Microsoft word - gilardi_berkel

Spurious and Symbolic Diffusion of Independent Regulatory Agencies in Western Europe Independent regulatory agencies (IRAs) have become the most widespread form of organisation for regulatory policies in Western Europe. Their pattern of emergence suggests that a diffusion process may have been at work, namely a process where the decisions to set up IRAs have not been independent. This paper investigates whether the spread of IRAs is related to diffusion. It draws a distinction between spurious diffusion, which is due to the concomitant reaction of independent actors to similar pressures, and interdependent diffusion, where actors are influenced by the behaviour of others. Eight mechanisms of interdependent diffusion are discussed, which can be differentiated on the basis of how they are related to the improvement of the problem-solving capacity of actors. While learning and regulatory interdependence assume that actors are sensitive to the behaviour of other because it can make them better off, symbolic diffusion stresses that policies and institutions can be diffused irrespective from the functions they perform. The empirical analysis relies on an original data set comprising information for regulators in seven regulatory domains (telecoms, electricity, financial markets, competition, food safety, pharmaceuticals, and environment) in seventeen countries (EU member states plus Switzerland and Norway). It does not examine all mechanisms but, more modestly, aims at determining, through an event history model, the role of spurious and symbolic diffusion. Results show that both have Paper for presentation at the workshop "The Internationalization of Regulatory Reforms", Center for the Study of Law and Society, University of California at Berkeley, During the last fifteen years or so, independent regulatory agencies have been created in all West European countries and most regulatory domains. Figures 1 and 2 show the trend in economic and social regulation2 respectively. The number of IRAs has increased very sharply since about the mid-80s for economic regulation and the early 90s for social regulation. In some domains, such as telecoms, all countries have set up an IRA. At first this trend strikes as surprising, but a closer look reveals that it is all but exceptional. In fact, many phenomena could be described by a similar curve. These include the spread of technological innovations, diseases, social actions such as joining a strike or a riot, and other phenomena commonly regarded as being subject to diffusion. Indeed, this strongly suggests that a diffusion process may be at work behind the creation of IRAs.
The purpose of this paper is to examine this hypothesis by differentiating between spurious and symbolic diffusion. Spurious diffusion is simply due to the concomitant reaction of independent actors to similar functional pressures. In symbolic diffusion, on the other hand, actors are influenced by the behaviour of others. More specifically, actors are more inclined to set up an IRA if others have already been so, but, unlike in learning for example, this happens independently from the problem-solving properties of IRAs. As stressed in the sociological institutionalist literature, the set-up of IRAs may occur because they have become taken-for-granted as an appropriate way to organise regulation, or because they supply legitimacy to other policy choices. Schematically, the symbolic diffusion hypothesis suggests that the decision to establish an IRA at time t should be somehow (positively) related to the number of IRAs existing at time t-1.
The paper is structured as follows. I will first try to make clear which aspects of diffusion I consider in the analysis, and which others I leave out. To this end, I develop a 1 I wish to thank Victor Lapuente Giné, David Levi-Faur and Covadonga Meseguer for their comments ona previous version of this paper. I also gratefully acknowledge the generous financial support of theFondation du 450ème anniversaire de l’Université de Lausanne.
2 Regulation is conventionally termed "economic" when it deals with the price, entry, exit and service ofan industry, while it is termed "social" when it concerns non-economic issues such as safety and health(Meier 1985: 3).
tentative typology of diffusion mechanisms based on the existing literature, to which I add a dimension often neglected, namely symbolic diffusion. In particular, the typology differentiates between spurious diffusion and interdependent diffusion, of which symbolic diffusion is one component, along with learning and regulatory interdependence. In the second part I carry out an empirical analysis where I try to determine to what extent the spread of IRAs has been driven by spurious and symbolic diffusion. In a first step IRA creations are studied at the aggregate level through a Poisson model; in a second I examine individual IRA creations through event history analysis. The analysis is based on an original data set comprising information for regulators in the seven regulatory domains shown in figures 1 and 2 and seventeen European countries (EU member states plus Switzerland and Norway). Results show that there are good reasons to believe that both spurious and symbolic diffusion explain The political science literature on diffusion is rapidly growing and, though a unified approach does not exist, several typologies have been developed. In this section I put forward my own classification. Although I hope that it can add something to the existing discussions, the primary goal is to make clear what is included and, more importantly still, what is left out in the present analysis.
