In brief

How to read this

A municipality (opština) is a unit of local government; Serbia without Kosovo has 167, counting the city municipalities of Belgrade, Niš and other cities. A settlement (naselje) is a census place: a village, a small town or a city, 4,721 in all. A polling station has a list of addresses assigned to it; one school often hosts several, and then they form one polling place.

The type of area comes from GHS-SMOD, the European Commission's (JRC) classification built from satellite-based population data: every 1×1 km square belongs to one of seven classes, from urban centre to very low-density rural. A station takes the class of the square it stands in.

Shares here are of valid votes, so they run 0.5–0.7 points above the official ones: RIK counts against all ballots found in the box. "pp" means percentage points, the difference between two shares. "Vučić's list" is "Aleksandar Vučić – Ujedinjena Srbija" in 2026 and "Aleksandar Vučić – Srbija ne sme da stane" in 2023.

Turnout decided it: Vučić's list has nearly the same votes, the Student list 2.6 times what the main opposition had in 2023

This is the study's main answer, and it needs no map. In 2023 Vučić's list won 1.78 million votes, in 2026 1.73 million: minus 52,000, or 3%. 624,000 more ballots were cast, and the opposition, which ran as eight lists in 2023, ended up almost entirely in one. These are differences of totals: aggregate results cannot say which voters came for the first time, which switched lists and which stayed home.

What this means in plain words. Vučić's list did not lose because it shed votes: it has almost as many. It lost because 624,000 more ballots were cast and the opposition vote gathered in one list. Who came for the first time, who switched lists and who stayed home cannot be seen from stations; that takes surveys.

Who won where: cities and the north for the students, the rural south and east for Vučić

The Student list came first in 90 of 167 municipalities of Serbia without Kosovo, Vučić's list in 73, and minority lists in four: Kanjiža (Alliance of Vojvodina Hungarians), Tutin (SDA of Sandžak), Preševo and Bujanovac (Albanian Coalition). The winner here is the list with the most votes, not necessarily a majority.

By region the picture looks like north against south: Belgrade 63.6% for the students against 32.0% for Vučić's list, Vojvodina 53.0 against 40.1, Šumadija and Western Serbia 51.5 against 40.5. Southern and Eastern Serbia is the only region where Vučić's list leads, and only narrowly, 46.4 against 45.0. The station map shows that this lead rests on villages. In the towns of the south and east the students are ahead: 63.3% in Niš, 58.9% in Zaječar, 55.4% in Pirot, 53.2% in Leskovac, and in Vranje it is almost a tie, 48.1 against 46.7.

Outside the core territory people voted very differently. The 93 stations abroad gave the Student list 72.8% at a turnout of 77.9%. At the 60 stations for voters from Kosovo and Metohija, Vučić's list took 73.8% at a turnout of 35.5%. At 29 stations in prisons turnout was 90.5% and Vučić's list took 66.4%.

Land and people: two thirds of the land for Vučić's list, but a quarter of the voters

There are no polling-station boundaries in open data: by law the local election commissions set the stations on the proposal of municipal administrations, and a station is defined by a list of streets and house numbers, not a polygon (Articles 56 and 57 of the Law on the Election of Members of Parliament). To draw stations as areas, each polling place gets the inhabited land closer to it than to any other place of the same municipality. These are modelled catchments, not official boundaries.

Minorities: the Hungarian list lost half its votes, the Albanian one grew

In three parts of the country voting works differently, because Serbia's largest minorities live there and have their own lists: Hungarians in northern Bačka, Bosniaks in Sandžak, Albanians in the Preševo valley. These lists have a special status. In 2026 nine of the 18 lists received it, but only five have a real territorial base: the Alliance of Vojvodina Hungarians, three Bosniak lists (SDA of Sandžak, Zukorlić's SPP and Ljajić's "Ljudski") and the Albanian Coalition. The other four lists with minority status (Russian Party, Russian–Serbian Unity Party, the Bulić coalition and "Mother Serbia – Gruda") won 0.35% between them. Minority-list status is a legal category: ODIHR writes that RIK granted it to every list nominated by parties registered as minority parties, on formal registration and without assessing whether the lists met the substantive criteria. So three things are kept apart here: the status of a list, the territory where it collects votes, and the minority's share in the census.

