Changing the Dutch electoral system

2 October 2020

The Dutch government wants to change the way in which representatives, including Members of Parliament, are elected. Currently, the position of candidates on the party’s list of candidates plays a key role. If a party wins seven seats, then the first seven persons on the list of candidates will usually be elected. The new system, for which a consultation will be launched this autumn, will make it easier for candidates who get many votes to bypass the list order decided by their party (see electoral system).

To get an idea of the possible impact of such a system, I’ll explore what happens if candidates are ranked by the number of votes they get. I’ll use data from the 2017 Lower House election. Caveats apply, as explained in the Method section.

I’ll use slope charts to illustrate how the position of candidates would change. For example, if someone had position number 3 on the list of candidates, but got the second-highest number of votes among their party, then this person will be represented by a line from position 3 on the left to position 2 on the right. The colour of the lines shows whether they move up, down, or are flat (i.e., the vote rank is equal to the position on the list).

Gender

The chart below groups results by gender.

There are almost twice as many male as female candidates. If candidates are ranked by the number of votes they get, female candidates tend to move up. This is partly because some voters vote for a female candidate as a matter of principle. In addition, a successful campaign was launched in 2017 to encourage tactical voting for female candidates (Stem op een vrouw).

The table below shows the potential impact on who gets elected.

in out
female 25 4
male 12 33

In a hypothetical situation where seats would be assigned purely based on vote rank, some candidates would lose their seat, whereas others would now be elected. As a result, 21 more female candidates would have been elected. This would mean that exactly half the elected members of the Lower House would have been female.

An example is Vivianne Heijnen of the CDA, which won 19 seats. She was number 26 on her party’s list of candidates, but with 15,821 votes she would have been number 9 if ranked by number of votes. She was not elected, and has since become alderwoman in Maastricht.

The findings discussed above suggest that female candidates might benefit if more weight is given to the number of votes candidates get (emphasis on might because of caveats discussed below).

Incumbents

The chart below groups the results by whether the candidate is an incumbent (i.e., someone who already was a member of the Lower House or the government before the election).

Previous research found that incumbents get more votes than new candidates. Incumbents tend to have a higher position on the list of candidates, which may reflect how much power they have within their party. Of course, when you start with a high position on the list, it’s difficult to move further up. On average, incumbents have a vote rank that is lower than their position on the list, but the situation is not as clear-cut as with gender.

Regional following

The chart below groups the results by whether candidates strongly rely on votes from one district.

Some candidates get their votes from across the country; others get a high share of their votes from one district. Candidates who got at least one-third of their votes from one district are shown to the left; others to the right. On average, those with a strong regional following would move up if they were ranked on the number of votes they get.

The table below shows the potential impact on who gets elected.

in out
at least one-third 22 12
under one-third 15 25

In a hypothetical situation where seats would be assigned purely based on vote rank, 10 more candidates with a strong regional following would have been elected. An example is Christa Oosterbaan from Terschelling. Her party, the PvdA, won 9 seats. She was number 20 on the list of candidates, but with 5,512 votes she had a vote rank of 7. 79% of her votes were from the Leeuwarden voting district. Oosterbaan is still active as party leader on the Terschelling city council.

This suggests that candidates with a strong regional following might benefit if more weight is given to the number of votes candidates get.

Note that this is an imperfect measure because the size of districts varies considerably. Further, I’ve excluded candidates who didn’t run in all the districts. This explains why lower positions are missing from the slope graphs above.

Randstad

The chart below groups results by whether candidates live in the Randstad region.

On average, candidates who live outside the Randstad region of the Netherlands (here pragmatically operationalised as the provinces Noord-Holland, Zuid-Holland and Utrecht) would move up if they were ranked on the number of votes they get.

The table below shows the potential impact on who gets elected.

in out
Randstad 15 31
other 21 5

In a hypothetical situation where seats would be assigned purely based on vote rank, 16 more candidates from outside the Randstad would have been elected. An example is Kristie Lamers from Nijmegen. Her party, D66, won 19 seats. She was number 29 on the list of candidates, but with 3,425 votes her vote rank was 19.