As Eyestone stresses, "any pattern of successive adoptions of a policy innovation can be called diffusion" (Eyestone 1977: 441). The challenge is to discriminate diffusion processes where policy choices of given actors are affected by the prior choices of other actors from processes where the outcome is driven by the reaction of independent actors to similar functional pressures. Only the former can be characterised as diffusion, while the latter should be seen as spurious diffusion. This view is consistent with the argument of Dobbin, Garrett and Simmons (2003), who distinguish between diffusion and a "null hypothesis" where the spread of policies can be explained by the independent behaviour of actors. In contrast to this formulation, the concept of spurious diffusion explicitly acknowledges that actors behaving independently can, under some conditions, lead to a pattern of policy adoptions with the same features as an interdependent diffusion Spurious diffusion of IRAs may derive from the roughly concomitant emergence of functional pressures for delegation, and from governments' response to it. The logic is similar to that of spurious diffusion of central banks. The main rationale of delegating powers to independent central banks is that they are an effective means to keep inflation low. Evidence that governments set-up independent central banks after experiencing high inflation would indicate the presence of spurious diffusion. Interestingly, Castro and McNamara (2003) find that inflation has no effect on the decision to grant independence to central banks. In the case of IRAs, spurious diffusion could be driven by the need to achieve credible commitment capacity and/or the attempt to cope with political uncertainty problems. Firstly, when opening markets to competition, as in the case of telecom or electricity, governments need to be able to convince investors that the regulation of the market will not be biased in favour of the former state-owned incumbent (Levy and Spiller 1994). By delegating regulatory powers to IRAs, governments can improve the credibility of this commitment to fair regulation (Majone 1997, 2001). The reason why this is a particularly strong explanation is that, unlike most others, it addresses the really original feature of IRAs, namely their independence from politics. A second explanation that goes in the same direction is the political uncertainty hypothesis (Moe 1990; de Figueiredo 2002), which claims that politicians insulate policy from politics to make the former last beyond their term of office. In reason of the democratic process, political property right over policy are always uncertain. New governments can undo what old ones have done. Delegation to independent agencies is a means to make this more difficult. As there is evidence that both credibility and political uncertainty matter in explaining cross-sectional differences in delegation to IRAs (Gilardi 2002, 2003), these two arguments are good candidates as causes of Turning to non-spurious diffusion, the most prominent scholars agree that it is driven by the interdependence of actors, but there is no clear consensus on its different sub- types. Dobbin, Garrett and Simmons (2003) stress the role of dominant actors, economic competition, rational learning, and social emulation. Weyland (2002) emphasises external pressure, rational learning, cognitive heuristic (bounded learning), and symbolic imitation. Simmons and Elkins (n.d.) differentiate between economic competition, learning (through communication channels), and cultural emulation. Brune and Garrett (2000) similarly speak of competitive emulation, learning, and social emulation. On the basis of these studies, I have tried to develop my own typology. As stressed above, the main goal is to clarify what is included in the empirical analysis, and what is left out, though hopefully the discussion also supplies a useful synthetic view of Figure 2 presents the typology. It is explicitly acknowledged that diffusion proper occurs only when actos behave interdependently. When this does not occur, a diffusion- like pattern of adoptions can still emerge, notably if actors react to similar functional pressures, but in this case diffusion should be considered spurious. One of the main tasks of empirical analyses, as stressed by Dobbin, Garrett and Simmons (2003), is to assess if, and to what extent, diffusion is non-spurious. The second task, if interdependent diffusion is found, is to determine which mechanism is at the root of the diffusion process. Neither task is easy. In effect, the exercise is plagued by the problem of observational equivalence, as all mechanisms, including spurious diffusion, lead to the same outcome (or at least to very similar outcomes).
In my view, the main distinction is between diffusion mechanisms where problem- solving is the primary rationale for action, and, on the other hand, those where policies spread irrespectively from their problem-solving capacity. In the first category we find learning, which can be fully rational or only boundedly so. Rational learning (mechanism 1) is best conceptualised in bayesian terms (Meseguer 2002; Breen 1999). Here, governments (as well as other actors) are assumed to act after updating their beliefs about the benefits of a given policy by looking at the experience of others. For example, the mean and variance of relevant outcomes can supply information on the effects of a given policy, and this information can be used to update prior beliefs and eventually Bounded learning (mechanism 2), on the other hand, is a bounded rationality version of bayesian learning (Weyland 2002). In this case, actors try to gather relevant information from the observation of the behaviour of others, but, rather than on bayesian updating, they rely on "cognitive shortcuts" such as representativeness, availability and anchoring3 (Tversky and Kahneman 1974; McDermott 2001; Weyland 2002). Learning here can be much less effective than in the bayesian view. Actors do try to get new information from the experience of others, but use cognitive shortcuts rather than Bayes' rule to update their beliefs. An implication is that while in bayesian learning all relevant information is used to rationally update beliefs, in bounded learning only some relevant information is gathered and used through such cognitive shortcuts as representativeness, availability and anchoring. Brune and Garrett's (2000) and Dobbin, Garrett and Simmons' (2003) "social emulation" and Simmons and Elkins' (n.d.) "cultural emulation", in spite of their names, belong to bounded learning. As the authors make clear, in fact, these mechanisms involve information-gathering from a peer group, and are thus an imperfect (i.e. not fully rational, i.e. non-bayesian) way to learn about the In cooperative and competitive regulatory interdependence (Lazer 2001), on the other hand, the logic of diffusion remains problem-solving oriented, but is not grounded in the wish or need of actors to gather new relevant information that can help them in making better policy choices. Under cooperative interdependence (mechanism 4), diffusion is driven by the benefits that follow from having compatible policies, and under competitive interdependence (mechanism 3) by strategic responses to the 3 Representativeness refers to the process through which people link phenomena by evaluating the degreeto which one is representative of the other, that is by the degree to which one resembles the other. Forexample, it has been demonstrated that people assess the probability that a person is engaged in aparticular occupation by determining the degree to which his/her description, or appearance,corresponds to the stereotype of the occupation. This method can lead to serious misevaluation because itneglects several factors that affect the objective probability, such as prior probability, sample size, thereliability of the evidence (Tversky and Kahneman 1974). This heuristic device is incompatible withbayesian learning, where, for example, sample size matters in the formation of posterior beliefs (Meseguer2002: 27). Secondly, availability refers to the process through which "people assess the frequency of a classor the probability of an event by the ease with which instances or occurrences can be brought to mind"(Tversky and Kahneman 1974: 1127). An example would be determining the likelihood of getting cancerby recalling the number of cases of cancer among acquaintances. Availability also means that theevaluation of the risk of an activity, for example, is affected by the ease with which its dangers can beimagined. This process can introduce several biases in judgements, because instances which are moreeasily retrievable or imaginable, for example because of familiarity or salience, tend to be perceived asmore numerous or more frequent. This heuristic device is at odds with bayesian learning, which assumesa use of information consistent with statistical laws. Anchoring, finally, is the process through whichpeople make estimates by adjusting an initial value. The adjustment will typically be insufficient, meaningthat the estimate will remain anchored to the initial value, even when this does not convey relevantinformation. This means that external information has a bigger impact on judgement than it should.
behaviour of competitors. The latter mechanism is thus analogous to Brune and Garrett's (2000) "competitive emulation", Dobbin, Garrett and Simmons' (2003) "economic competition", and to Simmons and Elkins' (n.d.) "diffusion among economic competitors", while the former is not present in these studies.
