Critics of open-list PR (and I have sometimes been among them) often lament the near randomness in determining who is in and who is out among the candidates competing for any given party’s seats. This was mentioned in the comments to one of my “preserved fruit” posts, specifically the one in which I proposed a variant of an open list for an MMP system.

Alan put it well, by describing the problem thus:

the danger of the main party leaders getting a huge share of the votes and the priority of candidates further down the list becoming almost random because of the giant votes locked up by the leadership.

In this light, the experience of Colombia’s recent election is interesting. Parties could present lists that were either closed or open, and most (by far) opted for the open list. And when they did so, it was a fully open list; that is, there was no way for parties that permitted preference votes to ensure that certain of their candidates were elected ahead of others: preference votes alone ordered the list.

This was the first experience Colombian voters had with preference voting (aside from the 2003 local and departmental assembly elections).

So, did we see a pattern of one or two recognizable leaders soaking up most of the preference votes and the order of the rest of the list being essentially random? Not really, at least in the House. There is some evidence of that effect in the Senate.

First, the House. For lists that won two seats (regardless of district magnitude, which averages around six, and ranges from two to eighteen), the average individual preference-vote percentages (candidate vote divided by party vote in the district) were:

32.5, 21.3, 13.3, 8.2…

So the top candidate had a percentage that was about 1.5 times that of the second candidate, who was also elected. The ratio of the last elected to first loser was about the same (1.6), and the ratio of the first to second loser was also 1.6. In other words, the candidates fell off pretty clearly, suggesting that there might have been reasonably good knowledge about who was in the running and who was not. As a rule of thumb, I might expect a ratio of first to second loser to be 2.0 or greater, indicating really clear knowledge of who was hopeless, but this result seems pretty good for a first time.

Now, for lists that elected three candidates:

23.4, 17.6, 14.2, 9.9, 5.7

The top three–the elected candidates–are pretty close in votes, with ratios from one to two of 1.3 and from two to three of 1.2. The first loser is also close behind the last winner, with a ratio of 1.4. But the second loser is farther back, with a ratio of the first two losers of 1.7. This again suggests reasonably good information. There is no “giant” on the average list who soaks up most of the preference votes, and there is something of a race for the final seat, with just over four percentage points separating the last winner and first loser, on average. But the second loser averages more than eight percentage points short of victory. It was thus not the case that any of the top five candidates could have been elected, with “randomness” sorting out the winners (after the “giant”) from the losers.

For the Senate, on the other hand, the picture is quite different. There is not a great deal of difference here for any lists that won more than ten seats (there were five such parties). I am going to list the top three winners’ averages, then the last three, followed by the top two losers. The averages were about as follows, with the ratio of the previous candidate’s percentage to the next shown on the second line:

10, 5.8, 4.8,… 2.3, 2.2, 2.1, 2.0, 1.9

… 1.72, 1.21, … …1.05, 1.05, 1.05, 1.05

Now, the difference between the last three winners and the first two losers looks random!

Of course, it is to be expected that voters would have a harder time discerning who is in the running and who is not the larger the magnitude is. And the Senate has a tremendously high magnitude: 100, one of the largest districts within which open lists have ever been used. But for the lower-magnitude House, the candidates sorted surprisingly well for a first-ever congressional election with open lists.

Another interesting aspect concerns the percentage of voters who declined to give a preference vote at all. Voters were free to cast just a list vote, or a list vote and then a vote for a candidate within that list (in an earlier post on Colombia, I provided links to some ballot images). Only preference votes affect the identity of the legislators a party will actually elect, so voters who cast only a list vote are effectively delegating the selection of specific legislators to other voters (not to the party organization, as with so-called flexible lists, like those in Belgium and many other European countries).

The percentage of voters who cast list-only votes ranged generally from around 15% to 25%, which is rather high compared to other similar systems (e.g. Brazil). And this percentage was higher for some parties and in some districts. Specifically, in some multiple regression analysis I performed on the data, voters for the largest pro-Uribe party, known as La ‘U’, were more likely than voters for other parties to cast list-only votes. This may mean that these voters are more interested in the party program (to support Uribe’s policies) than they are in specific candidates (and the pork or other ‘goodies’ that they can offer as personal constituent servants). On the other hand, voters for the Conservative party (a traditional party that is also aligned with Uribe) were much more likely to cast preference votes, which may indicate that they are much more after the pork and personal connections than are the voters for La ‘U’.

Additionally, district magnitude (within the House) was a significant factor in predicting the percentage of voters who cast list-only votes (even when controlling for the party). This is not surprising, in that the higher the magnitude, the longer the list of names to sort through to pick a specific candidate.

There also is a pretty clear tendency for list-only votes percentages to be higher in the Senate than in the House, a we’d expect, given the huge difference in district magnitude. However, I did not check to see if this intercameral difference was statistically significant, because I have not pooled the House and Senate data.