Bon MMP, Bad MMP

The National Assembly of Wales, a bad example of MMP (image source)

Many Canadians are calling for our provincial and federal elections to be conducted using the Mixed Member Proportional (MMP) voting system.¹ If Canada does make this transition, it is extremely important that we adopt a good MMP model and not a bad one. What’s the difference? I claim that a simple rule distinguishes between good and bad MMP voting systems:

A good MMP model is one where a vote for a district candidate counts as a vote for their party.
A bad MMP model is one where a vote for a district candidate does not count as a vote for their party.

A bad MMP model will perpetuate if not exacerbate tactical voting. It will fail to make parties more accountable to the electorate in every region. It might create the perception that there are two classes of representatives, one superior to the other. It might leave many voters with less confidence in Canadian democracy.

A good MMP model, on the other hand, will simply improve Canadian democracy without introducing any serious adverse effects.

Why MMP?

MMP continues to gain traction as an alternative to First Past the Post (FPTP) for Canadian elections. In October 2017, Jagmeet Singh included a transition to MMP as a policy item in his successful bid for the leadership of the federal New Democratic Party (NDP).²

a Jagmeet Singh-led government will – as a first priority – establish a Mixed- Member Proportional system with regional lists. After two electoral cycles under this system, Canadians will be asked to ratify it in a referendum.

In December 2016, the Green Party of Canada passed a policy resolution supporting MMP.³

About 350 attending members passed an emergency policy resolution […] that supports a Mixed-Member Proportional (MMP) electoral system, while remaining open to other forms of proportional representation. Prior to this resolution, the GPC had favoured PR without specifying any particular PR system.

In a plebiscite that wrapped up in November of 2016, voters in Prince Edward Island were offered a choice of 5 voting systems to use for future provincial elections. The plebiscite made use of a ranked ballot, meaning that voters were able to express their 1st preference, 2nd preference, 3rd preference, etc. After 3 of the 5 options were eliminated, MMP accumulated a majority of the votes and was declared the winner ahead of FPTP.⁴

During the 2016 PEI plebiscite, I was actually hoping Islanders would select the other form of proportional representation on offer, Dual Member Proportional (DMP). Whereas MMP elects one MLA per district plus several regional MLAs, DMP elects two MLAs per district and no regional MLAs. One reason I preferred DMP is that it appealed to a few MLAs on the right of the political spectrum as well as the left. You’ll see this in the following quote, where Progressive Conservative MLA Brad Trivers speaks in the PEI legislature prior to the plebiscite.⁵

I personally don’t like the mixed member
proportional system because it creates two
different tiers of MLAs with two different
sets of responsibilities that don’t at all match
up with what we’ve seen over the last 150
years, and I don’t know how that would play
out.

I really like the dual member proportional
system because I think it will provide
representation across the province and it will
do that where every member is accountable
to specific constituents within a specific
district.

But as much as I like DMP, the choice Islanders made is perfectly reasonable. MMP is a fine electoral system for PEI and for Canada…

…as long as it satisfies one simple rule.

Examples of MMP

As I claimed at the outset, a good MMP model is one where a vote for a district candidate counts as a vote for their party, and a bad MMP model is one where a district vote does not count as a vote for a party. Let’s look at some examples.

Example #1: Louis Massicotte’s single-vote MMP for Quebec (Good!). Professor Massicotte recommended a model in which Quebec would be divided into several small regions with a fixed number of seats. About 60% of MLAs would be elected by FPTP to represent single-member districts (also known as “constituencies” or “ridings”). The remaining 40% of MLAs would be elected from closed party lists to represent a region encompassing several districts. Importantly, voters would only get one vote. A vote for a party-affiliated district candidate is interpreted as a vote for the party; it helps increase the party’s overall share of the seats in the legislature, including both district seats and regional seats.⁶

