The Problem With Referendums That Bundle Logically Separate Propositions Into Single Votes: A Decision Tree for Identifying Question Coherence Failures

A referendum question that asks voters to approve or reject a package of logically distinct policy choices in a single Yes/No vote doesn’t measure public opinion. It obscures it. Bundling — nesting multiple propositions within one referendum question — is among the most persistent design flaws in direct democracy. It survives because it serves legislative drafters who want to force consensus where none exists. The voter gets a ballot question they cannot answer honestly without affirming something they oppose or rejecting something they support.

Two referendums illustrate this problem from different angles. California’s Proposition 63 (1986) bundled an official-language designation with sweeping government-operations mandates. Switzerland’s June 2021 referendum on the COVID-19 Act folded emergency executive authority, financial compensation schemes, and contact-tracing infrastructure into a single up-or-down vote. Both cases show how bundling creates comprehension barriers during the campaign and implementation ambiguities after the vote. What follows is a decision tree that referendum administrators can use to identify when a question has crossed from comprehensive to incoherent — and to act before the ballot prints.

What Bundling Looks Like in Practice

Bundling is not the same as comprehensiveness. A referendum question can address all elements of a single, coherent policy change without bundling. Bundling occurs when a ballot question contains two or more propositions that a voter could reasonably support or oppose independently but is forced to treat as a single unit. The structural test is straightforward: if you can articulate a coherent policy position that requires answering Yes to one component and No to another, the question is bundled.

Consider the ballot title and summary for California’s Proposition 63, which appeared on the November 1986 ballot. Placed before voters by initiative petition, the measure proposed to amend state law to designate English as the official language of California. But the operative provisions extended well beyond a symbolic declaration. The proposition also restricted the use of non-English languages in government operations, limited bilingual education services, and required the state to enforce these provisions through litigation. A voter who supported a symbolic official-language designation but opposed restricting bilingual services had no way to express that position. One binary choice fused declaration with enforcement, symbolism with policy, and language status with service delivery.

The Swiss case operates on a different institutional track but presents the same structural problem. Switzerland’s Federal Act on the Legal Framework for the Implementation of Measures to Combat the COVID-19 Epidemic (COVID-19 Act) was challenged by referendum under the optional referendum procedure, which allows 50,000 signatures within 100 days of a law’s publication to force a popular vote. The Federal Assembly passed the Act in March 2021. It consolidated the legal basis for a wide range of pandemic measures: restrictions on gatherings and venue capacity, financial compensation for businesses and cultural organizations, contact-tracing infrastructure, testing protocols, and the certification system for COVID-19 documents. When voters went to the polls on June 13, 2021, they faced a single question — approve or reject the Act — that bundled public health restrictions with economic support, surveillance infrastructure with cultural-sector relief, and emergency executive authority with parliamentary oversight mechanisms.

The Cognitive Load Problem

Bundling imposes a specific cognitive burden that differs from the ordinary difficulty of understanding complex policy. A voter facing a complex but single-issue referendum can seek information, weigh trade-offs, and reach a decision through a linear reasoning process. A voter facing a bundled question must perform a multi-dimensional optimization instead: assess each component separately, assign weights to each, determine whether the package as a whole is preferable to the status quo, and then compress all of that analysis into a single binary output. This is not just hard. It is structurally incompatible with how most voters process political information under real-world conditions of limited time and attention.

Research on voter cognition and public opinion formation around complex policy questions, as documented by the Pew Research Center’s nonpartisan polling and data-driven social science research, shows that voters rely heavily on simplifying heuristics when facing ballot measures — cueing off trusted endorsers, party signals, or campaign slogans rather than reading the full text of propositions. Bundling amplifies this problem because the heuristics become less reliable. An endorsement from a trusted organization may reflect that group’s position on one component of the package while obscuring its stance on others. The voter who follows the heuristic is not making an informed choice about the bundled components. They are making a choice about the endorsement, which may not map cleanly onto the policy content.

In the Proposition 63 case, campaign materials on both sides largely ignored the enforcement provisions and focused on the symbolic question of whether English should be the official language. Voters who supported the symbolic designation but were unaware of the operational restrictions on government multilingualism had no reason to investigate further — the campaign framing suggested a simple status question, not a complex policy mandate. In the Swiss COVID-19 referendum, campaign discourse split between those framing the vote as a referendum on government overreach and those framing it as a vote on economic support for affected businesses. Neither framing captured the full content of the Act. Voters who engaged with only one framing were making a partially informed decision about a multi-component package.

Implementation Ambiguity After the Vote

Bundling creates a second-order problem that manifests after the votes are counted: implementation ambiguity. When a bundled referendum passes or fails, the mandate is unclear because the result cannot be decomposed into component-level signals. Proposition 63 passed by 61%. Did that mean 61% of voters supported restricting bilingual government services? Or did it mean 61% supported a symbolic official-language designation and tolerated the enforcement provisions as a secondary consequence? The answer matters for implementation. Legislators and state agencies interpreting the mandate face a signal aggregated across logically distinct propositions, and the aggregation destroys the information that would tell them which components had genuine public support.

