Much of the macro cross-national literature in the field of political
science focuses on different (and often competing) explanations and
predictions of welfare state development.1
However, there is one
overwhelming concern within the current discussion on the future of the
welfare state: the importance and degree of the impact of international
factors, often phrased in terms of globalization, on national policies.
Within this discussion, there is a controversial debate on whether
globalization changes the capacity of nation states to determine the
political processes within their borders. Some observers claim that
globalization leads to a convergence of policies among countries,
resulting in a “race to the bottom” of social standards that
individual governments can do little about. Extreme interpretations even
see the end of the nation-state as a political entity.
2 In contrast, others argue
that globalization does not undermine the capacity of individual states
and that national policies differ now as much as they used to.
3 The way
states respond to globalization may differ: some may even attempt to
compensate for negative effects of internationalization, thereby
increasing domestic state activity.
4 In this article I argue that one source of confusion about the impact
of globalization on domestic policy is the inappropriate analytical and
methodological treatment of international interaction: “The one area
in which the development of methods has lagged drastically behind the
practical needs is in the analysis of effects of interdependence. Indeed,
inadequate methodological attention to interdependence is the most
damaging weakness of cross-national studies.”5
Przeworski 1987, 42. In the now
classic study, Collier and Messick (1975, 1314)
claim that “it is clearly time that comparative political analysis
devoted more attention to the role of diffusion in political
change.” This statement has been repeated since then, particularly
in reference to diffusion and the impact of globalization in
macro-comparative studies. Teorell and Hadenius (2004, 9) state, “To systematically assess such
external diffusion or demonstration effects with large-n data is a fairly
novel enterprise in the field.” For reference to the lack of
methodologically informed works on globalization, see Moses 2001, 1. However, there are recent studies
addressing this problem to which I will refer to later in this
article.
International interdependence and the impact resulting
from it have been discussed as “Galton's problem” in
social sciences. It poses substantial methodological problems for
comparative analysis, so that Goldthorpe recently concluded that
“the Galton problem could be regarded as potentially more damaging
at the present time than ever before.”
6
Therefore it is surprising that recent comparative research on welfare
states does not even discuss diffusion as a methodological problem.
The major focus of this article is twofold. First, I contribute to the
debate on the effects of globalization on nation-states by using updated
statistics on social expenditure. New data are important because the
effects of globalization are in flux. The direction of the trends is
unclear, however, and cannot be fully comprehended at the moment.
Conclusions about the rejection of a convergence of social expenditure,
which many cross-national studies have identified, need to be reevaluated
in light of new information. The second, more important, focus of this
article is to contribute to the analytical and methodological treatment of
globalization. All established studies in the field treat globalization as
a functional variable without highlighting its specific character. In this
article, I consider globalization as a process of international diffusion,
which requires a particular analytical treatment and data analysis.
Goldthorpe identifies globalization as diffusion and claims that the
variable-oriented approach has the potential to deal with it by including
appropriate historical and cultural variables.7
However, in this article I go
further and claim that the analysis of diffusion needs an alternative
logic of analysis. This alternative logic rests on solutions to
Galton's problem as applied to the impact of globalization on
domestic policies. This analysis illustrates that a diffusional
perspective is highly significant and potentially provides a more
comprehensive understanding of the operation of modern states. While the
functional impact of globalization is still ambiguous, this analysis
demonstrates that the whole logic of politics may have shifted from a
domestic to an international orientation in the past decade.
The disagreement over the impact of globalization on the capacity of
nation states to handle domestic policy depends on the type of policy in
question. While some claim that globalization “has undercut the
policy capacity of all but a few areas,”8
Mosley's review of current empirical studies of advanced capitalist
democracies is more subtle: “Substantial cross-national diversity
remains in areas such as government consumption spending, government
transfer payments, public employment, and public taxation, but growing
cross-national similarity exists in aggregate monetary and fiscal
policies.”
9 Goldthorpe
10 reconfirms Mosley's
findings in general but stresses that globalization may have a substantial
impact on public policy, particularly on economic and social policy.
11 One could postulate that even within social
policy there are substantial and important differences among various
sectors of social programs (see Burgoon 2001;
and Castles 2002). This aspect is important,
however, it would extend the analysis of this article.
This
disagreement shows that social policy is at the center of the dispute
about globalization's domestic impacts, making it a suitable basis
for an analysis of the impacts of globalization.
The article is divided into four parts. First, by referring to the
established hypotheses of “compensation” and “race to
the bottom” of welfare state activity, I identify two distinct
periods of welfare state development with a turning point in 1990. These
periods give an indication that there is a break point in the development
of welfare states, which requires special attention and analysis. Second,
I show how cross-national data analysis, normally poorly equipped to
conceptualize diffusion, can actually do so when globalization is
analytically defined as a reorientation of policymakers and market actors.
Third, I elaborate on this issue methodologically, by referring to the
literature on the solution of Galton's problem. Fourth, after having
thus laid out the tools for analyzing globalization as diffusion, I
conduct an empirical analysis using the “globalization as
diffusion” approach and interpreting the results in this light. The
findings indicate that there has been a sustained shift in politics and
policy in highly industrialized countries.
Convergence or Divergence of State
Activity?
The relationship between globalization and state activity is hotly
debated in political science.12
As with
many others, I use social expenditure over time as the dependent variable,
as it is a centerpiece of state activity and a “most commonly used
measure of welfare state effort in past quantitative analyses.”
13 Interventionary states tend to spend more
than liberal states. Even if social expenditure is also dependent on other
factors, it is an appropriate indicator for state activity.
There are three lines of argument in the debate on the impact of
globalization on welfare state activity. These can be summarized by the
terms efficiency, compensation, and convergence. Efficiency and
compensation both address the disagreement on whether globalization leads
to a decrease or increase of state expenditure. According to the
efficiency argument, increasing international interaction leads to
pressure on national economies, so national governments lower costs and
cut social expenditure to stay competitive on the world market. As a
consequence, there is a “race to the bottom” in social
expenditure that may affect large welfare states more strongly than
smaller ones.14
See Swank
2002, 32. Burgoon 2001 distinguishes
different impacts of globalization according to the type of the regime of
welfare states.
The compensation hypothesis, however, comes to
the opposite conclusion: increasing pressure from world markets forces
national governments to provide ample social compensation to those
affected by greater internationalization.
15
These state responses result in equal or even increased social
expenditure.
The third hypothesis, which is related to the efficiency hypothesis
but not incompatible with the compensation hypothesis,16
Hays 2003, 82, also stresses the
point that convergence is independent from the race-to-the-bottom
hypothesis and may stand alone as an indicator for the impact of
globalization.
postulates that different countries' domestic
policies become more similar. International interaction leads to market
integration, which in turn forces national political actors to assimilate
their policies to those of national actors from other countries.
