Four thoughts on the future of #policydiffusion research

Last week I had the pleasure of being on a roundtable on “Transnational Diffusion: Concepts and Mechanisms” at the ISA conference in San Francisco, together with Etel Solingen, Zachary Elkins, Detlef Jahn, and Covadonga Meseguer. In fact, diffusion was the general theme of the conference, which pissed off some tasteless discerning people.

It was a great occasion to think freely about the future direction of diffusion research. Here is a summary of my contribution to the discussion. A more traditional review of the literature is here, and on Twitter I use #policydiffusion for tweets related to this topic.

1) Concepts are clear, worry about measures

We have reached a consensus on the definition of diffusion. Diffusion is a consequence of interdependence and is not defined exclusively (or even primarily) by the fact that something has spread. This implies that, when studying diffusion, we are interested more in the process than in the outcome. Convergence, for instance, can be a useful complement to a diffusion analysis, or it can motivate the research in the first place, but is not what we are actually studying.

Moreover, there is consensus on three broad classes of diffusion mechanisms: learning, emulation, and competition (some add coercion, but I disagree). For definitions, see this chapter.

This means that, conceptually speaking, it is pretty clear what we talk about when we talk about diffusion. There is definitely room for some improvement, but not much. Most new conceptual distinctions are hairsplitting. Where we have real problems is with operationalization. In this paper (still at draft stage), Martino Maggetti and I have done a meta-analysis of 100+ diffusion studies and have found that there is a lot of confusion on what indicators are appropriate for the different mechanisms. The same indicators are used for different mechanisms, and different indicators are used for the same mechanism. A mess. This has to improve if we want to generate more cumulative knowledge.

2) Learn about diffusion, or use diffusion to learn about something else?

There are two types of research questions that are worth asking at this point.

First, we can try to make a contribution to the diffusion literature itself. It is becoming harder and harder to pull this off successfully. The n-th study showing that a unit is more likely to adopt a policy if its neighbors/competitors/etc. have done so is not going to cut it. What is required is better, more focused questions, which themselves require better, more focused theory. As the building blocks of diffusion are fairly clear, theoretical advances should aim to explain more precisely how they operate in different contexts. For instance, Susan Hyde has been working on how practices diffuse and become norms in virtue of the signals they send. For instance, refusing to invite observers to monitor elections has become an unambiguous sign that a country is not democratic, which is why even clearly non-democratic states do it. In a panel at last week’s ISA conference, Hyde suggested that many other phenomena may fit this argument, such as sovereign credit ratings. In my own work I have tried to move theory forward by, for instance, arguing that different policy makers learn from different policy outcomes (including implications for their re-election) and that socialization attenuates tax competition.

Second, we can try to use the insights of diffusion research to learn something new about other phenomena. Diffusion often gives an original angle to do this. For instance, traditional work on competition focuses on, well, competition, but diffusion research tells us that this is just one type of interdependence among others: there is more to competition than just competition. Or, the literature on women’s representation has identified many types of spillovers, but it turns out that, until women’s participation in politics becomes a well-established norm, the number of women candidates in one unit increases with the number of women elected in other units.

3) We need better research designs

Standard research designs have almost fulfilled their potential. Adoption in one country = f(adoption in other countries, controls) has been done to death. Mostly for good reasons: it is a good approach to show that something diffuses. But if we want to push things forward, at this point we need something new (and better).

First, we need better data. Often this means moving away from cross-national analysis, which is also a trend in political science in general. In many cases sub-national units offer data of higher quality, are more comparable, and are closer to the level at which the action is really going on. We should also think creatively about data sources. Automated text analysis seems an especially promising avenue in this respect.

Second, research designs should be tailored to the specific questions asked. This is a truism that applies to any area, of course, but the problem seems particularly acute in diffusion research. There is a clear template with which we can study almost anything, so there is the temptation to actually do it. Which is fine, except that we cannot expect significant new insights to follow.

Third, we should take causal inference more seriously. This is a big fad trend in political science right now, but one needs not be an identification Taleban to say that very, very few diffusion studies in political science pay any attention at all to this issue. Although in our context the problem is even thornier than usual (some say there is no hope), the status quo is not OK and we should do our best to improve on this front.

4) So what? Enter the “diffusion multiplier”

If you are into diffusion, you cannot get enough of it. But why should others care?

Well, assume that you are an advocate of marriage equality. Same-sex marriage has already spread a good deal, but it is still by no means the norm. Would it not be useful to know which states or countries one should persuade in order to accelerate the process? We can call this the “diffusion multiplier” (by analogy with the “social multiplier”): if the “right” units adopts a policy, others will be more likely to do the same. Thus, by influencing one unit directly, many more are reached indirectly.