The second broad category of diffusion mechanisms is characterised by the fact that behaviour is not oriented towards problem solving. In coercive and normative isomorphism (DiMaggio and Powell 1991), the spread of organisations and policies depends much more on their advocacy by powerful or authoritative actors than on the problems they permit to solve4. Coercive isomorphism (mechanism 5) results from the presence of pressures, both formal and informal, exerted on organisations by other organisations upon which they depend. This mechanisms includes then Dobbin, Garrett and Simmons' (2003) "dominant actors" that, through power relations, can impose the adoption of policies and thus contribute to their diffusion. Normative isomorphism (mechanism 6), on the other hand, arises from processes of professionalisation and socialisation within networks, where persuasion may occur through the development of conceptual model that gain authority through their advocacy by prominent actors The last two mechanisms of diffusion are related to symbolic imitation, and will be at the centre of the empirical analysis. First, the set up of an organisation, or the adoption of a policy, can be a ceremony intended to provide legitimacy to certain decisions by diverting the attention from more substantial concerns (mechanism 7) (Meyer and Rowan 1977: 349). In the case of IRAs, governments may create independent regulators so as to legitimate other decisions, such as liberalisation of utilities. As IRAs become valued by the broader institutional environment (which includes norms and values), establish them may enhance the legitimacy of certain policy choices. Second, over time some organisational forms can become “taken for granted”, while others disappear from the “domain of possible” (mechanism 8) (Hannan and Carroll 1992; Baum and Oliver 1992).
In this perspective, organisations are not established as legitimation devices, but simply because they have become the normal or obvious thing to do in given contexts, while other options are not even considered. Over time, some policies become the most Again, this is different from bayesian learning, where the use of the new information, which is used totransform priors into posteriors, does not lead to bias in beliefs.
widespread. As a result, potential new adopters will tend to see those policies as the natural or obvious way to deal with a given problem, and tend to unconsciously exclude other solutions that could also be viable in principle. The outcome of this process is that new adopters tend to choose the dominant policy or organisation, which then becomes even more dominant. Technically, an organisational form that reaches this stage is said to be "taken-for-granted", and it has been argued that "taken-for-grantedness" is directly related to the total number of similar organisations that exist. The main hypothesis that is derived from these arguments is that the relationship between the number of existing organisations and the number of new adoptions has an inverted-U shape: the impact of the former on new adoptions is at first positive (because of taken-for-grantedness), but then becomes negative (because of the increased competition between organisations).
This pattern has been tested empirically and is extremely robust. It can be found in a wide range of organisations, including public schools (Rowan 1982), the multidivisional form of management structure (Fligstein 1985), health care organisations (Ruef 2000), newspapers (Carroll and Hannan 1989), and banks (Ranger-Moore et al. 1991), just to To sum up, I suggest that interdependent (i.e. non-spurious) diffusion mechanisms can be divided in two main categories. In the first, the behaviour of actors is problem- solving oriented, and can be found mechanisms such as rational (Bayesian) and bounded learning, and competitive and cooperative interdependence. In the second, policies are diffused independently from their problem-solving value. The mechanisms are here coercive and normative isomorphism, as well as symbolic diffusion, which can be divided into legitimacy-seeking and taken-for-grantedness. In the next section I present the hypotheses that will drive the empirical analysis.
I will not be able to examine all eight diffusion mechanisms. Rather, I will try to determine the relative importance of spurious and symbolic diffusion (mechanisms 7 and 8) in the spread of IRAs. There are several reasons to focus on symbolic imitation.
4 In the third form of isomorphism, mimetic isomorphism, copying occurs because of uncertainty. It can The first is that, while it is central in sociology, it has been neglected in political science studies of diffusion. As mentioned in section 2, when political scientists speak of social or cultural emulation, they actually mean a process of bounded learning. Secondly, some mechanism, seem less relevant for IRAs than for policies such as foreign economic policy, privatisation or welfare state reform. Learning, both rational and bounded, are likely to very important for IRAs, but the difficulties of an empirical analysis are daunting at this stage. For Bayesian learning, it is crucial to identify clearly (and measure) the outcomes that governments observe in order to update their preferences.
This is far from straightforward given that I compare not only countries but also sectors.
For bounded learning, it should be at the very least determined which is the “peer group”, or group of reference, not only for every country but also for every sector. In effect, it is very much possible that peers are not the same in all sectors. For example, governments could look at the UK for utilities regulation, but at Scandinavian countries for environmental policy. In addition, the homogeneity of the countries under enquiry means that meaningful geographical groups can less easily be identified than in global studies (e.g. Brune and Garrett 2000; Simmons and Elkins n.d.). Finally, normative isomorphism is both interesting and problematic. In effect, it is interesting to note that networks of regulators have been created, and it is very much plausible that within them a common professional culture is created that can eventually lead to isomorphism.
However, the problem is that in most cases regulators can enter these networks only if they are IRAs. Moreover, most such networks are more recent than many IRAs.
Coercive isomorphism is also problematic. Two institutions (or “dominant actors”) can potentially have led to coercive isomorphism for IRAs, namely the OECD and the EU.
The OECD, however, has only recently become interested in the promotion of IRAs, and it has not yet issued explicit recommendations in this direction. The impact of the EU, on the other hand, is partly taken into account in the analysis, as explained below.
In the empirical analysis I will then simply try to determine to what extent the spread of IRAs is due to spurious diffusion, and to what extent to symbolic diffusion. The H1: privatisation and liberalisation have a positive impact on the likelihood that an IRA is created (credibility hypothesis, spurious diffusion); be conceptualised as a form of bounded learning, and is thus not treated separately here.