Massicotte’s proposal shows that MMP doesn’t require two votes. In fact, when the first MMP system was implemented in Germany after World War II, there was only a single vote. The 2nd vote was added later, for partisan reasons I highly suspect.⁷ Lesotho first adopted two-vote MMP, but after the system was gamed in 2007 they dropped the 2nd vote and counted the district vote as a vote for the party. Now Lesotho’s system is exceptionally proportional and difficult to game. German professors Joachim Behnke and Friedrich Pukelsheim recommended that Canada adopt a single-vote MMP model with exactly the ballot that is currently used under FPTP.⁸

Example #2: The Welsh Assembly’s two-vote MMP (Bad!). Wales uses small regions, each with roughly 8 district members elected by FPTP and 4 regional members elected from closed lists. Voters have both a district vote and a regional vote, and only the regional vote counts as a vote for the party.⁹

The Welsh implementation of MMP is so deceptively simple, I worry it will be copied in Canada. The problem is that it can produce highly irregular outcomes, a case in point being the 2003 election where Labour won 75% of the district seats and none of the regional seats.¹⁰ The resulting seat distribution was highly disproportional, but that’s not the main concern. The real danger with this sort of model is the fact that many voters have a strong incentive to vote tactically. We’ll revisit this later.

Example #3: Bavaria’s two-vote, open-list MMP (Good!). This is the only implementation of MMP that allows voters to choose not only the district representative, but also the regional representative. Both of an elector’s votes — the district vote and the regional vote — count as a vote for a specific candidate and as a vote for the candidate’s party.¹¹

Although many MMP implementations feature two votes, I consider the Bavarian model the only one for which the 2nd vote is truly justified. The model succeeds because each party’s district votes and regional votes are added together to determine how many seats the party should receive. In other two-vote models, only the regional votes are intended to determine the overall seat distribution.

Example #4: The Law Commission’s two-vote, flexible-list MMP (Bad!). The Law Commission of Canada recommended a model very similar to that used in Wales, but with flexible lists instead of closed lists. With a flexible list, a voter may choose to support a specific regional candidate; however, regional candidates must surpass a certain percentage of the votes in order to rise above the list position assigned to them by their party.¹²

The Law Commission’s report was a fantastic, ground-breaking report. And the question of whether to use closed, open, or flexible lists is an interesting discussion topic. But the bottom line is that the Law Commission’s model shares the same critical flaw as the Welsh model: that district votes do not count for the party. It would probably produce the same tactical voting issues in Canada as seen in Wales.

Example #5: Baden-Württemberg’s single-vote, list-free MMP (Good!). The German state of Baden-Württemberg uses a unique form of MMP with no party lists. Regional seats are awarded to the candidates who receive the most district votes, but who do not succeed in winning their districts.¹³

The list-free method has a number of advantages and disadvantages compared with closed, open, and flexible lists. But the relevant point is that the Baden-Württemberg model has only one vote, a vote that counts for a candidate and their party. It does not introduce any serious issues as far as tactical voting is concerned.

Ticket splitting under bad MMP models

There is a particular type of tactical voting one must worry about with certain two-vote MMP models: ticket splitting. Simply put, the idea is for voters to increase their influence by “splitting” their support between two parties: giving the district vote to a large party and the regional vote to a small party. Ticket splitting is not necessarily a bad thing. Suppose a voter genuinely wants a large party’s candidate to be their district representative, but prefers a smaller party’s platform. In that case, voting for both parties is completely okay. The problem is that voters may face a strong tactical incentive to vote in this manner, even if it does not reflect their true preferences.

Let’s see how ticket splitting tactics come about. We’ll use data from the 2015 federal election, and focus specifically on the province of New Brunswick.

