A CRITIQUE OF THE USE OF PATH

DEPENDENCY IN POLICY STUDIES

ADRIAN KAY

Path dependency is an important notion in diachronic approaches to understanding

social and political processes. The first section of this paper examines the application of
path dependency to policy studies; the advantages of the concept in understanding
policy development are highlighted by examples from pension policy and social
housing policy in the UK, and the EU budget. The next section considers several
criticisms of path dependency: (1) it is a fashionable label for the intuition that ‘history
matters’ without a clear and convincing account of decision-making over time; (2) it
explains only stability and not change; (3) its normative implications are confused and
mostly left unexplored. The final section concludes that path dependency, despite
being theoretically inchoate and difficult to operationalize empirically, is a valid and
useful concept for policy studies. However, its proper application demands sensitivity
from scholars to other temporal dynamics that may operate in policy development.

There is an increasing emphasis on the role of temporality in the analysis of

social and political processes; indeed some claim this is a part of a ‘historic
turn’ in the social sciences (McDonald 1996). This is a return rather than a
new trend; for example, Hodgson (2001) charts the rise and fall of the
question of the role and nature of historical specificity in mainstream
economics. An emphasis on history is a defining feature of heterodox
traditions in economics, for example, in the works of Georgescu-Roegen
(1967, 1971) on the possibility of an entropy law and Kaldor (1970) on
cumulative causation. Earlier and beyond economics, writers such as Marx,
Hegel and Toynbee all believed in some form of historicism. Pierson (2000a,
p. 72) puts the contemporary form of this ambition for temporal analysis
pithily: we want to move to a world of moving pictures not static snapshots.
A central concept in the recent theoretically informed investigation of historical
causation is path dependency (Krasner 1984; North 1990; Tilly 1991; Thelen
1999; Goldstone 1998; Hay and Wincott 1998; Pierson 2000a, b, c).
A process is path dependent if initial moves in one direction elicit further
moves in that same direction; in other words the order in which things happen
affects how they happen; the trajectory of change up to a certain point constrains
the trajectory after that point. As the Nobel Laureate Douglass North (1990,
pp. 98–9) puts it: path dependency is a process that constrains future choice sets:
At every step along the way there are choices – political and economic –
that provide…real alternatives. Path dependence is a way to narrow

Adrian Kay is in the Department of Politics and Public Policy, Griffith University, Brisbane.

Public Administration Vol. 83, No. 3, 2005 (553–571)

© Blackwell Publishing Ltd. 2005, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street,
Malden, MA 02148, USA.
554 ADRIAN KAY

conceptually the choice set and link decision-making through time. It is

not a story of inevitability in which the past neatly predicts the future.’
The reference to choice sets and decision making reveals the origins of the
concept in economics. Indeed, path dependency is problematic for that
discipline because, at a broad level, a path-dependent decision is one that is
informed by its historical setting such that the rationality at issue is
‘bounded’ in some relevant respect and thus likely to produce sub optimal
outcomes. The relationship between path dependency and efficiency in
policy studies is examined in more detail in the section on criticisms.
The concept of path dependency is neither a framework nor a theory or
model in the terms of Ostrom (1999, pp. 39–41): it does not provide a general
list of variables that can be used to organize ‘diagnostic and prescriptive
inquiry’; nor does it provide hypotheses about specific links between vari-
ables or particular parameters of those links. Instead, path dependency is an
empirical category, an organizing concept which can be used to label a cer-
tain type of temporal process. The application of this label to a phenomenon
is a form of explanation; it competes with alternatives – such as the particu-
lar political circumstances pertaining at different times – to provide the best
explanation of that phenomenon. Importantly, however, the concept of path
dependency does not per se provide necessary or sufficient conditions to
understand or explain that which it labels: path dependent processes, even
when identified, require theorizing. This is considered as a criticism of the
concept in the second section of the paper.
The empirical difficulties of identification are significant. The observation
that change has been limited over a period of time is not sufficient to infer a
path-dependent process: an unconstrained but stable series of choices
would be consistent with this observation. In order to demonstrate
constrained change, and thus identify a path-dependent process, it is also
necessary to show that what did not happen could not have happened – that
is, that certain options were not feasible because of earlier sequences of deci-
sions. The construction of such a counterfactual is a formidable challenge;
most empirical policy applications of the concept concentrate on identifying
constraints that proscribe certain options.
The section that follows considers the application of path dependency to
the analysis of policy development and its potential advantages in under-
standing the dynamics of that development. The next section considers
several criticisms of the concept: (1) that it lacks a convincing account of
decision-making over time, both of the accumulation of constraints and
context bound rationality; (2) that it is incapable of dealing with policy
change; and (3) that it lacks a clear normative focus. The paper’s conclusion
argues that despite its theoretical underdevelopment and relatively limited
number of successful empirical applications, the concept of path depend-
ency does have potential utility in the field of policy studies in terms of
explaining not only why policies might be difficult to reform but also why

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 555

they may become more complex over time. However, for that potential to be
realized requires explicit recognition of some of the limitations of and
assumptions behind the concept.

PATH DEPENDENCY AND POLICY STUDIES

Path dependency has been used within political science almost exclusively
within a broad institutionalist framework. It is institutions that are path
dependent and the challenge for scholars is to discover the relevant explan-
ations in the operation of systems of institutions and their interaction with
the wider socio-economic and political environment. Raadschelders (1998,
p. 569) states that: ‘whatever the discipline…contemporary neo-institutional
analysis has one feature in common: the notion of path dependency’. The
concept is particularly associated with historical institutionalism (see Hall
and Taylor 1996; Thelen 1999; Hay 2002) and is one of the defining features
of that school of thought. Historical institutionalism embraces the idea that
individuals act within institutional arrangements, the present structure and
functioning of which are understood only partially when not embedded in a
historical perspective.
I follow North (1990), Hall and Taylor (1996), Goodin (1998), Ostrom
(1999) and Pierson (2000 a, b, c), in my use of the term institution in this
paper. Individual agents and groups act in a context that is collectively
constrained and these constraints take the form of institutions. The category
of institutions can be widely drawn as in the definition offered by Hall and
Taylor (1996, p. 938):
the formal or informal procedures, routines, norms and conventions
embedded in the organizational structure of the polity or political
economy. They can range from the rules of a constitutional order to the
conventions governing trade union behaviour or bank–firm relations. In
general, historical institutionalists associate institutions with organiza-
tions and the rules or conventions promulgated by formal organization.
Three different levels of institutions are regularly distinguished: the
macro or constitutional level; the collective choice or policy decision level;
and the operational level of individual decisions. Pierson (1993) argues that
most institutionalist scholars have tended to focus their attentions on formal
institutions at the macro or constitutional level. For example, in this journal,
Holzinger and Knill (2002) considered path dependency of institutions at
the EU constitutional level.
However, the present paper is concerned with the policy level; as Pierson
(1993, p. 596) states:
major public policies also constitute important rules of the game, influ-
encing the allocation of economic and political resources, modifying the
costs and benefits associated with alternative political strategies, and
consequently altering ensuing political development.

