Reading Time: 5 minutes

1. Introduction: Ada, the referendum question, and the political context

Ada is a municipality in the north of Serbia, in Vojvodina, and one of the local features that matters politically is its nationally mixed population, with a large share of ethnic Hungarians. That does not mechanically determine electoral or referendum outcomes, but it does mean that local political behavior is often shaped by a more specific social and institutional context than one finds in a more homogeneous municipality.

The referendum held in Ada on 7 June 2026 asked voters to decide on the introduction or continuation of local self-contribution. This was therefore not a party election in the strict sense, but a vote on a concrete local public issue with direct material implications. Even so, such referendums rarely exist outside politics. They are organized in an institutional setting, they are advocated or opposed by actors with political weight, and their outcomes can send signals that go beyond the narrow legal wording of the question.

That broader context matters in Ada because the local ruling bloc had, at least in formal electoral terms, entered this referendum from a position of strength. On the latest local elections, the ruling list associated with Aleksandar Vučić and local leadership in Ada won a strong result. If the same local governing structure publicly stood behind the “Yes” option in this referendum, then the fact that the municipality voted predominantly “No” cannot be dismissed as politically irrelevant. It does not automatically mean that every “No” vote was an anti-government vote. Referendums are often more issue-specific than elections. But it does raise the question of whether a local ruling structure that had demonstrated strong electoral capacity failed to translate that strength into majority support on a concrete policy question.

This is the point at which a polling-station analysis becomes useful. The aggregate result already tells us who won the referendum. But it does not tell us whether that result was uniform across the municipality, whether it was driven by a small number of stations, or whether there are meaningful internal differences worth discussing. That is the main purpose of this two-part post.

2. Referendum forensics is not the same as election forensics

Before turning to the numbers, one methodological clarification is essential. Referendum forensics is not the same thing as election forensics in the usual sense. In a multiparty election, analysts often examine competition across parties or candidates, vote concentration, turnout-vote relationships, and the distribution of support across several political actors. A referendum has a much simpler formal structure: here the central split is between “Yes” and “No”, with invalid ballots as a much smaller third category.

That difference matters for interpretation. In a referendum, the aim of a forensic-style analysis is not necessarily to prove manipulation. On a small dataset such as this one, that would be neither methodologically realistic nor analytically responsible. The main goal is more modest and, in some ways, more useful: to summarize the result clearly, to identify internal patterns, and to offer a sensible interpretation of how the result was produced across polling stations.

In that sense, referendum forensics here is best understood as a structured descriptive and exploratory exercise. It asks whether the internal distribution of the result is smooth or uneven, whether some stations stand out sharply, whether stations with objections look different from those without them, and whether turnout appears to be associated with one side of the referendum more than the other. Those are meaningful questions even when the dataset is too small for large-scale forensic claims.

3. What this small dataset can and cannot show

The Ada dataset contains results from 21 polling stations. That is enough for a serious descriptive analysis, but not enough for a “full” election-forensics exercise of the kind used in larger elections. On a sample of this size, a small number of observations can meaningfully affect correlations, rankings, and even the visual impression produced by a graph.

That does not make the dataset weak. It simply defines the correct scope of the analysis. What this dataset can show reasonably well is the basic descriptive structure of the referendum, the spread of turnout and Yes/No results across polling stations, the relationship between objections and selected indicators, and the presence of unusually high or low stations. What it cannot do, at least not responsibly, is bear the weight of strong inferential claims by itself.

This is why the analysis in this post and the next one is intentionally limited to descriptive statistics, exploratory relationships, simple group comparisons, and outlier detection. That is not a limitation to be apologetic about. It is the right level of ambition for the evidence at hand.

4. Basic descriptive picture of the referendum

The aggregate result was clear. Out of 14,925 registered voters, 4,899 voted. That means that turnout, while not extremely high, was substantial enough to make the referendum politically meaningful. Of the ballots cast, 4,874 were valid and only 25 were invalid. The share of invalid ballots was therefore very small.

Table 1. Detailed descriptive statistics

Statistic	Value
Number of polling stations	21
Total registered	14925
Total votes	4899
Total valid	4874
Total invalid	25
Total Yes	1334
Total No	3540
Total number of people who voted outside the polling stations	21
Total received help	7
Average turnout	33.72
Median turnout	32.83
Standard deviation of turnout	10.51
Minimum turnout	3.48
Maximum turnout	60.50
Average percentage of valid votes Yes	27.79
Median percentage of valid votes No	26.14
Standard deviation of the percentage of valid votes Yes	10.96
Average percentage of valid votes No	72.21
Median percentage of valid votes No	73.86
Standard deviation of the percentage of valid votes No	10.96
Mean invalid vote rate	0.43
Median value of invalid vote rate	0.42
The mean value of the rate of votes outside the polling station	0.41
The median value of the vote rate outside the polling station	0.27
Mean rate of votes that received assistance	0.19
Median value of the rate of votes that received help	0.00
The mean value of the ratio of Yes and No votes	0.42
Median value of the ratio of Yes and No votes	0.35
Ratio of Yes votes to valid votes	0.27
Ratio of no votes to valid votes voices	0.73
Ratio of invalid ballots to total votes	0.01

Among valid votes, 1,334 were cast for “Yes” and 3,540 for “No”. Put in relative terms, that means “Yes” received 27.37% of valid votes and “No” received 72.63%. The ratio between the two sides was therefore strongly tilted toward “No”. This is not a close referendum result. It is a clearly one-sided one.

Still, the aggregate result should not be confused with uniformity. The average polling-station turnout was 33.72%, but the range was wide. Some stations had far lower participation, while others approached or exceeded the upper half of the turnout scale. The average “Yes” share across polling stations was 27.79%, but again with visible dispersion. So even before one goes deeper into the dataset, the basic descriptive picture already suggests that the municipality-wide result was produced through a non-uniform local structure.

Table 2. Key descriptive statistics

Statistic	Value
The ratio of Yes to No	37.7%
Ratio Yes to all valid	27.4%
Ratio of No to all valid	72.6%
The ratio of invalids to all casted	0.5%

5. Does the dominant “No” send a political message?

At the most cautious level, the answer is yes. A municipality in which the local ruling structure had recently shown strong electoral strength did not produce majority support for the referendum proposal if that proposal was associated with the local governing centre. That, by itself, is politically meaningful.

The stronger claim would be that the referendum result directly measures support for or opposition to the ruling party. That claim would go too far. Referendums are often shaped by issue-specific considerations: material burden, local trust, understanding of the proposal, and even fatigue or skepticism about local decision-making. A “No” vote may therefore express opposition to the proposal without necessarily expressing stable opposition to the party structure behind it.

But even with that caution, it would be analytically weak to deny the signal entirely. If the ruling side supported the “Yes” option, then a municipality-wide result dominated by “No” does send a message: electoral strength at one moment does not guarantee policy legitimacy at another. In that limited but important sense, the referendum result can be read as a warning sign for local power-holders.

6. Conclusion of Part I

Part I establishes three things. First, the referendum result in Ada was clearly dominated by “No”. Second, that result emerged in a municipality with a specific local social and political profile, including a nationally mixed population and a recent history of strong local performance by the ruling bloc. Third, the referendum result should not be reduced either to a pure technical outcome or to a simple party election in disguise. It is something in between: a policy vote with clear political implications. The next part moves from this aggregate picture to the internal structure of the result. It asks how turnout, objections, invalid ballots, outside voting, and unusually behaving polling stations help explain the way this referendum result was actually produced.