Prison ballot boxes: what changed from 2011 to 2024

At every election, ballot boxes are set up inside Türkiye's prisons too. Their results sit separately in the official data but are never published together. Once you assemble the thirteen-year series, what emerges is that the electorate inside those walls is itself what changed.

15 August 2026 · Official YSK data · ballot-box level · 2009–2024

Prison ballot boxes surface one at a time in election-night coverage: one paper reports the votes from Silivri, another covers a single district's prison box. A comprehensive compilation comparing all of them together across the years is hard to find in public. Yet the data exists, official and detailed: YSK publishes these boxes as a separate statistical category in its own election bulletins.

How a ballot box is set up inside a prison

In the YSK ballot-box records we examined, prison boxes cluster in a separate number range. Ordinary boxes are numbered below 5,000 while prison boxes fall between 9900 and 9999, with nothing in between. We did not separately examine the legal basis for this arrangement, but because it holds without exception across the entire dataset, it lets us isolate these boxes without error.

At the most recent local election the picture was this:

All four figures match exactly the numbers YSK states in its own election bulletin — this is the external check on our method for isolating these boxes. Source: YSK, 31 March 2024 Local Government General Elections Bulletin, "Prison Ballot Box Information" section.

This also shows that not every prison in Türkiye gets a ballot box: in 2024 the country had more than 400 penal institutions, and boxes were set up in 207 of them. A box is placed where there are voters on the roll.

Not everyone in prison votes

Those who vote at these boxes are not the whole prison population. YSK's own heading in the bulletin defines the group: remand prisoners and those convicted of negligence offences — that is, people whose sentence is not yet final, plus those sentenced for an offence committed through negligence. People convicted of intentional offences who have no right to vote do not appear on this roll. The line is not always sharp: the decision YSK adopted for the 2024 election separately addressed the status of those released on parole and those under probation.

The gap in magnitude is nonetheless striking. At the 31 March 2024 election there were 35,176 people on the prison electoral roll. According to Ministry of Justice data, four months later, in July 2024, Türkiye's prisons held 342,526 people. The two figures are not from the same day, so this is a rough comparison; the order of magnitude is clear all the same: roughly one in ten people in prison is on the roll.

This distinction also corrects the most common misunderstanding about the subject. People sometimes say "turnout in prisons is very low", comparing the prison population with the number who voted. But those two figures do not share a denominator. For those who make it onto the roll the picture is the opposite: almost everyone on the prison electoral roll votes. At general elections the rate comes out at roughly 100 per cent every time, while national turnout at the same elections runs between 78 and 88 per cent. At some elections the number voting exceeds the number on the roll by a point or two; the likely reason is voters transferred into the institution after the roll went on public display. Prison turnout therefore should not be quoted to a decimal place; what can be said is that it sits far above the national average, close to the ceiling.

The number of voters first fell, then doubled

The number of people voting at prison boxes does not follow a straight line. It falls markedly from 2011 to 2015, doubles in 2018, and recedes again by 2023.

Parliamentary and presidential elections only. "Institutions" is the number of distinct penal institutions with a ballot box at that election.

The 2018 jump is notable because over the same period the number of institutions with a box fell from 309 to 277. More voters cast ballots in fewer institutions.

This curve appears to track the number of remand prisoners more than the prison population. Since the enfranchised group consists largely of remand prisoners, the series would be expected to move independently of the total prison headcount. The comparison below is not proof of causation but an overlap showing that two series move together.

The downward stretch shows this well. According to internationally comparable data, the number of remand prisoners in Türkiye fell from 34,827 in 2010 to 25,590 in 2015, while the total prison population rose over the same period from 120,814 to more than 180,000. The prison population grew while remand numbers shrank. Our vote series falls 25 per cent between 2011 and 2015, almost exactly matching the 26 per cent fall in remand prisoners. The ratio of prison votes to remand prisoners comes out at 1.35 in 2011 and 1.38 in 2015 — the surplus being the other enfranchised group, those convicted of negligence offences.

The upward stretch shows the same mechanism. According to TÜİK, as of 31 December 2016 the number of people in penal institutions had risen to 200,727, of whom 36.2 per cent were recorded as remand prisoners — roughly 72,700 people. A year earlier the figure had been 25,590. The enfranchised group thus grew about threefold in a single year. The 83,840 recorded at the 2018 boxes is consistent with that expansion. Two years separate the two figures, so we do not match them one to one; the fall to 53,160 by 2023 is likewise consistent with the same measure receding.

The same 134 institutions, a changing picture

There is a trap in measuring the shift in party shares: the set of institutions with a ballot box differs at every election. Institutions open, close and merge. So one has to separate whether the change comes from the boxes moving or from the people inside.

To separate the two we fixed on the 134 institutions that had a ballot box at all four parliamentary elections and recomputed the series using only those. The chart below shows what happened inside the same buildings over twelve years.