H2: when controlling for privatisation and liberalisation, the likelihood that an IRA is created is more likely in competition and financial markets than in other regulatory domains (credibility hypothesis, spurious diffusion); H3: the risk for a government of being replaced by a coalition with different preferences has a positive impact on the likelihood of IRA creation (political uncertainty hypothesis, spurious diffusion) H4: the impact of replacement risk depends on whether this high replacement risk is common or exceptional in a given country; its impact is bigger when political uncertainty is unusual (see Gilardi 2003) (political uncertainty, spurious diffusion); H5: veto players increase policy stability (Tsebelis 2002) and are thus a functional equivalent for delegation for both credibility and political uncertainty. Thus, veto players mediate the impact of privatisation, liberalisation, competition and financial markets regulation, and replacement risk (spurious diffusion); H6: the number of IRAs existing at time t-1 have a positive impact on the likelihood that an IRA is created at time t (symbolic diffusion); H7: the number of IRAs existing at time t-1 interacts with functional pressures for the creation of IRAs (H1, H2, H3) and enhances their positive impact on the likelihood that an IRA is created (symbolic diffusion); H8: EU regulations requiring the establishment of separate regulators have a positive impact on the likelihood that an IRA is created (coercive isomorphism).
I will examine the diffusion of IRAs in two steps. Firstly, I follow the methods used in the population ecology literature and analyse the pattern of creation of IRAs at the macro level, by focusing on the evolution of the population of IRAs disaggregated into sub-populations (i.e. regulatory domains). This will constitute a first test of the relevance of symbolic diffusion, but some limitations will be discussed that make this test insufficient. In the second step, then, I shift to the micro level to examine individual IRA creations through an event history analysis. In both cases, the analysis5 relies on an 5 The variables, their operationalisation and source are summarised in Appendix 1.
original data set comprising information for regulators in seven regulatory domains (telecom, electricity, financial markets, competition, food safety, pharmaceuticals, and environment) in seventeen countries (EU member states plus Norway and Switzerland).
In the population ecology literature (Hannan and Carroll 1992; Baum and Oliver 1992), the evolution of organisational populations is analysed by looking at how many new organisations of the same type are created in a given period of time. When sub- types of organisations can be identified, the analysis is carried out at the sub-population level. In this first step, I embrace this technique and examine how many new IRAs are created each year in each of the seven regulatory domains covered by my database. The dependent variable consists of event counts, and this kind of data can be best analysed through Poisson regression models (King 1988; Greene 2003: 740-752). The model used E(yi|xi) = exp(βX), where E(yi|xi) is the expected number of events per period, conditional on the independent variables, and βX is a vector of the regression coefficients and the Table 1 shows the results. The three models investigate the pattern of emergence of IRAs in Western Europe at different levels of analysis. The first model is the most general and undifferentiated. It does not differentiate between regulatory domains and studies only the overall IRAs population. The second model divides the overall population in two subpopulations by considering social and economic regulation domains separately, but without differentiating between, for example, electricity and financial markets, or food safety and environment. In the first model, the number of IRAs is scarcely significant. In the second, the overall number of IRAs has significant positive first-order and negative second-order effects, as predicted by the theory. The number of IRAs at the regulatory type level, on the other hand, is not significant.
Model 3 tests the main argument of this first step of the analysis, namely that the creation of IRAs in different regulatory domains is affected by the number of existing IRAs. It can be seen that number of IRAs at both the overall and regulatory domain level are significant in the expected direction, namely with positive first-order and a negative second-order effect, but not at the regulatory type level. Economic regulation, the dummy used to account for functional pressures for the creation of IRAs, is significant and positive. These results show that the number of IRAs has an impact at both the overall and the regulatory domain level, but not at the regulatory type level. In other words, the creation of an IRA for competition policy, for example, is affected by the existence of other IRAs in the same domain, but also by the existence of IRAs in general.
It does not matter however, whether existing IRAs outside the specific regulatory domain are of the same regulatory type (i.e. economic or social regulation). For the creation of an IRA for competition policy, the existence of IRAs for financial markets or food safety matter equally. Another result is that, in spite of the relevance of the number of IRAs, functional pressures also play a role in explaining the creation of IRAs. In effect, economic regulation, where credibility problems are more acute than in social regulation and thus the incentives to delegate regulatory competencies to IRAs are higher (Gilardi 2002), is positively associated to the creation of IRAs after controlling for the impact of the number of existing IRAs. Symbolic diffusion seems to be at work, but Figure 4 depicts graphically the relationship between the total number of IRAs and the creation of new IRAs6. It can be seen that, as predicted by the theory, the relationship is non-monotonic. When few IRAs exist, the effect of the total number of IRAs on new creations is relatively small, but grows quickly up to the point where the fact that many of the regulators have become IRAs reduces the likelihood that new IRAs are created. A second point of interest in Figure 2 is the difference between the curves for economic 6 To the values on the y axis, representing predicted foundings, should not be attributed too muchimportance as their value depends strongly on the values at which the independent variables not includedin the graph are kept constant.
and social regulation. The curve for social regulation has always lower values than that for economic regulation. This reflects the fact that functional pressures for delegation, notably in terms of credibility, are different between the two types of regulatory This result strongly suggests that a symbolic diffusion process at least partly explains the pattern of creation of IRAs. However, the methods used in this first step of the analysis have several problems that limit its usefulness. The first is that the S-shaped pattern of diffusion shown in Figure 1 can be identified in countless other domains, most of which have nothing to do with politics. Ironically, the same pattern has been followed by the diffusion of research on diffusion (Rogers 1995: 45). It can thus be suspected that the inverted-U relationship between the number of existing IRAs and new IRA creations (or other adoptions) is simply due to the “inescapable mathematics of musical chairs” (Schelling 1978) that characterises many social phenomena, i.e. the fact that they are true by definition. The second is that the focus on whole populations (or sub-populations) prevents the researcher from controlling for many potentially important variables. In the case of IRAs, for example, one cannot account for variables that vary cross-nationally. It is also difficult to integrate time-varying explanations in a consistent way, because they typically vary also cross-nationally. The clearest illustration of this problem is the fact that, in the analysis above, functional pressures for delegation were taken into account only through a time-invariant dummy for economic and social regulation. Some improvement could be achieved, but not much. In other words, most of the hypotheses presented in section 3 cannot be tested, and notably those In the second step of the analysis, then, I move to event history analysis. The unit of analysis is no longer the sub-population/year, but the single regulatory domain in a single country. Concretely, the information contained in the dependent variable tells when a single IRA was set up in a given regulatory domain of a given country: for example, when an IRA for telecom was established in the UK. This shift to the micro level solves both problems. First, the impact of the number of existing IRAs is now studied on individual IRA creations and not on the evolution of the population. If it is found to be significant, it will be much more difficult to argue that it is a mere statistical phenomenon common to all sorts of diffusions. Second, the limits in the capacity to control for variables are due only to data limitations and not to limitations built into the model. The results of the event history analysis will be a much stronger test of the relevance of symbolic diffusion in explaining the observed pattern of IRA creations.
