From votes to seats: the Bundestag seat law
Abstract. We implement the Bundestag seat allocation under the 2023 electoral law as modified by the Constitutional Court on 30 July 2024: the 5% threshold with the three-Wahlkreis and minority exceptions, 630 seats by Sainte-Laguë/Schepers, Land lists, and Wahlkreis winners seated in order of Erststimmen share until the party's Land seats run out. Given the official 2025 votes, the calculation reproduces every party's seats, all 81 Land lists and the seated or unseated status of all 299 Wahlkreis winners exactly. National polls reach the Länder and Wahlkreise by proportional swing from 2025, a projection that has not yet been backtested.
Summary
This paper describes how Polling Forecast turns national vote shares into a Bundestag. The calculation has two halves. The first is the law: the rules under which the 23 February 2025 election was held, which are those of the 2023 reform of the Federal Electoral Act as modified by the Federal Constitutional Court on 30 July 2024. Our implementation of those rules, given the official 2025 votes, reproduces the official result exactly: the seats of every party, the seats of all 81 party lists by Land, and the seated or unseated status of every one of the 299 Wahlkreis winners. The second half is a projection: from today's national polls to Zweitstimmen in each Land and Erststimmen in each Wahlkreis. That half is ours, it is a simple proportional swing from 2025, and it has not been tested against a past election in the way our British seat model was.
The law is where German seats are won and lost. A party at 4.9% of the Zweitstimmen gets nothing, and at 5.0% it gets about thirty-five seats. A party can win a Wahlkreis and not take the seat. Both effects are sharp, and both are modelled exactly.
What has changed since this paper was written
Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.
- The map (November). The map can show every Bundestag election since 1998 on the Wahlkreis boundaries of its day, and any census figure; each census figure also has its own page (census maps). It can also replay a past polling miss ("The polls miss again") or take a vote you set.
Two votes, one allocation
Each voter has two votes. The Erststimme (first vote) goes to a candidate in the voter's Wahlkreis, of which there are 299; the candidate with the most votes wins the Wahlkreis. The Zweitstimme (second vote) goes to a party's list in the voter's Land. The Zweitstimme decides how many seats each party gets. The Erststimme decides, within that number, which of the party's candidates sit for a Wahlkreis rather than from the list.
Before 2023 a party that won more Wahlkreise in a Land than its Zweitstimmen entitled it to kept the extra seats (overhang seats), and the other parties received compensation seats so that the whole chamber stayed proportional. The Bundestag grew well beyond its nominal size as a result. The 2023 reform fixed the size at 630 and removed both overhang and compensation. The price is that a Wahlkreis winner is no longer guaranteed a seat.
The rules are in §§ 1, 4, 5 and 6 of the Bundeswahlgesetz (BWG), as amended by the Act of 8 June 2023. The Constitutional Court's judgment of 30 July 2024 (2 BvF 1/23) upheld the new allocation, including Zweitstimmendeckung, but held the 5% threshold, in the form the reform gave it, incompatible with the Basic Law. Until Parliament legislates, the threshold applies with the condition that a party below 5% still qualifies if its candidates win at least three Wahlkreise, the rule the reform had tried to abolish. This is the law under which the 2025 election was held, and it is the law we apply to the next one. If Parliament changes it before 2029, the calculation will have to change too.
Step 1: which parties qualify
Let \(Z_p\) be party \(p\)'s national Zweitstimmen and \(Z = \sum_q Z_q\) the valid Zweitstimmen of all parties. Let \(W_p\) be the number of Wahlkreise its candidates win. Party \(p\) takes part in the allocation if
where \(\mathcal{M}\) is the set of parties of national minorities, in practice the SSW, the party of the Danish and Frisian minorities in Schleswig-Holstein. Winning three Wahlkreise is enough; the winners do not have to be seated.
In 2025 seven parties qualified: the CDU, the AfD, the SPD, the Greens, the Linke, the CSU and the SSW. The CSU cleared both routes, with 5.97% of the Zweitstimmen and 47 Wahlkreise. The SSW qualified through the exemption alone, on 0.15%. Two parties fell short: the BSW on 4.981%, about 9,500 votes below the line out of 49.6 million valid votes, and the FDP on 4.33%. Neither won a Wahlkreis. Altogether 13.7% of the Zweitstimmen went to parties that won no seats.
The three-Wahlkreis route last mattered in 2021, under the earlier law, when the Linke won 4.9% and three Wahlkreise and so took seats in proportion to its whole vote. The CSU is the party for which it could matter next. Because the CSU stands only in Bavaria, its national share is roughly its Bavarian share times Bavaria's weight in the electorate, and in 2025 it was only about one point above the line. Under our projection the CSU's share moves in proportion to the Union's, so a Union polling below roughly 24% nationally puts the CSU under 5% on its own. It would still qualify as long as it won three Bavarian Wahlkreise, which it does comfortably in today's projection. The calculation checks both routes in every run.