Economists refer to the law of one price: “If identical goods and
services in different economies have the same or nearly equal prices, the
economists consider these economies to be closely integrated with each
other.”
17 Regarding welfare policy, this conclusion would
lead one to expect that the level of social expenditure of countries
within the Organization for Economic Cooperation and Development (OECD) is
converging under globalization.
Most established empirical investigations do not observe a substantial
decline in government spending. Along with Swank they conclude that
“it is clearly the case that by most measures of social welfare
spending, programmatic characteristics (e.g., social insurance replacement
rates) and public-private mixes, the welfare states of advanced
democracies have not been dismantled or dramatically
retrenched.”18
Swank goes so far as to postulate that “the
conventionally hypothesized globalization dynamics are absent.
Internationalization has no systematic impact on welfare policy
change.”
19 Huber and Stephens also confirm this
general pattern but see some indicators in specific areas of retrenchment
in the second half of the 1990s.
20 However, do these statements hold true when one
continues the analysis beyond the early 1990s?
The following figure shows the average social expenditure of sixteen
OECD countries, and the three countries with the lowest and the highest
social expenditure levels from 1980 until 2001.21
In the latest data set, the OECD changed its calculation
methods so that researchers can only trace the process back to 1980. As a
consequence, I cannot reanalyze the data back to the 1960s as most other
studies did. Instead, I am able to continue the analysis from the
mid-1990s until 2001, the latest year with available data. However, the
new OECD data set does not contain sufficient data for Austria and Norway,
so the analysis treats only sixteen countries. For the country selection,
see fn. 50.
The latter measure is an indicator for
“large” or “small” welfare states and therefore
indicates whether there is a convergence of expenditure levels among
welfare states.
22 The variation index
(standard deviation/mean) shows a continuing trend toward convergence
with strong periods in the early 1980s and mid-1990s. This conclusion will
also be confirmed later on in the more elaborated analysis.
As Figure 1 shows, social expenditure was
relatively stable with a slight increase in the 1980s. This increase was
mainly caused by the moderate increase of the countries with the lowest
social expenditures. However, the picture changed fundamentally during the
early 1990s. From 1990 until 1993, social expenditure increased
substantially in OECD countries. In particular, the countries with the
highest expenditures increased their share dramatically. During the late
1990s, social expenditure changed once again: between 1993 and 2000 the
countries with the highest expenditure reduced it substantially, which led
to a decrease in the OECD-wide average during this period after the
substantial increase between 1990 and 1993. However, in contrast to the
high-spending countries that increased their expenditure in the early
1990s but then decreased it to the level of 1990 again, the low-spending
countries increased the share of social expenditure so that in 2000 the
gap between the highest and lowest social spender had become smaller. In
2001 this downward trend came to an end.
Development of social expenditure as percentage of the gross domestic
product between 1980 and 2001 in 16 OECD countries
How can one interpret these findings in light of the three hypotheses
above? First, there are two clearly distinguishable periods: the stable
1980s, where social expenditure did not change much, and the turbulent
1990s with its volatility. The 1980s are characterized by a stable
development without clear indicators for any hypotheses. Therefore, it is
not surprising that all established studies whose data analyses end in the
early mid-1990s conclude that there is no clear confirmation of the
efficiency hypothesis, and that if globalization had any effect, it can be
explained with the compensation hypothesis. The updated OECD data
demonstrates, however, that such conclusions are likely the result of
temporal selection bias: if data from the later 1990s is included, those
conclusions may not follow.
If there has been a substantive shift in the policy and politics of
industrialized countries from the 1980s to post-1990, and if globalization
is a diffusion process in which international interaction and orientation
intensifies, researchers need concepts and measurements to grasp this
change and interpret the empirical results in this light. If this
hypothesis is correct, then one can expect a diffusion variable to have
little or no effect in the 1980s, but a significant one in the 1990s. But
how can one interpret globalization as diffusion in this light?
Theoretical Statement: Globalization as
Reorientation
The literature concerning the impact of globalization on domestic
policy is ambiguous. Functionally, it appears that economic openness led
to an expansion of the welfare state in the past23
and a
retrenchment in current years.
24 Katzenstein argues that small
states with open economies used to compensate for the harmful effects of
international competition with social welfare benefits.
25 Katzenstein 1985. In fact, the
correlation between export and import rates, on one hand, and the size of
a country (measured in population), on the other, is high and significant
in the 1970s, 1980s, and 1990s.
Why should these states not do
the same today? The causal mechanism that directs economic openness must
have changed, for there are certainly differences in aggregate social
spending of OECD countries from one time period to another. These temporal
effects that may influence the degree and impact of diffusion or may even
be caused by it have not been highlighted sufficiently in the current
debate on the impact of globalization on domestic policies.
26 Pierson 1994
identifies two logics for welfare state development in his case studies:
one for welfare state expansion, and another for welfare state
retrenchment. The analysis here aims to identify a similar shift in policy
on the aggregated level of analysis.
It is not even clear whether
globalization is a cause or at least associated with retrenchment.
There is currently no one theory of mechanisms of globalization: the
best available are “lines of argument,” as Huber and Stephens
phrase it.27
However, there are several
explanations concerning mechanisms of diffusion, including coercion,
economic competition, learning, and emulation.
28 For an overview of these approaches, see Simmons, Dobbin, and
Garrett forthcoming.
Drawing on these, I attempt to combine an
economic competition model with the elements of learning theory and
emulation that are most suitable in the institutionally thin and
nonhierarchical environment of OECD countries. In such a model,
governments would intervene extensively in their economies for social and
political purposes if there were little or no international competition.
Factors that influence government policy in this situation are domestic:
strength of different interest groups and parties, or economic and
structural conditions, for instance. However, under the pressure of
international competition and continuing economic openness, governments
take the strategies and decisions of other governments into account. A
nation will consider particularly the policies of other governments to
which it is closely linked economically. This is so because there is close
communication (learning through communication) and dependency (control
through resource dependence) between those countries.
29 See, for instance, Rogers 1995, for
learning through communication; and Pfeffer and
Salancik 1978, for organizational adjustment under conditions of
resource dependence. These approaches go well together with emulation
(following leaders, epistemic communities, psychological proximity).
However, because I define globalization in primarily economic terms, I may
not consider “governmental learning” as much as structural
adjustments (see for this distinction and concepts Levy
1994, 296–98). It is difficult to disentangle these different
influences in an aggregated analysis and case studies may shed more light
on the specific processes.
As Kurzer illustrates in her
qualitative study:
In the past, governments could spend lavishly on public
programs to reconcile the conflicting demands of labor and business.
However, such expansionary programs produce expectations among financial
asset holders that future inflation rates will drift above the rates of
the country's main trading partners. This perception triggers capital
outflows and foreign currency speculation…. Governments must
reverse their policies to arrest future outflows.”30
Kurzer 1993, 12.