A minority list's share in a municipality closely follows that minority's share in the 2022 census: across the 26 municipalities where Hungarians make up at least 5% the rank correlation is 0.95, and across the six Sandžak municipalities with at least 5% Bosniaks and Muslims it is 0.94. This is aggregates lining up on a map, not data on how people of a given ethnicity voted: Hungarians are 83% of Kanjiža's population, the Hungarian list won 40% of its votes, and the map cannot tell who voted for the other lists.

Vučić's list gained ground since 2023 in exactly three municipalities, Kanjiža, Ada and Bačka Topola, and all three are Hungarian. In Novi Pazar it went the other way, losing 21.9 points, more than almost anywhere else: from 41.4% to 19.5%, and the Student list came first there.

How votes become seats: one nationwide constituency, so winning a municipality earns no seat

Winning a municipality does not give a list an MP from that municipality. All of Serbia is a single constituency.

The rules are in Articles 5, 128, 129 and 140 of the Law on the Election of Members of Parliament (Official Gazette 14/2022, amended by 47/2026). RIK adopted the final report on 5 November and allocated the seats on 6 November 2026: Student list 140 seats (2,354,417 votes), Vučić's list 103 (1,731,323), Alliance of Vojvodina Hungarians 2 (35,461), Albanian Coalition 2 (27,683), Ljajić's SDP–CDP ("Ljudski") 1 (20,638), SDA of Sandžak 1 (12,596), Zukorlić's SPP 1 (12,440); the repeat vote of 1 November moved one seat from the Student list to SPP. All five minority lists with seats stayed below 3%, so they entered under the special rules: no threshold and quotients raised by 35%.

This is an important difference from countries with single-member districts. The urban concentration of votes that costs a party seats in the United States (the main argument of Jonathan Rodden's "Why Cities Lose", 2019) costs nothing in Serbia: a vote in Belgrade and a vote in a village weigh the same. The city–village divide here is about who lives where and how they vote, not about how districts are drawn.

City and village: the smaller the place, the stronger Vučić's list

The spread inside each group is wide, but the middle halves do not overlap. The middle half of stations in urban centres gave the students 60.0% to 70.9%, in the emptiest villages 27.3% to 48.6%. Resampling whole municipalities many times gives a 95% interval of 62.4–66.3% for urban centres and 34.5–40.5% for the emptiest villages. The other 16 lists together take 5.5% in urban centres and about 9.4% in villages.

The next question is whether there is a threshold. Taken one settlement at a time, the transition turns out to be smooth, with no step, and there is no single threshold for the country.

In the 3,010 villages under 2,000 people the students took 40.1%, and they won a majority in only 13.8% of them.

One caveat applies to the whole section. The type of area and the settlement are taken at the point where the station stands, not where its voters live. For a village these are almost the same thing: the station sits in the local school or community hall and the village votes there. In a large city a school in the centre can serve surrounding neighbourhoods too, so the urban classes are somewhat blurred. That is why "stations in urban centres" is more precise than "city dwellers": the shares describe stations, not residents.

The town–village split is not new: it existed in 2000 too, only on the challengers' side it has since become three times steeper.

Turnout: the city–village gap in turnout is far smaller than the gap in the vote

Turnout is measured against the voter register, and observers had doubts about it. The register closed on 9 October with 6,450,523 voters; on 23 October RIK set the final number at 6,451,292, because the law allows changes until 72 hours before election day. The register audit was not completed before the election (CRTA, ODIHR). Eastern Serbia has a long history of labour emigration; if some of the people on its register live abroad, turnout there is understated, but open data cannot test this.

The repeat vote took place on 1 November at 53 stations in 24 municipalities: at 31 of them the 25 October result could not be determined, at 22 the vote was annulled (RIK decision of 28 October). These stations have 24,438 voters; 10,978 came, or 44.9%. The students took 42.7% there and Vučić's list 43.8%, and 39 of the 53 stations stand in rural squares. The RIK export does not flag these stations; here they are matched from RIK's decision and shown as a separate layer on the interactive map.

Since 2023 Vučić's list lost almost everywhere, and by similar margins

The results of the election of 17 December 2023 were collected for all 8,273 stations from the same source, RIK, and the totals match the official ones. The comparison is between lists that held the same place in politics, not between the same organisations. Vučić's list was called "Srbija ne sme da stane" in 2023 and "Ujedinjena Srbija" in 2026. The main opposition list in 2023 was the coalition "Srbija protiv nasilja", in 2026 the Student list, and these are different structures.