This works also the other way round: if you want to prevent the spread of a policy, it would be useful to know on which units you should concentrate the efforts. For instance, not all states are equally effective as firewalls against the spread of soda bans.

Of course, we first need to know which units are influential and why, which is where diffusion research has something to say. We are nowhere near being able to make such specific recommendations, but this is certainly one of the potential practical payoffs of this literature.

Can cooperation limit tax competition? Three lessons from Switzerland

This post is co-authored with Fabio Wasserfallen and is cross-posted (with a different title) at the LSE European Politics and Policy Blog.


The creation of the single market is widely believed to have strengthened tax competition among European countries; with few remaining barriers to the movement of capital and people, some member states reduce levels of tax to attract more investment at the expense of others. Despite several efforts, policy makers have not been able to agree on effective political actions constraining this dynamic, which many consider to be harmful. Evidence from tax competition in Switzerland suggests that institutional cooperation in fiscal affairs between selected countries in the EU might help to limit the negative externalities of tax competition.

Tax competition is a key characteristic of Swiss federalism. Switzerland’s twenty-six cantons enjoy almost unlimited freedom to set taxes and the general consensus is that they compete with one another to attract tax payers. There is disagreement on whether tax competition is a good or a bad thing, but all agree that it is an important phenomenon. In our study of tax competition between Swiss cantons, we started from this premise but then looked at the factors constraining competition. In particular, we analysed how institutionalised forms of cooperation between cantons limit the extent of competition between those cantons that participate in them.

In the Swiss context, an important role is played by the so-called regional conferences of cantonal ministers. These institutions have been in place for several decades and serve as a forum for discussing policy problems and elaborating common positions for negotiations with the federal government. An important goal of these conferences is the defense of cantonal autonomy. Therefore, they do not pursue stronger cooperation in tax policy. However, constraints on tax competition emerge as a byproduct of cooperation on other issues. The fact that finance ministers work together on a regular basis and develop personal relations makes them more sensitive to how their tax policies affect their colleagues. By no means does this form of socialization erase tax competition. However, it sets limit to it.

Our statistical analysis shows that, controlling for many other factors, tax competition is stronger between cantons that do not work together in the same regional conference. Specifically, we measured the extent to which two cantons are in competition with one another by looking at the commuting patterns between them (that is, how many people live in one canton but work in another). Most studies assume that competition occurs between neighbours. Our approach is similar but, we argue, more precise. In effect, in some cases using neighbourhood as a measure does not accurately show the connections between two cantons; moreover, the binary nature of neighbourhood (either two cantons are neighbours, or they are not) gives no information on the scale of connectedness. By contrast, commuting patterns give a fine-grained picture of the extent to which cantons compete with one another for taxpayers. An intuitive way to read the results of the statistical analysis is that, if a canton has two comparable competitors, it will react more strongly to the tax policies of the one with which it does not work together in the regional conference. Our interpretation of this pattern in terms of socialisation is consistent with the information we obtained in interviews with policy makers. For instance, they explained to us that, while the regional conferences strive to project a public image of consensus, in fact policy makers can be quite outspoken about their dissatisfaction with the policies of their colleagues.

Our research suggests that, in the Swiss case, cooperation can limit tax competition. What can the EU learn from the Swiss example? While there are enormous differences between these two political systems, we think that three conclusions may be helpful:

1) Most importantly, cooperation should take place between the relevant subgroup of countries. One reason why the Swiss regional conferences help to limit tax competition is that they bring together precisely those cantons that are most likely to compete with one another. In the EU, a more clustered and specialised structure of intergovernmental relations might increase the effectiveness of cooperation also in other policy areas.

2) Less intense tax competition does not have to be the goal of cooperation. It can emerge as a byproduct of cooperation on other issues because working together fosters social influence.

3) The unintended benefits of cooperation likely emerge only in the long term. Cantons have a prolonged history of cooperation and know exactly with whom they will have to work in the future. Newly established institutions cannot be expected to lead immediately to similar outcomes.

All in all, these lessons do not suggest that tax competition within the EU can be reshaped simply by enhancing a few cooperative arrangements. Nor would such a claim be credible. What our research implies, however, is that, over time, effective constraints on tax competition can emerge, provided that the structure of cooperation matches that of competition. Whether policy makers want to foster competition or rein it in, this conclusion can inform their choices.

Diffusion everywhere: Regulatory independence reaches the Vatican

The diffusion of independent regulatory authorities has been well documented (e.g., here, here). Now the tide has reached even the Vatican:

The Vatican is seeking to make its new financial watchdog agency fully independent by relieving its president of his other job in the Holy See’s administration.