This assumption refers to rational learning and policy convergence (Jacoby 2000, 9, 24) and may meet the demand of Moses
(2001, 10) that “students of globalization
need to employ a method that allows us to evaluate the motives and
objectives (both explicit and implicit) of policy-maker and market
actors.” Though if Moses denies that large-scale cross-national
statistical analysis has the potential to address globalization, the
procedure presented in this article may be a remedy for some of his
concerns.
This conclusion supports the argument that governments redirect their
orientation in light of globalization. Whereas domestic actors and
conditions were the main reference point in the past, and international
aspects were subordinated to domestic ones, this whole dynamic reversed in
the period of increasing globalization: international factors became a
major driving force for policy orientation and domestic factors became
subordinated to them. The universal decline of corporatist arrangements in
the OECD countries may support this view.31
One
important reference point—as indicated by Kurzer—for this
process is the increasing importance of a nation's main trading
partners. The efficiency of national policy is increasingly judged in
comparison to the efficiency of other states. Thus domestic policymakers
prioritize the policies of a country's main trading partners.
In order to specify the causal links and importance of diffusion among
the OECD countries, I pursue three models through which to identify when
and through which mechanisms diffusion became salient in the welfare
policies of highly industrialized countries. It is important to identify
the periodicity of diffusion moments to better understand diffusion
processes.32
This aspect has often been
neglected in research of diffusion and globalization. As the research on
international diffusion develops, researchers should more systematically
consider the periodicity of diffusion. Are there historical conjunctions
that give raise to diffusion processes? Are diffusion processes more
likely after certain events such as regime changes and wars? Which factors
hamper diffusion?
In the first model, diffusion is the cause of
the politics of retrenchment of welfare. Expansion of social expenditure
has been guided by domestic (class) policy, while retrenchment is forced
on the countries by external forces. In this model, diffusion is a
condition for retrenchment; therefore diffusion must have happened before
retrenchment processes began. A second model puts emphasis on the
financial openness of countries to the world market. As expressed by Quinn
and Inclán, “structural dependence might be a
consequence of financial openness.”
33 National governments become more
sensitive to the policies of their trading partners after they passed a
threshold of economic openness. This in turn may result in a reorientation
of politics. Under the third scenario, change in policy orientation is a
consequence of uncertainty and fluctuation. A well-established hypothesis
in organizational theory postulates that actors emulate other actors who
are familiar or with whom they share a dependency relationship in
turbulent times more than in stable periods.
34 For all three models, one may predict a policy of the “focus
country” by taking as an explanatory variable the same policy in
that nation's trading partners.35
A
“focus country” is the country whose dependent variable is to
be explained. Although focus on the main trading partners can be only a
proxy for empirical analysis, it can serve as an indicator for a test of
the hypothesis that policy orientation has changed.
The
explanatory power of such a variable should be more profound after a break
point defined by retrenchment, economic openness, or turbulent
development. Taking such an analytical perspective implies a fundamental
change in methodology from functional to diffusional analysis. In order to
elaborate on this aspect, I wish to refer to the literature that offers a
solution to the Galton's problem.
Methodological Statement: Globalization as
“Galton's Problem”
In methodological terms, diffusion cannot be explained by
characteristics of states acting independently, but rather by the
interactions between and among states. How to analyze diffusion as a
process has long been considered as a methodological problem in the social
sciences. The classic example is a nineteenth-century study by Tylor that
examined the relationship between marriage laws and descent patterns in
tribal cultures using data from a cross-cultural sample.36
In a critique of the paper, Sir Francis Galton noted that the correlation
found by Tylor might have been a result of contacts between the cultures
in the sample and not findings based on truly independent cases. Galton
noted: “It might be that some of the tribes had derived [the
traits being studied] from a common source, so that they were
duplicate copies of the same original.”
37 This statement had profound effects in
that it obstructed, with few exceptions, quantitative comparative studies
in anthropology for the next fifty years. The methodological consequences
of diffusion have thus been called Galton's problem.
Most often the solution of the Galton's problem is to avoid it
altogether by selecting cases that have no or limited contact with each
other. However, this sampling solution is not appropriate for the study of
the OECD countries.38
The sampling solution
aims to create a sample of cases that is free of diffusion. Geographical
proximity is highly correlated with interaction and diffusion. Therefore,
units in the sample should be selected from separate geographical areas.
The “most different systems design,” as advocated by Przeworski and Teune 1970, can also be considered a
solution to “Galton's problem,” but it is more suitable
for research dealing with human behavior than for research dealing with
“middle range theories” that are most pertinent to the study
of OECD countries.
First, selecting specific cases for analysis
(which implies dropping others) is inappropriate because all cases in a
most similar systems design—which is the most pertinent to the study
of OECD countries—potentially interact with each other. Second,
dropping cases to facilitate a pure functional analysis manipulates the
sample so that diffusion processes are eliminated. If the effects of
diffusion are what one wishes to analyze, one cannot eliminate diffusion
from the research design; one needs a way to cope with diffusion rather
than avoid it.
This can be done with so-called additional variable solutions. One of
these, still relying on pure functional logic, includes variables that
measure globalization at the level of nation states. This solution has
been favored by Goldthorpe and most others in the field. Established
variables with respect to economic globalization are trade openness,
foreign direct investments (FDI), interest rate differentials, portfolio
investment inflows, or an index of financial openness. Including a sample
of these variables is an established strategy in almost all studies of
welfare policy that measure the nexus between globalization and politics.
If globalization is to be treated as diffusion, however, measuring only
the functional impact is insufficient.
Another additional variable strategy, originally advocated by
Naroll,39
suggests pairing or linking societies
according to their geographic proximity. Ross and Homer take Naroll's
idea of paired cases as a starting point, but they replace geographic
proximity with substantial factors such as common language and trade as
criteria for pairing cases.
40 This idea of analyzing diffusion through linking countries to each
other has been forgotten for the last quarter-century but has popped up
again at the center of recent research in comparative political science
dealing with diffusion.41
I would like to
thank one reviewer of a former version of this article for pointing out to
me research in progress that uses the same logic of analysis as the older
anthropological studies. Although these current approaches are
statistically much more sophisticated than the anthropological studies
cited in this article, they do not refer to the intellectual root of
dealing with the Galton's problem in anthropology. See, for instance,
Simmons and Elkins 2004; Basinger and Hallerberg 2004; Franzese and Hays 2004; and Beck,
Gleditsch, and Beardsley 2005.
The basic idea of analyzing
diffusion is to link countries that are in close contact to each other by
using the value of their dependent variable as an independent variable for
the focus country. The first step is to identify a variable that links
cases. The selection of such a variable is an analytical question: one
should choose variables theoretically useful for the issue in question.