A station counts as the same if the municipality and number or a unique name match and the number of registered voters differs by no more than 15% (the rule podatak.rs uses). This gives 6,295 pairs out of 8,392 stations. The right-hand panel shows 5,882 of them, those with a coordinate, since the type of area cannot be set without one. Comparable stations vote like all stations of their class: the share of Vučić's list among them differs from the whole class by at most 0.6 points, the students' share by at most 1.3.

The opposition share grew in both cities and villages. At comparable stations "Srbija protiv nasilja" took 35.3% in urban centres and 12.3% in the emptiest villages in 2023; the Student list took 64.1% and 39.0% in 2026. That is growth of 1.8 times in cities and 3.2 times in villages, while the gap between them in percentage points barely moved: 23 and 25.

In absolute numbers Vučić's list lost almost nothing: 1.73 million votes against 1.78 million in 2023, minus 3%. At comparable stations its votes fell by 2–8% in every type of area, while the main opposition list's votes grew 2.1 times in urban centres and 3.4 times in the emptiest villages. By the totals of list groups it adds up like this: 624,000 more ballots cast; 555,000 votes of eight small 2023 lists that all but vanished in 2026 (NADA, "Mi – glas iz naroda", Zavetnici, SRS and others together won 81,000 instead of 636,000); 155,000 lost by Dačić's SPS; 52,000 for Vučić's list, 51,000 from fewer invalid ballots and 15,000 for minority lists.

The share of Vučić's list fell by about the same amount everywhere, yet the mechanism differed. In the quarter of municipalities where turnout rose most (by 13.7 points on average), its vote count actually grew by 1%, but its share fell by 8.2 points: more ballots were added than votes it gained. In the quarter where turnout barely rose (+1.6 points), its votes fell by 15% and its share by the same 7.4 points. One outcome, two routes, but aggregates show only this much: which people came for the first time, switched lists or stayed home cannot be told from stations; that takes surveys.

Vučić's list lost most in Žabalj (−23.9 points), Novi Pazar (−21.9) and Bački Petrovac (−17.8). The biggest swing was abroad: 57.1% for Vučić's list in 2023 and 24.9% in 2026. But the people voting abroad were not the same: registered voters abroad rose from 39,270 to 102,147, or 2.6 times, and the number of stations from 81 to 93. This is a change in who votes as much as a change in preferences. Kosovo did not move: 74.0% and 73.8%.

Then and now: September 2000 resembles 2026, but as a historical analogy

The last time those in power lost an election in Serbia was in 2000. The comparison is a historical analogy, not the same kind of election: on 24 September 2000 the vote was federal, for the president of the FRY and the Chamber of Republics, not for Serbia's parliament, and it was surrounded by a dispute over the count, street mobilisation and 5 October. Milošević's SPS–JUL coalition took 1.48 million votes in Serbia, 60,000 more than in 1997, but 4.70 million people voted instead of 4.13 million, and 18 opposition parties ran as one coalition, DOS, gathering 2.09 million; turnout in Serbia was 73.4%. The arithmetic of 2026 is similar: Vučić's list 1.73 million against 1.78 million, 624,000 more ballots, a single Student list with 2.35 million, turnout 68.9%.

A model of place: a place's profile predicts about three quarters of the differences between stations

The question behind the model is simple. If all you know about a place is its profile (settlement size, density around it, education and employment in its municipality, what is nearby), how well can you guess how it voted?

First, what the model leaves out. Of 5,427 polling places with a precise coordinate, 4,673 with at least 150 registered voters went into it; 754 places, or 978 stations, were excluded: they hold 70,000 voters, about 1% of the register, 99% of them in rural classes, and Vučić's list took 60.7% there. How they vote is covered at the end of this section, but it has to be kept in mind from the start: the model describes Serbia without its smallest villages.

Each of the 4,673 places has 60 traits: 34 from the 2022 census by municipality and settlement, and 26 about surroundings, namely the GHS-SMOD type of area, voters and stations within 1 and 3 km, and shops, schools, health facilities, stops, banks and other OpenStreetMap objects within 500 m and the distance to the nearest of each. The model is a random forest, a set of 400 decision trees, and every prediction comes from a model that has not seen that place. The 4,673 places are not 4,673 independent census observations: the 34 census traits are identical for all places of one of the 161 units, and only the surroundings and settlement traits differ between places. To draw the pattern inside cities, the territory is split into H3 hexagons of about 0.1 km², each taking the result of its nearest polling place.