© Blackwell Publishing Ltd. 2005

556 ADRIAN KAY

In spite of this step, Pierson appears hesitant about the claim that public
policies may be path dependent: it appears as footnote 50 in Pierson (1993)
and is elsewhere described as ‘unorthodox’ (Pierson 2000b, p. 255). There is
justification for such hesitancy: one common and essential element in most
writers’ use of the term policy is purposiveness of some kind (Parsons 1995,
pp. 13–16). Policy expresses a general set of objectives or a desired state of
affairs. In a well-known definition of policy, public policy is: ‘anything a
government chooses to do or not to do’ (Dye 1972, p. 2). Policy is about
choice; the choice of reasons for (in)action, the choice of policy instruments,
the choice of how to respond to the consequences of policy outputs. Institu-
tions are different; no matter how broadly defined they are or whatever
intentional goal underlies their introduction or any function that may be
attributed to them, institutions count as such by their collective acceptance
as rules or constraints to govern behaviour. This holds whether or not institu-
tions are respected or whether or not they are followed, consciously or
unconsciously, as in internalized habits and routines.
The conceptual distinction between a policy and an institution is significant
in terms of the use of path dependency in policy studies and provides two
options for progress. The first is to proceed by analogy from institutions to
policy as the subject of path dependency and so allow the existing body of
work in historical institutionalism to be used to support theory-building and
empirical testing in specific studies of policy. The second option would be to
apply and theorize the concept in policy development sui generis. Almost all
recent work has proceeded using the first route since the second would
separate the concept of path dependency from historical institutionalism.
This is possible, as shown in the section on the origins of the concept, but
would require undertaking significant new theoretical work with uncertain
benefits. The issue of the robustness of the argument by analogy from
institutions to policy is considered in the section that follows. The initial
question considered here is: what does it mean to say that a policy is path
dependent?
There is no unique policy level or scale but rather several levels that may
be examined as ‘policy’, as Heclo (1972, p. 84) puts it:
As commonly used, the term policy is usually considered to apply to
something ‘bigger’ that particular decisions, but ‘smaller’ than general
social movements. Thus, policy, in terms of levels of analysis, is a concept
placed roughly in the middle range.
The first scale in this middle range is that of the policy system. In these
terms, policy is a ‘whole’ or system as in discussions of health policy,
defence policy or housing policy. It is not simply journalistic shorthand to
talk about policy development, there is a ‘whole’ or a ‘system’ at a policy
level that can be the subject of active and passive verbs and the object of
empirical investigation without stretching the limits of our imagination too
far. A policy system is a relevant and valid unit of analysis for the application

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 557

of path dependency. However, while the conception of policy as a single,

point variable tracing a path in policy space is adequate in terms of making
theoretical progress, but the empirical challenge remains: what specifically
about a policy system is path dependent. In other words, how do we oper-
ationalize a policy system as a variable?
A policy system is a complex, composite variable consisting of many
interrelated elements. Within a policy system there are several policy sub-
systems (or elements), each with their set of actors, organizations, goals and
instruments (Baumgartner and Jones 2002, Chapter 1). For example, within
the health policy system there is the primary care policy subsystem and the
public health policy subsystem. The development of policy subsystems may
equally be understood, using path dependency, as the policy ‘whole’ itself.
Rose and Davies (1994) argue the policy programme level should be the
basic unit of analysis in policy studies because it is most readily observable.
A policy programme refers to a specific combination of laws, commitments,
appropriations, organizations and personnel directed towards a more or
less clearly defined set of goals. In other terms, this is a policy instrument; an
identifiable tool or resource of government used for a specific set of pur-
poses. This is a more finely grained perspective than the policy subsystem.
One of the corollaries of different policy scales is a lack of precise frame-
works for ‘measuring’ policy or, alternatively, locating policy continuity and
policy change. Judging policy change is difficult because even if, at the macro-
scopic or policy system level, there is limited change and policy is con-
sidered stable, there may be concurrent change occurring at the policy
subsystem level or programme/policy instrument level. In the context of
economic policy, Hall (1993) sets out three orders of policy change: change
in the level of policy instruments (first order); change in the instruments
actually used (second order change); and change in the overall policy para-
digm (third order change which occurs at the policy system level). Under
this categorization, policy change and policy stability may be simultan-
eously observed. Hall’s work is important because it allows the crucial logical
distinction between changes in things and changes in kind to be made in
terms of policy. It gives us categories of things (policy instruments and their
different levels) and kind (policy paradigm). This allows us to assess
whether an existing policy is changing or whether a new policy is being
introduced.
As noted earlier, within the policy system ‘whole’, there are various struc-
tures at different scales which act as institutions in shaping agents’ decision-
making in the formulation and implementation of policy. These are not
reducible either to individual level agents or elements in the policy process.
Examples of such policy institutions are budget rules, policy networks and
standard operating procedures in government departments and agencies.
Most importantly in terms of understanding policy development as path
dependent, past policy decisions are institutions in terms of current policy deci-
sions: they can act as structures that can limit or shape current policy options.

© Blackwell Publishing Ltd. 2005

558 ADRIAN KAY

The question of what it is about a policy that is path dependent does not
admit a single, conclusive answer; rather, it remains an open and empirical
question for scholars applying the concept. It is incumbent upon scholars
using the concept, whether theoretically or empirically, to be clear about in
which sense the term policy is being used: in particular, which scale of per-
spective is being adopted – the level of the whole or particular constituent
elements of a policy or to policy level institutions?