Individual IRA creations can be analysed through event history analysis models.
Event history analysis is a statistical technique explicitly devised to study the pattern and determinants of the occurrence of events (Allison 1984; Yamaguchi 1991). The establishment of an IRA is the relevant event in the context of this study. I employ here the widely-used Weibull model, which takes the form (Box-Steffensmeier and Jones h(t) = h α(ht)α-1 and which is estimated through the equation h(t) = exp(βX + αlnt) where h(t) is the hazard rate, βX is a vector of the independent variables and their coefficients, and α is the shape parameter, whose estimate indicates the effect of time.
The hazard rate is a key concept in event history analysis. It is defined as "the rate at which a duration or episode ends in the interval [t, t + ∆t], given that the duration has not terminated prior to the beginning of this interval." (Box-Steffensmeier and Jones h(t) = lim∆t→0 P(t + ∆t > T ≥ t | T ≥ t) / ∆t.
If the baseline rate α and the explanatory variables are taken into account, the hazard h(t) = lim∆t→0 P(t + ∆t > T ≥ t | T ≥ t; α; βX) / ∆t.
From these two equations it appears that the hazard rate is basically a probability (though it can be bigger than one), and precisely the instantaneous probability that the event occurs in the interval [t, t + ∆t].
The results of the analysis are presented in Table 2. The difference between the three models lies in the level at which the number of IRAs existing at time t-1 (the proxy used for symbolic diffusion) is computed. In the first model (whose results will be examined in detail), the number of IRAs refers to IRAs in economic or social regulation; in the second, to the overall number of IRAs; and in the third, to the number of IRAs in the same regulatory domain. It can be seen that estimates do not change much between the first and the second model: signs do not change at all and significance levels change only slightly. In the third, on the other hand, the impact of the number of IRAs is What do these results tell about diffusion of IRAs? The first model shows that the number of IRAs of the same type (i.e. economic or social regulation) has a positive and significant impact on the hazard of IRA creation. This evidence supports the hypothesis that symbolic imitation has been at work (hypothesis 6). On the other hand, the interaction between the number of IRAs and liberalisation (i.e. a functional pressure for delegation) (hypothesis 7) is unexpectedly negative, but significant only at the 10% level.
Additional analysis (not shown) also indicates that here is no interaction between privatisation and the number of IRAs, nor between it and competition / financial markets. This is not consistent the symbolic imitation hypothesis, which is however supported by the significant independent effect of the number of IRAs. Further, EU directive 97/51 (amending dir. 90/387 and 92/44 for the purpose of adaptation to a competitive environment in telecommunications) has a strongly significant positive impact on the hazard of IRA creation, thus suggesting the existence of coercive isomorphism (hypothesis 8). On the other hand, additional analysis (not presented here) shows that EU directive 96/92 (common rules for the internal market in electricity) has no such effect. It must be said that, although both the telecom and electricity directives leave considerable manoeuvring room to member states with respect to the set-up of IRAs, the electricity directive is less explicit on this point.
There is also abundant evidence that the spread of IRAs has been partially driven by spurious diffusion. First, both privatisation and liberalisation have a significant positive impact on the likelihood that an IRA is created, thus supporting hypothesis 1. The set- up of IRAs is also more likely in competition and financial markets than in other regulatory domains (controlling for liberalisation and privatisation in electricity and telecoms), which is consistent with the prediction of hypothesis 2. These two findings confirm that one of the reasons of the creation of IRAs is the need to improve the credibility of policy commitments. A second spurious diffusion mechanism is related to political uncertainty. The risk for a government of being replaced by a coalition with different preferences increases the hazard of IRA creation (hypothesis 3). Further, its impact depends on the average level of replacement risk in a given country, as predicted in hypothesis 4. On the other hand, the role of veto players / political constraints (hypothesis 5) is only partially confirmed. A significant mediating effect is found only for competition and financial markets, and to some extent for replacement risk (significant at the 10% level), but not for liberalisation and privatisation (as indicated by additional analysis not shown here).
Finally, the model includes three variables that were originally intended to be for control only, but which turned out to supply interesting insights. In effect, the partisan composition of governments seems to be a relevant explanatory variable, in particular in mediating the effects of liberalisation. The sensitivity of governments to symbolic imitation also seems to depend on their partisan composition. Although I had no specific hypotheses on these effects, the results are interesting and add to our understanding of the diffusion of IRAs.
Looking at coefficients is useful, but their direct interpretation is not always straightforward. Thus, I have drawn figures of predicted hazards of IRA creation for some of the most interesting results.
Figure 5 shows symbolic and spurious (liberalisation / privatisation) diffusion effects in the choice to set-up IRAs. It appears clearly that both types of diffusion are present.
Both liberalisation and privatisation increase the hazard of IRA creation, but the impact of the latter is much stronger, as indicated by the vertical distance between the curves.
There are also strong symbolic imitation effects. The likelihood that an IRA is created increases tremendously when the number of existing IRAs goes up. This is a clear manifestation of symbolic diffusion. Governments are influenced by concerns that are not related to problem-solving when deciding to set-up IRAs. However, it is not possible to discriminate legitimacy-seeking from taken-for-grantedness.