Imagine that Canada has adopted a bad two-vote MMP model where district votes do not count as votes for a party. New Brunswick’s 10 FPTP districts have been replaced with 6 MMP districts, and the remaining 4 MPs are now elected from party lists (either open lists or closed lists; it does not matter for this example). In order to use votes cast in the 2015 federal election, we’ll merge 4 pairs of existing FPTP districts.¹⁴ The resulting 6 MMP districts are listed in the table below along with the number of Liberal, Conservative, and NDP supporters in each district.¹⁵ At the bottom of the table, we see that the Liberal Party wins all 6 district seats with 54% of the popular vote, the Conservatives win 2 regional seats with 27%, and the NDP wins 2 regional seats with 19%.¹⁶

Party Supporters

MMP District	Liberal	Conservative	NDP
Acadie	46623	10003	29749
Beausejour	36534	6017	8009
Fredericton	39672	27905	9390
Miramichi	35111	26701	9922
Moncton	30054	11168	8420
Saint John	39770	30276	15615
*Popular Vote:*	54%	27%	19%
*District Seats:*	6 seats (60%)	0 seats	0 seats
*Regional Seats:*	0 seats	2 seats (20%)	2 seats (20%)

Let’s assume, for the sake of argument, that the numbers in the table above reflect how people honestly want to vote. What happens if district votes do not count as votes for a party? The answer is that there’s little benefit in voting for a district candidate who cannot win. In Acadie and Beausejour, the Conservative candidates will not win, so many Conservative voters will consider supporting the Liberal candidate to prevent the NDP from winning the district. Similarly, NDP candidates in Fredericton, Miramichi, Moncton, and Saint John will not win, so many NDP voters may support Liberal candidates to ensure Conservatives are shut out of the district races.

In the table below, I indicate that 50% of Conservative and NDP voters may have an incentive to give their district votes to Liberal candidates. Observe, for example, that 5001 out of 10,003 Conservative voters in Acadie are indicated as having a tactical incentive to vote Liberal. This is based on the assumption that 50% of Conservative voters prefer the Liberals to the NDP, and 50% of NDP voters prefer the Liberals to the Conservatives. I may actually be underestimating these numbers, as the centrist Liberal Party is likely to be the 2nd choice of more than half of Conservative and NDP supporters. In any case, the table suggests that a significant number of voters might consider making a tactical decision for their district vote.

District Votes

MMP District	Liberal	Conservative	NDP
Acadie	46623 + 5001	10003 – 5001	29749
Beausejour	36534 + 3008	6017 – 3008	8009
Fredericton	39672 + 4695	27905	9390 – 4695
Miramichi	35111 + 4961	26701	9922 – 4961
Moncton	30054 + 4210	11168	8420 – 4210
Saint John	39770 + 7807	30276	15615 – 7807

Now here’s the key argument.

If many voters choose — for tactical reasons — to give their district votes to candidates who are leading in the polls, these candidates will almost certainly win at the district level. In some regions, this trend will result in one or two large parties winning so many district seats, it will become predictable that they will not win any regional seats. In these regions, there is no benefit in giving a regional vote to such a party. Therefore all of the large party’s supporters will have an incentive to give their regional votes to smaller parties.

In our New Brunswick example, polls may predict that the Liberal Party will win every district seat and no regional seats. If voters trust the polls, they may realize that a regional vote for the Liberal Party is a wasted vote. They will then have a very strong incentive to give their regional votes to either the Conservatives or the NDP. Let’s assume for simplicity that 50% of Liberal voters prefer the Conservatives, and 50% prefer the NDP. The table below indicates the numbers of Liberal voters affected by this tactical incentive, and where their regional votes may end up.

Regional Votes

MMP District	Liberal	Conservative	NDP
Acadie	46623 – 46623	10003 + 23311	29749 + 23312
Beausejour	36534 – 36534	6017 + 18267	8009 + 18267
Fredericton	39672 – 39672	27905 + 19836	9390 + 19836
Miramichi	35111 – 35111	26701 + 17556	9922 + 17555
Moncton	30054 – 30054	11168 + 15027	8420 + 15027
Saint John	39770 – 39770	30276 + 19885	15615 + 19885

In the following table we add up all the voters who have an incentive to vote tactically with either their district or regional vote. The column on the right indicates the percentage of voters in each district who have a logical reason to vote against at least one of their true preferences.

