“Matt Shugart writes”:http://fruitsandvotes.com/?p=2418 about how he would have voted in yesterday’s election in Israel had he been eligible.
bq. That should be an easy question, with some 30 parties to choose from in a system that will give all those that win at least 2% of the national vote a share of the 120 Knesset seats quite close to their share of the votes. Yet it isn’t easy. Even with extreme party-based (closed list) PR and a parliamentary form of government, strategic and personal-leadership criteria enter in. … Of course, one starting point is to determine one’s position in the multi-dimensional ideological space. For this task, the “Israel Electoral Compass”:http://israel.kieskompas.nl/ was set up during the current campaign. Unlike the other versions of the Compass out there, this one plots your position in three dimensions–probably the very smallest number that can make sense of the tangled Israeli ideological and partisan scene. (And, yes, you can take the test in English as well as Hebrew–or Dutch.)
Matt touches on an interesting phenomenon that gets no attention in the US – how political scientists are starting to play God in multi-party systems. Websites designed to tell voters how they should vote given their policy preferences have begun to play a significant role in elections in Continental Europe which, like Israel, have lots of political parties, generating possible voter confusion. But do they provide good advice? In a “recent paper”:http://webh01.ua.ac.be/m2p/publications/1216814370.pdf, Stefaan Walgrave, Michiel Nuytemans and Koen Pepermans argue that they very often don’t. Where a country has competing voter advice websites, they often give dramatically different recommendations. Their advice is not only inconsistent across site, but across time. Choices over statement selection and statement wording can have very substantial consequences in skewing the advice given to voters. The authors of the paper perform a set of simulations based on a survey poll and find very large differences between the advice offered by the website and actual voting outcomes in the 2004 and 2007 elections – the former substantially under-rates some parties and over-rates others. They conclude:
bq. We established that statements matter, and that they matter a lot. The outputs of voting advices to vote for certain party differ extensively across configurations. Some configurations generate hardly any advices to vote for a certain party, other configurations boost a party’s advices. Some parties’ vote shares more than sextuple between the least and the most favourable statement configuration. The real electoral strength of parties does not seem to matter. Distances between the vote advice distributions and real electoral distributions are large. Our simulations show that the real electoral score of a party is mostly situated in the extreme ends of the advice distribution indicating that the mean output of a random configuration would be entirely different than what happens in the real world. The great majority of simulations did not approximate the electoral results at all. There is a central tendency in the statements: there is variation in advice distributions but the variation is clustered around a dominant mean. This suggests that an indiscriminate batch of statements has an inbuilt tendency to favour certain parties in contrast to others. Another batch of statements may have an entirely different built-in partisan bias. It does not seem to be a good idea for VAA-builders to neglect these effects and to act as if statements are just statements and that all selections would inevitably lead to the same or to a similar advice. This clearly is not the case.
bq. Our results show that statement selection is the crux of the VAA-building exercise and that it should be undertaken with the largest possible care; statements are too important to be selected light-heartedly. The carefulness with which political scientists design their scientific surveys stands in sharp contrast to the lightheartedness with which some of them engage in devising VAAs. As VAAs may have real world consequences one might expect rather the opposite to be the case. …
bq. Responsible VAA-builders—to a large extent political scientists—need to know exactly what the statements incorporated in their VAA entail in terms of advice output. If it is unclear and remains untested what voting advices a VAA produces, it is unjustified and even irresponsible to present a VAA as an instrument that helps voters to make their choice. Any other VAA produces entirely different results. The advice given is then just an arbitrary and subjective output that has no scientific grounding and that cannot claim reliability or validity. If VAAs are not benchmarked their output is not valid or, at the very least, it cannot be considered a valuable advice. One might as well ask Madame Soleil what party to vote for in stead of asking for a VAA advice. VAAs without proper benchmarking are closer to charlatanism than to political science.
It should be noted that these websites may have significant effects – at least in some countries (the Netherlands), large numbers of voters consult them (although of course we don’t know whether the information they get shapes their actual voting behaviour). One could also view this in a different light – perhaps political scientists should pump up the irresponsibility rather than toning it down. These websites potentially offer _awesome_ possibilities for conducting experiments on voting behaviour (and hey, if a few small Continental European countries end up having to figure out how to create coalition governments between the neo-Nazis, the dolphin huggers, and the Natural Law Party, thems the costs of conducting good social science, right?)