Machine-only voting: who wins and who loses

Published: 2026-07-21

An interactive what-if on Bulgaria's own election data: remove paper ballots in every large polling section, and watch national vote share and parliamentary seats recompute live.

This article is interactive. On the live page you can set a section-size threshold and a turnout drop-off, and the vote-share bars, the parliamentary seat allocation and a district-by-district map recompute instantly against the most recent election. What follows is the method and the headline findings.

Since 2021 every Bulgarian parliamentary section has recorded machine votes and paper votes side by side — the same electorate, on the same day, split only by the medium each voter chose. That split is a natural experiment: within a single section, machine-voters and paper-voters picked parties in measurably different proportions.

This lets us ask a concrete reform question. What if paper ballots were removed in every "large" section — above some number of registered voters — forcing those voters onto the machine?

This is not merely hypothetical. On 7 July 2026 the governing party, Progressive Bulgaria (ПрБ), submitted a bill to parliament proposing exactly this — fully machine voting, with paper kept only in sections under 300 voters (plus mobile, hospital and ship sections) — as reported here. The interactive tool tests what that reform would have done to the most recent result; ПрБ's own bill draws the line at 300.

The model

For each section above the chosen threshold that also has real machine votes, we treat the observed machine-voters as revealing that section's preference, and recast its paper-voters onto that same machine-vote distribution:

machineShare_p = machineVotes_p / Σ machineVotes
paperShare_p   = paperVotes_p   / Σ paperVotes
votes_p(d)     = machineVotes_p + (1 − d) · (paperTotal · machineShare_p + invalid · paperShare_p)

The turnout dial d is the share of a section's paper-voters who abstain rather than switch to a machine. At d = 0 turnout is held constant; at d = 100% every paper-voter in a large section stays home and only the original machine-voters count.

A machine also won't accept a spoiled ballot, so the model recovers each affected section's invalid paper ballots (about 3.9% of paper in 2026) and adds them as valid votes, distributed by that section's paper-vote shares — because the voters who mis-mark ballots belong to the same older, paper-preferring demographic (Fujiwara 2015). This feeds the paper-heavy parties and softens the swing (at the default it turns ПП-ДБ's seat gain from +12 to +11 and ГЕРБ-СДС's loss from −12 to −11).

Smaller sections — and the few large sections whose machine failed — are left exactly as recorded. We then re-aggregate nationally and run a Hare-quota seat allocation (4% threshold) on both the actual and the modelled totals.

What we already excluded

Three parliamentary elections (July 2021, November 2021, October 2022) were fully machine-only — they have no paper votes to redistribute, so they are outside this exercise. The five elections in this analysis (April 2021, April 2023, June 2024, October 2024, April 2026) all had real voter choice between the two media; the live tool runs the scenario on the most recent of them.

The headline finding

The direction is remarkably stable across every election: machine voting favours urban/reform parties, paper voting favours ГЕРБ and ДПС. Forcing large sections onto the machine consistently:

• cuts ГЕРБ-СДС by roughly 4–6 points and ДПС by 1–3 points;
• lifts ПП-ДБ by roughly 4–6 points, plus smaller gains for reform/protest parties;
• in 2023 and 2026 it flips the second-place finisher (ПП-ДБ overtakes ГЕРБ); in October 2024 it knocks a party below the 4% line.

This direction is backed by a separate real-data analysis: in a recent election machine voting was chosen by 80% of ИТН, 79% of ПП-ДБ and 70% of Възраждане voters, while ГЕРБ, ДПС and БСП voters leaned to paper (Radio Free Europe).

The likely reason is who each group is: the machine is preferred by younger, urban voters, paper by older voters and people in smaller places — so removing paper amplifies the urban/reform vote. By district, ПП-ДБ's gain is biggest in Sofia and Plovdiv (up 5–7 points), and there is exactly one winner flip — Sofia's 23rd district, from ПрБ to ПП-ДБ. The regional map colours each district by its projected winner.

What this is not

• The model combines two channels — the composition (selection) effect (from which medium each voter chose) and the spoiled-ballot recovery above. It still is not the full effect of mandating machines, which would also shift turnout and carry the machine's own influence on the vote.
• It assumes paper-voters would vote like their section's machine-voters — it captures the selection skew between the groups. The machine's on-screen flow nudges behaviour (the explicit "I support no one" button is picked by 1.8% of machine-voters vs 1.3% on paper) and can create errors of its own (Zucco & Nicolau, 2016).
• The spoiled-ballot recovery rests on an assumption: recovered invalid ballots are distributed by each section's paper-vote shares because mis-marking voters resemble the paper-preferring demographic (Fujiwara 2015), but a spoiled ballot carries no recorded preference, so the true split is unknowable.
• The natural experiment is only as strong as machine adoption. In April 2021 machine was a freely-chosen minority (~29% of affected voters), so its voters skew young/urban/early-adopter — read that election as the most aggressive extrapolation, not the most reliable.
• The turnout drop-off is a scenario dial, not an estimate. In Bulgaria's three machine-only elections (2021–2022) turnout was 39–42%, within the range of the mixed elections around them (34–51%), so a large machine-driven turnout collapse is not visible in the data — though repeat-election fatigue confounds the comparison. Small drop-offs are the more likely case.
• Seats are allocated by the national Hare-Niemeyer quota with the 4% threshold — the same method Bulgaria uses to fix each party's national total (it reproduces the official result for every election since 2013). Bulgaria then spreads those totals across the 31 districts, which only changes where seats land, not how many each party wins; the meaningful figure is the actual-vs-model difference.

Further reading

• Fujiwara, T. (2015). Voting Technology, Political Responsiveness, and Infant Health: Evidence from Brazil. Econometrica 83(2).
• Zucco, C. & Nicolau, J. (2016). Trading old errors for new errors? The impact of electronic voting technology on party label votes in Brazil. Electoral Studies 43.

Data: per-section protocols from the Central Election Commission, as processed by electionsbg.com. This is an analytical scenario, not a forecast.