R² is the share of the differences between places that the model explains. A random split flatters the model, because neighbouring places in one municipality are alike and the model recognises them. Holding out whole municipalities and districts is fairer: the model then predicts about three quarters of the differences (R² 0.74, mean error 5.3 points) with municipalities held out and about two thirds (0.67; 5.8 points) with districts held out. This is predictive accuracy, not an explanation of why people vote as they do. Census data by municipality transfer to municipalities the model has not seen; what stays outside the model is mostly the difference between neighbourhoods inside a city, which Serbia's census does not record.

The strongest predictors are settlement population, unemployment in the municipality and the number of voters within 3 km. Part of the link with the census runs through urbanisation. Once the urban share of the population and density are held constant, the link between the students' share and higher education falls from 0.58 to 0.24, with primary education or less from −0.67 to −0.37, and the link with the share of people over 65 disappears (from −0.26 to +0.11, within chance). Unemployment holds: −0.58 and −0.45. In other words, the link with age is almost entirely accounted for by urbanisation, while the link with unemployment holds at the same level of urbanisation. Feature importance in a random forest is not a causal explanation: population, density, voters nearby and the number of shops are closely related, and the importance is shared among them.

Votes cluster strongly in space: neighbouring municipalities resemble each other (Moran's I for the students' share is 0.61). The model explains almost all of this clustering: neighbouring polling places have an index of 0.73 for the students' share and 0.03 for the model's errors. At the municipality level the errors still cluster slightly (0.18), so some regional feature remains outside the traits.

What the model leaves out: 978 tiny stations where Vučić's list is even stronger

By law a station is meant for 100 to 2,500 voters; fewer than 100 is allowed only where another station is too far away or the terrain is difficult (Article 57). Serbia without Kosovo has 526 such stations, and 978 with fewer than 150 voters: these are the mountain and remote villages, mostly in the south and east.

Statistical tests: no stuffing-type anomalies in favour of Vučić's list found in 2026

Polling-station results can be searched for statistical traces of padded protocols and stuffed ballots. The best-known approach is Sergey Shpilkin's method, which he has applied to Russian elections for many years, including with Dmitry Kobak and Maxim Pshenichnikov. The method looks for an unusual pairing of high turnout and a raised share for one list relative to a comparison group. But turnout and preferences can be linked without any irregularity: cities and small villages both vote and turn out differently, so the result depends on how the comparison group is formed and how comparable the stations are. Stuffed ballots and rewritten protocols raise both the turnout of a station and the share of one list at once. That list then gets a heavy "tail" at stations with abnormally high turnout, and the protocols get too many round percentages.

This is not a test of the election's integrity as a whole and not an assessment of every kind of irregularity: only four statistical signs at the level of polling stations are examined. They are applied to the 8,210 stations of Serbia without Kosovo, the diaspora and prisons (these groups have their own turnout and their own choice), separately for 2026 and 2023.

2023: there was a tail, but it rested on whole municipalities

The naive estimate for the whole country gave Vučić's list an excess of 93,000 votes, 66,000 within regions, and the excess disappeared when stations were compared within a municipality (−44,000). So the 2023 tail comes from whole municipalities where both turnout and the share of Vučić's list are higher. That can be an honest difference (small villages in the south and east vote together and almost everyone turns out) or irregularities that cover a whole municipality. Station-level statistics cannot tell these apart.

The threshold below which stations count as the reference changes the size of the estimate but not its sign: with the threshold of the paper by Kobak and co-authors (the reference is the lowest-turnout stations holding 20% of the ballots) it comes to −208,000 for 2026 and +29,000 for 2023. The authors themselves call such an estimate an upper bound. There were 16 stations with 100% turnout in 2023 and five in 2026, and neither group leans to one list.

A check of the method: it "finds" stuffing for the opposition too

The same procedure gives the opposition an excess of 107,000 to 285,000 votes in 2026. That is not stuffing: within municipalities, stations with higher turnout voted more for the Student list. Shpilkin's method picks up any link between turnout and choice, so honest mobilisation looks the same as stuffing in it. Harvey (2016) showed this on Russia's 2011 election, and Rozenas (2017) showed that round shares appear by themselves at small stations. The method's conclusions hold only when the excess is large for one list and absent for the others.