Advantages of path dependency for policy studies

Path dependency is an appealing concept for understanding public policy
development; it provides a label for the empirical observations and intui-
tions that policies, once established, can be difficult to change or reform.
Recent examples of the use of path dependency to understanding policy
development include health care policy in the US (Wilsford 1994; Hacker
1998, 2002 and the UK (Greener 2002); the reform of housing benefit in the
UK (Kemp 2000); to UK pension policy (Pemberton 2003); and the Common
Agricultural Policy (CAP) of the EU (Kay 2003).
Path dependency encapsulates the insight that policy decisions accumu-
late over time; a process of accretion can occur in a policy area that restricts
options for future policy-makers. In this sense, path dependency arguments
can ‘provide an important caution against a too easy conclusion of the inevit-
ability, “naturalness”, or functionality of observed outcomes’ (Pierson
2000b, p. 252). For example, Pemberton (2003) argues that the pensions
‘crisis’ in the UK is not primarily demographic but rather due to a low
savings rate; further, this low rate is a function of the path dependency and
increasing complexity of pension policy. The system of pension provision in
the UK has shifted over the last 20 years from one dominated by state provi-
sion to one in which the state pension plays a residual welfare role. Despite
this large change at the policy system level, there is evidence of path
dependency in particular policy subsystems. In the case of UK pensions,
policy subsystems exist around specific pension schemes. An individual
contract established under a particular pension scheme at a particular time
is costly to change: there are large sunk costs; increasing returns associated
with rising numbers of contributors and pensioners in a particular scheme;
further, there may be significant learning effects. All of these factors contribute
to significant switching costs for the abolition of one scheme and the
transferring of that set of individual contracts into a superseding scheme.
Particular schemes are ‘locked in’ for particular individuals. Nonetheless,
pension reform has been possible but change has come in the form of the
addition of new schemes or elements to the system. This amounts not to a
single, path-dependent policy trajectory but rather to a widening array of
‘locked in’ subsystems over time. This accretion of new subsystems or
schemes has led to the increasing complexity of the overall system of
pensions and raised questions of effectiveness at a policy ‘whole’ level.

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 559

Kemp (2001) reports similar dynamics of increasing complexity with regard

to housing benefit policy in the UK.
Path dependency encourages explicit attempts at dynamic analysis. In this
sense, dynamic means that time is an independent variable in the explan-
ation of change. This contrasts with comparative static explanations of
change and development where time, if it is considered at all, is simply a
dependent variable in a theory or model where change results from some
change in initial parameters or an exogenous shock. Dynamic analysis is
superior in that it allows for the possibility of temporal distance between
causal factors and subsequent observed change; it insists on an account
of how the drivers of policy change unravel themselves over time; it allows
the possibility of change as a function of time and therefore not just
different orders of policy change as in Hall (1993), but different rates of policy
change.
The concept of path dependency is valuable in understanding issues such
as the development of the EU budget system. The establishment of set of
budget rules in the early period of the EU have limited options for future
budgetary institutions. Institutional change has consisted of additions to the
initial set of rules which have remained as fixed and immutable in the
budget process. Further, the choice of each subsequent budget rule has been
determined by the need to maintain and validate existing budgetary institu-
tions. The path-dependent nature of budget rules in the EU has had the
cumulative effect of limiting the set of budgetary options available to EU
policy-makers, with the corollary of significant limits on EU policy activism.
The Treaty of Rome, A.268, introduced a balanced budget rule (BBR) for
the EU budget; and after the 1970 Treaty of Luxembourg, EU spending has
been classified either as Compulsory Expenditure (CE) or Non-Compulsory
Expenditure (NCE). These are the two principal budgetary rules in the EU.
This CE-NCE distinction has no accounting or administrative logic, but
means that once expenditure-generating policies in the CE category have
been agreed, all resulting expenditure obligations are required to be met.
CAP spending was assigned to the CE category and dominated the entire
EU budget up until the late 1980s because the Council of Agriculture Ministers
(CoAM) could decide the CAP and its levels of support without significant
reference to any wider EU budget process.
The main instrument of the CAP prior to 1992 was price support. This
generated two main types of budget expenditure: export subsidy and the
costs of physical storage. Given the CE budget rule, such expenditure obli-
gations had to be met regardless of how much it cost. The problem for the
EU was that the combination of higher prices and guaranteed outlets (the
government would buy if the market would not) encouraged farmers to
produce ever greater quantities. This ‘open-ended’ support resulted in CAP
spending rising over time as production and, increasingly, surpluses, rose.
The net budgetary burden of the CAP threatened the BBR and the financial
viability of the EU from the late 1970s onwards.

© Blackwell Publishing Ltd. 2005

560 ADRIAN KAY

This tension provided the backdrop to a series of ineffective CAP reforms

in the early to mid-1980s as the EU went through various budget crises. The
budget system was significantly reformed in 1988; the central element was
the introduction of Financial Perspectives which set out, for several years at
a time, permitted EU spending by main policy area. These were part of the
first of several subsequent budgetary inter-institutional Agreements (IIAs)
designed to introduce greater budgetary discipline. However, the tension
between the two main, path-dependent budget rules continued since it
remained possible for the rule embedded in the definition of CE to over-ride
the constraint of the BBR. Under the 1988 budget system, without changes to
the basis of spending under the CAP, respecting budget limits would
remain a fundamental challenge to the EU. Indeed, it was only when the
CAP was reformed in 1992 to introduce an element of direct payments
alongside price support mechanisms that the CAP budget was able to be
subjected to some form of multi-period financial planning, that is, the open
ended nature of the policy was, to some degree, checked. This has allowed
the 1988 budgetary system to be developed in subsequent IIAs to bring a
substantial degree of financial control in the budget, at a global, policy and
operational level.
The key rules under which the EU budget operates are path dependent.
The main two rules are established in EU treaties and change would require
unanimous agreement among the member states. In the initial budgetary
framework designed by the EU, the BBR was insufficient to maintain bud-
getary discipline because of the ability of the CoAM to commit to ‘open-
ended’ support under the CAP. This tension between the two basic fiscal
rules of the EU created a series of EU budget crises. These led to CAP reform
in 1992 to allow the policy to be part of a multi-annual financial planning
system and successive budget reforms to reduce policy flexibility for other
Councils. The concept of path dependency can help understand how the
current EU budgetary framework has evolved around the fixed and immut-
able constraints of the CE category and BBR.
The concept of path dependency has the further advantage of flexibility
for policy scholars. It is capable of being used in terms of first and second
order policy change; even when there may be simultaneous third order
change at the level of a policy paradigm that is explained by quite different
theories. The insights of path dependency can complement rather than rival
other accounts of policy change. Social housing in the UK is an example of
where path dependency can hold at the subsystem level, with interesting
consequences, but where the policy system as a whole has changed
profoundly. Since the mid-1970s there has been a series of failed initiatives
by central government to directly control the rents charged in the social
housing sector and ensure equity between local authorities and housing
associations. This particular element of social housing policy is path depend-
ent; each local authority has an established policy for rent calculation, often
determined by initial decisions made in the immediate post-war era. That

