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.

The metamorphosis of an old political institution

(with Karima Bousbah)
The Swiss political system provides a few institutional veto points, which were originally thought to protect the catholic minority, organised around the Christian Democratic Party (CVP). After a re-configuration of the political conflict lines, the ‘Ständemehr’ now seems to serve the interest of the Swiss People’s Party (SVP), at the detriment of the CVP.  

In Switzerland, all constitutional amendments need to be approved in compulsory referendums by both the majority of people (Volksmehr) and the majority of cantons (Ständemehr). This “cantonal veto” was originally introduced in order to accommodate the catholic minority, and as a veto card of their political arm, the Christian Democratic Party (CVP). Historically, the Christian Democrats (CVP) could count on very large majorities among voters in the catholic cantons of Switzerland. On average, the catholic cantons are considerably smaller than the protestant cantons, and thus, they can use the Ständemehr to bloc constitutional amendments.

In the past 165 years, there have been only nine instances in which the Ständemehr had a direct effect. Meaning that, so far, in the history of Swiss direct democracy the majority of cantons has overruled the majority of the voters on nine different referendums on constitutional articles. However, it is possible that the Ständemehr had furthermore very strong anticipatory effects on parliamentary decision-making: in the last 100 years (this is the period for which voting recommendations by the political parties are available), there has been only one constitutional amendment opposed by the CVP, which was subject to a popular vote. It reached voters’ majority approval only in three (mainly protestant) cantons.

In a newspaper article, which appears on Tuesday, I am suggesting that the Ständemehr might, over time, have evolved into a veto card working in favour of the Swiss People’s Party (SVP). Since the 1990s, the constitutional amendments on which small and rural cantons have differed most strongly from the will of the voting population, dealt with questions of a conservative-liberal nature, or were related to the issue of political opening (European Union, United Nations, etc). Nowadays, on all these issues, the CVP is usually (and increasingly) aligned with both the parliament’s and the government’s , whereas the SVP stands in clear opposition. The SVP, however, increasingly finds support in the small, rural cantons, and while the CVP still predominates in electoral terms, the SVP has increased its influence in referendum votes in these cantons. This implies that the Ständemehr could turn into a veto card in the hands of the Swiss People’s Party (SVP).

Among the nine constitutional referendum instances, in which there has indeed been a difference between the majority of the people and the majority of the cantons, two occurred in 1994, and one occurred this last Sunday (3rd of March 2013). In 1994, the constitutional referendum addressed the promotion of arts; it was opposed only by the SVP and the small Liberal Party (representing 3% of the electorate). The constitutional amendment was blocked by the majority of the cantons.

This Sunday, the Ständemehr struck back again. The referendum was on the issue of a constitutional article on families (the controversy was mainly about child care). Also in this case it was the SVP (supported by parts of the Free Democratic Party), which politicised the issue along a conservative-liberal dimension. In the end, the SVP managed to defeat the article with the help of the Ständemehr. Given that the Swiss People’s Party (SVP) is nowadays quite strong, especially in the small cantons, and is often in opposition with parliament on conservative-liberal issues, we could actually expect an increase of referendums, where the Ständemehr is decisive, i.e. overruling the population’s majority. One thing, however, is particularly ironic about the failed referendum of Sunday: the veto card – originally thought to function as a the veto of the catholic minority (i.e. CVP) – was now used by the SVP to defeat a bill initiated by the CVP.

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.

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.

Who should elect the Swiss government? The Swiss People’s Party’s strategy as a boomerang

The Swiss People’s Party (SVP) wants Swiss voters to elect the Swiss government (Federal Council) in a direct vote. Today, they submitted a popular initiative, proposing an amendment to the constitution, to the authorities. They were required to collect the signatures of 100,000 citizens for this aim, and now, Swiss voters will be given the opportunity to decide on this proposal in a popular vote in the years to come – unless the party should withdraw its initiative in the meanwhile. As of now, it’s the legislative assembly that elects the government every four years, without any possibility of removing the government from office.

Usually, amendments to the electoral systems are proposed by political parties that expect to profit from such changes. It is not very surprising that the proposition to change the election modus for the Swiss government comes from the Swiss People’s Party. Although the Swiss government is usually composed of a large coalition, and mandates are almost proportionally allocated to the largest parties according to their share of parliamentary seats, in recent elections, the SVP has experienced difficulties to get represented in proportional terms, and/or the parliament refused to elect their official candidates (as they often did, in earlier periods, with Social Democratic candidates).