The Swiss case produced a different but equally instructive implementation ambiguity. The COVID-19 Act was approved by 60.2% of voters, but this result could not tell the Federal Council or the Federal Assembly whether public support extended equally to the public health restrictions, the financial compensation schemes, and the certification infrastructure. In the months following the vote, the Federal Council faced decisions about which provisions to extend, modify, or sunset as the pandemic evolved — and the referendum result provided no granular mandate for any individual component. The approval was a mandate for the package, not for its parts, and the package was a product of legislative negotiation rather than a coherent policy position that voters had evaluated as a whole.

This implementation gap is not incidental. It is a direct consequence of the information loss that bundling produces. A referendum result is a data point, and bundling is a form of data compression that discards the component-level signals. When implementers try to reconstruct those signals from a bundled result, they are forced to make interpretive assumptions that may or may not reflect what voters actually intended — and the referendum process, which is supposed to produce clarity, instead produces a mandate that is less informative than a well-designed opinion poll would have been.

How Drafters and Commissions Have Attempted Mitigation

Some jurisdictions have recognized the bundling problem and developed partial mitigations. The most common approach is question splitting: dividing a multi-component proposal into separate referendum questions that voters can answer independently. Switzerland itself uses this practice frequently at the federal level, where distinct constitutional amendments or laws appear as separate questions on the same ballot. California’s initiative system, however, provides no structural mechanism for splitting an initiative petition that has qualified as a single measure. The initiative’s sponsors control the scope, and the electoral commission has limited authority to demand division.

A second mitigation is sequenced voting, where voters approve a framework principle before being asked to approve specific implementation measures. New Zealand’s 2015–2016 flag referendums attempted this approach by first asking voters to rank five alternative flag designs, then pitting the winning alternative against the current flag in a second referendum. The design was flawed in execution — the sequential structure created strategic voting opportunities and the two-stage process extended campaign fatigue — but the underlying principle of decomposing a complex choice into sequenced decisions has merit for bundled propositions.

A third mitigation is enhanced explanatory materials: voter information booklets that break down a bundled proposition into its components and describe each separately. This is the approach Switzerland’s Federal Chancellery takes with its voter information campaigns, which distribute booklets describing the content and arguments for and against each referendum question. The limitation is that explanatory materials can clarify what a bundled proposition contains, but they cannot change the fact that the voter must still compress their assessment into a single Yes/No answer. Better information about a structurally flawed question does not fix the structural flaw.

Institutional research on how ballot measure complexity affects voter behavior and comprehension, as analyzed by the Brookings Institution’s research on U.S. elections and direct democracy processes, underscores that information provision alone cannot compensate for structural question design failures — voters need coherent questions, not just longer pamphlets. The finding is consistent across jurisdictions: when the ballot question itself is the problem, no amount of supplementary material fully resolves the comprehension barrier.

A Decision Tree for Identifying Question Coherence Failures

The following decision tree is designed for referendum administrators, electoral commission counsel, and legislative drafters who review referendum questions before they appear on ballots. It proceeds through five diagnostic questions, each answerable through text analysis and cognitive pre-testing. If a referendum question fails at any node, the administrator should consider question splitting, sequenced voting, or returning the proposal to its drafting body for revision.

Node 1: Component Independence Test. Can a voter coherently support one component of the proposal while opposing another? To apply this test, identify each operative provision of the referendum text. For each pair of provisions, ask whether a reasonable person could hold a policy position that affirms one and rejects the other. If the answer is yes for any pair, the question contains logically separable propositions and may be bundled. Proceed to Node 2. If no pair of provisions is independently affirmable or rejectable, the question is comprehensive rather than bundled, and no further testing is required on this dimension.

Node 2: Component Salience Test. Are the logically separable components each significant enough that a voter’s position on one could determine their vote? A bundled question that contains a minor consequential provision alongside a major policy change may not produce meaningful comprehension barriers. To apply this test, assess whether each independently affirmable component addresses a policy area that voters are likely to have distinct opinions about. If all separable components are minor or technical, the bundling may be acceptable. If one or more separable components are politically salient — meaning voters are likely to have formed or to form distinct preferences — proceed to Node 3.

Node 3: Campaign Framing Divergence Test. Is it likely that campaign organizations on either side of the referendum will frame the question around different components? If the Yes campaign emphasizes Component A while the No campaign emphasizes Component B, voters will receive asymmetric information about the package’s content, and the referendum result will not reflect a common understanding of what was voted on. To apply this test, examine the political coalition structure around the referendum. If different organizations within the same side disagree about which component is the primary issue, the question is likely to produce campaign framing divergence. If divergence is likely, proceed to Node 4.