Tactical Incentives

MMP District	Liberal	Conservative	NDP	% of voters facing tactical incentives
Acadie	46623/46623	5001/10003	0/29749	60%
Beausejour	36534/36534	3008/6017	0/8009	78%
Fredericton	39672/39672	0/27905	4695/9390	58%
Miramichi	35111/35111	0/26701	4961/9922	56%
Moncton	30054/30054	0/11168	4210/8420	69%
Saint John	39770/39770	0/30276	7807/15615	56%

As we see in the table above, more than half the voters in every district have a reason to vote tactically. In Beausejour, more than 3 quarters of the population are affected. Far from eliminating tactical voting in Canada, a bad MMP model could actually lead to more Canadians voting dishonestly.

But how many of these 56% to 78% of voters would actually engage in ticket splitting for tactical reasons? The truth is no one knows. But here’s a troubling fact. If voters on one side of the political spectrum embrace these tactics, then voters on the other side must follow suit.

In the New Brunswick example, suppose that all Liberal-leaning Conservative supporters and all Conservative-leaning Liberal supporters split their tickets. They all vote Liberal at the district level, and Conservative regionally. In that case, assuming everyone else votes honestly, the Conservative Party would gain a 3rd regional seat at the expense of the NDP. Similarly, if only NDP supporters and NDP-leaning Liberal supporters split their votes, then the NDP would win 3 regional seats and the Conservatives would drop down to 1 seat. However, if voters on both sides of the spectrum vote tactically, then the regional seat distribution reverts to 2 seats for the Conservatives and 2 for the NDP. If one side votes tactically, the other must do the same or wind up severely under-represented.

If this example has made you concerned about the potential for tactical voting in MMP, rest assured there is a solution. Canada can adopt a good MMP model where votes for district candidates count as votes for their party. If district votes are guaranteed to count, then voters can simply choose their preferred option at the district level. And when voters are honest with their district votes, it becomes very difficult to predict if there’s any advantage in being dishonest with the regional votes. In fact you may recall that a number of good MMP models only have one vote, which makes ticket splitting impossible.

Ticket splitting in the real world

Ticket splitting tactics are not just a theoretical problem. They actually occur in the real world under bad MMP models. Here we focus on the UK, where MMP is used for the assemblies of Scotland and Wales.

Our first piece of evidence is a 3-minute clip featuring Tommy Sheridan, the leader of a small Scottish pro-independence party called Solidarity. He refers several times to Scotland’s most popular pro-independence party, the SNP. Pay attention to the logic behind Mr. Sheridan’s argument, which is based on the peculiarities of the electoral system. Note that in this part of the world, the “1st vote” is always the district (constituency) vote and the “2nd vote” is always the regional (list) vote. Only 2nd votes count as votes for a party.

Paraphrasing, Mr. Sheridan’s argument is as follows: A 2nd vote for the SNP is a wasted vote, so give your 2nd vote to me. Yes, some of his claims were exaggerated, as is typical of any politician. But the tactic he promotes is based on sound logic.

Let me be clear that this type appeal has nothing to do with the issue of Scottish independence, and everything to do with the electoral system. The Green party in Wales makes an almost identical appeal in the video below.

Paraphrasing, Jake Griffiths is saying a 2nd vote for Labour is a wasted vote, so give your 2nd vote to me. It’s exactly the same logic that Tony Sheridan outlined in the previous video, though the country and the emphasized issue are different. Note that as party leaders, Mr. Sheridan and Mr. Griffiths would have been at the top of their parties’ lists, so they would have been personally elected if enough voters had followed their advice.¹⁷

It’s not just politicians who advocate for tactical voting in Scotland and Wales. Some academics also advise people to split their votes. The following quote on voting in Scotland is from an article by Professor Patrick Dunleavy.¹⁸

But bear in mind that at the 2015 general election the SNP won 56 out of 59 local Westminster constituencies, and this year they again face a very fragmented opposition. So the SNP are almost certain to sweep the board in most of the Scottish Parliament local seats using at the first past the post voting stage. If they do so, then voting SNP also in the regional list part of the election would effectively be to waste your second vote. The SNP could be so drastically over-represented already in terms of the more numerous local seats that they cannot possibly win any more seats at top-up stage.