Putnam used the amount of military spending of neighbor countries as an
independent variable to explain military spending in the focus country,
for example.
42 A similar strategy has recently been applied by
Mosley,
43 who uses the U.S. long-term interest rate
to explain long-term government bond rates in highly developed nations.
Together with the inflation rate and current account balance, this
indicator has the most pronounced and statistically significant influence.
Others use competitors, geographical distance, or cultural variables to
connect countries.
44 In the context of economic globalization, trade is an appropriate
variable, as it is an indicator of interaction among countries, and
interaction in turn may be considered a catalyst for diffusion. In
contrast to other variables such as cultural or geographical distance,
which are symmetric for each case (New Zealand is as far away from
Australia as Australia is from New Zealand), trade can be used to measure
the relative importance of another state by its proportion of the focus
country's total trade. This weights large trading partners more
heavily than smaller ones and is not mutually equal: the United States may
be important for Panama, but Panama is a relatively trivial trading
partner for the United States. This situation can be mapped by an N
× N spatial weights matrix labeled W. Analytically this can
be considered a “spatial lag,” which is in principle analogous
to the temporal lag of a conventional lagged dependent variable. However,
instead of lagging the value of the dependent variable one unit in time,
one lags it one unit in space. For country i the spatial lag
is:
where W is the spatial weights matrix and yj is
the dependent variable for country j. This results in a matrix of
ρWy, where ρ (rho) is the spatial autoregressive
coefficient to be estimated, W is an N × N weighting
matrix, and y is an N × 1 vector of values of the dependent
variable. The following equation includes the dependent variable of the
linked country as an independent variable:
The term ε is the vector of errors for all units, and x
is the vector of nondiffusion regressors with the coefficient β. This
model can easily be transformed for time-series cross-section (TSCS)
analysis. However, for TSCS analysis one can also include a time variant
term, making W to N × N × T and y to N
× T:
One can estimate such a model in several ways.45
See Franzese and Hays 2004. I do
not consider spatial error models here because they treat spatial
dependence as a nuisance that biases the interpretation of the parameter
of interest. When spatial dependence itself is the focus of research, as
in this article, spatial terms must be included as regressors in the
model.
The simplest one is to estimate ρ and β by
ordinary least square (OLS) regression of a model that includes
Wy together with
x on the right side. This
solution is called spatial OLS or S-OLS. Including a variable of diffusion
in this way, however, inherently implies a simultaneity bias. This is
because the social expenditure of country A explains that of country B,
and in turn country B's social expenditure explains that of country
A, and so on. Thus spatial lags are correlated with the model's error
terms. Such a situation is a textbook illustration of endogeneity,
however. In the case that the diffusion process is positive and/or
reinforcing (which is most often the case) ρ inflates. The remedy
suggested by Franzese and Hays is to ensure that the model be specified,
so that other exogenous variables are also important and that external
common shock is also part of the model.
46 Ibid. 43; pp. 19–23 lay out the problem and the
conclusions for their recommendations.
S-OLS is most effective
when diffusion is significant but not dominant. Although it is an
empirical judgment, one should become suspicious if the diffusion
coefficient becomes very large. At the end, one must decide whether the
model suffers from omitted variable bias, in which case one neglects the
interaction between cases, or whether the model has a simultaneity bias,
which suppresses the significance of other variables in the model.
There are other ways to deal with this model, however. First, one may
use maximum likelihood (ML) to estimate ρ and β that specifies the
endogeneity of Wy. The S-ML model, however,
underestimates the strength of interdependence, particularly in small-N
studies. Furthermore S-ML models can be difficult to implement.47
Another strategy, called spatial two-stage least squares, instrumental
variables (S-2SLS-IV or just S-2SLS), is to instrument for
Wy using
x and
Wx. This
strategy works only if the
xj are indeed
exogenous but related to
yi. While none of
the solutions is perfect, S-OLS has advantages for my analysis over the
other solutions. Franzese and Hays demonstrate in their Monte Carlo
statistical simulation comparison among OLS without spatial lag, S-OLS,
S-ML, and S-2SLS, that S-OLS dramatically improves estimates in comparison
to OLS without spatial lags, because simple OLS suffers from omitted
variable bias. S-ML produces good estimates, but it is computationally
demanding and tends to underestimate the strength of interdependence,
particularly in small-N samples and when the true diffusion strength is
modest. S-2SLS yields good results under two conditions: when sample size
is large and when instruments are fully exogenous. Small samples, modest
diffusion strength, and imperfect exogeneity of instruments, which are
common in most cases in comparative politics and international political
economy, favor S-OLS in comparison to S-2SLS. Although none of the models
is perfect, Franzese and Hays conclude: “Simple S-OLS type
approaches can perform as or nearly as well as more-sophisticated
estimators under fairly wide ranges of reasonable sample-sizes and
parameter levels.”
48 Beck, Gleditsch,
and Beardsley favor S-OLS over the more complicated S-ML model and point
out that the latter is almost impossible to calculate with a lagged
dependent variable.
49 A
solution for using a lagged dependent variable together with a spatial lag
would be a time lag for
Wy. This makes theoretical sense
in most cases, because one cannot expect that diffusion takes place
instantly. Thus I apply a simple S-OLS regression with a time lag for
Wy to estimate the diffusion strength in welfare
expenditure among the OECD countries.
An Application of the Analysis of
Globalization as “Galton's Problem”
Estimating the effect of globalization on national policy requires a
model that includes both international factors and factors that determine
domestic policy. For that I use the annual social expenditure
rates of sixteen OECD countries as the dependent variable.50
These are the EU member states (except Spain,
Portugal, Greece, Luxembourg, and Austria) along with Switzerland, United
States, Canada, New Zealand, Australia, and Japan. Austria and Norway were
excluded because no sufficient expenditure data is available for the
1980s. Imputing the data for these two countries, as similar
investigations do, is problematic in this case because predicted values
would smooth the trends for the 1980s, which in turn would bias the
results in favor of my argument. However, analyses conducted with these
two countries came to similar results. I excluded Spain, Portugal, Greece,
Iceland, and Luxembourg because they were missing numerous data, and
because they were not considered in most of the other established studies
in the field.
To interpret globalization as diffusion, I base the
analysis on established approaches and their variables. Conventional
explanations of welfare state development address aspects of social
pressure, state properties, and globalization.
51 State property has been analyzed by using resource
mobilization and institutional approaches.
Resource mobilization approaches have been applied to the development
of the welfare state in various ways. Some authors see resource
mobilization of the working class as a determining factor in the
development of the welfare state.52
If the working class is
able to mobilize its resources (as measured in terms of electoral success
of left parties and union density), welfare states are more extensive and
are of a universalistic and egalitarian style. A variation of the resource
mobilization approach is the “parties matter” hypothesis.
