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.

Advertisements

Getting Closer When It’s Close

A guest post by Oliver Strijbis, Sveinung Arnesen, Kjetil Thuen, and Lucas Rachow*

*****

Two weeks ago we published on this blog the predictions taken from our prediction market on the outcome of the ballots from March 3. The predictions were taken on February 22 – nine days before the official voting day. Our predictions deviated from the official results from 0.3% to 6% depending on the ballot. After one year of experience with prediction markets we come to the conclusion that this result is largely representative of our predictions over the last year.

The prediction on the Yes-vote for the “Bundesbeschluss über die Familienpolitik” was a strike. The prediction was that 54.6% of the voters would be in favor of the law while on voting day it was 54.3% (0.3% deviation). With only 3.2% deviation also our prediction on the “Raumplanungsgesetz” was rather accurate: while we expected a Yes-share of 59.7% on voting day it was 62.9%. The least precise prediction was on the share of Yes-votes for the “Abzocker-Initiative” – the most hotly debated proposal – where we expected 61.9% Yes-votes against the resulting 67.9% (6% deviation).

The predictions were rather representative of our experience with the prediction market since our first try one year ago. First, the predictions have been more precise the closer the outcome was. This is clearly reflected in our experience with the 26 ballots (15 national and 11 cantonal) for which we made predictions during the last year. Secondly, the accuracy of our predictions were within the range that we found for the predictions made previously. While the accuracy is typically within a 5% margin of error when the Yes-votes are between 40% and 60% it gets larger when the outcome is more clear (as with the “Abzocker-Initiative”).

Overall, after one year of applying our prediction markets to Swiss direct democratic decisions we can conclude that they have considerable potential. While they might not necessarily be equally precise as electoral forecasts they clearly allow to get a good feeling about the probable outcome of ballots at an early stage of the campaign and in particular when the race is close.

* Oliver Strijbis is a postdoctoral researcher at the University of Hamburg, Sveinung Arnesen is a postdoctoral researcher at the University of Bergen. Together with Kjetil Thuen and Lucas Rachow they are founders of politikprognosen.ch.

Predictions for the Ballots on March 3

This is a guest post by Oliver Strijbis, Sveinung Arnesen, Kjetil Thuen, and Lucas Rachow*

***

About one year ago we published on this blog for the first time predictions for direct democratic votes taken from our “prediction market”. Prediction markets are designed to aggregate information and produce predictions about future events. Prediction markets are markets for contracts that yield payments based on the outcome of an uncertain future event, such as an election or a referendum. A considerable bulk of literature has shown that prediction markets can easily compete with surveys in forecasting election outcomes. This has also been shown for the Swiss parliamentary elections in 2011 where the forecasts of the prediction markets were more accurate than those of the surveys (Tagesanzeiger online, 27th October 2011).

Predicting election results, however, is clearly easier than results from direct democratic votes. And indeed, the accuracy of our forecasts published one year ago were rather mixed. We identified several reasons why this might have been the case. One reason is that predictions are easier if there is abundant information, which is clearly more so for national elections than for initiatives and referenda. There is little to do about that and predictions on direct democratic votes might always be somewhat less precise than election forecasts.

However, as another major reason for the rather large variance in the accuracy of our predictions we hypothesized that the participants need to learn. As a consequence, we decided to further develop our prediction market and applied it to the ballots of May, September, and November 2012. This allowed us to maintain a rather small though faithful community of traders. In order to test our hypothesis that for the prediction of direct democratic votes the ability of the traders is particularly important, we also made use of the knowledge about their behavior in previous rounds. In order to give the best traders more influence, they could now keep playing with the raised overall amount of money from the previous prediction cycle (all participants would win between 20 and 150 Swiss francs). Hence, the ballots from March 3 will allow us to test how important the ability of the traders in the market actually is for the accuracy of our predictions.