So if you are an SNP supporter across most of Scotland, there is just no risk at all to the SNP’s chances of winning seats in looking at the other parties and possibly giving your regional preference to one of them. You can (in effect) get two bites at the cherry and maximize the representation of parties closest to your views.

The quote below is from a letter signed by a number of Labour-supporting professors in Wales.¹⁹

Where Labour does well in the constituency section, it fares poorly in the regional list section, which is why so many people think that a second vote for Labour in such areas – like south Wales for example – is a wasted vote.

In these areas it makes sense for voters to cast their vote in a purposeful fashion – to aid a progressive party like Plaid Cymru and to block any party that trades in fear and prejudice.

The videos and quotes above all impress upon large-party supporters that a 2nd for their favourite party is a wasted vote, and therefore everyone should give their 2nd vote to a smaller party. But what about the 1st vote? If we revisit the video of Tommy Sheridan, he’ll describe the predicament the 1st vote presents for small-party supporters.

In short, Mr. Sheridan advises his supporters to give their 1st vote to the largest pro-independence party, even if they don’t particularly like the SNP. This is the of course the same form of tactical voting we experience in Canada.

In summary, real-world evidence from Scotland and Wales suggests that a bad MMP model will not only maintain the type of tactical voting we currently experience in Canada, but will likely augment it with a new form tactical voting associated with the 2nd vote. Yet a good MMP model will avoid all of these problems. If district votes count as votes for a party, voters will rarely if ever encounter appeals such as those in the above videos and quoted articles. Such tactics wouldn’t make sense, since all votes would contribute to the overall seat distribution and all votes would be of value to every party.

Conclusion

Perhaps you have read this far and are starting to question the suitability of any form of MMP for Canada. If bad MMP models lead to tactical voting and a number of other problems, are not all MMP models affected to some extent? Rest assured there is nothing wrong with MMP in general. A good MMP model will improve upon First Past the Post while either minimizing or completely avoiding all of the negative consequences described in this article.

If district votes count as party votes, then small parties will have to compete for district votes. And by taking district votes away from large parties, the large parties will not be able to rely completely on their success at the local level; they will have to compete for regional votes. With all parties competing in all districts for both types of votes, accountability will increase, the two types of representatives will be regarded as equal (since they will campaign in a similar manner), and disproportionality will be reduced to an acceptable level. Most importantly, confidence in Canadian democracy will not be subject to any serious risk.

Early in the article we saw that there are a variety of good MMP models where district votes count as party votes. I personally prefer the single-vote models, which render ticket splitting an absolute impossibility. However many advocates of proportional representation have a strong preference for MMP models with two votes and open lists. Given this preference, the Bavarian model is a perfectly decent tried-and-tested system in which district votes and regional votes are added together to determine the make-up of the legislative assembly.

Just because a country’s voting system is a form of Mixed Member Proportional doesn’t mean the system is genuinely proportional. And it certainly does not mean that the system will achieve the many other goals of proportional representation, such as reducing tactical voting, improving accountability, and ensuring as many votes as possible factor into the final seat distribution. The details of how MMP is implemented are extremely important, and now is the time to ensure Canadian decision-makers understand how to get the details right.²³

Proportional representation advocates say that every vote should count. If that’s the case, and if Canada adopts an MMP model with two votes, then shouldn’t both votes count? A vote for a district candidate may not change who wins the district, but it can and should be counted as a vote for the candidate’s party.