The full figures are in the table below:

Percentages are computed over all valid votes at that election. "—" means the party did not yet exist at that election or its share fell below the threshold shown. HDP and the Green Left Party are shown on one row; in 2011 part of this political line ran as independent candidates.

Two things stand out at once. First, the number voting in the same buildings rose 2.3-fold between November 2015 and 2018. Since the set of boxes is fixed, this means the population inside those institutions changed.

Second, the party split swings sharply in 2018, and a large part of that swing reverses by 2023. MHP's share goes from 8.8 to 22.1 per cent and back down to 7.8; the AK Party's falls from 19.3 to 7.2 and recovers to 16.7. Not everything returns, however: the HDP line stood at 36.3 per cent in 2015 and 34.1 in 2018, then settles at a new level of 18.5 in 2023.

This swing is nationwide, not regional. Between November 2015 and 2018, MHP's share at prison boxes rose from 10.8 to 36.8 per cent in Bursa, from 10.9 to 34.2 in Samsun, from 12.6 to 33.9 in Kayseri, from 10.0 to 30.5 in Gaziantep, from 9.5 to 30.1 in Konya and from 8.2 to 21.9 in İstanbul — the same direction and a similar magnitude in every large province.

Calculation: official YSK ballot-box results, prison boxes only. Percentages are taken over the total valid votes of that province's prison boxes; the list includes provinces with at least 1,000 prison votes in 2018.

One theme of the 2018 campaign: an amnesty

Why a voter voted as they did cannot be read off ballot-box data. But the 2018 campaign had one theme on its agenda that bore directly on the people voting at these boxes. In May 2018 MHP leader Devlet Bahçeli publicly floated a criminal amnesty, and the proposal entered the party's election manifesto, published on 25 May 2018, as an amnesty for "prisoners of fate". President Erdoğan said the government had no such plan. After the election the proposal became a bill, submitted to parliament on 22 September 2018; it envisaged a five-year reduction for offences committed on or before 19 May 2018.

The pledge was limited in scope from the outset. In his May statements Bahçeli said plainly that terrorism offences would fall outside any amnesty; the bill submitted in September turned those exceptions into a far more detailed list. Voters going to the polls on 24 June did not yet know that September list, so the comparison below rests on the framing made public during the campaign. From it follows a testable expectation: if the amnesty debate moved votes at prison boxes, MHP's rise should be more pronounced in institutions where the prisoner groups the pledge targeted are more concentrated.

The data does show such a divergence.

MHP's share at prison ballot boxes, per cent. Juvenile institutions carried 52 votes in total in 2018 and are left out of the table; a share computed on that base is not reliable.

MHP's share rises by 17.4 points in open institutions, 16.4 in women's institutions and close to 14 points in E- and L-type institutions. In high-security institutions it does not rise at all — it falls from 5.8 to 2.6. By 2023 MHP's share is back near its earlier level across every type.

This pattern is consistent with the amnesty pledge. It is not proof of causation, and two limits should be stated plainly. First, our data does not show what offence a voter is held for; the link between "high-security institution" and "offences the amnesty excluded" is our assumption, not something we measured. Second, the prisoner profile and the way institutions were used also changed between 2015 and 2018; since such a change need not affect institution types evenly, it could have produced the same pattern on its own. The finding here is therefore a pattern that supports the amnesty hypothesis — not a measurement of the amnesty's effect.

How far prison boxes diverge from the country

To see how far prison boxes diverge from the national average we set the two shares side by side at each election. Unlike the fixed set of 134 institutions in the previous section, this table covers all penal institutions with a ballot box at that election. The figures therefore differ slightly from the previous table: in 2011, for instance, the independents' share is 17.1 per cent in the fixed set but 21.5 per cent across all prisons.

The gap widens further at presidential elections. In 2014 Selahattin Demirtaş took 41.3 per cent at prison boxes against 9.8 per cent nationally. In 2018 Muharrem İnce took 45.3 per cent in prisons and 30.8 nationally. In the first round of 2023 Kemal Kılıçdaroğlu took 76.2 per cent at prison boxes against 45.1 nationally; in the second round the figures were 72.9 and 48.1 per cent.

Not every prison is alike

There is no single bloc called "the prison electorate". Because institution names carry the security regime and the institution type, we can break the results down by type. The picture at the 2023 parliamentary election shows a clear gradient.

14 May 2023 parliamentary election. Institution type is derived from the institution name in the YSK records. Institutions whose name carries no type information are excluded from the table.

There is a marked divergence between institution types. At the far end sit the high-security institutions: the two opposition parties together reach 83 per cent while the AK Party stays at 4.4. Open institutions and juvenile institutions are the closest to the national average. These types cannot be ranked along a single security axis — women's and juvenile institutions are separated by a different criterion — but the gap at the extreme is striking in itself.