This suggests that candidates who live outside the Randstad might benefit if more weight is given to the number of votes candidates get.

What happened to those who would have been elected

37 persons were not directly elected, but would have been if seats had been assigned based on the number of votes they got. Of these 37 persons, twelve have since joined the Lower House to replace members who became members of the government or who stepped down over scandals. Three were probably lijstduwers (literally: list pushers): people who have been put on the list to attract votes but who have no intention to take up a seat.

Background: the electoral system

In the current electoral system in the Netherlands, voters vote for a person on the list of candidates of a party. Many voters simply pick the first candidate of the party they support, but others will vote for the highest woman on the list or for some other candidate. Votes for other candidates than the first on the list are referred to as ‘preference votes’.

Seats are assigned to parties based on the total number of votes for all candidates of that party. Normally, the position on the list of candidates will determine which candidates of that party are elected. Lower-ranking candidates can only bypass the order of the list of candidates if they pass a votes threshold (for the Lower House, this threshold is one-fourth of the number of votes per seat). Usually, only one or two Lower House candidates are elected as a result of passing this threshold. In 2017, the number of candidates elected through preference votes was relatively high (4), likely as a result of a campaign to promote tactical voting for female candidates who might not be elected otherwise (Stem op een vrouw).

In a study published in 2012, Joop van Holsteyn and Rudy Andeweg found that preference voting is increasing. The extent to which this happens seems to depend on the person who leads the list of candidates: a popular number 1 will attract more votes, leaving fewer preference votes. Preference votes are cast more often by higher educated, politically interested and efficacious female voters. Preference votes are often cast for persons who have a high position on the list and would be elected anyway.

Van Holsteyn and Andeweg also analysed how characteristics of candidates correlate with the number of votes they get. The best predictor was found to be incumbency: incumbents get more votes.

The government has announced that it will introduce a proposal for a new electoral system this autumn. In the new system, based on the so-called Burgerforum Kiesstelsel proposal, voters can either vote for a party, or for a specific candidate on the list of that party. Suppose a party wins 6 seats; two based on votes for the party as such and four based on votes for individual candidates. In that case, the first two seats will be assigned based on list position (i.e., numbers 1 and 2 on the list of candidates); the remaining four seats will be assigned to the remaining four candidates with the highest number of votes.

The Burgerforum Kiesstelsel proposal was recommended by the Staatscommissie parlementair stelsel in 2018, with the intention to ‘facilitate personal and regional preferences of voters more than is the case now’. They argue this might have a positive impact on the ‘30% of voters who think they are badly represented’. The Burgerforum system was originally proposed in 2006 and rejected by the government in 2008.

Method

I used data from the Kiesraad (Electoral Council) to analyse votes in the 2017 Lower House election. For most candidates, gender and residence are available. I used this Python package to parse the election data and this package to guess the municipality from the residence provided; this way I could also link candidates to a region and a voting district. For pragmatic reasons, I limited the analysis to parties that won seats.

A number of parties have different candidates in different districts for the lowest positions on the list. It is even possible that a candidate has different positions on the list in different districts. In the latter case, I assigned them the mean of the positions they had. Per party, I ranked the positions on the list and assigned the mean value if multiple candidates held the same position. For example, the PvdA has 19 different candidates who each have position 76 in their district. They were all assigned a list position of 85. This explains why, in the slope charts above, there are a number of positions to the lower end of the candidate list where various lines start.

For a number of reasons, vote ranks calculated with this dataset cannot simply be considered a prediction of what would happen under a system where vote rank would determine who gets elected. If you change the system, parties might adopt different strategies for composing their lists of candidates, and voters might adopt different voting strategies.

In a more general sense, note that I’m not making any claims about causality. Incidentally, if you’d try to construct a predictive model with the variables discussed, a considerable share of variation would remain unexplained.

2 October 2020 | Categories: data, election