Now corrected from its original version (which contained some errors in the Czech data reported). Originally planted 18 July.
In discussing the proposed threshold of 5% of party preference votes to guarantee election from “flexible” lists in Catalonia, I expressed skepticism that this would result in many members being elected by their own votes.
Well, I just happen to have some data that might shed a little ray of light on this question, for those few of you who actually find this kind of stuff interesting. I opened up my merged data set of legislators from several open list systems, and restricting the analysis to districts that are about the size proposed for Catalonia (specifically, districts of 10 to 25 seats), I find that the mean preference vote share (preference votes divided by list votes in the district) of the elected members is .174. That is a lot more than .05, but the standard deviation is almost as big as the mean: .162. The median share is .115, and about 15% of elected candidates in these districts won with under .05.
So, perhaps I turned the old skepticism meter a bit too high when I expressed doubt that many candidates would be elected on their preference votes under a 5% threshold. Still, if the legislators in my data set had needed 5% to be elected, 15% of them would not have been–unless they had a high enough list rank provided by the party to be elected anyway. (Remember, in a flexible list system, seats not filled by preference votes default to the list rank, unlike in an open list, where there is no list rank and thus preference votes alone determine election.)
For larger parties, those that win at least the mean for this set of districts, which is 5 seats, the result is more favorable to my skepticism: Mean prefshare of .097, standard deviation of .10; median of .068; and around 28% elected with less than 5%. Not surprisingly, larger parties divide their votes among more candidates, successful or otherwise, while smaller parties are somewhat more likely to have a single vote-puller.*
However, all the above may overestimate the number of members who would be elected on preference votes if lists were flexible rather than fully open. My (more limited) information on existing flexible-list systems suggests that a very high share of total preference votes in such systems are given to the list head or one of the other top-ranked candidates (in other words to those who would win easily anyway). Of course, the more that voters do that, the smaller the pool of preference votes left to go around for other candidates with lower ranks, and thus the fewer candidates win on “their own” efforts and the more that are elected as if the list were closed.
To probe this further, we can look at data that I have from one flexible-list case in which many legislators are elected in districts of magnitudes ranging from 10 to 25: The Czech Republic (2002 and 2006). There, for winning legislators in these districts, the mean prefshare is only .017, the standard deviation is .022, and the median is .028. How many of those elected had preference shares under 5% of their list’s vote? 71.4%!
A quick check of the data appears to show only 17 of the 374 members elected on their preference votes who would not have been elected based on their party-provided rank. Thus the lists do not prove to be very “flexible” in practice.
The Czech rules in 2002 and 2006 allow the voter to cast up to two candidate preferences and require a 7% threshold of list votes for the candidate to be assured of election (i.e. even if the party has not given a high enough pre-election rank to the candidate).
Surprisingly (to me, anyway), Czech voters do not appear to concentrate their preference votes on a few top-ranked candidates. Even the winners who had the first rank on their list average less than 9% of their list’s votes. Nonetheless, the result is the same as what I expected–lots of candidates getting less than 5% of their list’s votes–even if the way that result came about is quite different from what I expected. (Looks like some research into flexible lists is in order!)
Is any of this relevant to Catalonia? Well, if Catalonian voters vote like Czech voters when it comes to using the preference vote, not very many candidates are going to clear the 5% preference-vote threshold.
Roman Chytilek, a political scientist at the Masarykova University of Brno, reports that for the next Czech elections the intraparty threshold will be reduced to 5% and a voter will be allowed up to four preference votes. (Before 2002, the rules provided for four preference votes and a 10% intraparty threshold.)
___
*In fact, each additional seat won by a party reduces the estimated prefshare of its winning candidates by .02, and the relationship is highly significant and the R-squared is more than 20%.
Data sample sizes: N=767 for the open-list sample; N=374 for the Czech Republic.
Another data note: In the Czech data, there is a much smaller relationship between the number elected off the list and the prefshare of those who are elected. It is “significant,” but barely more than zero anyway (-.003) and the R–squared is only around 7%.