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 561

these have proved resistant to central control or influence significantly

constrains the ability of central government to pursue some of its objectives
for social housing. However, over the same period as this path-dependent
process, there has been a clear shift in the social housing policy paradigm.
One of the five ‘pillars’ of the UK welfare state has been, to a large extent,
removed. Between 1971 and 2002 home ownership increased from 49 per cent
to 69 per cent, with most of the increase occurring in the 1980s. The ‘right to
buy’ scheme introduced in the early 1980s contributed to the increase in
home ownership, since it allowed local authority tenants to buy their own
homes. Corresponding to this, the percentage of households renting council
homes increased from 31 per cent in 1971 to 34 per cent in 1981, but then
gradually declined during the 1980s to 24 per cent in 1991. This percentage
has continued to decrease and in 2002 14 per cent of households rented from
the council.

CRITICISMS OF THE CONCEPT

Accounts of decision making over time
The criticism that the concept path dependency lacks explanatory power is
well expressed by Raadschelders (1998, p. 576):
it is only by virtue of retrospect that we are aware of stages or paths of
development. ‘Path dependency’ refers to a string of related events: cau-
sality in retrospect. The concept does not come even close to pinpointing
a mechanism or the mechanisms that propel social change.
The quotation contains two criticisms. The first is that the concept cannot be
used for current or future phenomena. This is, of course, not a singular
feature of path dependency but common to many concepts that are useful
for retrospective, ‘thick’ historical description in the social sciences. In
addition, such concepts are candidates for explanation by abduction, a
notion closely associated with the American pragmatist tradition. The
formal structure of abduction is: a theory (T) would, if true, explain a collection
of data (D); no alternative theory can explain D as well as T does; therefore T
is probably true or ‘more credible’ in the terms set out by Stinchcombe (1968,
pp. 17–19).
The more important criticism is that even if one accepts path dependency
as a possible candidate for explanation qua abduction it is unlikely to be con-
vincing because the notion does not provide any fine-grained mechanisms
that might provide necessary and sufficient conditions for the process
observed. According to Elster (1998, p. 45), mechanisms are: ‘…frequently
occurring and generally recognizable causal patterns that are triggered
under generally unknown conditions or with indeterminate consequences’.
The ubiquity of the language of mechanisms owes something to the persist-
ent ambition for covering law explanations (or deductive-nomological
explanations – see Hempel 1965) in the social sciences; combined with a

© Blackwell Publishing Ltd. 2005

562 ADRIAN KAY

persistent failure to uncover generalizations that might be formulated as a

law in historical development. Mechanism are not desirable per se; rather,
they are good only because they allow explanation when law-like general-
izations do not appear. To offer an explanation of an outcome or event (the
explanandum) in these terms requires the statement, after the event, of a set of
initial conditions and a mechanism that connects initial conditions to
outcome that is more general than the phenomenon it subsumes. That is,
mechanisms do not contribute to explanation because they operate at a
lower level to that being explained: policy as a ‘whole’ may be explained in
terms of other factors at that level of abstraction without demanding any
reduction to a more micro-scale level.
The ambition to gain some degree of generalizability highlights a poten-
tial schism between the social scientific view of society and historical
perspectives. Social science is essentially a nomothetic exercise: it endeav-
ours to establish laws or mechanisms, based on the fact that events in nature
are repeatable and recurrent. In contrast, history does not repeat itself. It
occurs only once in a certain context and at a certain time. Thus, any examin-
ation of history can only be idiographic, that is, a description of events that
have already occurred and are unique. The question of whether macro-scale
events can be reduced to antecedent micro-scale events that exhibit some
regularity across time and political contexts is current and live amongst
historians (see, for example, Roberts 1996; Roth 1999). However, a positive
answer to this question is central to the utility of path dependency in terms
of explaining policy development.
One influential strand of the literature on path dependency has worked
on the micro-foundations of the concept, using the insights of new institutional
economics (David 1985; North 1990; Arthur 1994). Much of this literature,
particularly that by US-based scholars, proceeds by analogy from technolog-
ical development to institutional development. In simple terms, imagine
two technologies, A and B, both of which are subject to increasing returns
but there is uncertainty over the rate of increasing returns. Initial adoptions
of one technology, say B, which may occur for a number of small or chance
reasons, beget further adoptions of B in the market because of increasing
returns, that is, it becomes cheaper for future firms to adopt technology B
rather than A. The interesting results from the models built on these
assumptions (for economists at least) are that you might get inefficient tech-
nologies adopted by markets. The normative implication of this borrowing
from economics is considered in more detail below.
Arthur (1994) states the circumstances in which path dependence qua an
increasing returns process is likely as the presence of: (1) large fixed (and
sunk) costs; (2) potential network effects; (3) potential learning effects, and
(4) adaptive expectations. As noted, these factors have been used at a macro,
constitutional level to make arguments about path dependency in institu-
tional development (see Pierson 2000b for a summary). However, a focus on
increasing returns is only a partial interpretation of the economics of path

