Lessons from International Multi‐Option Referendum Experiences

Referendums are regularly criticised for reducing complex policy decisions to two maximally opposed options. This reduces opportunities for voter expression and can polarise debates. Alternative referendum designs which present more than two ballot options can offer innovative opportunities, but also raise new challenges. We can benefit by learning from previous experiences with multi-option referendum voting. Discussions of such experiences are rare and have often focussed on a limited number of cases. This article provides an overview of over 100 multi-option referendum experiences around the world. It discusses the topics on which they were held and the ballot options that were offered. It then analyses the variety in ballot design in terms of questions posed and voting methods applied. Drawing on the experiences of multi-option referendums, the article concludes with lessons that can be learned in relation to initiating and designing these referendums.

REFERENDUMS ARE increasing in popularity and frequency across the globe. Critics point toward the polarising nature of the two-option choice that is commonly offered. Referendums could, however, also be designed to include more than two ballot options. So far, little has been written about practical experiences with such referendums. To address this gap, I have compiled a dataset of experiences which can inform our understanding of variety in multi-option referendum designs. In order to learn from experience, this article describes the empirical evidence and draws several lessons which can inform practical referendum design choices.
Particularly in cases in which more than two distinct scenarios are conceivable-for example in the Brexit referendum-a binary choice might be too restrictive. Framing the vote in multiple options has been suggested as an alternative approach. 1 For the Scottish independence referendum, such a procedure received serious attention by policy makers prior to the vote. 2 A ballot containing several scenarios empowers voters to express their opinion in more detail rather than forcing them to select one of two extremes. Ballot options are more specified than broadly interpretable options such as 'leave'. Providing more detailed options helps voters to understand the policy consequences of the referendum result, as is recommended in international standards for referendums. 3 In multi-option referendums, 'voters are presented with more than two options addressing the same issue, each of which is distinctive, leading to one outcome'. 4 Whereas binary referendums commonly pose a single option-a new policy proposal or scenarioagainst the status quo, multi-option referendums offer a wider range of alternative policies. An example would be voting on a new electoral system with a choice between first past the post, mixed member proportional, or alternative vote systems. Extending choice beyond two options requires additional design choices in two important respects: the selection of options to appear on the ballot (ballot content) and the way voters are questioned on the options (ballot design).
The next section considers the prevalence of multi-option referendums over time and space; the following two sections focus on multi-option referendum experiences with respect to the design aspects that set multioption referendums apart from their binary 1 counterparts: first, ballot content and, second, ballot design. This empirical overview is followed by several lessons drawn from experience. Most of the existing literature, such as work inspired by social choice, has discussed extensively the theoretical possibilities and implications of multi-option voting designs. However, few studies have reflected on how multi-option referendum voting has played out in practice. This article explores empirically the effects of different multi-option ballot design choices as observed in actual referendum practice.

Prevalence of multi-option referendums
Discussions on the applicability of multi-option designs can benefit by learning from practical experiences elsewhere. Whilst binary referendums are the norm, there have been over 100 experiences of referendums offering at least three options at the highest government level in countries and dependent territories around the world (see Appendix 1). In existing literature, there is no comprehensive overview of such cases. Analyses of experiences in academic and popular literature are scattered and often limited to sporadic single-case studies or a discussion of a small set of better-known cases. 5 This article seeks to provide a complete overview based on a dataset of 106 cases compiled by the author. 6 The dataset includes national-level referendums in sovereign states-at the time of the vote-and referendums at the highest government level in dependencies and territories formally related to another state. It excludes further experiences on local and regional levels. Voting data were largely compiled using existing datasets and were verified using official sources. The earliest documented cases date back to 1848, when several sovereign Italian city states voted on mergers.
In the 1960s, multi-option referendums notably increased in frequency, peaking in the 1970s with an overrepresentation of referendums in Switzerland and New Zealand. Contrary to a prevalence of constitutional status questions in the previous century, more recent decades represent a shift towards a broader range of policy questions. This illustrates the continued or even renewed relevance of the multi-option design. Moreover, aside from actual empirical experiences, the debate on the possible use of multi-option referendum designs has gathered pace in recent times.
Since the start of the twenty-first century, nineteen multi-option referendums have been held worldwide (see Figure 1). The most recent cases were in Guernsey (2018), Puerto Rico (2017), New Zealand (2015) and the Dutch island of St. Eustatius (2014). The latter three had experienced such a referendum before; New Zealand in 1992 and 2011 on electoral system reform, as well as twenty-eight repeated referendums-alongside each general election-on liquor licensing between 1894 and 1987. Puerto Rico voted on its status in relation to the US five times since 1967, each time with a differently designed multi-option referendum. Five islands of the former Netherlands Antilles witnessed a total of eleven multi-option referendums on their relative constitutional status in relation to the Kingdom of the Netherlands.
A considerable number of multi-option referendums were also held in dependent island states of the US, UK, France and Australia over the past century. Experiences in UK territories include referendums in the British Crown Dependencies of Guernsey