What the tests cannot see

• Vote buying and pressure on voters. These are real votes with a signature and a ballot; in the protocols they look like any other votes.
• Carousel voting and voting in the name of people who left, from an inflated register, as long as a station's turnout does not stand out from its neighbours.
• Ballots taken away. At 577 stations there were fewer ballots in the box than signatures (2,661 ballots): this happens in carousel voting, but also with ordinary mistakes by the board.

ODIHR writes precisely about what such statistics do not measure: numerous credible reports of pressure on public-sector employees, attempts to buy the votes of Roma and other vulnerable groups, group voting at 7% of observed stations. The day after the election CRTA judged that systemic manipulation at polling stations had largely been kept in check, unlike in 2023.

Why the election was early and how it went

The preliminary statement of the OSCE/ODIHR mission of 26 October found that equality of opportunity in the campaign was seriously compromised. It describes misuse of state resources, pressure on voters, candidates and civil society, bias in the monitored media, a lack of confidence in the accuracy of the voter register, shortcomings in counting and some cases of physical violence. Independent access for voters with physical disabilities was not provided at 61% of the 1,485 polling stations where the mission observed voting. These findings do not change the results of this study and are not directly linked to them; they show where electoral geography could go next, for instance into the accessibility of polling stations themselves.

What the two main lists promised

A full programme text was found only for the Student list; the promises of Vučić's list are collected from campaign speeches.

Both sides made promises to the countryside, and the results still diverged: promises do not explain why villages vote differently from cities. Sources: Mašina, 15.09.2026; RTS and Telegraf, 20.10.2026; the English Wikipedia articles "Student List", "United Serbia (2026 coalition)" and "2026 Serbian parliamentary election".

Limitations: places, not people, and unofficial coordinates

• These are places, not people. A station's share describes those who came to vote; census traits describe everyone living in the municipality. That a list is stronger where more people have degrees does not mean those people voted for it.
• Association, not cause. Correlations and feature importance show what goes along with a vote, not why people vote as they do.
• A station stands in one place and its voters live in another. Type of area and settlement are taken at the station's point; in big cities this blurs the classes.
• Hexagons add no data. All hexagons of one polling place repeat its result; the fine pattern inside a city is a way of drawing, not 84,000 observations.
• The census is coarse and not new: 161 units for 6.6 million people, data from 2022, an election in 2026.
• Station coordinates are unofficial. Of 8,210 stations in Serbia without Kosovo, 7,823 (95.3%) have a coordinate; 100 of those fell in another municipality, so the maps show 7,723 stations (94.1%); 7,499 are precise enough for the model, which makes 4,673 polling places of at least 150 voters. A spot check of 20 stations found all of them where their address says, but no single point can be guaranteed.
• The voter register has not been independently audited, so turnout is a share of the register, not of the voters actually living in the country.
• Status of the data: RIK results per polling station as of 4 November 2026, after the repeat vote of 1 November; this is the freeze date of the analysis. RIK's final report of 5 November (75th session) and the seat allocation of 6 November (76th session) match these data to the vote.

What follows

For researchers in electoral geography: Serbia's city–village divide is a smooth gradient of settlement size and density, not capital against province. Municipal census data transfer to municipalities a model has not seen, but explaining differences inside cities needs a census by neighbourhood, which does not exist.

For observers such as CRTA and CeSID: 6,295 comparable pairs of stations for 2023–2026 and the map of model residuals show where results depart from the profile of a place. These are directions for further checking, not a list of irregularities and not a ranking of suspicious stations. The statistical tests find no stuffing-type anomalies in 2026; vote buying and pressure are not measured by them and need observers on the ground. A separate task is the accessibility of polling stations: according to ODIHR, 61% of observed stations offered no independent access to people with physical disabilities.

For journalists, three checkable numbers: 64.6% against 37.7% by type of area, a 27-point gap in the vote with 3 points in turnout, and the absence of a single threshold: the students lead from 2,000 people around Belgrade and only from 12,000 in the south and east. The station zones on the map are modelled, and the caption has to say so.

What stronger conclusions would need: official boundaries for polling stations (they can be rebuilt from the address register, as is done in France), a census by neighbourhood, and surveys linked to place that could separate the effect of a place from the make-up of its residents.

Earlier work on the topic

By 4 November at least five teams had mapped the results by municipality, so here municipalities get one panel, and the step further is type of area, settlement size, station catchments, minorities and a model of place.