I thank Joel Johnson for correcting my misreading of the data output and Roman Chytilek for his help in data acquisition as well as interpretation (including his comments in this thread since its original 18 July “plant” date.)
How are you considering the differing maximum number of preferences per voter? This paper suggests that there are only two preference votes allowed in the Czech Republic, less than the 2-5 (20% of 10-25 seats) in the Catalonian proposal. Obviously more allowed preference votes will cause higher ratios against the total party votes. Having a number proportional to region magnitude seems especially wise.
I only skimmed the paper so far, but it also says that a Czech candidate needs 7% of the party’s vote share in preferences to move up, which doesn’t jibe with your average of .038. Was there a rule change since 2002, the last year the paper considers?
I’ve probably mentioned before what I consider the greatest fault with flexible lists: they usually make it much easier to move a candidate up than down. To remove an undeserving candidate with a high list rank from the legislature would require a coordinated effort to give preferences to a number of lower ranked candidates; as well as a pretty good guess at how many seats the party will earn, to decide just how many lower candidates need to be supported. I’m not sure why designers of political systems make this so much harder than moving a low-ranked candidate into a seat-earning position—especially in proportional systems with many parties, where someone given a lower-than-deserved rank could join or create another party with relative ease.
I think you once mentioned a system where voters had “negative preference votes”, allowing them to cross out candidates they particularly disliked instead of circling those they liked. I’d love to see that become more widespread, what do you think?
Seed planted by Vasi — 19 July 2007 @ 22:21
Vasi, all excellent points. And the answer to your question:
How are you considering the differing maximum number of preferences per voter?
…is that I am not. (Might as well be honest.) Obviously one should!
Yes, I thought the threshold was higher, but the data sure imply it is not, at least in the two most recent elections.
I will see what I can find out, as the Czech case is one I will be using for upcoming research. The “research” above is based on fairly fresh data. It is, after all, a blog post and not a conference paper, let alone a journal submission.
Thanks much for the comment and the link to the paper.
(And the case with “negative preference votes” is Latvia. If you click “PR/party lists” at the top of the post, and scroll down a bit, you should be able to find the Latvian discussion (October), and within it, there is a link to a far more extensive post at Colomer’s blog.)
Seed planted by MSS — 20 July 2007 @ 16:54
MSS, in addition, unless it is required, not all who vote for a list will actually cast preference votes. This in effect makes the threshold even harder to reach, since the pool of preference votes/voters is smaller than the total list vote on which the official threshold is calculated.
A probable effect of having a higher threshold, both in a formal sense and in such an “effective” sense, is that fewer voters will cast preference votes. This then makes the effective hurdle even higher.
Perhaps the threshold should be calculated as a percentage of those who actually cast preference votes, not all list voters. This probably still prevents very small numbers of voters from deciding which candidates get the last seats, while not potentially discouraging preference voting itself.
An example: Sweden has since 1998 had a preference vote system with a high 8% threshold (at parliamentary elections, at lower levels the hurdle is 5%), and each voter is allowed to cast one vote. But only a minority will usually bother, and many of those who do will vote for high-ranking candidates, some of whom stand (and soak up preference votes) in multiple constituencies. I believe the number of candidates who would not otherwise be elected, but who are due to preference votes, hovers around the teens, in a Riksdag of 349 members.
In my own country of Norway (which often can be caught copying from its bigger brother to the East) this was sought corrected not by lowering the 8% hurdle, but by making the number of preference votes each voter could cast unlimited, except that only one could be cast per candidate, and all must be cast within the same list. In other words, approval voting, subject to a threshold. In the end this change was not approved by the Storting (parliament) for elections to itself, but will be used for the second time at county council elections this autumn. Officially, there are already flexible lists at Storting elections, although in practice they are entirely closed.