Figure 6 shows that the impact of symbolic diffusion varies with the partisan composition of governments. Centre-right governments are more sensitive to the taken- for-grantedness or legitimacy-enhancing effects of IRAs than are centre-left governments. After a certain threshold (around 30 IRAs), the creation of an IRA even becomes more likely under a centre-right government without privatisation than under a centre-left government with privatisation. The question remains of why are centre-right governments more sensitive to symbolic diffusion than centre-left governments. A possible answer could be that the values that are diffused are more in line with the preferences of centre-right governments than is the case with centre-left governments. If this is true, a similar effect could be found also for central banks.
Figure 7 further examines the effects of spurious diffusion, and notably privatisation / liberalisation and the partisan composition of governments on the set-up of IRAs.
Under centre-right governments the hazard of IRA creation is always higher than under centre left-governments, but interaction effects exist only for liberalisation. Privatisation increases the likelihood of IRA creation for both centre-right and centre-left governments. Liberalisation, on the other hand, increases the likelihood for centre-right governments, but decreases it for centre-left governments. This is an indication that when centre-left governments liberalise telecoms or electricity (for reasons that are beyond the scope of this paper, see e.g. Levi-Faur 2002), they want to keep a more direct control over these sectors than does the centre-right. They are more afraid of the possible negative effects of the free market, and are less willing to reduce their intervention possibilities. When it comes to privatisation, on the other hand, both centre-right and centre-left governments acknowledge that delegation is necessary for a credible commitment vis-à-vis those who buy shares of the privatised company, who appreciate guarantees against expropriation dangers (Spiller 1993). In addition, short of liberalisation most of the fears associated to the free market do not emerge, which explains the willingness of centre-left governments to give up some of their direct Figure 8 examines the last source of spurious diffusion, namely political uncertainty, and more precisely the risk for a government of being replaced by a coalition with different preferences. It can be seen that the impact of replacement risk depends on how uncommon is a situation of high replacement risk in a given country. Generally, replacement risk has a positive impact, as predicted by hypothesis 3 and 4, except in countries where average replacement risk is high (i.e. above the mean of all countries, which covers 1/3 of the sample). In these countries replacement risk has a negative impact. This is surprising because, though hypothesis 5 predicts that in these countries the impact of replacement risk is not the same as in countries with low average replacement risk, it does not predict that it is negative. These findings seem partially consistent with recent results on the strategic manipulation of public debt, which, not unlike IRAs, can be a means to constrain the choices of the opposition when it will gain power. Franzese (2002) finds that incumbents who expect frequent oscillation in government may be unwilling to bind themselves to constrain their opposition because they expect to hold office again soon, while with less frequent oscillations the larger expected time between offices leads them to consider strategic manipulations of budget deficits. The findings also seem consistent with de Figueiredo's (2002, 2003) results, which suggest that governments are not willing to pay the cost of insulating their policies if, when they face replacement risk, they expect to be back in power soon. What remains puzzling is that, when expecting frequent oscillations, governments are not insensitive to replacement risk but, rather, are more willing to impose constraints to their opponents when current replacement risk is low.
The results of the analysis can be summarised quite briefly: interdependent diffusion, in the form of symbolic diffusion, has driven the spread of IRAs in Western Europe, but only in part: spurious diffusion, in the form of credible commitment and political uncertainty, has played an important role too.
This paper sought to conceptualise and explain the spread of IRAs in Western Europe. It started from the observation that the empirical pattern of the creation of IRAs has an S-shape that is one of the main constants of diffusion studies. This suggests that the spread of IRAs may have been driven by a diffusion process, i.e. a process where the main actors, namely governments, act interdependently rather than independently.
The conceptualisation of the diffusion of IRAs put forward here (see figure 3) firstly distinguishes between spurious and interdependent diffusion. Spurious diffusion occurs when actors respond in a similar way to similar functional pressures at roughly the same time, like people opening umbrella when it starts raining. In the case of IRAs, such pressures are best identified with liberalisation, privatisation, and more generally economic regulation, which are domains where the need to achieve credible commitment capacity gives governments incentives to delegate some of their regulatory powers. A second element that may give such incentives is political uncertainty, and more precisely the risk for a government of being replaced by a coalition with different preferences. In this case, IRAs are a means to constrain the policy choices of future Spurious diffusion is to be distinguished from interdependent diffusion because actors do not behave in response to each other actions. It still deserves to be called diffusion, albeit with a strong qualification, because it may lead to an outcome which is very similar to that of interdependent diffusion. In the latter, actors are interdependent and do behave in response to each other's actions. Interdependent diffusion mechanisms can be distinguished with respect to how they are related to problem-solving.
Sometimes actors look at each other to improve their problem-solving capacity. In this case, we found mechanisms such as learning (rational / bayesian or bounded) and regulatory interdependence (cooperative and competitive). On the other hand, if actors are influenced by the behaviour of other actors independently from the problem-solving benefits they can get, diffusion can be conceptualised as isomorphism and as symbolic imitation. In the latter, mechanisms such as taken-for-grantedness and legitimacy- The analysis was limited to an examination of the relative importance of spurious and symbolic diffusion in the spread of IRAs. An implication of the latter is that the number of agencies existing at time t-1 should, ceteris paribus, have an impact on the choice to set-up an IRA at time t. This hypothesis has found empirical support, thus corroborating the claim that there has indeed been symbolic diffusion of IRAs. On the other hand, the analysis has also clearly shown that spurious diffusion has also been at work: the need to improve credible commitment capacity and political uncertainty increase the In conclusion, it should not be overlooked that other diffusion mechanisms have not been included in the empirical analysis. Learning in particular, both rational /bayesian and bounded, has a prima facie high relevance, and future research should try to deal with it. The empirical difficulties of such a task make the enterprise all the more number of IRAs created in each year and author's data set 1 = economic regulation (electricity,telecom, financial markets, competition),0 = social regulation (food safety,pharmaceuticals, environment) years since the creation of the first IRA years since the creation of the first IRA in the PoliticalContraints Data Set(Henisz 2000) Parliament and of the Council of 6October 1997 amending CouncilDirectives 90/387/EEC and 92/44/EECfor the purpose of adaptation to acompetitive environment intelecommunications.