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 563

dependency. Increasing returns are sufficient (but not necessary) for path
dependency. As Arrow (2000) points out, significant sunk costs and
sequencing arguments are sufficient to construct path dependency stories in
the economics of technology. Further, Arthur Stinchcombe’s celebrated
work (1968, Chapter 5) on constructing theories of historical causation
emphasizes the central importance of sunk costs. Thus, work should not focus
exclusively on increasing returns processes and it is moot whether this mechan-
ism type should be privileged over other sources of path dependency.
This is an important point since a number of non-increasing returns mech-
anisms have been suggested as underlying path dependency in policy
development. These include: (1) the effect of policy on interest groups as
when policies constrain some groups and enable others (Pierson 2000b); (2)
policies involve investment or disinvestments in administrative infrastructure
this transforms governmental capacity and the set of possible future policies
that may be enacted (Skocpol 1992); (3) policies involve the establishment of
formal or informal contracts with individuals (Kemp 2001; Pemberton 2003;
Kay 2003) which are costly to change. Further, there are network effects to
types of a contract; that is the transaction costs of agreeing another contract
of that type in that area of public policy will be considerably lower than a
different, and in particular bespoke, contract.
All these policy specific mechanisms are based on definite, conscious
choices that have the foreseeable consequence of high future switching
costs; none relies on an increasing returns process. Insofar as policy-making
does not involve small, unintentional and random actions that have signifi-
cant consequences because of increasing returns, the argument by analogy
from the economics of technology to policy is weak. There are, however,
examples that suggest that increasing returns processes can occur in policy
development. In the reform of the primary care sector in the UK after 1997, a
series of primary care models were piloted. By a series of chance factors, a
particular primary care trust model quickly became popular. This model
subsequently became the government’s template for all future combinations
of primary care organizations. There was no particular feature to this model
to recommend it over any of the others that were piloted between 1997 and
2000; instead, it was the case that this model was adopted early in the
government’s reform process, something that made it considerably easier
(or cheaper) for subsequent primary care groups to use it, and with such
a momentum it became the template adopted by the government for all
primary care agglomerations.
An emphasis on policy as choice raises important questions of rationality.
In the mechanisms noted above, agents make choices according to the costs
and benefits of the consequences of different options. Any borrowing from
economics inevitably situates the agent in terms of responding to these costs
and benefits in a manner consistent with straightforward parametric ration-
ality. The assumption of this type of rationality serves certain purposes in
formal economic modelling, but to use it in more informal, intuitive and

© Blackwell Publishing Ltd. 2005

564 ADRIAN KAY

post-positivist accounts of path dependency in public policy is problematic.

Hay (2004) provides an extensive critique of the issues surrounding rational
choice theory in public administration; it is sufficient to note here that this
version of rationality is central to the development of the concept of path
dependency by analogy to the recent economics of technological change.
The notion of context-bound rationality is at the heart decision making in
a path-dependent process. A borrowing from new institutional economics
sees institutions as altering the payoffs to agents of various strategies or
actions. However, Nooteboom (1997) describes the manner in which
markets lock-in to certain technologies in similar terms to how philosophers
of science characterize the entrenchment of scientific theories. He cites
Kuhn’s famous account of how scientific theories develop according to
paradigms, a set of tacit and unarticulated guiding assumptions, rather than
the standard conceptions of ‘pure’ rationality at the heart of a scientific
approach. Further, just as the Kuhnian model challenges the conception of
scientific activity approximating to certain canons of rationality, so will any
parallel model applied to the economic case.
It is worth noting that there are alternative routes to this analogy from
science to economics that may be of value to scholars of public administra-
tion who wish to use the concept of path dependency without engaging
with the literature from economics. One is to proceed directly by an analogy
from science to policy: Hall (1993) invokes the notion of a policy paradigm
in a similar vein to Kuhn’s work in science. For Hall, a policy paradigm is an
interpretative framework that operates in the policy-making process. Speci-
fically, it refers to the framework of ideas and standards that specifies the
goals, instruments and the very nature of a policy issue. On this approach,
how policy-makers interpret and use evidence to construct the notion of a
policy and options for reform greatly affects any policy change. This inter-
pretive framework can be path dependent. Alternatively, Pritchard (2002)
argues it is possible to generalize away from science and economics by using
Wittgenstein’s notion of ‘hinge propositions’. These are the guiding assump-
tions of a certain activity, assumptions that inform and restrict the choices
taken. These guiding assumptions are forms of tacit knowledge; it is not that
agents do not know the assumption at issue, namely, that this is not situ-
ation of incomplete information; instead, these assumptions are not recog-
nized as such. They are ingrained in how agents construct their situations,
their decisions and their actions. It follows from both these alternative con-
ceptions of decision-making procedures that these key guiding assumptions
are not themselves evaluated and thus that the range of possible alternative
courses of action that are scrutinized in the decision-making process is lim-
ited. Path dependency in the policy process is thus a form of context bound
rationality among policy actors.
An alternative account of decision-making across time in the literature on
path dependency proceeds by metaphor: ‘they [historical institutionalists]
have been strong proponents of an image of social causation that is path

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 565

dependent in the sense that it rejects the traditional postulate that the same
operative forces will generate the same results everywhere in favour of the
view that the effect of such forces will be mediated by the contextual fea-
tures of a given situation often inherited from the past’ (Hall and Taylor
1996, p. 941). In these terms, path dependency is a metaphor, but this does
not imply any attenuation in the explanatory force of the concept. Indeed,
this may reflect the ‘history’ part of historical institutionalism. Metaphors
can contribute to the narrative of unique policy paths. A narrative conveys
not just information but understanding. It arranges events in a way that
renders them intelligible. It is therefore tempting to describe narrative as a
form of explanation. This is something accepted by many historians (Roth
1999).
A metaphor can help improve a policy story by providing reasons. A rea-
son is a rationale, which provides the explanatory context for the action.
Reasons can explain in the absence of mechanisms. Some historians will
accept this claim (Turner 1999; Roth 1999), often as a sine qua non of conducting
historical inquiry. However, the concept of path dependency is ambiguous in
these terms. For some narrative explanation is a reduced form of causal explan-
ation, useful when there is uncertainty about the mechanism operating or
where a reason is accepted as existing that implies the mechanism.
For others, narratives can never explain. For example, in a widely cited
paper, Goldstone (1998) examines path dependency from the perspective of
historical sociology and asserts that narratives tend to be based on ‘Dr
Suess-like explanatory principles’ (Goldstone 1998, p. 832): events are
wholly contingent and unique and they just happen to happen this way and
are not very likely to happen that way again. Narratives cannot be causal
(and therefore enjoy explanatory power) ‘unless there is some assertion of a
necessary or highly probable connection between events A, B, and C in the
series of events that make up the narrative’ (p. 835). This demand for well-
established mechanisms serves to highlight their scarcity outside the intel-
lectual borrowings from the economics of institutions and provides a clear
challenge for policy scholars.
The notion of context bound rationality is at the heart of path dependency
yet remains underdeveloped in terms of the policy literature. This does not
defeat the concept but highlights a challenge for scholars wishing to employ
the term. Path dependency literature is most developed around the follow-
ing two claims: (1) that the analogy from economics to institutions can be
extended to policy; and (2) that micro economics can be borrowed as the
micro foundations of path dependency in policy development. However,
these claims require a rational choice actor making the decisions. The assump-
tion of this type of rationality is a strict corollary of claims (1) and (2).