Vasi: On the crossing-out of candidates, have you seen the old Hagenbach-Bischoff system, versions of which are used at Swiss and Luxembourg elections, in many German local elections, and for about a hundred years until 1999, at Norwegian local elections as well? Basically, each voter has as many votes as there are seats, which he or she can place among all candidates standing in the district. In practice, parties and other groupings submit lists, one of which voters may adjust by adding a vote to some candidates, striking out others (this would in practice be a negative vote), and writing in names from other lists. Votes for candidates from each list are grouped together, deciding the proportional representation of each party. Within each party the candidates with the largest numbers of personal votes are elected – a majoritarian process. The latter is problematic, since controversial candidates with much support can still fail to get elected, if even more voters strike them off the party list. This candidate could very well lose out to a relatively unknown candidate with few additional positive votes, but also few negative votes (resulting in a nearly average number of personal votes). Although such cases were not common in Norway, partially because parties could list high-ranking candidates twice, it was among the arguments for the change to a local election system without negative votes.
This would be a valid argument against having negative votes in PR, I think, that a “seat’s worth” proportion of voters potentially could be denied representation by their favourite candidate because a different set of voters disliked this person.
Seed planted by Espen Bjerke — 01 August 2007 @ 23:50
Vasi, you are right about the 7% threshold+2 votes provision for the 2002 and 2006 Czech elections (for the next -probably 2010-election, threshold of 5% and 4 votes stipulation has been already agreed upon. This may indeed add some flexibility to the lists!).
I can more or less confirm the 71.4% figure but the number of deputies elected solely on preferences was much lower (18 of out 400) in the 2002 and 2006 elections.
When and if the Czech voters dare (and they dare not too much- only about 20% available preference votes is being utilized) to cast preference votes then it is usually in order to support the leader of the list. Exceptions to this rule that eventually led to the election of lower seeded candidates have been often framed to some story. In 2002, e.g., the voters of the christian-rural KDU-CSL cast -partially successfully- their preference votes in order to elect their lower ranked candidates instead of higher ranked candidates of the liberal-urban Freedom Union, with which KDU-CSL formed preelectoral alliance (“Koalice”). In 2006 the voters of Communist party (usually portrayed as believers of the democratic centralism) reversed the decision of the party politbyro and returned into parliament several incumbents that were moved down on the lists by the party selectors.
Seed planted by Roman Chytilek — 18 August 2007 @ 20:18
… ie, Roman, it’s not purely random and spontaneous, but organised by some kind of “counter-hierarchy” who’ve been shut out by the official party list-making hierarchy?
Generally, by the way, do any parties in List-PR polities use some form of PR (STV or lists for sub-factions) when drawing up their lists? I’m aware that Israeli parties usually use Limited Vote with some regional and gender quotas tacked on, and the German parties (or at least the Greens) fill each slot on the list, top first, by separate majority vote.
I once came up with a back-of-the-envelope system where the party convention (or regional general meeting) votes in an exhaustive (and probably exhausting) series of ballots where -
(a) on each ballot, you can vote for up to [NUM REMAINING SEATS] candidates
(b) your ballot has a value of 1,000,000 points which is divided equally among all the candidates you vote for (ie, the “Peoria” version of cumulative voting – if you tick 3 candidates, eg, each gets 333,333 points)
(c) on each ballot, the candidate with the fewest points is eliminated, and assigned the lowest unfilled “slot”
and
(d) the final list is composed of the last [NUM NAMES ON LIST] candidates, in reverse order of elimination.
This, ISTM, would produce STV style prop rep of factions within parties, without having to rely on either official or unofficial reservation of particular slots for particular factions (as with Bundestag party lists, or Australian parties’ mainland upper house [*] tickets).
[*] Federal Senate, and NSW, Vic, SA and WA State Legislative Councils.
Seed planted by Tom Round — 18 August 2007 @ 23:27
PS: Actually, some State branches of Australian parties use STV-PR to select their team of Senate candidates, but then rank them (which de facto pre-determines who’s elected) using the “first-past-the-quota” method, which to my mind is even worse than majority vote.
Seed planted by Tom Round — 19 August 2007 @ 00:03
Thanks to Roman and Tom for the “revival” of this planting. I have the corrected data (with the help of my RA, who in turn received the original data from Roman).