Footnotes

If you’re not familiar with MMP and how it would likely be implemented in Canada, I suggest watching Fair Vote Canada’s video on MMP before reading this article. A Canadian implementation of MMP would likely involve relatively small regions with fixed numbers of seats. My rule that differentiates “good” MMP models from “bad” ones is less relevant to implementations such as Germany’s where the calculations are performed nationwide and the number of seats varies from election to election to ensure proportionality.
See Jagmeet Singh’s policy on electoral reform.
See the Green Party of Canada’s post on the endorsement of MMP.
See the results of the 2016 plebiscite. Despite the outcome, a majority of PEI MLAs decided to maintain FPTP for another election with the idea of having Islanders vote again on the reform.
See Hansard, PEI Legislative Assembly (May 13, 2016).
Louis Massicotte’s MMP model is described in his 2004 report, In Search of Compensatory Mixed Electoral System for Québec (see PDF).
By introducing a 2nd vote in such a way that the district vote no longer counts for the party, tactics emerge that help incumbents retain their seats. Thus it is generally in a governing coalition’s best interests to switch from a good single-vote MMP model to a bad two-vote model.
Joachim Behnke and Friedrich Pukelsheim’s recommended MMP model can be found in a brief submitted to the Special Committee on Electoral Reform of the House of Commons of Canada (see the PDF).
See the Welsh Assembly’s description of how its members are elected.
See the Wikipedia table summarizing the 2003 Welsh Assembly Election.
An excellent description of the Bavarian open-list MMP model can be found on Wilfred Day’s January 2016 blog post, which also outlines how the system could be adapted for use in Canada.
The Law Commission of Canada’s recommended MMP model is described in their 2004 report (see the PDF).
An excellent explanation of list-free MMP can be found in this YouTube video, where the system is referred to as “Near Winner Proportional”.
Please note that while I have merged several existing New Brunswick districts for the sake of convenience, what would actually happen is that the district boundaries would be completely redrawn for MMP.
To keep the New Brunswick example simple, I’m ignoring votes cast for the Green Party of Canada.
I am translating votes into seats using the D’Hondt method, which slightly favours large parties. There are other methods Canada might adopt, but all of them would give rise to the same tactical voting issues illustrated in the example.
Neither Jake Griffiths nor Tony Sheridan won seats in the elections following the videos (Wales 2011, Scotland 2016). Since neither of their parties ran district candidates in the relevant regions, it is difficult to say whether their supporters (5.6% in the case of Jake Griffiths) voted honestly or based on the promoted tactics. That said, ticket splitting by Labour supporters in Wales is consistently more evident in regions where the party dominates the local races, suggesting that tactical voting is influencing voters’ decisions.
See Professor Dunleavy’s article on how to vote tactically in Scotland.
See the BBC article on a letter advocating for tactical voting in Wales.
See the 2016 November 22 Hansard of the Prince Edward Island Legislative Assembly.
See the 2016 August 30 ERRE Evidence containing Professor Massicotte’s testimony.
See Wikipedia page referencing past Welsh elections.
At the time of writing, a process is underway to choose one or more proportional systems for a 2018 provincial referendum in British Columbia. Consider filling out the Citizen Engagement questionnaire, or contact an MLA to stress that any MMP model on the referendum must minimize the risk of ticket splitting tactics by making all votes count as party votes. Choose your favourite good MMP model from this article as an example.

Election Year	Con. Votes (%)	Con. Seats (%)	Plaid Votes (%)	Plaid Seats (%)	Labour Votes (%)	Labour Seats (%)
1999	16.5	15.0	30.5	28.3	35.4	46.7
2003	19.2	18.3	19.7	20.0	36.6	50.0
2007	21.5	20.0	21.0	25.0	29.6	43.3
2011	17.9	18.3	22.5	23.3	36.9	50.0
2016	18.8	18.3	20.8	20.0	31.5	48.3

Rhys Goldstein's Blog