Nicoletti (2000), Levi-Faur (2002), OECDregulatory database average of replacement risk over allavailable years (roughly 1953-1999) situation, 4 = left-centre complexion, 5 = Appendix 2. Operationalisation of political uncertainty Following Franzese (2002), replacement risk is operationalised as the product of hazard rate and a standard deviation of the partisan "centre of gravity" of governments across The hazard rate is operationalised as the inverse of actual duration of governments.
Raw data on the actual duration of governments are taken from Woldendorp et al.
(2000), who give the duration in days of each government from about 1945 to 1998. Data for more recent governments are taken from various special issues "Political Data Yearbook" of the European Journal of Political Research", as well as from several online sources, mainly BBC's "country profiles" (http://news.bbc.co.uk/2/shared/bsp/hi/country_profiles/html/default.stm) and The Economist's "country briefings" (http://www.economist.com/countries/).
The units of the raw data are governments. For example, Woldendorp et al. (2000) report that Germany's 22nd post-war government began on 30 March 1983 and lasted 1442 days. The relevant units for my purposes are years, so the data must be transformed. I have first translated duration from days into years by dividing it by 365, and then calculated its inverse (1 / duration in years). The problems remains of how to deal with years where there is a change of government. I follow Franzese (2002) and take mean of the duration of the governments that were in place, weighted by the share of the year that each holded office. For example, in France there was a change of government on the 28th of June 1988. The hazard rate for 1988, then, is 50% that of the first government, and 50% that of the second. For attributing weights, I have considered only months, and not days. Months are attributed to a given government if it was in office for at least 15 days of that month.
The centre of gravity of governments is measured following Woldendorp et al. (2000), who, in their data set, have an indicator called "ideological complexion of government and parliament" which accounts for the relative strength of parties in government with reference to the left-right dimension, through a five-point scale in which the proportional shares of left, centre and right parties are coded 1 to 5. Scores represent the degree of dominance of either party both in parliament and government (1 is right wing and 5 is left-wing dominance). This is admittedly a rough measure, but, for my purposes, I prefer it to alternative, more refined measures such as that of Budge et al.
(2001). These authors have developed measures based on party manifestos and collected an impressive data set where detailed information on the policy preferences of parties over fifty years. Using these data to measure the centre of gravity of government, however, is problematic since the policy position of parties may change over time, often does change and sometimes changes dramatically. For my purposes, this can lead to the paradoxical result that replacement risk increases (through centre of gravity) even though there has been no party change in government. For this reason, I have used the Woldendorp et al. (2000) measure and data. Their data are until 1998 at best; I have updated them from various sources, mainly the same I used for the hazard rate (see Secondly, there is the problem of what standard deviation of the centre of gravity is more appropriate. Franzese uses a moving nine years standard deviation centred on present. This implies that governments base their estimates on the experience of the previous four years as well as on a perfect foresight four years ahead. The idea is that governments are not fully backward-looking, but have some guesses with respect as how their re-election prospects will evolve in the near future. I keep this assumption, but use a seven-year standard deviation with five years back and one forward. The main reason for this choice is that the most recent year are among the most important, and certainly more important than the 1950s. Including many years forward would have prevented me to take most of the 1990s into account. On the other hand, I wanted to keep the assumption that governments have guesses about their fate in the near future.
Assuming that they can see one year ahead seems reasonable.
7 I would like to thank Robert Franzese, who kindly answered some questions I had on hisoperationalisation of replacement risk.
Figure 3. Eight mechanisms of policy diffusion Figure 4. Impact of the total number of IRAs on new Figure 5. Symbolic and spurious diffusion of privatis. onlyboth liberalis. and privatis.
Number of IRAs (economic or social regulation) Figure 6. Symbolic and spurious diffusion of Number of IRAs (economic or social regulation) (liberalisation / privatisation) (Weibull low average repl.
risk (Swi, Ger, Aut) high average repl.
risk (Ita, Gre, Fra) Table 1. Spurious and symbolic diffusion of IRAs: Poisson model Note: * p<0.10, ** p<0.05, *** p<0.01; standard errors in parentheses. Maximum likelihoodestimation.
Table 2. Spurious and symbolic diffusion of IRAs: Weibull model (compet. / fin. markets) x (pol. constr.) Note: Maximum likelihood estimation (streg command in Stata). Robust standard errors in parentheses.
* z < 0.1, ** z < 0.05, *** z < 0.01.
Allison, Paul D. (1984), Event History Analysis, QASS series, London, SAGE Publications Baum, Joel A. C. and Christine Oliver (1992), "Institutional Embeddedness and the Dynamics of Organizational Populations", American Sociological Review, 7, pp. 540-559 Box-Steffensmeier, Janet and Bradford S. Jones (1997), "Time is of the Essence: Event History Models in Political Science", American Journal of Political Science, 41(4), pp. 1414-1461 Boylaud, Olivier and Giuseppe Nicoletti (2000), "Regulation, Market Structure and Performance in Telecommunications", Economic Department Working Papers No 237 Breen, Richard (1999), "Beliefs, Rational Choice and Bayesian Learning", Rationality and Society, Brune, Nancy and Geoffrey Garrett (2000), "The Diffusion of Privatization in the Developing World", Paper presented at the APSA Annual Meeting , Washington DC, 30 August - 3 Budge, Ian Hans-Dieter Klingemann, Andrea Volkens, Judith Bara and eric Tanenbaum (2001), Mapping Policy Preferences. Estimates for Parties, Electors, and Governments 1945-1998, Oxford, Castro, Eduardo and Kathleen McNamara (2003), "The Diffusion of Central Bank Independence", paper for the conference "Interdependence, Diffusion and Sovereignty", Carroll, Glenn R. and Michael T. Hannan (1989), “Density Dependence in the Evolution of Populations of Newspaper Organizations”, American Sociological Review, 54, pp. 524-541 de Figueiredo, Rui J. P. (2002), "Electoral Competition, Political Uncertainty, and Policy Insulation", American Political Science Review, 96 (2), pp. 321-333 de Figueiredo, Rui J. P. (2003), "Budget Institutions and Political Insulation: Why States Adopt the Item Veto", Journal of Public Economics, forthcoming DiMaggio, Paul J. and Walter W. Powell (1991), "The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields", in DiMaggio, Paul J.