What, then, does our market foresee for the ballot of March 3? Here are our predictions from February 22: 61.9% yes for the “Abzocker-Initiative”, 54.6% yes for the “Bundesbeschluss über die Familienpolitik”, and 59.7% yes for the “Änderung des Bundesgesetzes über die Raumplanung”. Hence, for all three proposals we anticipate a rather clear victory.

* Oliver Strijbis is a postdoctoral researcher at the University of Hamburg, Sveinung Arnesen is a postdoctoral researcher at the University of Bergen. Together with Kjetil Thuen and Lucas Rachow they are founders of politikprognosen.ch.

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.

Gaddafi’s Demise Is Not The Only Reason for The Military Coup in Mali

A guest post by Jonathan van Eerd, a PhD candidate at the University of Zurich. He recently completed field work in Mali’s capital city, Bamako, and is currently a visiting scholar at Cornell University.

****

Mali was considered to be one of the few functioning democracies in West Africa. It never experienced a military coup since the introduction of multiparty democracy in 1992. Considering that, last Wednesday’s coup comes as a surprise.

The group under the lead of so far unknown Capt. Amadou Sonago claims that they have overthrown the democratically elected government of Amadou Toumani Touré (short: ATT) because of its “incompetence” in handling the Tuareg Rebellion in Mali’s North.

The Malian army was indeed poorly prepared for its newest task of defending the nation’s unity. The soldiers are badly trained, have outdated weaponry and not enough supply. There were failures in informing soldier’s families about fatalities in combat.

Mali is one of West Africa’s few fairly working democracies. Why was there no national or international political force that pleaded for the strengthening of the Malian forces in the combat against the rebellion?

Being one of the least developed countries in the world, the internationally supported downfall of the Gaddafi’s regime in Libya caught Mali on the wrong foot. Many of Gaddafi’s former Tuareg-soldiers became jobless and went back to the Sahel region, of which Mali’s North is a part. They have not been disarmed by anyone. And in January of this year they started a new Tuareg-rebellion in Mali’s North. Its goal is the independence of Mali’s northern regions.

Mali has experienced recurring Tuareg-rebellions since the sixties. However, the intensity of this new rebellion was unprecedented. Along with that appeared a new generation of well-armed Tuareg-fighters, which came back from Libya. Together with some factions of older Tuareg-rebellions, they formed the National Movement for the Liberation of Azawad (MNLA).

The Malian government and its army were taken by surprise. They were ill-prepared because the peace treaty they have signed after the last Tuareg-rebellion in 2008 with the old generation of Tuareg-rebells and the generous “development aid” of the Malian government for the Tuareg to keep them at ease gave the government a misleading feeling of security. Additionally, the foreign, most notably French, diplomatic and military aid for Mali was rather weak or even counterproductive: The French engaged in direct talks with the MNLA, because they hoped to gain their help in France’s battle against the terror-organization Al-Qaeda Organization in the Islamic Maghreb
(AQMI), which kidnapped and murdered French and other Western tourists and expats in the region. This boosted the MNLA’s self-confidence and led to diplomatic tension between France and its former colony.

But these are not the only causes of the coup. One important cause lies in the nature of ATT’s governing style and Mali’s political culture itself. ATT, who was without any partisan affiliation, preferred to govern with a consensual-styled all-party cabinet. Since his re-election in 2007 he included every important party – with one exception – in one way or another in his government: Out of 160 parliamentarians, only 4 of a small socialist party were in the opposition before the coup.

In such an ethnically heterogeneous country like Mali the inclusion of every important power-base has many advantages. The fact that Mali experienced relative stability or no ethnic conflicts in contrast to its neighbors Ivory Coast, Niger or other West African countries in the last 20 years proves this point.