There had been questions about possible conflicts of interests when Cardinal Attilio Nicora was named president of the Vatican’s Financial Information Authority earlier this year.

The watchdog agency was created to ensure all Vatican financial transactions comply with EU and international anti-money laundering and anti-terror financing laws.

Vatican spokesman the Rev. Ciro Benedettini said Thursday that “common sense” dictated that as chief watchdog, Nicora couldn’t be responsible for checking compliance of his other office, which administers Vatican personnel and other assets.

Policy diffusion and survey experiments

When studying policy diffusion, establishing causality is even trickier than usual because interdependence violates an important assumption of the Rubin causal model, namely the so-called “stable unit treatment value assumption” (SUTVA), which requires that the outcomes in one unit do not depend on the treatment status of other units.

With observational data, going around this problem is very hard. In an forthcoming article just published in the American Journal of Political Science, Katerina Linos uses a survey experiment to address this issue. Survey experiments have become increasingly popular in political science but, as far as I am aware, Linos is the first to apply this method to policy diffusion.

The study asked respondents (a representative sample of Americans) whether they agreed or disagreed that “the United States should increase taxes in order to provide mothers of newborn children with paid leave from work.” Respondents were assigned randomly to either this baseline question or to one of four treatments, which consisted of statements that (1) “Canada” or (2) “most Western countries” already have similar policies, or that the policy is recommended by (3) the United Nations or (4) American family policy experts. The main results are shown in this graph (own elaboration of Table 1 in the article):

The effect of all four treatments is large and also remarkably similar. In additional analyses, Linos shows that these effects decrease with the level of information of respondents, which suggests that the experience of other countries and the advice of authoritative people or organizations influence opinions by supplying new information and not, for instance, by setting certain normative standards to which it feels appropriate to conform.

Interesting work, which should encourage other diffusion scholars to use these tools both to replicate these findings and to explore other questions.

(Lack of) diffusion everywhere: negative interest rates


Way back on August 27, 2009 the FT reported:

“Bankers Watch as Sweden Goes Negative”.

At issue was Sweden’s embrace of unorthodox monetary policy, specifically their decision to cut the interest rate paid on bank reserves below zero. In other words, they were charging a penalty. In the USA, by contrast, the Federal Reserve actually raised the interest rate paid on reserves.

Yesterday, the FT reported:

“Sweden records fastest quarterly growth”

That’s the fastest growth Sweden has ever recorded. Ever. They’ve already tightened monetary policy from where it was in the depths of the crisis and, naturally, are poised to do some additional tightening. So are bankers actually watching this? It seems to me they’re not. When Sweden did something unorthodox, people said the world’s central bankers were going to watch. It worked. It worked really really well.

But is anyone paying attention?

Diffusion everywhere: Uprising in North Africa


Egyptian police have used tear gas and water cannon to break up rare anti-government protests in the capital. Thousands of people had joined the protests in Cairo, inspired by the uprising in Tunisia and vowing to stay in place until the government fell.

In a recent article on the diffusion of regime contention in Europe between 1830 and 1940, Kurt Weyland writes:

Tapping in the dark in their own country, oppositionists are easily impressed by a shining example elsewhere. The unexpected fall of a foreign autocrat suggests to disaffected groups in other polities that the time has come for the decisive move. If a seemingly powerful regime is suddenly revealed as brittle, they are tempted to believe that their own ruler is equally weak and that their compatriots are willing and able to shake off the yoke of nondemocracy as well. Given that citizens of the first country achieved an unexpected success, they should manage to accomplish the same feat! Given the imperfect information and high uncertainty prevailing under autocracy, an external precedent can exert great impact, prompting an immediate updating of situational judgments, suggesting a propitious opening, and thus triggering a rash of challenges. The precedent of a foreign success can acquire disproportionate importance and suddenly reshape political actors’ assessments of opportunities and risks. In these ways, a stunning precedent can inspire many efforts at emulation and thus provide a powerful impetus for change.

Moreover, Weyland argues that the outcomes of the diffusion process can be quite diverse. In addition to the successful replication of the protests (which is quite rare) and to their failure, diffusion can lead to:

– preemptive reforms, with which authorities attempt, often successfully, to prevent their downfall, but which can bring significant progress towards democracy;

– determined repression, which reinforces the authorities’ stronghold on power and reduces the prospects democratization prospects in the medium term.

The latter scenario implies that outcomes in different countries are negatively correlated, which is a good reminder that diffusion needs not lead to convergence.