and Walter W. Powell (eds), The New Institutionalism in Organizational Analysis, Chicago, Dobbin, Frank, Geoffrey Garrett and Beth Simmons (2003), "The International Diffusion of Democracy and Markets", Paper for the conference "Interdependence, Diffusion and Eyestone, Robert (1977), "Confusion, Diffusion, and Innovation", American Political Science Fligstein, Neil (1985), "The Spread of the Multidivisional Form among Large Firms, 1919-1979", American Sociological Review, 50, pp. 377-391 Franzese, Robert J. (2002), Macroeconomic Policies of Developed Democracies, Cambridge, Gilardi, Fabrizio (2002), "Policy Credibility and Delegation to Independent Regulatory Agencies: A Comparative Empirical Analysis", Journal of European Public Policy, 9 (6), pp. 873-893 Gilardi, Fabrizio (2003), "Delegation to Independent Regulatory Agencies in Western Europe: A Cross-Sectional Comparison", Paper presented at the workshop "Delegation in Contemporary Democracies", ECPR Joint Sessions of Workshops, Edinburgh (UK), 29 Greene, William H. (2003), Econometric Analysis, 5th edition, Englewood Cliffs, Prentice Hall Hannan, Michael T. and Glenn R. Carroll (1992), Dynamics of Organizational Populations. Density, Legitimation, and Competition, Oxford, Oxford University Press Henisz, Witold J. (2000), "The Institutional Environment for Economic Growth", Economics and King, Gary (1988), "Statistical Models for Political Science Event Counts: Bias in Conventional Procedures and Evidence for the Exponential Poisson Regression Model", American Journal of Political Science, 32, pp. 838-863 Lazer, David (2001), "Regulatory Interdependence and International Governance", Journal of European Public Policy, 8 (3), pp. 474-492 Levi-Faur, David (2002), "The Politics of Liberalisation: Privatisation and Regulation-for- Competition in Europe's and Latin America's Telecoms and Electricity Industries", Nuffield College Working Paper in Politics 2002-W13, University of Oxford Levy, Brian and Pablo T. Spiller (1994), “The Institutional Foundation of Regulatory Commitment: A Comparative Analysis of Telecommunications Regulation”, Journal of Law, Economics, and Organization, 10 (2), pp. 201-246 Majone, Giandomenico (1997), "Independent Agencies and the Delegation Problem: Theoretical and Normative Dimensions", in: Steuenberg, Bernard and Frans van Vught (eds.), Political Institutions and Public Policy, Dordrecht, Kluwer Academic Publishers, pp. 139-156 Majone, Giandomenico (2001), "Nonmajoritarian Institutions and the Limits of democratic Governance: A Political Transaction-Cost Approach", Journal of Institutional and Theoretical McDermott, Rose (2001), "The Psychological Ideas of Amos Tversky and Their Relevance for Political Science", Journal of Theoretical Politics, 13 (1), pp. 5-33 Meier, Kenneth J. (1985), Regulation. Politics, Bureaucracy, and Economics, New York, St. Martin's Meseguer, Covadonga (2002), Bayesian Learning about Policies, Madrid, Centro de Estudios Avanzados en Ciencias Sociales, Instituto Juan March de Estudios e Investigaciones Meyer, John W. and Brian Rowan (1977), "Institutionalized Organizations: Formal Structure as Myth and Ceremony", American Journal of Sociology, 83 (2), pp. 340-363 Moe, Terry M. (1990), "Political Institutions: The Neglected Side of the Story", Journal of Law, Economics, and Organisation, 6, Special Issue, pp. 213-105 Ranger-Moore, James, Jane Banaszak-Holl and Michael T. Hannan (1991), "Density-Dependent Dynamics in Regulated Industries: Founding Rates of Banks and Life Insurance Companies", Administrative Science Quarterly, 36, pp. 36-65 Rogers, Everett M. (1995), Diffusion of Innovations, 4th edition, New York, The Free Press Rowan, Brian (1982), Organizational Structure and the Institutional Environment: the Case of Public Schools”, Administrative Science Quarterly, 27, pp. 259-279 Ruef, Martin (2000), "The emergence of Organizational Forms: A Community Ecology Approach", American Journal of Sociology, 106 (3), pp. 658-714 Schelling, Thomas C. (1978), Micromotives and Macrobehavior, New York, W.W. Norton Simmons, Beth and Zachary Elkins (n.d.), "Competition, Communication or Culture? Explaining Three Decades of Foreign Economic Policy Diffusion", mimeo, University of California, Spiller, Pablo (1993), "Institutions and Regulatory Commitment in Utilities' Privatization", Industrial and Corporate Change, 2 (3), pp. 387-450 Steiner, Faye (2000), "Regulation, Industry Structure and Performance in the Electricity Supply Industry", Economic Department Working Papers No 238 ECO/WKP(2000)11, Paris, Tsebelis, George (2002), Veto Players. How Political Institutions Work, Princeton, Princeton Tversky, Amos and Daniel Kahneman (1974), "Judgement under Uncertainty: Heuristics and Biases", Science, 185(4157), pp. 1124-1131 Weyland, Kurt (2002), "The Diffusion of Innovations: A Theoretical Analysis", paper presented at the APSA Annual Meeting, Boston, 29 August - 1 September 2002 Yamaguchi, Kazuo (1991), Event History Analysis, London, SAGE Publications Woldendorp, Jaap, Hans Keman and Ian Budge (2000), Party Government in 48 Democracies. Composition - Duration - Personnel, Dordrecht, Kluwer Academic Publishers

Source: http://poli.haifa.ac.il/~levi/res/GilardiB.pdf