53 This perspective extends resource
mobilization and its consequences for the welfare state to other social
actors. While leftist parties promote a universalistic and egalitarian
welfare state, centrist parties favor a kind of welfare state that is also
strong but less universalistic and egalitarian.
54 In contrast, secularized
conservative parties support a “lean” welfare state. Once in
power, these parties attempt to create their respective type of welfare
state.
55 The “parties matter” hypothesis uses as
indicators the strength of left, centrist, and right parties, measured
most often in number of cabinet seats. I referred to the “parties
matters” hypothesis by using the
strength of left parties
and
strength of centrist parties variables, respectively. Left
(Social Democratic, Socialist, and Communist) and centrist parties
(Catholic, Christian Democratic) are known to support welfare state
expansion.
56 However, there is a
strong dispute on their position in periods of welfare state retrenchment
and economic openness. Some argue that leftist parties hinder cuts in
social expenditure,
57 while others see
leftist parties as promoters of lower increases and even of cuts in the
1990s.
58 Under institutional approaches, characteristics of the polity of the
state are the focus of analysis. In this view, state structures and
institutions matter. Different authors emphasize different aspects of this
argument: however, some focus on government structures (presidentialism
versus parliamentarism) and veto points,59
while others emphasize
inclusiveness of election systems.
60 In the field of welfare state
research, interest intermediation is a crucial variable. Often measured on
a continuum from pluralism to neocorporatism, it appears to have a
significant impact: countries with a high degree of
corporatism
support an extensive welfare state more strongly than pluralist states
61 and
they are more reluctant to cut back the services rendered by the welfare
state.
One of the most sophisticated institutional approaches is
Lijphart's concept of patterns of democracy.62
Lijphart
includes many key variables such as the inclusiveness of the election
system, presidentialism and parliamentarism, neocorporatism, and so on, in
an index that ranges from majoritarian to consensus democracy. Even if
Lijphart states that consensus democracies rather than majoritarian
democracies lead to a “kinder and gentler society” (which
would imply a larger welfare state and less social cuts in times of
retrenchment), there is not much evidence for this conclusion within
current welfare state research.
63
Another problem is the use of his approach for empirical analysis.
Lijphart's complex variable is difficult to measure over time: one is
left with substitutes such as the
effective number of parties
variable.
Institutional variables are often considered to be the basic indicator
of the strength of domestic factors. Some think their increasing
importance indicates the low influence of globalization on domestic
policies: “If anything, cross-national variations in democratic
political institutions become more (not less) important in structuring
domestic policy choices as we move into the contemporary era of global
markets.”64
The effects of economic openness or globalization have been included
in the models in different ways. First, I employed the established
variables: trade = (import + export)/gross domestic product
(GDP) and Quinn's index of the liberalization of financial flows
(financial openness).65
I also
experimented with data for foreign direct investment and
capital flows, but these did not alter the substantive results
when used as variables, and they confirmed the trends for trade and
openness. Because there were missing data for these variables, I excluded
them from the final analysis.
Second, I included a European Union
(EU) membership dummy.
EU membership may foster trade and policy
convergence. Scholars in the field of macro-comparative studies have
largely neglected the role of the EU when analyzing the effect of
international factors on domestic welfare policy.
66 This is surprising because the EU
is the major agent that aims to harmonize policy conditions in Europe and
a major player in promoting globalization through deregulation.
67 If globalization reduces countries' social
expenditure, all the variables in this section should have a negative
correlation with social expenditure. More specifically, it is at the break
points that globalization's impact should go from insignificant to
significant or should at least increase in a relative sense.
In order to identify the strength of diffusion, I constructed a
diffusion variable by linking countries in the above-described
manner, according to the major trading partners of the focus country. This
was done by weighting the social expenditure of the sixteen countries
according to their share of the total trade (imports and exports
combined).68
The weighting is limited to the
sixteen countries under investigation because I have uninterrupted time
series data for social expenditure only for these countries. Even if some
countries may have stronger trade exchanges with countries outside this
group, trading is strong among OECD countries. Other studies also limit
the number of investigated trading partners for practical reasons. Simmons and Elkins 2004, 179, use the weighted mean of
the ten most important partners. The weighted mean of the sum of the
sixteen countries (which resulted in fifteen trading partners for each
country in the sample since the focus country itself is excluded as
trading partner) has been divided by fifteen. I have not pursued other
ways of measuring diffusion (culture, geographic distance, etc.) because
the focus of this article lies in the linking technique. Including
regional dummy variables, such as the families of nations, led to severe
multicollinearity that cannot be controlled for without extending the
article substantially.
One alternative reaction to international pressure that is not
grounded in policy diffusion but needs to be controlled for is external
shock. Diffusion connotes international interdependence in policymaking
and is distinct from common policy responses to correlated external
shocks. An example is a global recession, in which government deficits
increase across the OECD. This common response (deficit spending) to a
common external shock (global recession) is not diffusion.69
I am indebted to one of the reviewers for this
aspect and example.
In order to control for common shocks, I used
two strategies: first, I included the average misery index (unemployment
plus inflation) for all OECD countries as
economic shock
variable.
70 I experimented with economic
growth, inflation, and unemployment individually and in combination; once
for all the OECD countries and once for the most important (Group of 7
countries). The aggregated misery index—though theoretically
controversial—had the clearest statistical impact. Therefore I use
this index for the variable economic shock.
Second, I
included a complete set of year dummies for all analyses.
Finally, I included some control variables into the model. As Hicks
and Zorn indicate, social expenditure is dependent on pressure.71
Pressure has been taken into account by including
the unemployment rate (
unemployment) and the number of elderly
people (
pension). Even if both expenditures are a target of cuts
(and are variables of welfare state retrenchment) they have significant
effects on social expenditure. Furthermore, these two variables are
indicators for the validity of either the compensation or efficiency
hypotheses. I also include
economic growth72 I do not use the level of GDP because this variable may be a
nonstationary “unit root” (Beck
2001, 280). See also below for further discussion of this problem
in the context of research on social spending.
as a control
variable. Growth is normally negatively related to social expenditure per
GDP, because high growth increases the denominator of social expenditure.
In accordance with common practice in the field, I use a one-year lag for
all independent variables (t−1).
Table 1
summarizes the variables and their hypothetical impact on social
spending.
Summary statistics, hypothetical directions, and sources for variables
included in the analysis
Specifying the Model
The analysis was conducted in the standard tradition of TSCS data with
panel-corrected standard errors (PCSE) and a for first-order
autoregression correction. I also included a full set of country and year
dummies, which is particularly important when analyzing welfare
states.73
For the inclusion of country and
year dummies in welfare state research, see Garrett and
Mitchell 2001, 162–65. The downside of including
fixed-effects and year dummies is that one can underestimate the effects
of relatively invariant variables such as corporatism and use many degrees
of freedom for these variables.