I just have not gotten around to correcting the post. I hope to do so soon. When I do, I will “replant” it with the then-current date.I have now done it! And I am making use of the new edit-comment feature at the same time!
Seed planted by MSS — 19 August 2007 @ 15:46
Tom: to answer the spontaneity question- the communists case was undoubtely spontaneous- their voters are in some respects more conservative than the party elites and probably disliked the idea of young guns being elected.
The KDU-CSL vs. Freedom Union affair is a bit more complicated- it was speculated that the KDU-CSL voters were mobilized via their church attendance. Anyhow, the voters of the KDU-CSL were always the far most active in the use of preference votes so the result was no surprise (but discouraged any further preelectoral alliances with that party).
Seed planted by Roman Chytilek — 20 August 2007 @ 04:52
Espen, thanks for the response! I seem to recall a different sort of thing being called Hagenbach-Bischoff, but it wouldn’t be the first time the same name was applied to two concepts.
Anyhow, I get the impression that the ‘personal votes’ you describe are the difference between positive and negative votes, is that right? While that does accomplish my goal of making it easy to punish corrupt or otherwise undesirable candidates, you’re quite right that it could have strange side effects. I’m not really sure what I would like to happen if a candidate got 10% positive votes and 40% negative. Giving him a seat despite overwhelming dislike (among his own party!) seems ridiculous. But denying that 10% of his party’s voters is also not entirely fair. Hmm…I suspect this is something that I’ll have to ruminate on.
PS: Happy bloggiversary(?) Matthew!
Seed planted by Vasi — 30 August 2007 @ 08:56
Vasi,
I think the term “Hagenbach-Bischoff” is most commonly used for the method for doing a quicker d’Hondt seat calculation, developed by the Swiss professor behind the name. His more elaborate method of distributing personal and simultaneously list votes, which I mentioned, is often left out. The latter ought perhaps to have a different name, but if it already has, I haven’t heard of it.
You are right, the amount of personal votes for each candidate results from the difference between positive and negative corrections to the original lists. But unlike in a true negative vote scheme, before corrections, each candidate will start off with a number of votes equal to how many voters used his or her particular list. This is because casting an uncorrected list would yield each candidate on it one vote (in Norway it was possible for a party to rank a certain proportion of its candidates twice, meaning these would start off with two votes per used list). If a voter’s corrections would have lead to more votes being cast than the number of seats to be filled, those ranked the lowest would automatically lose one vote (voters could also alter the proposed ranking).
But the analogy with negative votes is really only valid if voters are presented with physically separate party lists which they then can proceed to correct. Having a starting point for candidates of one vote per ballot instead of zero makes no difference, though officially, no negative votes are then cast. This was the version which operated in Norway. But if instead there is just one common ballot for all parties with no preset rankings, as is the case in other places, there is no analogy to having negative votes.
There are also other variations among Hagenbach-Bischoff systems, such as how many extra votes can be cast per candidate, what possibilities there are for parties to favour certain candidates or to enter into apparentements, et cetera.
Happy ruminating!
Seed planted by Espen Bjerke — 30 August 2007 @ 14:55
According to an old publication I found (click my signature), it is actually called the “free list” system. I may have remembered something incorrectly, remembered correctly something incorrect, or actually been correct as well, though the latter seems unlikely.
Seed planted by Espen Bjerke — 23 March 2008 @ 14:51
So, “free list” is another name for panachage, right?
Thanks for that link, which is to a book from 1901 that I had not been aware of before:
Seed planted by MSS — 24 March 2008 @ 13:21
I suppose a system with many variants can cope with having a few more names!
Perhaps panachage is a slightly more general term though. A “free list” system will certainly always have a panachage feature. But I am less sure that every panachage system – even limited to proportional ones – will necessarily be a “free list” system, which is also akin to the cumulative vote (but with votes counted party by party). That is the basic setup; various restrictions are often imposed (how many extra votes to one candidate, how many votes to other lists, etc). It typically also has one vote per seat.
Seed planted by Espen Bjerke — 25 March 2008 @ 19:39