Benjamin Forest, in his review "Electoral geography: From mapping votes to representing power" (Geography Compass, 2018), divides the field into three parts: where and how people vote, how place affects voting, and how votes turn into representation. This study is the first part with a step into the second. Luke Temple ("Reckoning with the digital turn in electoral geography", Progress in Human Geography, 2023) describes exactly this kind of turn: open station-level results, volunteer geocoding, population rasters and OSM instead of surveys. The closest academic work for the region is Sven Hojski's master's thesis on Croatia's 2020 election (PMF Zagreb, 2022). Bursać and Vučićević analysed the 2023 election as political scientists ("Electioneering strategies in the hybrid regime", 2025), without a spatial analysis. The third part, the geography of representation, has been studied for Serbia by Dejan Bursać (Politička revija, 2023): where the members of the National Assembly come from over nine elections since 2000 under the single constituency, which districts are under-represented and why. A search found no peer-reviewed electoral geography of Serbia at the level of polling stations.

Data and method

The station totals match RIK's results to the vote: 2,354,417 votes for the Student list and 1,731,323 for Vučić's list, 4,441,636 ballots in the box, 4,385,065 valid, turnout 68.89% (4,444,310 of 6,451,292). The totals were checked twice, in pandas and separately in SQL with DuckDB. Every station coordinate was checked against the polygon of its municipality and settlement. The same figures stand in RIK's final report of 5 November 2026 (75th session); the seats were allocated on 6 November (76th session).

The comparison with 2023 uses comparable stations (same municipality and number or a unique name, registered voters within ±15%). The 7,000-people line comes from a smoothed curve (LOWESS) of the students' share against the logarithm of settlement population, with an interval from bootstrapping settlements. Partial correlations are computed on ranks across 161 census units, holding the urban share of the population and density constant. Moran's I uses municipality contiguity (shared border) and the 8 nearest polling places.

Ballot-stuffing tests. Votes by turnout and the estimate of "anomalous" votes follow a variant of Shpilkin's method (Kobak, Shpilkin, Pshenichnikov, 2012): stations with turnout no higher than the peak (within a region or municipality, no higher than the median) are the reference, and elsewhere the leader's votes are compared with its ratio to the other lists at the reference stations. The fingerprints are a two-dimensional histogram of turnout and the leader's share, weighted by voters (Klimek et al., 2012). Whole percentages are counted as in Kobak, Shpilkin and Pshenichnikov (2016): the number of stations whose turnout lies within 0.05 points of a whole percent is compared with 400 random samples in which each station's number of voters who voted is redrawn at the same number of registered voters; stations with fewer than 100 voters are left out. The protocol check compares ballots in the box with signatures in the voter list.

Tools: Python (pandas, geopandas, H3, scikit-learn, statsmodels, PySAL), DuckDB for the cross-check, matplotlib for the figures and Leaflet for the interactive map.

Sources the text refers to:

• RIK, results by polling station: https://www.rik.parlament.gov.rs/542645/rezultati/
• RIK, decision on the repeat vote at 53 stations, 28.10.2026: https://www.rik.parlament.gov.rs/extfile/sr/files/additionalDocuments/817/835/Re%C5%A1enje%20o%20sprovo%C4%91enju%20ponovnog%20glasanja.pdf
• RIK, 66th session, final number of voters, 23.10.2026: https://www.rik.parlament.gov.rs/vest/809141/66-sednica-republicke-izborne-komisije.php
• RIK, 75th session, final report on the results and seats by list, 05.11.2026: https://www.rik.parlament.gov.rs/vest/822434/75-sednica-republicke-izborne-komisije.php
• RIK, 76th session, decision on the allocation of seats, 06.11.2026: https://www.rik.parlament.gov.rs/vest/822560/76-sednica-republicke-izborne-komisije.php
• RIK, 73rd session, voting abroad, 29.10.2026: https://www.rik.parlament.gov.rs/vest/818648/73-sednica-republicke-izborne-komisije.php
• Government of Serbia, results from all 8,392 stations, 02.11.2026: https://www.srbija.gov.rs/vest/1010157/rik-objavio-rezultate-sa-svih-8392-birackih-mesta.php
• Law on the Election of Members of Parliament (text on RIK's site): https://www.rik.parlament.gov.rs/extfile/sr/41907/Zakon%20o%20izboru%20narodnih%20poslanika%20lat.pdf
• ODIHR, Statement of Preliminary Findings and Conclusions, 26.10.2026: https://odihr.osce.org/odihr/669165
• CRTA, campaign report (register closed at 6,450,523), 21.10.2026: https://crta.rs/wp-content/uploads/2026/10/EOM3-SR-izvestaj_final.pdf
• CRTA, repeat vote and seats, 02.11.2026: https://crta.plus/updates/repeat-vote-at-53-polling-stations-brings-one-seat-to-zukorlics-list/
• JRC, Global Human Settlement Layer: https://human-settlement.emergency.copernicus.eu/
• RZS, census 2022 database: https://data.stat.gov.rs
• RZS, "Izbori za narodne poslanike Narodne skupštine Republike Srbije, održani 23.12.2000. i 10.01.2001." (Belgrade, 2001), tables 2.1–2.2 by municipality: https://pod2.stat.gov.rs/ObjavljenePublikacije/G2001/Pdf/G20016002a.pdf
• arhivaizbora.rs, parliamentary elections of 1997 and 2000 by municipality (2000 cross-checked against the RZS publication to the vote): https://arhivaizbora.rs/teritorija/RS/srbija/izbori/par2000, https://arhivaizbora.rs/teritorija/RS/srbija/izbori/par1997
• Federal Election Commission, results of 24.09.2000 ("Službeni list SRJ" 50, 55 and 56/2000) as reprinted by CeSID, "Oko izbora 4" (Belgrade, 2000), pp. 60–81: http://www.cesid.rs/pdfovi/OKO%20IZBORA%204.pdf
• Goati V., Izbori u Srbiji i Crnoj Gori od 1990. do 2013. i u SRJ od 1992. do 2003., CeSID, Belgrade, 2013: https://www.cesid.rs/images/1372054311_Vladimir%20Goati_knjiga.pdf
• RIK, archive of the 1997 and 2000 elections: https://arhiva.rik.parlament.gov.rs/arhiva-izbori-za-narodne-poslanike-2000.php
• Forest B., Electoral geography: From mapping votes to representing power, Geography Compass, 2018, doi:10.1111/gec3.12352
• Temple L., Reckoning with the digital turn in electoral geography, Progress in Human Geography, 2023, doi:10.1177/03091325231170328
• Rodden J., Why Cities Lose: The Deep Roots of the Urban-Rural Political Divide, Basic Books, 2019
• Kobak D., Shpilkin S., Pshenichnikov M., Statistical anomalies in 2011–2012 Russian elections revealed by 2D correlation analysis, 2012, arXiv:1205.0741
• Kobak D., Shpilkin S., Pshenichnikov M., Integer percentages as electoral falsification fingerprints, The Annals of Applied Statistics, 2016, 10(1), 54–73, doi:10.1214/16-AOAS904
• Klimek P., Yegorov Y., Hanel R., Thurner S., Statistical detection of systematic election irregularities, PNAS, 2012, 109(41), 16469–16473, doi:10.1073/pnas.1210722109
• Harvey C. J., Changes in the menu of manipulation: Electoral fraud, ballot stuffing, and voter pressure in the 2011 Russian election, Electoral Studies, 2016, 41, 105–117, doi:10.1016/j.electstud.2015.11.004
• Rozenas A., Detecting election fraud from irregularities in vote-share distributions, Political Analysis, 2017, 25(1), 41–56, doi:10.1017/pan.2016.9
• Kobak D., Shpilkin S., data and code on Russian elections: https://github.com/dkobak/elections
• Kovačić Z., Statistička analiza krađe na izborima: Kosjerić i Zaječar, Radar, 08.07.2025: https://radar.nova.rs/politika/statistika-kosjeric-i-zajecar-lokalni-izbori/
• CRTA on election day, N1, 26.10.2026: https://n1info.rs/english/news/crtas-nedeljkov-massive-turnout-proves-serbian-government-can-be-voted-out-despite-uneven-playing-field/
• Bursać D., Faktori teritorijalne reprezentativnosti u zatvorenom proporcionalnom izbornom sistemu: slučaj Srbije, Politička revija, 2023, 77(3), doi:10.5937/pr77-44022
• Bursać, Vučićević, Electioneering strategies in the hybrid regime: Another regular snap election in Serbia, 2025, doi:10.1177/2336825X241308429