Ballot content: topic and options
Political and electoral changes, as well as constitutional status questions, are popular topics for multi-option referendums. Non-sovereign territories have-sometimes repeatedly-used the multi-option design to propose a range of different options to voters with respect to the territory's relationship to the sovereign state, offering such options as independence, commonwealth, statehood, municipal or provincial status and free association. Examples include five referendums in Puerto Rico on its relationship to the United States. The options that were offered in the referendums differed. Independence and US statehood were always on the ballot, together with one or two alternative options (commonwealth and free association). The 1998 referendum featured an explicit 'none of the above' option, which received an absolute majority of votes as a result of societal dissatisfaction with the absence of the status quo-commonwealth with the US-on the ballot.
In the two most recent referendums in 2012 and 2017, an absolute majority of voters preferred to become a US state, although the latter suffered from low turnout (an all-time low at 23 per cent, compared to 78 per cent in 2012). So far, the US Senate has not approved Puerto Rican statehood. Guam is in a similar situation and voted for a US commonwealth status in 1982, with no implementation to date. These experiences point to an important contextual element in status referendums: an international dimension. The dependence on US Congress approval requires both pre-referendum and post-referendum coordination.
An example of when such coordination was achieved is the 1948 Newfoundland referendum. A dominion of the UK at the time, the referendum questioned whether the territory would prefer to maintain its current status, become part of Canada or obtain what was described as 'responsible government'. The ballot options were coordinated with both the UK and Canada before being put to voters. Since the two latter options each received over 40 per cent of the votes, a second-stage ballot was held to decide the most popular option overall, and Newfoundland joined Canada as part of a new province. The status issue was thus successfully resolved. The Puerto Rico, Guam and Newfoundland examples illustrate how international coordination can influence referendum effectiveness.
Further status-related referendums were held in the 1990s on the five islands of what were then the Netherlands Antilles (Bonaire, Curac ßao, Saba, St. Maarten and St. Eustatius). The referendums offered four options: independence, autonomous area, status quo or connection to the Netherlands. Each of the islands voted overwhelmingly to retain the status quo. Following several years of reform of the islands, a second round of referendums was held a decade later. Curac ßao and St. Maarten assumed the status of a country within the Kingdom of the Netherlands and the other three islands became a Dutch municipality with special status. At different stages, multi-option ballots were used to measure support for different change scenarios prior to reforms and to vote on status preferences in order to reflect the new situation after the reforms.
Other popular topics for multi-option referendums are electoral reforms. Since electoral systems can be described as a multitude of distinct options, they are well suited for multi-option voting. New Zealand voted to change the first past the post (FPTP) electoral system in 1992, electing mixed member proportional representation (MMP) from a choice of four alternative systems. In 2011 the electorate voted on the issue again, this time on whether the new MMP system should be sustained, which was approved. Andorra (1982) offered the electorate a choice between a majority, proportional, or mixed system. Further referendums have been held on changing electoral rules within the existing system, such as the number of constituencies and legislators (Guernsey, 2018 and Jersey, 2013); the number of MPs (Liechtenstein, 1985); voting age (Virgin Islands, 1970); and legislative terms (Cook Islands, 1994). As these examples demonstrate, the added benefit of a multi-option design comes into its own on issues that potentially entail more than two scenarios-as is often the case for electoral and political reforms.
Constitutional changes were voted on in Chile (1925), Uruguay (1958, 1964 and 1966), and Benin (1990). In addition, popular initiatives affecting specific constitutional articles have led to several multi-option ballots in Switzerland after the addition of a legislative counter-proposal. Other constitutional issues included political reforms (Andorra, 1977 and 1978), sexual equality (Liechtenstein, 1985), capital punishment (Virgin Islands, 1978) and cultural issues such as a national hymn (Australia, 1977) or flag (New Zealand, 2015). There is also great variation in multi-option votes on non-constitutional legislation. Sweden used the multi-option referendum instrument twice to address innovations in pensions policy (1957) and nuclear power (1980). Other electorates voted on port construction (Pitcairn, 2009), fishing licences (New Zealand, 1954) and prohibition (Finland, 1931). Repeated referendums on prohibition and liquor licensing were held in New Zealand alongside each general election between 1894 and 1987. On two occasions a multi-option referendum ballot was used to vote on personal appointments, de facto functioning as elections: Cambodia, 1960, on the governor, and Greece, 1862, on the head of state. Both yielded an almost unanimous outcome.