Policy change and stability

At the heart of any account of path dependency is stability: observations of
change challenge the notion. This is a common criticism of the historical

© Blackwell Publishing Ltd. 2005

566 ADRIAN KAY

institutionalist school, ‘in its emphasis upon path dependence and historical
legacies it is rather better at explaining stability than change’ (Hay 2002,
p. 15). Thelen (1999) argues that path dependency is also too deterministic in
that once the initial choice is made, then the argument for future develop-
ment becomes mechanical. There is one fork in the road, and after that, the
path only narrows.
In terms of policy studies, one possible counterargument is based on the
interpretation of stability in path dependency. Specifically, the notion does
allow policy change; policy legacies constrain rather than determine current
policy. Policy does change but within a particular set of change options; thus
the policy may be said to exhibit stability. There are two main implications
of the constrained change argument. The first is that these choice sets may
be large or the paths wide; and the wider they are the less the notion of path
dependency can account for current policy development. In alternative
terms, the weaker is the ‘echo’ of past policy developments in the present,
the more other concepts, framework and theories are required.
The second issue is that, explicit in the notion of a policy path, is the
notion that policy may be represented as a single point variable in policy
space. As such, it is a vector, that is, it has a velocity and has a direction.
Because policy has a velocity that can change, policy time can unfold
unevenly and nonlinearly. As such, policy change is often characterized by
moments of crisis (Hay 2002). The problem for the notion of path depend-
ency as constrained change, is the velocity of change. Where the velocity of
policy is subject to change – acceleration or deceleration – this can assist the
formulation of accounts of crises that precipitate policy change. However, it
is difficult to reconcile acceleration in policy time with policy stability.
On the other hand, the notion of policy direction may assist constrained
change accounts of policy development. A stable policy path when
projected into policy space may well imply significant cumulative policy
change over time or, in other terms, a significant distance from the initial
position in policy space and time. A stable equilibrium path does not neces-
sarily imply a fixed point in policy space; change occurs as a function of
time. Rose and Davies (1994) show the importance of compounding effects
as small, incremental and constrained changes in annual budget allocations
can accumulate to significant policy shifts over a period of a decade or more.
Further, a change in direction may appear at one distance to be a small
perturbation but by shifting the direction of the policy may turn out in retro-
spect to have been a critical juncture and a problem for the validity of path
dependency as a description. Kay (2003) argues that the development of the
EU’s CAP is an example of this possibility: although considered minor at the
time the reforms of 1988, rather than 1992, were the critical juncture in shift-
ing the direction of the CAP from a price support to direct income payment
system.
Nonetheless, the limitations of path dependency as a conception of policy
change have been highlighted in recent empirical applications in public

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 567

policy: Kemp (2001) with respect to housing benefit reform in the UK;
Pemberton (2003) with regard to pensions; Greener (2002) on the NHS; and
Kay (2003) with respect to the CAP. Each of these studies finds path depend-
ency in policy development alongside some policy change. They consider
change as a reaction to the unintended consequences or side effects of
policy, or from pressure for reform due to exogenous shifts in the wider
policy environment; for example, where the distribution of power between
interested groups moves. These studies show that once a dualism between
policy stability and policy change is established, the notion of path depend-
ency is only useful for accounting for the former.
The dualism between stability and change can be avoided by considering
the sedimentation of policy decisions or the growing complexity of policy
space that is implied by the notion of path dependency. As noted earlier, the
dynamics of subsystem accumulation are theoretically underdeveloped in
the literature but seem promising in terms of path dependency in public
policy (see, for example, Pemberton 2003 and the UK pensions’ policy). The
development of UK pharmaceutical policy since the 1980s is another
example of new policies being added on as a ‘patch’ or ‘fix’ to satisfy pres-
sure to mitigate the consequences of the original policy: relatively high
prices for medicines were agreed by the government to reward innovation
by the industry under the extant Pharmaceutical Price Regulation Scheme
(PPRS); this contributed to the rapid increases in public expenditure on
medicines since the late 1980s. The PPRS remains unchanged and poten-
tially path dependent but its budgetary consequences precipitated a series
of new policies aimed at controlling the demand for medicines in the NHS –
for example, cash-limited prescribing budgets. The accumulation of these
policy patches is an example of policy change and the path dependency of
particular policy subsystem is a contributory factor to their explanation. The
idea of policy space becoming more complex as new policy elements accrete
around an existing path-dependent subsystem may encourage some theo-
retical work to examine linkages between policies. However, currently no
such work appears in the literature.