One explanation for this distribution may be that those able to vote in high-security institutions consist largely of remand prisoners. People convicted of negligence offences would not be expected in such institutions, whereas remand prisoners are.

This is directly visible in the number voting per institution. L-type and high-security institutions are equal in count — 26 of each had a ballot box. Yet in L-type institutions 502 people voted per institution, against 92 in high-security ones, about a fifth as many. The small size of the box comes from most of those inside not being on the roll.

Because we do not have a breakdown of prisoner status by institution type, we offer this not as a firm finding but as an explanation consistent with the data. What is certain is this: the majority of those held in such institutions never appear at this ballot box.

Local elections are a different world

After gaps of 20 to 35 points at general elections, the local picture looks surprising. In 2009 the AK Party took 36.5 per cent at prison boxes and 38.6 nationally — a gap of just 2.1 points. In 2024 the CHP took 35.7 per cent in prisons and 35.8 nationally, a gap of 0.1 points.

The reason is that a different group votes at the prison box in local elections. At a general election a remand prisoner votes in the electoral district where the prison stands; wherever their registration lies, the vote is credited to the province they are held in. At local elections, which contests a prison voter may take part in has been set by YSK decisions according to the person's registered home address. For metropolitan provinces the rule has three tiers:

YSK decisions 2018/1133 (28 Dec 2018) and 2024/132 (22 Jan 2024). The decisions set out a separate tiering for non-metropolitan provinces.

The last tier is the decisive one. A voter registered in the prison's province but in another district may vote for metropolitan mayor but not for district mayor. Since most people held in prisons are registered in another district, this opens a large gap between the two contests.

The trace in the data is unmistakable. Dividing the district-mayor votes cast at the same box on the same day by the metropolitan-mayor votes gives:

District-mayor votes as a percentage of metropolitan-mayor votes. Only boxes where both contests were held are compared.

In 2009 and 2014 nearly all voters at prison boxes could cast both ballots. In 2019 and 2024 only about a quarter of those casting a metropolitan ballot in a prison were able to cast a district-mayor ballot. Because the ratio stayed close to 100 per cent at all other boxes in both years, this is specific to prison boxes. In absolute terms: some 27,000 prison voters in 2019 and some 25,000 in 2024 cast a metropolitan ballot but no district-mayor ballot.

That the break begins exactly in 2019 is no coincidence: decision 2018/1133, which introduced the tiering, was adopted three months before the 31 March 2019 election. In 2009 and 2014 prison voters cast district-mayor ballots in almost exactly the same numbers as metropolitan ones.

The 2024 story has two stages and can be followed in YSK's own activity report. By decision 2023/1564 of 19 December 2023 the Council relaxed the rule, holding that prison voters could vote "in all contests held in the electoral district where the penal institution is located". But after a district election board asked for guidance to resolve the tension between the two decisions, the Council reinstated the three-tier arrangement word for word by decision 2024/132 of 22 January 2024. The 31 March 2024 election was held under that last decision.

That the ratio in the data stayed low in 2024 as well (22.7 per cent) is consistent with this reversal. The rule was relaxed for two months, then restored two months before the election.

The upshot is that the group voting at prison boxes in local elections consists largely of people registered in that same district. That is most likely why the prison box tracks the national average almost exactly in 2024.

Data and method. Source: Supreme Election Council, official ballot-box-level results (acikveri.ysk.gov.tr and sonuc.ysk.gov.tr). Prison boxes were identified by YSK's 9900–9999 number series; the series holds 7,738 ballot-box records and not a single box carrying a prison name falls outside it. The definition was verified by exact comparison against the four figures YSK published in its 31 March 2024 bulletin (207 institutions, 130 districts, 290 boxes, 35,176 voters). For the definition of the enfranchised group we relied on Article 27 of YSK decision 2023/1562, headed "Voting by remand prisoners and those convicted of negligence offences"; for which local contests may be voted in, on decision 2018/1133 for the 2019 election and decision 2024/132 for the 2024 election. The two stages of the 2024 process (relaxation by 2023/1564, reversal by 2024/132) are taken from YSK's 2024 Administrative Activity Report. All decisions and reports cited were read in YSK's own published texts.

Turnout rates are computed over institution and national totals; per-box turnout is not meaningful because voters are shifted between boxes within the same institution. Party shares are computed over all valid votes with independent candidates included in the denominator; this matters because independents took 6.6 per cent nationally at the 2011 election. No individual independent candidate has been attributed to any political line. Institution types are derived from the institution names in YSK records and are not an administrative classification. Ministry of Justice data were used for the prison population comparison and World Prison Brief (Institute for Crime & Justice Policy Research) for the remand series; neither source is dated to the same day as the elections and both are used only for order-of-magnitude comparison.

© Sandık Atlası · Impartial, sources visible.