Step 2: seats by party (Oberverteilung)
The 630 seats, less any won by independent candidates (§ 6(2)), are divided among the qualifying parties in proportion to their national Zweitstimmen, by the Sainte-Laguë/Schepers method (§ 5). In its divisor form, one finds a divisor \(D\) such that
where "round" is standard rounding (a fraction of exactly one half goes either way, and is settled by lot if more than one allocation fits). The same allocation is produced by the highest-averages form: give seats one at a time to the party with the largest quotient
where \(s_p\) is the number of seats it has so far. Our code uses equation (3), which needs no search for a divisor. Sainte-Laguë treats large and small parties even-handedly, unlike the D'Hondt method, which favours large parties; Balinski and Young (1982) and Pukelsheim (2017) give the theory.
The 2025 allocation shows how it works. The qualifying parties had 42,833,356 Zweitstimmen between them, about 67,989 per seat. Any divisor between 68,063.064 and 68,138.575 gives the official result; the Federal Returning Officer published 68,100.
| Party | Zweitstimmen | Quotient at \(D\) = 68,100 | Seats |
|---|---|---|---|
| CDU | 11,196,374 | 164.411 | 164 |
| AfD | 10,328,780 | 151.671 | 152 |
| SPD | 8,149,124 | 119.664 | 120 |
| Greens | 5,762,380 | 84.616 | 85 |
| Linke | 4,356,532 | 63.973 | 64 |
| CSU | 2,964,028 | 43.525 | 44 |
| SSW | 76,138 | 1.118 | 1 |
| Total | 42,833,356 | 630 |
Table: Oberverteilung, Bundestag 2025.
The law has one further rule here, the majority clause (§ 4(4)): if a party wins more than half the Zweitstimmen of all qualifying parties but not more than half the seats, it receives extra seats until it has one more than half, and the chamber grows accordingly. It was not triggered in 2025, but the code applies it.
Step 3: seats by Land list (Unterverteilung)
Each party's seats are then divided among its Land lists in proportion to its Zweitstimmen in each Land, by the same method. The CSU, which stands only in Bavaria, has one list and all 44 of its seats are Bavarian. The CDU's 164 were spread over the other fifteen Länder. In 2025 the seven qualifying parties had 81 Land lists between them.
Step 4: who takes the seats (Zweitstimmendeckung)
This is the step the 2023 reform added. In each Land, a party's Wahlkreis winners are ranked by their share of the Erststimmen in their own Wahlkreis, highest first:
where \(E_{k,c}\) is the Erststimmen of candidate \(c\) in Wahlkreis \(k\). They take seats in that order until the party's Land seats from step 3 are used up. Winners below the cut are not elected, and the Wahlkreis has no member of its own. Any Land seats left over go to the party's list in list order. Winners from a party that failed step 1 get no seat at all.
The ranking is by share, not by margin. A candidate who won with 30% in a fragmented Wahlkreis ranks below one who won with 35% elsewhere, however narrow the second win.
In 2025 the rule left 23 Wahlkreise without a member. Every case is in the table.
| Party | Land | Land seats | Wahlkreise won | Not seated | Lowest seated share | Highest unseated share |
|---|---|---|---|---|---|---|
| CDU | Baden-Württemberg | 29 | 35 | 6 | 34.98 | 34.42 |
| CDU | Hessen | 15 | 20 | 5 | 30.43 | 30.35 |
| CDU | Rheinland-Pfalz | 11 | 14 | 3 | 32.27 | 30.80 |
| CDU | Schleswig-Holstein | 8 | 9 | 1 | 31.76 | 26.51 |
| CSU | Bayern | 44 | 47 | 3 | 32.43 | 31.05 |
| AfD | Brandenburg | 8 | 9 | 1 | 33.61 | 30.84 |
| AfD | Mecklenburg-Vorpommern | 5 | 6 | 1 | 35.85 | 26.80 |
| AfD | Sachsen | 14 | 15 | 1 | 29.42 | 24.99 |
| AfD | Sachsen-Anhalt | 7 | 8 | 1 | 32.23 | 30.64 |
| SPD | Bremen | 1 | 2 | 1 | 30.27 | 25.20 |
Table: Zweitstimmendeckung in 2025. Shares are the winner's share of the Erststimmen in its Wahlkreis, in percent, from our calculation on the official votes.