Node 4: Implementation Decomposability Test. After the vote, will implementing authorities need to make separate decisions about each component? If the referendum passes, will the responsible ministry, agency, or legislative body need to determine which components to implement first, how to resource each, and whether to modify any? If the implementation process will require component-level decisions and the referendum result cannot provide component-level mandates, the bundling will produce implementation ambiguity. Proceed to Node 5.

Node 5: Structural Remedy Assessment. Given that the question is bundled, salient, framing-divergent, and implementation-decomposable, what structural remedy is available? The options, in order of preference, are: (a) question splitting — divide the bundled proposal into separate referendum questions on the same ballot; (b) sequenced voting — ask voters to approve the framework principle first, then vote on specific implementation components in a subsequent referendum; (c) revision referral — return the proposal to the drafting body with a requirement that it either justify the bundling on policy grounds or separate the components; (d) enhanced explanatory requirements — if no structural remedy is politically or legally available, mandate that the voter information booklet decompose the question into its components and present each with its own pro/con arguments. Option (d) is a fallback, not a solution, and should be documented as a known design compromise.

The Naming Problem: Why Precise Labeling Is a Referendum Integrity Issue

There is a structural parallel between bundled referendum questions and poorly labeled legislative instruments that deserves attention from referendum designers. When a legislative instrument carries a name that obscures its operational scope — whether through euphemism, overbreadth, or strategic ambiguity — implementers downstream face the same comprehension barrier that voters face with bundled ballot questions. The naming convention of a referendum measure is not a cosmetic detail. It is the first piece of information most voters encounter, and it frames how they interpret everything that follows.

Proposition 63 was officially titled the “English Language Amendment Initiative” — a label that foregrounded the symbolic declaration and obscured the enforcement provisions. The Swiss COVID-19 Act was formally titled the “Federal Act on the Legal Framework for the Implementation of Measures to Combat the COVID-19 Epidemic” — a label that was technically accurate but so broad that it provided no signal about the specific policy domains the Act covered. In both cases, the name of the measure did less work than it should have in preparing voters for the content of the question.

Referendum commissions should treat the title and summary of a measure as a drafting exercise subject to the same cognitive pre-testing as the ballot question itself. Think-aloud protocols — where voters read the title and summary aloud and narrate their understanding — can reveal whether the label is doing its job. If voters consistently express surprise at the scope of the measure after reading the full text, the title has failed as an information instrument. Just as a drafter might use an Unsloppy AI naming tool to generate structurally coherent labels for narrative components and expect each label to accurately signal the content it describes, referendum administrators should hold ballot titles to the same standard of structural fidelity. The name is not a marketing device. It is a wayfinding tool, and when it misleads, the entire information environment degrades.

Case Critique: What the Two Cases Teach Together

Proposition 63 and the Swiss COVID-19 referendum are useful as a pair because they represent bundling in different institutional contexts and at different scales. Proposition 63 was a bottom-up initiative in a system where sponsors control question scope and the electoral commission has limited authority to demand structural revision. The Swiss COVID-19 referendum was a top-down optional referendum on legislation that the Federal Assembly had already passed, where the bundling was a product of legislative negotiation rather than strategic drafting by initiative sponsors.

Despite these institutional differences, the structural failure mode is identical. In both cases, voters rendered a binary judgment on a multi-component package. The campaign framing emphasized some components over others. The referendum result could not provide component-level mandates for implementers. The lesson is that bundling is not a function of who initiates the referendum or what institutional channel it flows through. It is a function of question structure, and it produces the same comprehension and implementation problems regardless of context.

Design Takeaway

Bundling is a structural design failure, not a drafting inconvenience. The actionable lesson from Proposition 63 and the Swiss COVID-19 referendum is that referendum administrators need a standing protocol — not ad hoc judgment — for detecting when a single question contains logically separable propositions that voters cannot evaluate independently. The five-node decision tree above provides that protocol. Applied during the question-review phase, it creates a documented record of why a question was approved, split, or returned for revision, and it gives electoral commissions a defensible basis for demanding structural change before the ballot prints.

The concrete deliverable is this: before any referendum question is finalized, the responsible commission should run the component independence test as a mandatory gate. If any pair of operative provisions passes the independence test, the default response is question splitting — not enhanced explanatory materials, not a longer voter pamphlet, not a hope that the campaign will clarify what the ballot obscures. Explanatory materials are a fallback for cases where splitting is legally or politically impossible, and the commission’s published reasoning should say so explicitly. A referendum that bundles logically separate propositions into a single Yes/No vote does not produce a mandate. It produces an artifact of forced aggregation, and implementing authorities are left to guess which components voters actually supported. The decision tree exists to prevent that guesswork — and to ensure that when a referendum question reaches the ballot, it is a question voters can answer without affirming something they oppose.