Surprisingly, however, our analysis shows that the SVP might be an important loser of the change in the electoral rules. In a newspaper article (available in German and in Italian), Karima Bousbah and myself discussed whether the SVP might profit from such a change in the election mode. Therefore, we have analysed the elections of the governments in 22 Swiss cantons, which are already elected directly, according to the same rules as they would apply at the national level. (While in the newspaper articles, we solely report correlation analyses, and some descriptive data, we also used panel data over the last 20 years, in order to understand why and when the SVP manages to win seats in governments, and when not.)

Similarly as at the national level, and despite the direct election by majority rules, cantonal governments are large coalitions, which roughly follow the rules of proportional seat allocation. Only the SVP is severely under-represented, holding 23% of the mandates in cantonal parliaments, but only 14% of the governmental seats. The SVP addresses right-wing, conservative voters, but the position of the party branches used to vary across cantons, especially before a party split in 2008. The most radical branches of the SVP often have difficulties in building alliances to other parties. In order to get their candidates elected under majority vote systems, the SVP relies on the support of the other right-wing parties, the Liberals (FDP) and the Christian Democrats (CVP). Where the SVP has a very pronounced right-wing position, it has difficulties to do so.

Also, young party branches fail to get their candidates into governments. Before the 1990s, the party only existed in a few cantons. Elsewhere, the party has difficulties in finding experienced and credible candidates for governmental positions.

Certainly, dynamics in national elections might be slightly different than in cantonal ones, where public attention is low, and candidates are often unknown to the voters. Nevertheless, we argue that the cantonal elections might be the best available indicator of possible governmental elections at the national level.

So why does the SVP start such an initiative? Most likely, because the other parties are afraid of a direct election of the government, and the initiative serves the SVP as an instrument to receive certain concessions from the other parties, for instance in December this year, when the Swiss Parliament will elect the Federal Council. Probably this won’t be the last indirect election, though.

More on suicide rates and support for the gun control initiative

Some readers pointed out (please, leave a comment here next time!) that the relationship between suicide rates and support for the gun control initiative might be spurious. Well, it looks like it is:

	Estimate	Std. Error	t value	Pr(>|t|)
Suicide rate (females)	0.52	0.64	0.81	0.42
Suicide rate (males)	-0.09	0.26	-0.32	0.75
German speaking	-10.53	2.80	-3.77	0.00
Urbanization	0.09	0.05	1.76	0.09
% Catholics	-0.15	0.07	-2.08	0.05
(Intercept)	47.74	11.52	4.15	0.00

However, controlling for urbanization and share of Catholics, there seems to be a correlation between suicide rates and percent of yes to the initiative — but only in non-German-speaking cantons:

Suicide rates and support for the gun control initiative

The gun control initiative was rejected yesterday by more than 56% of voters and in all but 6 cantons. One of the arguments advanced by advocates of the initiative was that stricter gun ownership rules would reduce suicide rates. So, was there more support for the initiative in cantons with higher suicide rates? Surprisingly, there is a strong correlation between approval and female, but not male, suicide rates. The relationship holds after controlling for language region:

	Estimate	Std. Error	t value	Pr(>|t|)
Suicide rate (females)	1.94	0.57	3.41	0.00
Suicide rate (males)	-0.20	0.29	-0.68	0.50
German speaking	-9.36	3.18	-2.94	0.01
(Intercept)	31.86	8.03	3.97	0.00

(OLS regression; dependent variable: % yes; N=26; R²=0.51)

Any ideas how we can explain the difference?

UPDATE: more analyses here.

Some unsurprising but nonetheless interesting observations on the fair tax initiative

After the referendum vote on November 28, most attention was concentrated on the highly controversial deportation initiative, which pushed the fair tax initiative out of the spotlight, all the more so that the latter was soundly rejected.

Here are a few figures showing that, unsurprisingly, the fair tax initiative received less public approval where tax rates are lower. The first graph compares cantons on the basis of the tax rate (canton plus capital city) for a taxable income of CHF 200,000 per year:

The relationship is fairly strong, but it varies interestingly if we look at tax rates for different income categories. Points are regression coefficients (that is, the slope of the regression line in the previous graph), while bars show 95% confidence intervals:

Finally, a similar relation is found for municipalities in the canton of Zurich:

Nothing unexpected here, and obviously one cannot easily infer individual behavior from aggregate data, but the figures give some idea of the extent to which cost-benefit considerations may have impacted voting decisions in this case. If you have any thoughts, please do leave them in the comments.

The Effectiveness of Arguments in Direct-democratic Decisionmaking: Evidence from the Swiss Deportation Initiative?

On Sunday the Swiss decided to automatically expel foreign nationals guilty of serious crimes including murder and drug dealing. The vote on the deportation initiative has triggered strong reactions. Unsatisfied with the outcome of the referendum and allegedly ever more xenophobic Swiss immigration policies, opponents of the deportation initiative even engaged in partly violent demonstrations in several towns in Switzerland, a rarity in a country known for its peaceableness.