Ballot options
The selection of specific ballot options depends on the topic of the referendum. It is common for multi-option referendums to include the status quo, most often as a description of that situation. Some referendums (for example, referendums on the Netherlands Antilles between 1993 and 2004; Northern Mariana Islands, 1961; Guam, 1982) explicitly named one of the options 'status quo'. Some other referendums did not offer a status quo option, posing only change options. For example, the 1962 Singaporean status ballot only listed various association modes with Malaysia, as the government had already made the association decision.
Several referendums included an explicit blank, 'none of the above' or 'no to everything' option (for example, Andorra, 1978; Puerto Rico, 2012;Northern Mariana Islands, 1961;Liechtenstein, 1985) or an opportunity for voters to write their own preferred status (Guam, 1982;Micronesia, 1983). In some cases, this option replaced a status quo option. Only in Puerto Rico (1998) did a 'none of the above' option receive an absolute majority of votes (50.3 per cent). In Andorra (1978), a ballot containing two proposals for constitutional change and one 'no to both' option was decided in favour of the latter by a narrow plurality margin (35.6 per cent). Voting for no-options often results from dissatisfaction with the ballot options, in particular with the absence of a highly supported (status quo) option.
When several ballot options are highly similar or overlapping, first preference votes may spread over these options, producing an inconclusive result. In a 1978 referendum on the Virgin Islands, 41.5 per cent voted for capital punishment for all first degree murders; 13.8 per cent supported it under certain conditions; and 44.7 per cent rejected it under all circumstances. A capital punishment policy was not implemented, despite an absolute majority agreeing on a restricted variant of such a penalty. Similarly, in Andorra (1982), 24 per cent voted for a proportional system for both municipal and national elections and a further 43.3 per cent opted for a proportional system for municipal elections only. Because of an absolute majority threshold for referendum validity, the status quo-a majority system for all elections-prevailed by default, despite receiving just 32.8 per cent of the vote share. A two-thirds majority had expressed their preference for proportional elections at the local level to no avail. On the Cook Islands (1994), a majority of voters supported shortening parliamentary terms, but since their first preferences spread over three and four year terms, the status quo of five year parliamentary terms prevailed with a 41.8 per cent plurality. Following this experience, the government opted for a binary referendum on reducing parliamentary terms to four years in 1999-in which majority support lacked the necessary supermajority

Ballot design: questions and voting method
Besides decisions on the ballot content, multi-option referendums require decisions on their ballot questions and voting method. 'Ballot questions' refers to the structure in which the options are presented to voters. The ballot paper can entail either a single question or several questions of which the results are to be combined in one outcome. Referring back to the definition of multi-option referendums set out in the introduction, the one-outcome characteristic of multi-option referendums implies that only one explicitly described ballot option can be implemented. This characteristic distinguishes multi-option referendums using a multiple-question design from voting on multiple distinct-though related-binary questions, like the 1997 referendum on instituting a Scottish Parliament and whether it should have tax-varying powers. 'Voting method' refers to how voters mark their ballot papers-by marking one or several options-and how votes are aggregated into a referendum result. In some cases, voting may take place at multiple moments in time, similar to elections with multiple rounds.
Over three-quarters of all multi-option referendums were single-question referendums (see Figure 2). The vast majority of those were decided through first preference votes only, resulting in a plurality or absolute majority winner. Other strategies, both ensuring an absolute majority winner, are the inclusion of a provisional run-off stage (if no option wins absolute majority support, the two most popular options face each other in a second stage) and the use of ranking (voters number all options according to their relative preferences).
Some multi-option referendum ballots posed multiple questions. One-fifth posed alternative proposals on the ballot through multiple binary questions (in a series of binary questions, voters approve or reject each proposal separately). A few cases included two ballot questions in a gateway-filter structure (in a first question, voters opt for or against changing the status quo; in a second question they select their most favoured change option). This section considers empirical experiences with these different designs.