Normative aspects of the term path dependency

One of the consequences of constructing the explanatory foundations of
path dependency in public policy by analogy from the economics of techno-
logical development is to ‘import’ the normative result that inefficiencies
can be persist in path-dependent processes. This is a powerful result for neo-
classical economics: certain historical factors can ensure that inefficiencies
occur and markets do not eliminate these over time. Efficiency is understood
here as social efficiency, that is, a situation where both technical and alloca-
tive efficiency hold. The strength of this normative result depends on the
judgement as to whether the inefficiency could have been foreseen at some
point in the initial stages in the path-dependency process and corrected and,

© Blackwell Publishing Ltd. 2005

568 ADRIAN KAY

second, whether the inefficiency remains remediable, that is, that a pareto
improvement can be identified and is achievable from the current situation.
For some writers, the question of whether path dependency implies the
persistence of inefficient institutions is an open and empirical question (see,
for example, Hay and Wincott 1998); for others, path dependency is more
clearly something which inhibits the introduction of ‘better’ or perhaps
more rational policy or organizational form (see, for example, Greener 2002).
Overall, the normative implications of path dependency are less pressing for
scholars outside the boundaries of neo-classical economics; it is generally
accepted that inefficient policies or institutions may persist. However, it is a
much stronger claim that policies in a path-dependent process are necessarily
inefficient or, alternatively, that contained within the concept is the imputa-
tion of inefficiency. The claim is strong at a theoretical level since the
concept would require significant elaboration in terms of both policy design
and the pressures that sustain path dependent and inefficient policy. There
is also the problem of indeterminism: path dependency emphasizes that
policy paths are unique and arrived at by a series of small and contingent
moves. As such, it is difficult to say that there exists another path that could
have been arrived at which is more efficient and without such a relevant
counterfactual it is difficult to accept the imputation of inefficiency.
At an empirical level, the claim that policies are necessarily inefficient is
also strong. Developments in performance measurement in the public sector
might allow arguments that, for example, health care or education policy are
better in one system than another. Despite this, it is difficult to assert, within
a particular political system, that there exist policy options which represent
a welfare improvement over the current policy (net of switching costs and
increased transaction costs) and there is widespread recognition of this by
policy actors. Without these two conditions holding, the normative implications
of path dependency in terms of public policy are attenuated. Nonetheless,
empirical works on path dependency in policy development seem willing to
impute inefficiency to some degree (Wilsford 1994; Kemp 2001; Greener
2002; Pemberton 2003; Kay 2003). This is often not so stark as labelling poli-
cies in terms of efficiency and certainly involves no quantitative analysis;
however, it is not an over-interpretation of these works to tease out the
implicit assumption that a policy would be ‘better’ without path dependent
process acting as a barrier to effective reform.

CONCLUSION
This paper has examined the recent literature on path dependency in terms
of explaining public policy development. The first section highlighted the
problems with policy as a unit of analysis, in particular whether it can be
conceived as a composite whole or is it a series of decisions and actions at
different scales that do not cohere as a whole? Although the notion of path
dependency does not provide any definitive guide in this regard, I have
argued that it does make sense to talk about a policy ‘whole’ or system with

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 569

constituent elements or policy subsystems and instruments. The crucial

point emerging from this is discussion is that policy scholars wishing to
employ the term must be clear about in what sense the term policy is being
used and thus about what aspect is path dependent.
The next section reviewed some criticism of path dependency at a theoret-
ical level. None of these criticisms are fatal to the validity or utility of the
concept of path dependency in policy studies. However, they serve to high-
light the fact that the most theoretically developed work in the field has
borrowed extensively from the new institutional economics. This borrowing
is not costless and its consequences, in terms of explanatory mechanisms
and normative implications, often go unacknowledged. Further, an important
challenge for future work is to develop the notion of context bound rational-
ity, central to any path-dependent process, in ways appropriate to the pur-
pose of policy studies.
Theoretical work often accumulates more quickly than evidence from
robust empirical testing. The danger for path dependency is that it becomes
a concept in search of a case and all historical policy studies are stripped
down to parts that are, in some sense, path dependent. This should be
avoided. Instead, scholars should seek to deepen their theoretical and
empirical understanding of a larger set of historical dynamics that may
occur in policy development. The ambition should be to assemble a taxon-
omy of different means for understanding temporality – of which one will
be path dependency – alongside, for example, an evolutionary perspective
(John 2003) or a punctuated equilibrium model (Baumgartner and Jones 2002).

ACKNOWLEDGEMENT
I would like to thank three anonymous referees and Professor Rhodes for
their helpful and encouraging comments on an earlier version of this article.
All errors remain my own.

REFERENCES
Arrow, K.J. 2000. ‘Increasing Returns: Historiographic Issues and Path Dependence’, The European Journal of
the History of Economic Thought, 7, 1, 171–80.
Arthur, W.B. 1994. Increasing Returns and Path Dependence in the Economy. Ann Arbor, MI: University of
Michigan Press.
Baumgartner, F. and B. Jones (eds) 2002. Policy Dynamics. London: University of Chicago Press.
David, P.A. 1985. ‘Clio and the Economics of QWERTY’, American Economic Review Papers and Proceedings, 75,
332–7.
Denzau, A.D. and D.C. North. 1994. ‘Shared Mental Models: Ideologies and Institutions’, Kyklos, 47, 1, 3–31.
Dye, T.R. 1972. Understanding Public Policy. Englewood Cliffs, NJ: Prentice Hall.
Elster. J. 1976. ‘A Note on Hysteresis in the Social Sciences’, Synthese, 33, 371–91.
Elster, J. 1998. ‘A Plea for Mechanisms’, Chapter 3 in P. Hedstrom and R. Swedberg (eds), Social Mechanisms:
an Analytical Approach to Social Theory. Cambridge: Cambridge University Press.
Georgescu-Roegen, N. 1967. Analytical Economics. Cambridge, MA: Harvard University Press.
Georgescu-Roegen, N. 1971. The Entropy Law and the Economic Process. Cambridge, MA: Harvard University
Press.

