This page details the process by which the Pirate Party in the German state of Lower Saxony selected its candidate list (“nominees”) for the January 2013 state election. The process concluded with an election on August 25, 2012, which used Score Voting.

Above, second place finisher Catherine Nocuń and first place finisher Meinhart Ramaswamy, from the German newspaper Der Spiegel.

Results

The results were as follows (totals and averages):

André’s account

The following is an account of the process, by a party member named André. Highlighting ours.

On 21st/22nd of April 2012 in Nienburg (NI) we made the first attempt to create a party list.

We used a simple approval system with yes and no, and without the option of abstention (i.e. abstention=NO).

It took 6 rounds to get 42 people on the list. The results showed us, that a high number of voters were voting strategically/tactically and not honest¹. By law, a candidate must have more yes than no votes which results in an automatic 50% quota (without abstention), and the 50% yes were not easily achieved.

After the event 2 people tried to get the results annulled, their reasons being of a personal nature (because they weren’t elected), but their arguments created a legal problem. To determine whether their arguments were valid it would take a long time and by that time the deadline to enter the list for the election could have passed. So we might have ended up not having a list thus the whole election happening without us. ”Fortunately” it became apparent that 2 non-Germans had voted during the assembly, which is illegal, so the party executive committee “happily” annulled the party list and invited for a new assembly to elect a new list.

In preparation of this new assembly a lot of new ideas for a new election system sprung up, STV and IRV among them. People were really disappointed with Approval Voting, as they felt that they had no real influence on the order of the candidates². Only by voting strategically they could have an effect on the order of candidates which lead to a lot of rounds of ballots³.

It became apparent that most people had problems understanding voting systems and most of them had especially problems imagining how the counting of the votes would work. This is important, as computer/machine voting is strictly forbidden by law, so everything has to be counted manually!

One of the guys who tried to annul the previous list also entered a suggestion for a voring system which basically was approval voting with a 25% quota. A lot of people saw this as an attempt to make it possible for that guy to get the list annulled again, as a 25% quota may not be legal. (The guy is a former lawyer)

That was the moment we decided to create a system which was relatively easy to understand, fast enough to count and which was legally 100% safe.

We chose approval voting to decide who would get on the list without using the result to determine the order of the candidates. Also we added the option of abstention, because this would get us more candidates after the first round. As mentioned, by law we need more yes than no votes and the abstention would not be counted. We told the assembly to abstain whenever they did not know a candidate. Most of them understood and they did not vote strategically. We ended up with 33 people having more yes than no votes (needing 30 candidates, so we had 3 more in case anybody dropped out of the “race”).

This second assembly took place in Wolfenbüttel (WF) on the 21st/22nd of July.

Unfortunately the person responsible for the counting of the votes did not take the counting system that I had developed and he ended up using probably the most time consuming method imaginable to count approval voting ballots. He also made a lot of other mistakes, 2 initial counting rounds had to be annulled so we ended with a list of 30 candidates after 2 days and we had to adjourn the assembly.

So the assembly was continued on the 25th of August with three rounds of Score voting. We had 36 people for counting the ballots. This time we were prepared, had formed a counting team and had taken over the counting process completely.

1st round: places/positions 1 to 5 on the list, 20 candidates, score from 0 to 5
Counting would work this way: manual transfer of the scores from 10 ballots on to 1 counting sheet, this would happen 2 times, numbers were added per candidate, results were compared. If the results corresponded, 5 of those counting sheets would be transferred to summation sheets, again results were compared. In the last round the results per candidate were transferred on an individual candidate result sheet. 1st round took us over 2 hours, which was about 20% longer than expected.

2nd round: places/positions 6 to 15 on the list, 25 candidates, score from 0 to 9
we adjusted the counting system, most importantly only 5 ballots were transferred on 1 counting sheet. This way more counting sheets were created to be later added up further, but we could speed up the process significantly. We took exactly 2 hours though we had 5 more candidates.

3rd round: places/positions 16 to 30 on the list, 15 candidates, score from 0 to 9
exactly like the 2nd round, we had fewer ballots and fewer counters, but it went well, taking 1 hour 45 minutes.

Overall the counting proved that we had chosen the right system. We were only slightly slower than we had expected and most people were happy with the result. They felt that with the scoring they could express their will in a very effective way.

Conclusions that candidates need to draw: The order from the “preliminary” approval voting gives a candidate a very good indication where they will end up on the list. For the positions 1 to 5 everybody placed higher then 10 on the approval voting list shouldn’t have tried.

Overall nobody could improve their approval voting position more than 5 places (one guys improved 7 but only because 2 others had dropped out).

Questions you probably have:

Why these list ranges/blocks of 1-5, 6-15, 16-30?

We are aware that we could have used one list 1 to 30 or two times 15. With score this is not a problem, but score voting was completely new to the voters and we wanted not to ask too much from them. A concept like “The TOP 5” is easy to grasp, so we decided to use that. Then we decided on the 0 to 5 points range. If people wanted to clearly place their 5 top candidates in an order they could (no.1=5pts, no.2=4pts, etc). So the 1st round would also help them understand score voting. The 6 to 15 was somewhat logical: first 5 then 10 more. Also 15 seats is like the realistic maximum we can get, so these are the most important people to elect. As you can see in the results, after the 2nd round 34 people left, probably because they understood that 16 to 30 only have a marginal chance of winning a seat.

Why 0 to 5 and 0 to 9?

I partially answered that one. We decided on these ranges because most people seem to be able to work with them. Some test have shown that most people cannot put more than 7 candidates in a specific order, with score and 0 to 5/0 to 9 they have enough variation to judge 15 or more people (“these 2 are my favorites, they both get 9; these 3 a still good, they get 7, etc.”

Why not IRV or STV?

We have not found a counting method that would be fast enough for manual counting.

We have absolutely no experiences with either method when it comes to larger lists, we were not willing to risk not finishing the list. (Swiss pirates use STV for shorter lists, but they are allowed to use computers. IRV will be tested by the Hamburg Pirates, they want to try different counting systems. Most people doubt that it can be counted manually.)

A rated system is a lot easier to understand and apply than a ranked system. Rating 20 or 25 people is really hard but ranking the same amount is incredibly time consuming and absurdly difficult. Most people cannot rank more than 7 people, after that the order becomes more and more arbitrary.

Primary Data

Delmenhorst Results
First round ballot
Second round ballot
Third round ballot

Endnotes

1. Note this is not a general purpose indictment of Approval Voting with regard to tactical vulnerability. The specific problem here is that Approval Voting had been used to determine the order of candidates. Knowing that more rounds of voting would be held until a sufficiently long list of nominees had been selected, many voters approved a small number of candidates in order to achieve more control over list order.
2. It turns out that this is indeed more of a feeling than a reality. The original (unordered) slate of candidates, chosen with Approval Voting, turned out to be remarkably similar in order to the final list ordered by Score Voting. This is because, while the voters may feel that their individual Approval Voting ballots lack granularity (every candidate is either approved or not, with no ability to express degrees or orders of preference), there is a “statistical averaging” effect when all their ballots are aggregated. That said, we acknowledge the practical value of using a system which leaves voters feeling as though they’ve expressed themselves and been heard.
3. Actually, the tactic with Score Voting is the same as with Approval Voting (except in some very special circumstances), and so voterstechnically have just as much incentive to assign scores sparingly, so as to exert maximum influence on the ordering, leading to multiple rounds of voting. However it seems that the voter psychology comes into play here. This is related to a general finding by Warren D. Smith, that Score Voting users tend to vote more honestly than users of other types of voting methods which allow less expressiveness.