Single-question multi-option ballots
Three-quarters of multi-option referendums were conducted in a similar manner to binary referendums, with only a first preference vote cast. Other than in binary referendums, first preference votes do not guarantee an absolute majority winner when three or more options are on the ballot, risking an unclear or even controversial outcome. Most multi-option referendums (62 per cent) used first past the post rules to determine the winning option. Around half of those, a large proportion being the repeated New Zealand referendums, required an absolute majority outcome for implementation. Some multi-option referendums even required supermajorities of 60 per cent (St. Eustatius, 2014; New Zealand until 1908 for prohibition option only) or a two-thirds majority (Cook Islands, 1994; Singapore, 1962). More often than not, absolute majority or supermajority requirements resulted in invalid referendum outcomes.
In cases without absolute majority requirements, plurality winners regularly prevailed, sometimes with narrow margins. Examples include Sweden 1980 (39.1-38.7-18.9 per cent), Andorra 1978 (35.6-32.8-31.6 per cent) and Liechtenstein 1985 (43.6-39.0-17.4 per cent). Many of these referendums suffered from serious interpretation issues after the vote, often leading to a policy deadlock. The prevalence of single-question ballots with first preference votes results in a high rate of contested or invalid referendum outcomes. Such outcomes can severely diminish or even reverse the value of multi-option referendums as an instrument to decide on policy matters.
A number of multi-option referendums (11 per cent) provided for a run-off stage should no absolute majority winner emerge. For example, the 1982 seven-option status referendum in Guam ended in a 49.5 per cent plurality for US commonwealth status and was decided in favour of this option in a binary run-off against the next most popular option later the same year. The Newfoundland run-off stage featured the two most popular options of the 1948 ballot: responsible government (44.6 per cent) and confederation with Canada (41.1 per cent). This case demonstrates that the plurality winner under first preferences is not always the winning option when posed against the runner-up: confederation with Canada dominated the run-off result with 52.3 per cent. The referendums on the islands of the Netherlands Antilles also included provisions-which were never needed-for a run-off stage to be held only in the absence of an absolute majority winner. Slightly different from these cases with run-off provisions, the 1977 Andorran referendum, presenting six change options, was followed by an improvised second stage. The non-binary design of this 1978 referendum resulted in an inconclusive outcome: the 1977 plurality winner and a merged alternative of the losing options both lost to a 'no to both' option by a narrow plurality of votes. These examples illustrate that a binary run-off round can guarantee majority support and effectively settle the issue on the condition that meaningful options-which often includes a status quo option, at least in the first stage-are presented to voters in both rounds.
Ranking was used for referendums in Guernsey (2018), New Zealand (2015), Jersey (2013) and Australia (1997), with an alternative vote system electing the winning option. In Guernsey, Jersey, and Australia, the status quo was included on the ballot paper (respectively as a description of current electoral rules and the name of the current national hymn). The New Zealand case was different, as it only offered change options (five flag designs). The most popular flag design competed against the status quo flag in a later run-off stage. None of these cases yielded an absolute majority winner after counting first preferences alone. The votes of those people with a first preference for the least popular option were then redistributed to their respective second preferences and so forth. Such redistributions of votes yielded an absolute majority winner in each case. Ranking can therefore ensure majority support within a single voting stage and significantly simplify the interpretation of the referendum result.