© Blackwell Publishing Ltd. 2005

570 ADRIAN KAY

Goldstone, J.A. 1998. ‘Initial Conditions, General Laws, Path Dependence, and Explanation in Historical Soci-
ology’, American Journal of Sociology, 104, 3, 829–45.
Goodin, R.E. 1998. ‘Institutions and their Design’, in R.E. Goodin (ed.), The Theory of Institutional Design.
Cambridge: Cambridge University Press.
Greener, I. 2002. ‘Understanding NHS Reform: the Policy-transfer, Social Learning, and Path Dependency’,
Governance, 15, 2, 161–83.
Hacker, J.S. 1998. ‘The Historical Logic of National Health Insurance: Structure and Sequence in the Develop-
ment of British, Canadian, and US Medical Policy’, Studies in American Political Development, 12, 57–130.
Hacker, J. 2002. The Divided Welfare State: The Battle Over Public and Private Social Benefits in the United States.
Cambridge: Cambridge University Press.
Hall, P.A. 1993. ‘Policy Paradigms, Social Learning, and the State’, Comparative Politics, 25, 3, 275–96.
Hall, P.A., and R.C.R. Taylor. 1996. ‘Political Science and the Three New Institutionalisms’, Political Studies,
44, 4, 936–57.
Hall, P.A., and R.C.R. Taylor. 1998. ‘The Potential of Historical Institutionalism: a Response to Hay and
Wincott’, Political Studies, 46, 5, 958–62.
Hay, C. 2002. Political Analysis: a Critical Introduction. Basingstoke: Palgrave.
Hay, C. 2004. ‘Theory, Stylized Heuristic or Self-fulfilling Prophecy? The Status of Rational Choice Theory in
Public Administration’, Public Administration, 82, 1, 39–62.
Hay, C. and D. Wincott. 1998. ‘Structure, Agency and Historical Institutionalism’, Political Studies, 46, 5, 951–7.
Heclo, H. 1972. ‘Review Article: Policy Analysis’, British Journal of Political Science, 2, 83–108.
Hempel, C.G. 1965. Aspects of Scientific Explanation. New York: The Free Press.
Hodgson, G.M. 2001. How Economics Forgot History: the Problem of Historical Specificity in Social Science.
London: Routledge.
Holzinger, K. and C. Knill. 2002. ‘Path Dependencies in European Integration: a Constructive Response to
German Foreign Minister Joschka Fischer’, Public Administration, 80, 1, 125–52.
Howlett, M. and M. Ramesh. 2003, Studying Public Policy: Policy Cycles and Policy Subsystems. 2nd edn. Oxford:
Oxford University Press.
John, P. 2003. ‘Is There Life After Policy Streams, Advocacy Coalitions and Punctuated Equilibria: Using
Evolutionary Theory to Explain Policy Change?’ Policy Studies Journal, 31, 4.
Kaldor, N. 1970. ‘The Case for Regional Policies’, Scottish Journal of Political Economy, 18, 337–48.
Kay, A. 2003. ‘Path Dependency and the CAP’, Journal of European Public Policy, 10, 3, 405–21.
Kemp, P. 2000. ‘Housing Benefit and Welfare State Retrenchment in Britain’, Journal of Social Policy, 29, 2, 263–79.
Krasner, S. 1984. ‘Approaches to the State: Alternative Conceptions and Historical Dynamics’, Comparative
Politics, 16, 223–46.
Lindefeld, D.F. 1999. ‘Causality, Chaos Theory, and the End of the Weimar Republic: a Commentary on
“Henry Turner’s Hitler’s Thirty Days to Power”’, History and Theory, 38, 3, 281–99.
McDonald, T.J. (ed.). 1996. The Historic Turn in the Human Sciences. Ann Arbor, MI: University of Michigan Press.
Nooteboom, B. 1997. ‘Path Dependence of Knowledge: Implications for the Theory of the Firm’, in
L. Magnusson and J. Ottosson (eds), Evolutionary Economics and Path Dependence. Cheltenham: Edward Elgar.
North, D.C. 1990. Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge University
Press.
Ostrom, E. 1999. ‘Institutional Rational Choice: An Assessment of the Institutional Analysis and Development
Framework’, in P. Sabatier (ed.), Theories of the Policy Process, Boulder, CO: Westview Press, pp. 35–72.
Parsons, W. 1995. Public Policy. Cheltenham: Edward Elgar.
Pemberton, H. 2003. ‘Lock-in or Lock-out? Path Dependency and British Pensions’, Paper presented to PSA,
Leicester, April 2003.
Pierson, P. 1993. ‘When Effect Becomes Cause: Policy Feedback and Political Change’, World Politics, 45, 595–628.
Pierson, P. 2000a. ‘Not just What but When: Timing and Sequencing in Political Processes’, Studies in American
Political Development, 14, 1, 72–92.
Pierson, P. 2000b. ‘Increasing Returns, Path Dependence, and the Study of Politics’, American Political Science
Review, 94, 2, 251–67.
Pierson, P. 2000c. ‘The Limits of Design: Explaining Institutional Origins and Change’, Governance, 13, 4, 474–99.
Pritchard, D. 2002. ‘Are Economic Decisions Rational? Path Dependence, Lock-in and “Hinge” Propositions’,
Reason in Practice, 2, 3, 31–45.

© Blackwell Publishing Ltd. 2005

 THE USE OF PATH DEPENDENCY IN POLICY STUDIES 571

Raadschelders, J.C.N. 1998. ‘Evolution, Institutional Analysis and Path Dependency: an Administrative-
History Perspective on Fashionable Approaches and Concepts’, International Review of Administrative
Sciences, 64, 4, 565–82.
Roberts, C. 1996. The Logic of Historical Explanation. University Park, PA: Penn State University Press.
Rose, R. and P.L. Davies. 1994. Inheritance in Public Policy: Change Without Choice in Britain. London: Yale
University Press.
Roth, P.A. 1999. ‘The Full Hempel: a Review Essay of Clayton Roberts, The Logic of Historical Explanation’,
History and Theory, 38, 249–63.
Skocpol, T. 1992. Protecting Soldiers and Mothers: the Political Origins of Social Policy in the United States.
Cambridge, MA: Belknap Press of Harvard.
Stinchcombe, A.L. 1968. Constructing Social Theories. London: University of Chicago Press.
Thelen, K. 1999. ‘Historical Institutionalism in Comparative Politics’, Annual Review of Political Science, 2, 369–404.
Tholfsen, T.R. 1967. Historical Thinking: an Introduction. New York: Harper and Row.
Tilly, C. 1991. ‘How (and What) are Historians Doing?’, in D. Easton and C.S. Schelling (eds), Divided Know-
ledge. London: Sage.
Turner, H.A. 1999. ‘Human Agency and Impersonal Determinants in Historical Causation: a Response to
David Lindenfeld’, History and Theory, 38, 3, 300–06.
Wilsford, D. 1994. ‘Path Dependency, or Why History Makes it Difficult but Not Impossible to Reform Health
Care Services in a Big Way’, Journal of Public Policy, 14, 251–83.

Date received 10 June 2004. Date accepted 6 October 2004.

© Blackwell Publishing Ltd. 2005