Year dummies capture uncontrolled
time-specific effects such as common shocks. I deviate only in one respect
from the common standard in that I do not use a lagged dependent variable.
Although a lagged dependent variable may be necessary for the best-suited
model specification,
74 under conditions of heavily
trending variables, it may dominate the regression and attenuate the
effect of other variables.
75 Achen explains why not to include a lagged
dependent variable in the model when that variable has no causal effect
(as in the case of state expenditure) by pointing out that last
year's budget may strongly correlate with this year's budget but
did not cause this year's budget. Thus I will not include a lagged
dependent variable in my models for the level of
social
expenditure but use a correction for first-order autocorrelation
coupled with fixed effects.
76 Beck, Gleditsch, and Beardsley 2005, 32, point out
that spatial lags are not so important in models that include a lagged
dependent variable because that variable already contains prior spatial
effects. They assume that spatial lags have a strong impact in TSCS models
that do not include the lagged dependent variable in the specification.
Therefore, I ran all models with a lagged dependent variable for
comparison. The results for diffusion were identical in terms of
statistical significance in all the models except in Model 1, where the
diffusion variable remained insignificant. However, other
variables changed quite drastically: the coefficient for
unemployment was always negative, and the one for impact of
parties (strength of left parties and strength of centric
Parties) was stronger in the models with a lagged dependent variable.
It was also positive before and negative after the breaking point (except
for retrenchment).
Another methodological aspect that has received insufficient attention
in social expenditure research is the problem of nonstationary data (unit
roots). Apart from the substantial interest to test if the results are
robust as much for short-term as for long-term effects, this analysis is
also guided by methodological considerations. In general, analysis of
welfare state spending might be plagued by nonstationary data. Although
this problem has widely been ignored in the field and is hardly researched
in social sciences,77
Beck (2001, 280) points out: “We know little about
nonstationary TSCS data.”
some authors express serious
concern.
78 Kittel and
Winner 2005. In my models, unemployment, pension,
trade, and, above all, the dependent variable social
expenditure are nonstationary data. I applied the test developed by
Levin and Lin, and by Im, Pesaran, and Shin to identify nonstationary
data; see Banerjee 1999; Levin, Lin, and Chu 2002; and Maddala and Kim 1998.
To account for this
problem, I conduct analyses with models that use the first differences of
the variables.
The last methodological concern is modeling of periodicity. Based on
other studies in the field I use dummy variables to identify the impact of
break points.79
These
break points are:
retrenchment,
80 I
have taken the break point for retrenchment from Hicks and Zorn 2003, 35. Years before the break point,
that means years before the process of retrenchment began, I have coded 0
and afterwards I have coded the variable 1. The break points is 1982 for
Denmark, Germany, Ireland, Italy, and the United States; 1983 for Sweden;
1984 for Belgium, France, Japan, the Netherlands, and the United Kingdom;
1985 for Australia; 1987 for Canada; 1988 for Switzerland; 1991 for New
Zealand; and 1992 for Finland. On average retrenchment began in 1985 for
the sixteen OECD countries.
,
81 The break point for the openness
threshold has been defined, according to Quinn
1997 data, to be from 1960 to 1993. I used the year when the score
reached a level above the total mean for each country. This means that the
break point was 1981 for Germany; 1983 for the Netherlands; 1989 for
Canada, Denmark, New Zealand, Sweden, and the United States; 1990 for
Australia, Belgium, Finland, and Italy; 1991 for France; and 1992 for
Ireland and Japan. The United Kingdom and Switzerland had no break point
because they reached their respective value already in 1979 and 1980.
Taking the average of the OECD countries, financial openness exceeded the
mean in 1988.
and increasing
turbulence.
82 The beginning of the period of
turbulence has been defined by the mean of the absolute change
rate in social expenditure for all sixteen countries. The mean is 0.475
for the total period from 1980 until 2001; 0.359 for 1980 until 1989; and
0.552 for 1990 until 2001. 1990 was the first year where the mean exceeded
the double value of the total average (0.994). Therefore, 1990 has been
taken as a break point for all the countries. Using individual break
points for the countries according to the mentioned principle for
retrenchment and financial openness has less clear
though comparable results. However, theoretically the collective break
point is more reasonable because it captures the idea of herd
behavior.
In order to identify changes before and after the break
points, I let each of the substantive variables interact with the
respective break dummies.
83 Modeling
structural breaks and using weighted spatial lags may conflate the impact
of these two concepts on social expenditure. However, fixing one
of these estimated coefficients (either by not using structural breaks or
not weighting the spatial lags) would lead to similar results: estimates
without structural breaks are shown in Models 1 and 5. Those with fixed
spatial weights (not shown here) also do not alter the results
substantially. Structural breaks and changes in the patterns of trade may
be correlated. This, however, would be an interesting issue for further
research.
My general model is the following, where Σ is a
shorthand for several related variables measuring the same concept and SB
is a dummy variable that denotes the structural breaks:
Findings and Interpretations
I present the findings of seven different models, four concerning the
level of social expenditure and three dealing with the changes. Starting
with the basic model (without any structural break) I then consider
impacts before and after the various break points. Model 1 shows the
significant positive impact of unemployment and economic
shocks. It identifies significant negative relationships for
trade, eu membership, and growth. Countries
with a high number of effective parties (an indicator for
consensus democracies) have lower social expenditure rates. Parties
(strength of leftist parties and strength of centric
parties), corporatism, financial openness, and the
share of elderly (pension) have no significant effect. In
general, the results support the findings of other studies in the field,
with the exception of the low impact of the elderly. Even the significance
of diffusion does not depart from the findings in the few other
studies that include this variable in their model specification.84
However,
if one includes a lagged dependent variable into the model (not shown
here), the
diffusion variable is insignificant. The reason for
this may be that the lagged dependent variable already contains prior
spatial effects and therefore the spatial lag provides less independent
information in such a model.
85 For details
see explanation in fn. 76.
This is the only deviation in the
results between our models and models with a lagged dependent variable
concerning the
diffusion index. However, can one get a sharper
picture if one takes periodicity of globalization into account?
The impacts on social expenditure among sixteen OECD countries
(1980–2001)
There are some common features among all three models; for instance,
the models show that consensus democracies (effective number of
parties) changed their impact on social spending most dramatically.