The cut can be very fine. In Hessen the CDU's fifteenth-ranked winner had 30.43% of the Erststimmen and took a seat; its sixteenth had 30.35% and did not. The pattern is also telling. Unseated winners belong to parties that win more Wahlkreise in a Land than their list vote there covers: in 2025 the CDU and CSU in the south and west, the AfD in the east. A party whose Wahlkreis wins run ahead of its Zweitstimmen is the one that loses members this way.
What the calculation leaves out
Four details of the law are not modelled, and none arose in 2025.
Independent winners. A candidate standing without a party list who wins a Wahlkreis is elected outright, and the 630 is reduced by one (§ 6(2)). The Zweitstimmen of that candidate's voters are then left out of step 2. Our data group independents with the minor parties, and the published aggregates do not identify their voters' second votes, so neither can be done exactly. There were no such winners in 2025.
Ties. The law settles exact ties by lot, at every step. Our calculation does not; with shares computed to many decimal places, an exact tie in a projection is vanishingly unlikely.
Exhausted lists. If a party's list has fewer candidates than its list seats, the seats stay empty (§ 6(4)). List lengths are not in our data. It did not happen in 2025.
The majority clause in practice. It is implemented but has never been triggered.
The check: 2025 exactly
A seat calculation that cannot reproduce the last election cannot be trusted with the next. We wrote the calculation twice: first as a reference implementation in Python, using exact fractions throughout and returning the admissible divisor range, then as the C# code that runs on the site (BundestagSeats.cs). Both were given the official 2025 Zweitstimmen by party and Land and the official Erststimmen by party and Wahlkreis, from the Federal Returning Officer's open data, and compared with the official outcome.
Both match it exactly. Every party's seats match (CDU 164, AfD 152, SPD 120, Greens 85, Linke 64, CSU 44, SSW 1). All 81 party-by-Land seat counts match. All 299 Wahlkreis winners match, and so does the seated or unseated status of every one of them, giving the same 23 unseated winners. The reference implementation's divisor range for step 2 is identical to the one in the Federal Returning Officer's published calculation. The C# check runs as part of the site's data pipeline and reports any mismatch.
This check covers the law, not the forecast. It says that given the right votes, we produce the right Bundestag. It says nothing about whether our projected votes are right, which is the subject of the rest of this paper. It also applies only to 2025: the elections of 2005 to 2021 were held under the earlier law, with overhang and compensation seats, and our calculation does not reproduce them.
From national polls to Länder and Wahlkreise
The seat law needs Zweitstimmen by party and Land and Erststimmen by candidate and Wahlkreis. Polls give neither. They give national shares for the Union, the AfD, the SPD, the Greens, the Linke, the BSW, the FDP and "others" (see Forecasting the Bundestag: an overview for the average). The projection fills the gap by proportional swing from 2025.
Ratios
Let \(S_p\) be today's national share for party \(p\) and \(B_p\) its 2025 national Zweitstimmen share. Others is whatever the named parties leave, with a floor: \(S_{\text{oth}} = \max(0.5,\ 100 - \sum_p S_p)\). Each party's swing ratio is
The SSW keeps a ratio of 1: it is not polled nationally, and it stands only in Schleswig-Holstein.
Länder
Let \(b_{l,p}\) be party \(p\)'s 2025 Zweitstimmen share in Land \(l\), with the CDU and CSU counted together as the Union. The projected share is
and the projected Zweitstimmen are \(z_{l,p}/100\) times the Land's 2025 valid Zweitstimmen, so turnout in each Land is held at its 2025 level. The Union's projected votes are then given to the CSU in Bavaria and to the CDU everywhere else.
Wahlkreise
Let \(E_{k,c}\) be the 2025 Erststimmen of candidate \(c\) in Wahlkreis \(k\), and \(g(c)\) the candidate's party group (the Union for the CDU and CSU, the party itself for the other named parties, others for everyone else). The projected Erststimmen are
the winner is the candidate with the largest \(\hat{E}_{k,c}\), and the winner's share for the ranking in equation (4) is \(\hat{E}_{k,c} / \sum_{c'} \hat{E}_{k,c'}\). There is no separate model of the Erststimme: each party's first votes move in proportion to its national second votes. In the simulation each \(\hat{E}_{k,c}\) is also multiplied by a small random factor (Simulating German polling error: where the chances come from).
A worked illustration
Take round numbers. Suppose a party won 20% of the Zweitstimmen nationally in 2025 and polls 28% today, so \(r = 1.4\). In a Land where it won 35%, the unscaled product is \(35 \times 1.4 = 49\). The other parties' products depend on their own ratios; if, together, they had 65% and their ratios average 0.85, their products sum to about 55, and the party's projected Land share is \(49 / (49 + 55) \approx 47\%\). In a Land where it won 10%, the same arithmetic with the other parties at 90% gives \(14 / (14 + 76.5) \approx 15\%\). Its gain is about 12 points where it was strong and between 5 and 6 where it was weak. Its Wahlkreis candidates gain in the same proportion.