Multiple-question multi-option ballots
In 20 per cent of cases, alternative proposals were voted on in a series of binary approvereject questions. In Switzerland and Liechtenstein this is the common method of voting on citizen initiatives to which a counter-proposal has been drafted by policy makers. Since 1987, voters in both countries have been allowed to vote 'yes' on both the citizen-initiated proposal and the counter-proposal (referred to as the 'double yes' possibility), increasing each option's chances of obtaining an absolute majority. A third question, the Stichfrage or 'deciding question', is added to determine the winning change option in the event that both beat the status quo. The deciding question functions as a run-off between approved change proposals. By using a deciding question rather than directly electing the proposal with the highest support percentage, the relative preferences of voters approving or rejecting both proposals can also be taken into account.
The outcome of a three-question ballot potentially suffers an inconsistency when the change option approved by the largest number of voters loses in the deciding question against the change option approved by a smaller number of voters. This so-called 'vote cycling' occurred in one of six referendums in which a deciding question was used (namely Switzerland, 2010). It can logically follow from the relative preferences expressed in the deciding question by those supporting either both or neither of the proposals, or can result from different turnout rates for the different questions.
The status quo option wins if neither of the proposals for change receives an absolute majority of votes. Prior to the 'double yes' possibility, the spread of votes over two-sometimes similar-change proposals favoured the status quo. It led to the common practice in Switzerland for initiative committees to withdraw their proposal after a counter-proposal was launched, in order to boost the winning chances of the counter-proposal against the status quo. This issue is similar to the decreasing chances of similar change options winning absolute majority support under a first preference vote in a single-question referendum.
The problem of support spreading over multiple options when voters may only approve one option is exacerbated when more than two proposals for change are offered. In Slovenia (1996), three binary questions were used to vote on three separate proposals for electoral change. This case also illustrates how differences in turnout for the different questions can influence results. Voters could only choose one of the options for approval. One of the proposals was accepted by 65 per cent of those answering that particular ballot question. Those approving the option, however, only constituted 44 per cent of those voting in the referendum. The-belated-decision by the constitutional court to implement this option was highly contested.
A different multiple-question design is the gateway-filter structure, which includes a binary gateway question ('do you favour change or not?') and a multi-option filter question ('which alternative to the status quo would you prefer?'). Responses to the second question are discarded if a majority rejects change in the gateway question. New Zealand and Puerto Rico have used gateway-filter structures. In the 2012 referendum in Puerto Rico, 54 per cent voted in favour of changing the status quo in the gateway question. In the three-option filter question, 61 per cent voted in favour of statehood, which was deemed to be the winner of the referendum. Considering that this 61 per cent majority of a slim majority (54 per cent) constituted a minority of merely 33 per cent of voters overall, it would not have beaten the status quo in a run-off vote.
To prevent such an inconclusive outcome, referendum designs in New Zealand included a run-off stage. The winning option of the filter question (featuring four change options) in 1992 was posed against the status quo in a binary run-off stage in 1993 in order to ensure an absolute majority winner. The same design was provided for the 2011 referendum on the same topic, but this time a majority rejected change in the gateway question. Gateway-filter designs explicitly separate the desire for change from the type of change desired, which helps clarify whether a majority of voters actually prefer to change the status quo in the first place. The results of the gateway question could determine whether a run-off round is required.

Lessons from multi-option referendum experiences
Drawing on the empirical evidence of multioption referendum voting, several lessons can be distilled for good practice. Only those that are particularly relevant to multi-option referendums are discussed (as opposed to lessons that apply to all referendums, which are well-documented elsewhere).