However, Mali’s financially and organizationally weak political parties are not only in ATT’s all-party government due to political sanity, but also for their very political survival: In most African democracies and semi-democracies, access to state resources is crucial to win important client’s favor with gifts and other privileges, in order to make sure that they give support in elections. Consequently, no Malian political party was willing to occasionally play the indispensable part of the opposition, meaning that no important party pointed out to the deficiencies and failures of the ATT government in its handling of the newest Tuareg-rebellion. The parties did not want to risk their participation in the government and the consequential loss of access to state resources.

Switzerland is the world’s most famous example of a functioning consensus democracy. However, in contrast to Mali and most other democracies in the world, Switzerland’s democracy features extensive direct-democratic rights for Swiss citizens. This ensures that the citizens itself occasionally play the missing part of the opposition to a consensus government.

As no party wanted to harm its share in the Malian government, they criticized ATT only off the record for his lack of foresight on the fallout of the conflict in Libya, his hesitant diplomatic and military reaction, as well as his almost non-existent information policy regarding the rebellion in the North.

On the contrary, all political parties awaited ATT’s orderly replacement in April’s presidential elections. The unexpressed consensus was to first await the new president and only after that to strive for a solution of the conflict in the North. Until then, the parties concentrated on their preparations for the elections. Yet even while doing that, they did not consider to raise the issue of the Tuareg-rebellion as a topic for their individual election campaigns; again for the sake of the all-party consensus.

As a result, one part of the Malian army decided to take matters, or respectively, the part of the opposition in their own hands. In a drastic and non-democratic manner they pointed out the deficiencies and grievances of their army.

Prediction Markets and the Ballot of March 11: A First Evaluation

A guest post by Oliver Strijbis, Sveinung Arnesen, Kjetil Thuen, and Lucas Rachow. Oliver Strijbis is a postdoctoral researcher at the University of Hamburg, Sveinung Arnesen is a postdoctoral researcher at the University of Bergen. Together with Kjetil Thuen and Lucas Rachow they are founders of politikprognosen.ch.

****

Shortly before the voters went to the polls on March 11 we published on this blog the predictions taken from our prediction market. In short, the accuracy of our forecasts were mixed. Whether the predictions were accurate or not largely depended on the proposal. But let’s look at it in detail.

The good news is that the predictions for the ballot on the “Buchpreisbindung” and the “Bauspar-Initiative” were strikes. The prediction from March 7 (the latest prediction we published) was that the share of Yes-votes for the Buchpreisbindung would be 45.5% while it was 43.9% (1.6% deviation). The share of Yes-votes for the Bauspar-Initiative was expected to be 46.7% while it was 44.2% (2.5%).

Our predictions for the other proposals were less accurate. For the ballot on the “Zweitwohnungen” we predicted 45.5% of Yes-votes – deviating 5.1% from the final result of 50.6%. Hence, we did not predict the right winner in this instance. Our predictions for the referendum on the law on “Geldspiele” and the “6 Wochen Ferien für alle” initiative have foreseen the winner but were rather far off from the final results: while our prediction was 70.5% of “Yes” for the Geldspiele the result was 87.1%, and while we foresaw the share of Yes-votes for “6 Wochen Ferien für alle” to be 39.9% it was only 33.5%.

Why have our predictions concerning some ballots been far more accurate than others? It is generally argued that the accuracy of the predictions depend on what kind of relevant information the traders possess about the topic, and to which degree the traders are motivated and able to express their beliefs in the prediction market framework (e.g. Sunstein 2006). Based on these theoretical assumptions we would like to discuss a couple of points that might explain variance in the accuracy of the predictions of the March 11 ballots.

The proposals. Predictions are easier if there is abundant information on a proposal and if the campaign follows a well-known pattern. The fact that not less than five proposals were voted on at the same day had the effect that media coverage on some of the proposals was scarce. Furthermore, the initiative on the “Zweitwohnungen” and to some degree also the referendum on the law on “Geldspiele” did not produce constellations of political alliances which are typical for direct democratic decisions in Switzerland. Hence, experience helped only partially for the formation of well-informed expectations. However, in the case of the initiative on “6 Wochen Ferien für alle” the pattern followed a clear left-right division that has been experienced in many cases in the past. This means that also features concerning the topic and the number of the proposals can only partly explain the variance in the accuracy of our predictions.