Two arguments dominated the public debate prior to the referendum, each attached to one of the two political camps. The Swiss People’s Party (SVP), which instigated the deportation initiative, ran a costly and highly professionalized campaign, claiming that immigrants to Switzerland are disproportionately responsible for crime. The SVP argued that expelling foreign criminals would significantly reduce crime rates and thereby add to public security. It also played to stereotypes. The “Ivan S., rapist, soon a Swiss citizen?”-flyer invoked the well-known Swiss antipathy against immigrants from former Yugoslavia and presumably also alluded to Ivan IV of Russia known as Ivan the Terrible.

SVP Deportation Initiative Flyer. Source: http://www.kriminelle-nein.ch/downloads/index.html

 

Opponents like the Social Democratic Party (SP) said the deportation initiative was too harsh. They argued it would violate basic rights guaranteed in the Swiss constitution and even international law, since in some cases individuals would have to be deported to countries that practice torture or the death penalty.

Which of these arguments was more effective in changing people’s minds about the deportation initiative? This question is not only interesting for practitioners and scholarship on electoral behavior and immigration. Immigration constitutes a key component of globalization, a process in which countries remove legal restrictions on the flow of capital, goods, services, and, last but not least, labor. Immigration has therefore also become one of the rising themes in international political economy. The Swiss vote on the deportation initiative presents a valuable case to learn about the politics of immigration and the effectiveness of arguments in changing people’s minds about immigration policy.

From a political science perspective, examining the effectiveness of these two arguments poses a considerable challenge. Clearly, asking people about their preferences over the deportation initiative and having them indicate how convincing they find one argument or the other fails to provide any credible evidence. Individuals could hold an opinion because they find an argument convincing, but they could as well find an argument convincing because it supports their own opinion. We also have far too limited knowledge about the large number of other factors that potentially affect both citizens’ preferences over the deportation initiative and whether they find an argument convincing or not.

Jens Hainmueller, Dominik Hangartner, Marc Helbling and I have devised an empirical test that allows us to get an impression of each argument’s causal effects on citizens’ vote intentions. In the week prior to the referendum we conducted a large, representative telephone survey in which we embedded a randomized experiment.[1] We randomly allocated respondents into three groups. We exposed one group to the security argument and the second group to the non-discrimination argument. A third set of respondents received no argument and therefore served as a control group. To evaluate the effects of partisan cues we designed additional statements that individuals were exposed to randomly.

After providing respondents with an argument, we asked them to indicate their preferences over the deportation initiative and to answer more general questions about their attitudes toward immigrants. Any difference in respondents’ preferences between these different groups would be entirely attributable to the exposure to different arguments. Also, we are currently conducting a post-referendum survey in which we call the same individuals and ask them about their voting behavior and attitudes toward immigration to see whether the exposure to one of these arguments has any effects that last beyond the vote on the referendum.

We are currently conducting the last 300 interviews of the post-referendum survey and will post first results in this blog soon, so stay tuned.

---

[1] We thank the Wissenschaftszentrum Berlin and DemoSCOPE for financial support.

Xenophobic voters, or just strategic failure? The anti-immigrant vote in Switzerland as a Condorcet paradox

A guest post by Daniel Bochsler:

In yesterday’s referendum vote, Swiss voters have decided to expel foreign delinquents automatically. One year after the minaret ban, this new referendum further fuels the discussion whether direct democracy endangers minority rights. But the new referendum is also a rare observable occurrence of Condorcet’s paradox of majority cycles.

Majority cycles, which have been described by the French mathematician Marquis de Condorcet, are well known to students of political science and economics. However, they remain a phantom, as the literature is widely theoretically driven, and there empirical evidence remains scarce.

Let us define first what we speak about. Majority cycles can occur in any situation where voters have the choice between three (or more) options: Different groups of voters have their preference ordering about those propositions. Imagine that we have a first group of partisans of a reform, which however do not like an alternative counter-proposal, so that their preference order would be Reform > Status Quo SQ > counter proposal CP. A second group of voters is reform-friendly, but between both reforms they favour the counter proposal [CP > Reform > SQ]. A third group of voters does not like either of both reforms, but they would still prefer the counter-proposal to the reform proposal [SQ > CP > Reform]. Each group of voters has rationally plausible, transitive preferences. Jointly, the collective preference order results to be intransitive. Two out of three groups of voters favour Reform over SQ. Two groups, again, favour CP over Reform. Finally, two groups (although not the same), favour SQ over CP. Hence, (if none of the three groups counts an absolute majority of voters,) the collective preference order is intransitive, SQ > CP > Reform > SQ. Each of the three options is defeated by exactly one other option, as this table shows:

	Reform vs SQ	CP vs SQ	Reform vs CP
Voter 1	Reform	Status quo	Reform
Voter 2	Reform	Counter-proposal	Counter-proposal
Voter 3	Status quo	Status quo	Counter-proposal
Winner	Reform	Status quo	Counter-proposal

In yesterday’s referendum, the Swiss decided between two alternative reform proposals – the radical popular initiative by the Swiss People’s Party, a more moderate counter-proposal of the government, and the status quo. And they voted for the most radical option. While we lack information about the individual ballots, the result looks like a Condorcet cycle, as we see in this table:

	Yes	No	% for first option
Reform vs SQ	1,398,360	1,243,325	52.9
CP vs SQ	1,189,186	1,407,743	45.8
Reform vs CP	1,252,625	1,270,831	49.6

The occurrence of a majority cycle could be anticipated in the pre-referendum campaign, as a combination of the voting preferences of different groups of voters. Indeed, the referendum results shows that a majority of Swiss voters would have preferred the governmental counter-proposal (which respects international law and fundamental rights) to the radical initiative – but they voted instead for the initiative. This opens quite a few questions:

1) The counter-proposal contained many of the claims of the radical initiative. Having this in mind, why did a respectable number of those voters who voted in favour of the initiative reject the counter-proposal?

2) Public choice teaches us that in single-dimensional decisions, where we can rank all options on one axis, having two extreme poles and a solution between the two (more precisely, this implies that we should have single-peaked preferences), as in the present case, majority cycles should not occur. Apparently, it did, nevertheless.

The theory of majority cycles also teaches us how to vote strategically in multi-option decisions. Usually, committees or parliaments employ the amendment procedure for multi-option decisions. If there is more than one option for an amendment, then the proposed amendments are voted on in pairs. Acting strategically, a chair that anticipates a cycling preference order, can schedule his/her favourite option in the last round of voting, in order to make it pass.

In multi-option referendums in Switzerland, all pair-wise decisions appear on the same referendum ballot. In this situation, voters can strategically create artificial cyclic majorities in order to make their favourite proposal pass. In yesterday’s referendum, the Swiss People’s Party was probably inspired by this idea. The counter-proposal might have served as a compromise between the radical initiative and the status quo, and it was probably the most popular of the three referendum options. Also, it fulfilled a lot of the Swiss People’s Party’s requests. Nevertheless, the party encouraged its supporters to reject the counter-proposal. Lacking sufficient votes from anti-immigrant voters, the counter-proposal failed narrowly to win a majority of votes, and only the more radical initiative passed. According to the (reform-friendly) rules for multi-option referendums in Switzerland, the optional question (reform versus counter proposal) does not count in such cases, and the popular initiative is accepted.

Anticipating the outcome, the acceptance of the initiative might have easily been avoided. Indeed, in Condorcet cycles, one strategic vote can neutralise another. Radical anti-immigrant voters around the Swiss People’s Party probably rejected the counter-proposal for strategic reasons. Opponents of special laws for immigrants (equal-right-voters) might have applied a simple counter-strategy, and have hindered the radical initiative to pass: voting strategically in favour of the counter-proposal would have secured a majority of votes (and cantons) in favour of the less radical counter-proposal. This would have blackmailed the Swiss People’s Party campaign for the more radical option. Indeed, a group of Social Democratic MPs and several of their cantonal branches have given such a strategic voting recommendation. But the national party branch, along with the Green party, which campaigned for equal rights, decided against any strategic voting.

Switzerland appears like a dream world for the study of majority cycles. There is very little evidence of directly observable Condorcet paradoxes. In parliaments and committees, usually two consecutive simple majority decisions are made, so that MPs do not express their full preference orders, and majority cycles do not become visible. Swiss voting institutions, however, appear as quite unique in the world, as no other country, to our knowledge, applies multi-option referendums where voters can fully rank three options. After a change of the rules in 1987, Swiss voters have a short experience with multi-option referendums where they can fully rank-order all three proposals. Precisely six years before the referendum of yesterday, a majority cycle occurred in a multi-option referendum in the canton of Bern on 28 November 2004.

This new (supposed) Condorcet cycle might not only speak to Public Choice specialists. It might also speak to scholars who are concerned about minority rights in direct democracy. And the same Condorcet cycle also carries a message to equal-right-voters, who are now very concerned about the new slap of the Swiss voters in the face of the immigrant community: No, the referendum result is not only the expression of xenophobic values among many Swiss voters. It is also a consequence of the refusal of many equal-right-voters to vote strategically.