Ballot content: topic and options
Empirical evidence shows that many multioption votes took place on constitutional status questions and on electoral and political reforms. By their nature, such topics are often open to a range of different policy scenarios and therefore well suited to a multi-option design. Multi-option referendums are relevant for topics for which more than two alternative policies can count on realistic support in society. Several multi-option referendums have resulted in very high support for just one or two of the options, which raises questions about the added value of a multi-option design over a binary ballot. Whether the issue lends itself to a multi-option design and which options are considered to be realistically supported may be context dependent. It is therefore essential to have a basic understanding of societal preferences before designing the referendum, including the selection of a binary or multioption model. To select broadly supported ballot options, some governments have involved non-political actors, such as an electoral commission (Jersey, 2013) or expert panel (New Zealand, 2015) in the design process, or have obtained citizen input either directly or indirectly. Citizen assemblieslike the Irish and British Columbian examples for binary referendums-can also be employed for this purpose. 8 Offering voters a fair choice requires that the winning option is able to be implemented. Whilst this is also true for binary referendums, it deserves mentioning here because it applies in particular to the category of constitutional status referendums, which constitute a significant share of all multi-option referendum experiences. Where the approval of an external entity is required, it is necessary to have pre-referendum coordination of ballot options with governments of those countries to which particular statuses relate. The diverging trajectories of Newfoundland, on the one hand, and Puerto Rico and Guam, on the other, demonstrate how bilateral agreements ahead of the vote can increase the chances of successfully settling an issue.
On a more strategic level, the inclusion or exclusion of particular ballot options can affect the referendum process and result. Chances of successfully resolving the issue are highest if all relevant policy options are on the referendum ballot. 9 This often includes a status quo option. Two Puerto Rican cases (2017 and 1998) demonstrate how the exclusion of popular options can result in boycotts which jeopardise the legitimacy of the results. The Virgin Islands (1978) and Cook Islands (1994) experiences demonstrate how similar or overlapping options may distort the result by spreading supporters of related policies over different ballot options. The same was true in Switzerland and Lichtenstein for citizen initiatives and legislative counter-proposals with relatively similar content, prior to the introduction of the possibility to vote 'double yes'. In conclusion, it is best to avoid irrelevant or unimplementable options and to select a limited number of realistic and obtainable options. This facilitates voter understanding and avoids distortion of results in particular when each voter may only approve a single option.

Ballot design: questions and voting method
To ensure an uncontroversial outcome, multi-option referendums must not only offer a set of realistic and implementable options, but also ensure that the voting process results in a clearly supported outcome. Voting only on first preferences-as practised in most multi-option referendums to date-can render a clear majority outcome extremely difficult and the winning outcome might be contested by a majority of voters. As the Newfoundland (1948) referendum demonstrated, the plurality winner under first preferences is not always the winning option when posed against the runner-up in a binary vote. Similarly, the winning option emerging after a redistribution of ranked votes in the New Zealand (2015) referendum was not the plurality winner of first preference votes.
The most effective way to avoid controversies is to ensure that the referendum yields an absolute majority winner. This presents a challenge for referendums with more than two options, but can be achieved through run-off provisions, a series of binary questions or alternative voting methods, such as ranking and approval voting. Ranking has been successfully practised in multioption referendums, allowing voters not only to mark their first preference, but to express their order of preferences on all options. Another alternative voting method on a single-question ballot would be approval voting. It has not been used in official nationwide referendums, though effective local level examples exist, for example in the US and the Netherlands. Voting for more than one option could avoid a distortion of results when ballot options overlap or lie on one dimension, as with the Virgin Islands in 1978 and the Cook Islands in 1994. It also more generally increases the chances of yielding absolute majority or supermajority outcomes for multi-option ballots. For issues of particular importance, combining a (super)majority requirement with an alternative voting method could be effective.
The selection of a voting method and possible additional thresholds may depend on several context specific factors. These include the number of realistic policy options, voter experience with different methods and the scope of implications of the vote (for example, constitutional or regular policy; effects on international relations; reversibility). It is good practice to ensure that the voting method, possible staging or repetition and the conditions for the implementation of the referendum outcome are documented and communicated in advance. 10 This adds to the clarity and legitimacy of the voting process and helps to avoid allegations of changing the rules during the game.
With good design choices in terms of ballot options and voting processes, multi-option referendums can be used to maximum benefit. As with referendums in general, there is diversity in design types and in referendum contexts, and experiences with multi-option referendums represent a mix of effective and more challenged examples. They include cases where democratic values were under pressure, such as those in Cambodia (1960) and Singapore (1962) or where voting was vulnerable to boycotts (Puerto Rico, 1998 and2017). They also include very successful examples of solving policy problems, such as changing the New Zealand electoral system with evident majority support, and resolving the Newfoundland status question. The effect of design choices on the perceived legitimacy of the referendum process and the clarity of the result underlines the importance of learning from experiences elsewhere as part of a broader process of informed considerations about the most suitable referendum design.