The participants. Prediction markets work on the basis of the relevant information accumulated by the participants. We have tried to guarantee that our market profits from well-informed participants by recruiting mainly among social scientists in general and political scientists in particular. However, for many of them the mechanisms of a prediction market might be new. Furthermore, some of them might participate more out of goodwill than intrinsic motivation.

While some of our predictions have pointed to the potential of prediction markets for forecasting the results of direct democratic decisions in Switzerland, there remains room for improvement. Although, due to the often very short and spiritless campaigns, predictions on Swiss direct democratic decisions might always be fraught with risk, we think that this improvement is possible. Since the participants are undergoing a learning process and since in the future participants can be substituted we are confident that we can achieve more accurate predictions in a not too distant future. Even more, competent traders will also over time accumulate wealth in the market and thus become more powerful than those making wrong trades. In other words, we expect more influence to be transferred to those with good judgment, and less to those with bad judgment. Much will now depend on our ability to gain the most informed and motivated participants.

Posted in Uncategorized | 1 Reply

Predictions for the Ballots on March 11

A guest post by Oliver Strijbis, Sveinung Arnesen, Kjetil Thuen, and Lucas Rachow. Oliver Strijbis is a postdoctoral researcher at the University of Hamburg, Sveinung Arnesen is a postdoctoral researcher at the University of Bergen. Together with Kjetil Thuen and Lucas Rachow they are founders of politikprognosen.ch.

****

Who thinks in predictions of election and referenda outcomes thinks in surveys. However, an alternative method for forecasting has made its way into political science the last years. So called “prediction markets” are designed to aggregate information and produce predictions about future events. Prediction markets are markets for contracts that yield payments based on the outcome of an uncertain future event, such as an election or a referendum. A considerable bulk of literature has shown that prediction markets can easily compete with surveys in forecasting election outcomes. This has also been shown for the Swiss parliamentary elections in 2011 where the forecasts of the prediction markets were more accurate than those of the surveys (Tagesanzeiger online, 27th october 2011).

In an attempt to apply prediction markets in the context of Switzerland’s direct democracy we set up prediction markets for the ballots of March 11. We arrive at forecasts by setting up a market for each of the proposals. This means that on each of five markets the outcome of one proposal is treated as an asset. At voting day an asset pays the share of votes the proposal has received. For instance if the proposal “Schluss mit uferlosem Bau von Zweitwohnungen!” gets 45% of the votes, the final price of an asset of this proposal pays 45 units. Hence, a participant on the prediction market has an incentive to buy assets if the price is below 45 units and an incentive to sell if it is above. Consequently, rational players will buy assets if the current price is below the expected outcome and sell if it is above.

With the assistance of colleagues from various Swiss universities (special thanks to Laurent Bernhard) we were able to recruit 124 individuals of which 87 turned out to be active participants. In an attempt to recruit only the most talented players we were primarily approaching political scientists (students and professionals) and individuals trained in a related field. From the 87 participants 27% were political scientists, 21% economists, and 22% were trained social scientists from other disciplines. The participants are compensated with a small salary depending on their performance.

While theory tells us that our proceeding should yield accurate forecasts, only empirics can demonstrate it. So what do the prediction markets tell us for the ballots on March 11? Here are our predictions from March 3 (see Figure): 46.7% yes for the “Bauspar-Initiative”, 45.5% yes for the law on the “Buchpreisbindung”, 39.9% yes for the initiative “6 Wochen Ferien für alle!”, 70.5% yes for law on the “Neuregelung der Geldspiele”, and 45.5% yes for the initiative “Schluss mit uferlosem Bau von Zweitwohnungen!”. Hence, for all three proposals where a close race is expected we anticipate a narrow defeat.