In all three models, the relationship was negative before and positive
after the break point, meaning that consensus democracies pursue a
stronger welfare state policy than majoritarian societies after the break
points. This again indicates that they oppose cuts in social expenditure
more vehemently than majoritarian democracies. This may be because of the
fact that consensus democracies have more veto points and are therefore
slower in adjusting their policies. However, it could also be that they
simply pursue a more expansionist welfare policy. This interpretation of
the data refers to the prebreak level (where the coefficient is
significantly negative) and the postbreak effect (where it is
significantly positive). However, the net effect from the effective
number of parties is still negative.86
For the coefficient of the effective number of
parties, for instance, the calculation is: −0.804 + 0.560 =
−0.244.
Since I am mainly interested in the direction and
significance of this coefficient after the breaks, I emphasize on the
changes in the total effects less.
A striking finding is the impact of eu membership after the
break points. All three models depict a clear negative effect, meaning
that eu membership is associated with cuts in social spending
after the break point. The impact of eu membership was not
specific beforehand, as postulated in the general hypothesis for the
impact of international factors. All models also confirm the results of
other studies about the weak impact of parties on levels of social
expenditure.
Concerning diffusion, the retrenchment model deviates from
the other two: the diffusion variable is insignificant both
before and after the beginning of the retrenchment process. The
retrenchment model also differs from the other two models in that domestic
factors remain and become more important than international factors after
the break point. This indicates that the first wave of retrenchment in
welfare states is not caused by globalization. There are clear indicators
that retrenchment was mainly initiated by domestic issues
(unemployment, pension, growth (negative)) as
suggested by Hicks and Zorn.87
Hicks and
Zorn's (2003) indicator for the second
retrenchment phase, however, reaches different results. The findings of
that model come close to my two models in that international variables,
including diffusion (but also financial openness) become
significant after the second wave of retrenchment.
The other two models are quite similar in their results. Both show
clearly that diffusion was insignificant before the break point
and significant afterward. trade became negatively significant
after the break point in both models. This together with the significant
negative impact of eu membership shows the increasing importance
of international embeddedness for cuts in welfare spending. Considering
the relative impact of each variable on the level of social expenditure by
comparing the standardized coefficients reveals that the
diffusion variable has the highest (turbulence model) and second
highest impact (openness model) and trade has the third and
second highest.88
The strength of the
diffusion variable indicates that one does not have to worry that
this variable is inflated. In addition, the included exogenous variables
have substantial explanatory power in the model. This is particularly true
for the common shock variable.
has the highest impact in Model 3 but its relative strength is smaller in
Model 4. The reverse is true for
eu membership, which has the
lowest impact of all significant indicators in Model 3 but comes forth in
Model 4.
unemployment takes the fourth (Model 3) and third (Model
4) highest place and shows that pressure is still substantially
responsible for the level of social expenditure in the OECD countries. The
institutional variable
effective number of parties changed
direction of signs but is around ten times less influential than
trade and the
diffusion variable.
Both models also indicate a change in the impact of external
shocks. Before the break point, increasing unemployment and inflation
in the OECD countries resulted in increased spending; afterward the impact
of economic shocks decreased significantly. However, even if it
lost about 10 percent of its effect in Model 3 and around one third in
Model 4, the overall impact of economic shocks is still positive.
Even though national unemployment rates and social spending are
positively correlated, which documents the prevailing pressure of
unemployment on social spending levels, the significantly reduced
impact of the international variables (trade, economic
shock, eu membership) supports the efficiency
hypothesis.
In contrast to Model 2, Models 3 and 4 are dominated by the increasing
negative effect of the international variables on social expenditures.
diffusion becomes a significant variable after the break points.
Pressure on the welfare state (unemployment) still leads to
higher social spending levels. Partisan politics (strength of leftist
parties and strength of centric parties) has only limited
long-term effects, and the impact of institutional variables
(effective numbers of parties) on social expenditure
changes drastically. But what are the short-term effects of these
variables on social expenditure?
In order to analyze this question, I consider the changes in social
expenditure before and after the breaking points in the two models where
the diffusion variable increased in significance. The models for
change are identical to the level models except for the fact that I
included a lagged dependent variable (Δsocial
expendituret−1) into the model89
The use of a lagged dependent variable in a first difference
model is appropriate because it does not dominate the model and is an
indicator for path dependency and convergence.
and that the
effective number of parties variable was replaced by the number
of
veto points. I also use first differences for all variables
that change significantly from one year to the other.
90 Considering changes in social expenditure makes it feasible
to include the concept of veto points. The theoretically superior data set
of veto players by Tsebelis 2002 does not cover
all the years for some countries of interest here. In addition, it does
not contain information about the United States at all. However, an
analysis with the data from Tsebelis reaches similar conclusions. The
model of first difference also differs from the models of level in that I
did not include a set of country dummies. While the F-test for year
dummies indicates the significance of year effects, this was not true for
the set of country dummies.
The results of the basic Model 5 for the differences vary from the
level model in that only two variables are significant: the lagged changes
of social expenditure (Δsocial
expendituret−1) and common shocks. In
contrast to the basic level model, the diffusion variable is
insignificant. However, when one splits the data according to the two
break points for diffusion, the picture changes completely. Again
international factors appear to have become much more important after
economic openness reached a threshold point and social spending became
turbulent. This development confirms the impact of one key variable:
diffusion was insignificant before the break points but became
highly significant thereafter. Therefore, results concerning the
increasing impact of diffusion are statistically robust.
The impacts on changes in social expenditure among sixteen OECD
countries (1980–2001)
A comparison of the standardized coefficients shows clearly the
significant impact of the diffusion variable: in Models 6 and 7
diffusion is the variable with the highest impact. Another
international variable, trade, is third in both models. In
contrast to the level models, the first difference models show that
pressure has a significant but negative correlation with changes of social
expenditure after the respective break points. This time the impact of the
amount of elderly (pension) is particularly high (second in both
models). The relative strength of impact of domestic variables
(corporatism, veto points, and, above all, party
effects) is considerably lower. In Model 6 one might be concerned that the
diffusion variable is inflated. The standardized coefficient is
almost twice as high as the second largest coefficient.91
In Model 7, the standardized diffusion coefficient is
one-third higher than the second highest standardized coefficient. That
means the diffusion variable is higher in the first difference models than
in the level models. This is probably due to the fact that first
difference models generally have a lower R2 than level models,
and that the common shocks are less perfectly modeled than in the models
of levels.
However, numerous exogenous variables are also
significant and also have strong explanatory power in this model, so the
results are probably trustworthy. One is left with reservations about the
validity of the findings in Model 6 as regards simultaneity bias, however,
because there are no clear tests for simultaneity.
Other results are also worth commenting on: short-term effects of
previous social expenditure change signs before and after the breaks. That
means there was a continuing expansion of the welfare state before the
break points and a continuing reversal trend in social expenditure
thereafter. The net impact, however, was still positive. In this respect,
the two break points are real watersheds in the development of the welfare
state. This result can also be interpreted to show a trend of path
dependency before the break point and one of convergence afterward.92
For a similar interpretation of a lagged
dependent variable in a first difference model, see also Allan and Scruggs 2004, 507. The differences in
significant positive and negative results before and after the break
points prevail when we remove all variables from the analysis except the
lagged dependent variable, which means that there is an unconditional path
dependency before the break point and unconditional convergence after the
break point.
