3-4), 425–440.
doi: 10.1093/biomet/42.3-4.425
Simon, H. A. (1957). Models of man; social and rational. Oxford, England: Wiley.
Simon, H. A. (1969). The Sciences of the Artificial. MIT Press.
Simon, H. A. (1991). The Architecture of Complexity. In Facets of Systems Science (pp. 457–476).
Boston, MA: Springer US. doi: 10.1007/978-1-4899-0718-9 31
Simon, H. A., & March, J. (1976). Administrative behavior and organizations. New York: Free
Press.
Stauffer, M. (2021). Complexity Science and the Study of Armed Conflict: A Narrative Review. In
Complex Systems in the Social and Behavioral Sciences (University of Michigan Press ed.). E.
Elliott and L. D. Kiel.
21
Teisman, G. R., & Klijn, E.-H. (2008). Complexity Theory and Public Management. Public
Management Review,10 (3), 287–297. doi: 10.1080/14719030802002451
Thomas, H. F. (2017). Modeling contagion in policy systems. Cognitive Systems Research,44 ,
74–88. doi: 10.1016/j.cogsys.2017.03.003
Thurner, S., Hanel, R., & Klimek, P. (2018). Introduction to the theory of complex systems. Oxford
University Press.
Ubaldi, B. (2013). Open Government Data: Towards Empirical Analysis of Open Government Data
Initiatives.
doi: 10.1787/5k46bj4f03s7-en
van Veen, D.-J., Kudesia, R. S., & Heinimann, H. R. (2020). An Agent-Based Model of Col-
lective Decision-Making: How Information Sharing Strategies Scale With Information Over-
load. IEEE Transactions on Computational Social Systems,7(3), 751–767. doi: 10.1109/
TCSS.2020.2986161
Van Waarden, F. (1992). Dimensions and types of policy networks. European journal of political
research,21 (1-2), 29–52.
Watts, D. (2016). Computational Social Science: Exciting Progress and Future Challenges. In
Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery
and Data Mining (p. 419). New York, NY, USA: Association for Computing Machinery. doi:
10.1145/2939672.2945366
Weible, C. M., & Sabatier, P. A. (2018). Theories of the Policy Process. Routledge.
Weidmann, N. B., & Salehyan, I. (2013). Violence and ethnic segregation: A computational model
applied to Baghdad. International Studies Quarterly ,57 (1), 52–64.
22
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Armed conflict data display features consistent with scaling and universal dynamics in both social and physical properties like fatalities and geographic extent. We propose a randomly branching armed conflict model to relate the multiple properties to one another. The model incorporates a fractal lattice on which conflict spreads, uniform dynamics driving conflict growth, and regional virulence that modulates local conflict intensity. The quantitative constraints on scaling and universal dynamics we use to develop our minimal model serve more generally as a set of constraints for other models for armed conflict dynamics. We show how this approach akin to thermodynamics imparts mechanistic intuition and unifies multiple conflict properties, giving insight into causation, prediction, and intervention timing.
Article
Full-text available
Under what conditions does the USA adjust its defense spending dramatically? Scholars have identified many factors that affect military budgets, from international threats to domestic politics. Yet, most existing studies use regression analysis to estimate average marginal effects, thereby neglecting large-scale outlier “punctuations” that, though rare, supply theoretical insights, set institutional trajectories, and shape aspirations for future policy. Blending scholarship from public policy, international relations, and defense analysis, this article uses punctuated equilibrium theory (PET) and a mixed-method research design to argue that either a change in war policy or a presidential transition is necessary for a US defense spending punctuation. War debates and presidential transitions facilitate punctuations by triggering shifts in policymaker attention and policy subsystem structure, two mechanisms central to PET theorizing. In its quantitative section, the article uses a mathematical threshold to identify four punctuations since 1950: Truman’s Korean War buildup, Eisenhower’s post-Korean War drawdown, Kennedy’s peacetime civil defense buildup, and Bush I’s post-Gulf War, post-Cold War drawdown. War policy or a presidential transition figured prominently in each case. In its qualitative section, the article analyzes the Kennedy period in greater detail because, lacking a hot war, the case was least likely to witness a punctuation and therefore represents the hardest test for PET. In line with the theory’s expectations, Kennedy’s muscular agenda setting and subsystem shaping interacted with rising Cold War tensions to cause a dramatic-but-brief increase in civil defense funding to guard against a Soviet nuclear attack.
Chapter
The configuration of factors that lead to armed conflicts seems to bear the hallmarks of a complex system: actor heterogeneity, feedback loops, and nonlinear dynamics. This chapter offers a narrative review of the emergence and evolution of complexity science applied to the study of armed conflict, highlighting the conceptual, theoretical and empirical contributions made by past and present scholarship. It also examines future research avenues and emphasizes the needs to (I) explicitly anchor models in existing theory; (ii) increasingly leverage the vast amount of new data; and (iii) focus on policy-relevant contributions.
Article
The Advocacy Coalition Framework (ACF) is a theoretical framework widely applied by public policy scholars to study advocacy. Since its inception in 1986 by Paul Sabatier, the framework has been applied hundreds of times across the globe to various policy issues. One side effect of this growth and diversification is a lack of common methodological approaches. Inspired by historic calls to balance generalizable knowledge-of-policy process with contextualized knowledge-in-policy, as well as more recent calls for comprehensive accounting of trends in studies with such a diversity of application, this research supports social science dialogue and knowledge-building by offering a better understanding of the methods used in ACF applications and by highlighting model approaches. We used content analysis to categorize data sources and analytic methods of 161 English language peer-reviewed journal articles applying the ACF published between 2007 and 2014. We found that most ACF applications relied on qualitative analysis of interviews and documents, and analytic method varied according to theoretical focus. Multiple ACF applications modeled clear, compelling and varied data collection and analytic approaches consistent with theoretical conceptualization. Overall, this research provides a more detailed accounting and understanding of what methods social science scholars around the world use to study public advocacy and discusses some exemplary methods for theory testing.
Book
This collection illustrates how nonlinear methods can provide new insight into existing political questions. Politics is often characterized by unexpected consequences, sensitivity to small changes, non-equilibrium dynamics, the emergence of patterns, and sudden changes in outcomes. These are all attributes of nonlinear processes. Bringing together a variety of recent nonlinear modeling approaches, Political Complexity explores what happens when political actors operate in a dynamic and complex social environment. The contributions to this collection are organized in terms of three branches within non-linear theory: spatial nonlinearity, temporal nonlinearity, and functional nonlinearity. The chapters advance beyond analogy towards developing rigorous nonlinear models capable of empirical verification. Contributions to this volume cover the areas of landscape theory, computational modeling, time series analysis, cross-sectional analysis, dynamic game theory, duration models, neural networks, and hidden Markov models. They address such questions as: Is international cooperation necessary for effective economic sanctions? Is it possible to predict alliance configurations in the international system? Is a bureaucratic agency harder to remove as time goes on? Is it possible to predict which international crises will result in war and which will avoid conflict? Is decentralization in a federal system always beneficial?.