That is the character of proportional swing. A party that is gaining gains most, in points, where it is already strong, and a party that is losing loses most where it was strong. No share can fall below zero, which is the main reason to prefer it to uniform swing (adding the same number of points everywhere) in a system with seven parties of very uneven regional strength. Under uniform swing, a party losing eight points nationally would be pushed below zero in every Land where it had less than eight.
Why this matters for the seats
Two parts of the law depend on regional detail and not only on national shares. The Unterverteilung depends on each party's Land shares, and Zweitstimmendeckung depends on each Wahlkreis. The second is where the projection has most effect. As a party gains ground in its strongholds, it wins more Wahlkreise there, often with shares well below its list vote, and whether its Land seats cover them becomes a live question. On today's projection, 22 projected Wahlkreis winners would not be seated. The Wahlkreise page lists every projected winner, its margin and whether it is seated, and the Länder page shows each party's seats by Land with the change from 2025.
How good is the projection?
We do not know as well as we would like. The British seat model was chosen by a backtest: starting from the notional 2019 result on the 2024 boundaries, given the true 2024 nation shares, uniform swing called more seats correctly than proportional swing or a blend (From votes to Commons seats: UK swing models and their test). No equivalent test has yet been run for Germany. The data for one exist: the Federal Returning Officer has recalculated the 2021 result onto the 2025 Wahlkreise, so one could start from 2021, apply the true 2025 national shares by equation (5) to (7), and count how many Wahlkreis winners, how many unseated winners and how many Land seats came out right. Until that is done, the projection is a reasonable default, not a tested choice.
Some weaknesses are clear without a test.
East and West. Proportional swing assumes that every party's vote moves by the same proportion everywhere. Swings can differ between the eastern and western Länder. Regional Bundestag polls exist, most of them splitting East from West, but there are too few to build a regional layer, so none is used.
The Erststimme. Candidates, incumbency and tactical voting for the first vote all differ from Wahlkreis to Wahlkreis. Equation (7) moves every candidate by their party's national change in second votes and nothing else. Since the 2023 law, voters and parties may also behave differently in Wahlkreise where the winner may not be seated; 2025 is the only election held under that law, so there is nothing to learn such an effect from.
The CSU. The CSU is tied to the Union's national movement. If Bavaria moves differently from the rest of Germany, the CSU's seats and its qualification will be misjudged.
Turnout. Each Land keeps its 2025 turnout. A change in the balance of turnout between Länder moves a little weight between Land lists but does not change national seat totals, which depend only on national shares.
New parties. A party that did not stand in 2025 has no Land or Wahlkreis pattern to scale. The BSW stood in 2025 and has one; a new party would be spread in proportion to the 2025 "others".
The same engine everywhere
One function, GermanyForecast.Run, takes a set of national shares and returns the full Bundestag: Land Zweitstimmen, Wahlkreis Erststimmen, qualifying parties, seats, Land lists and every Wahlkreis winner, seated or not. The projection, the published polls, every past-miss lens (If we're wrong: replaying past Bundestag polling misses on today's polls), the reader's own error on the what-if page, the reader's own shares on the prediction page, each of the 2,000 simulated elections and each run of the polling-day forecast all go through it. Differences between any two of them are therefore differences in the vote shares they start from, never in how votes become seats.
Limitations
The legal half is as good as the check: exact for 2025, with the four omissions listed above. The projection half is the weak point. It is untested against a past election, it has no regional layer, it treats the first vote as a copy of the second, and it ties the CSU to the CDU. It is also built on a single past result, 2025, which was itself an unusual election: early, with a new party near the threshold and the first under the new law. What we don't know yet: limitations of the German forecast collects these points with the rest of the German forecast's weaknesses.
References
Balinski, M. L. and Young, H. P. (1982). Fair Representation: Meeting the Ideal of One Man, One Vote. New Haven: Yale University Press.
Bundesverfassungsgericht, judgment of 30 July 2024, 2 BvF 1/23.
Bundeswahlgesetz (BWG), as amended by the Act of 8 June 2023 (BGBl. 2023 I Nr. 147), §§ 1, 4, 5 and 6.
Die Bundeswahlleiterin (2025). Endgültige Sitzberechnung und Verteilung der Mandate, Bundestagswahl 2025. Wiesbaden.
Pukelsheim, F. (2017). Proportional Representation: Apportionment Methods and Their Applications, 2nd edition. Cham: Springer.
How to cite
Polling Forecast (2026). From votes to seats: the Bundestag seat law. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/germany-seat-law. Live figures change as new polls arrive; the date above is when the text was last revised.