The pressure variables also changed signs and are
significantly negative after the break points. All these findings strongly
support the efficiency hypothesis: pressure no longer leads to an increase
of social expenditure after the structural breaks but instead is
associated with a decrease, which means that social contributions must
have been severely cut. This conclusion is even more dramatic because
those decreases led to a reversal of direction of the total impacts
(change of signs).
The analysis of short-term effects reveals another interesting
finding. Strikingly, the impact of parties that once supported welfare
state spending is reversed; after the turning points these parties are
negatively associated with social spending. This result is clearly
significant for leftist parties and is only slightly below the level of
significance for centrist parties in Model 6. Though the party effects are
small in comparison to all other effects, this finding sheds some doubts
on the results of recent studies that conclude that left parties do not
cut welfare spending.93
Allan and Scruggs 2004, 505–7. However, their
conclusions refer to more specific aspects of welfare policy and might not
be comparable to general social spending changes.
Not surprisingly, corporatism and veto points appear
to hinder cuts in social expenditure. Many veto points associated
with consensus democracies obstruct speedy welfare retrenchments. However,
their relative impact, though more important than party effects, is
clearly less than that of international factors. That said, it is
surprising that common ecomimic shocks have no significant
impacts in the Models 6 and 7, though they were a dominant factor in Model
5.94
This could be explained by the fact
that the diffusion variable takes its explanatory power from the
common shock variable, as described by Franzese and Hays 2004, 43.
In stark contrast
to the level models, the EU variable has no short-term effects on models
of changes to social spending.
Table 4 gives an
overview of the most important findings of the empirical analysis in terms
of the changes in different variables' impacts and reveals to what
degree they support the efficiency and compensation hypotheses,
respectively.
Summary of main empirical results
Conclusions
The results of this study show that international factors have become
an increasingly important influence on domestic policy over time in the
industrialized world. This is true no matter how one measures
globalization, whether in terms of trade intensity or institutional
factors such as EU membership. Using diffusion as a proxy for
globalization processes, the empirical analysis confirms
globalization's influence on politics and policies. This finding in
turn carries with it serious methodological implications.
An appropriate analysis of the impact of globalization must take into
account causal relationships. Functional analysis is the dominant strategy
of most macro cross-national studies in comparative politics, but it is
not the only game in town. This has been known at least since
Galton's objections to Tylor's study, but unfortunately
comparativists ignored the role of diffusion too often. The impact of
diffusion ought to be an important factor, particularly for any analysis
performed in the tradition of the most similar systems design.
Previous analyses of globalization have concentrated on its functional
consequences and have produced ambiguous findings. Treating globalization
as a form of diffusion shows that it has increased significantly during
the past two decades. This in turn illustrates a new logic of politics of
industrialized nations: international, not domestic, imperatives
increasingly determine social policy. When one evaluates the impacts of
three potential break points, it becomes clear that retrenchment, which
took place before the mid-1980s for most of the OECD countries, is not
correlated with diffusion and globalization. The other two break points
come to rather similar conclusions: they indicate that diffusion (and
therefore globalization) became relevant after the respective break
points. However, one cannot be sure whether increasing economic openness
or turbulences in social spending are responsible for policy changes. It
could also be that domestic retrenchment policies had a long-term effect
and interacted with increasing openness that in turn led to turbulent
social spending. As mentioned above, statistical analysis cannot answer
these questions but gives hints for further analysis. Even though my
results cannot offer an unequivocal answer regarding the character of the
break, the analysis shows that something happened in the late 1980s that
altered policies in highly industrialized countries. This change is most
likely related to a reorientation of policy priorities from domestic
toward international concerns.95
An analysis
of party manifesto data—in Budge et al. 2001—shows that political parties increased
their international orientation and that a neoliberal discourse gained
importance during the 1980s and 1990s. These indicators support the
assumption that a reorientation has taken place in the OECD countries.
However, the causal relationships are complicated, and future research is
needed to shed more light on this process.
Even if the modeling
of the break points is not perfect, there is clear evidence that major
changes in the second half of the 1980s have had a continuing effect into
the new millennium. Altogether, the conclusions of this article challenge
claims that globalization is unimportant for domestic politics: it shows,
rather, that globalization has a systematic impact on domestic policy.
96 This conclusion stands in contrast to
Swank's claim (see fn. 19) that there is no systematic impact of
globalization on welfare policy change. However, the systematic impact is
diffusional rather than functional.
The findings of this article also challenge the claim that domestic
politics have no impact on policies anymore. The situation is far from
postulating the end of the nation-state or the idea that policy is only
created at a level beyond or above the nation-state. An entirely
functional analysis is blind to the impact of diffusion, but an analysis
that postulates the primacy of diffusion is unable to identify and specify
functional impacts. Only a combined analysis of function and diffusion is
able to identify the relationship and relevance of both factors.97
From a methodological point of view it must be
emphasized that many aspects need further attention, particularly the way
of estimating tolerable amounts of simultaneity bias and the modeling of
common shocks in first difference models. While TSCS analysis of level
data is saturated, implying lower risk of simultaneity bias, this is not
so for first difference models. The latter, however, are appropriate to
many research questions in social sciences with trending variables to
avoid the problems of nonstationarity.
What kind of change do increasing importance of international factors
and diffusion indicate? The findings of this article clearly confirm a
trend in social expenditure that favors the “race to the
bottom” hypothesis. The concept of diffusion, however, merely
implies that individual governments became similar to each other in their
social spending. It does not imply a universal decline of the welfare
state. In principle, diffusion can also work in the other direction.98
Factors that may influence the direction of
welfare state development are beyond the scope of this article. Public
discourse and the strength of political ideologies, for instance, are
influential in addition to structural changes. In the time after the break
points, a neoliberal discourse replaced a stronger Keynesian discourse of
the postwar period that even affected left parties. See, for example,
Hall 1993; and Jahn and Henn
2000.
However, the increasing importance of diffusion for
domestic politics and policies implies that highly industrialized
countries become more alike in their policies and outcomes. That means
that globalization is an amplifier for dominant strategies and processes
in the OECD countries.
Since internationalization is expected to increase in the future,
comparative studies should include a measure of the degree of diffusion as
a standard; otherwise, those studies suffer under a severe omitted
variable bias. This article is a contribution to such future analyses and
hopes to stimulate further debate. Considering globalization as
“Galton's problem” and using the available solutions for
it may be a way of improving research skills concerning the analysis of
diffusion and may help researchers to better understand development
processes in modern societies.
