How did the February 6 earthquake affect voter turnout?

Three months after the earthquake of February 6, 2023, Türkiye went to the polls to elect its president. Data from 972 districts shows where turnout fell in the earthquake zone and why: the decline tracks the scale of displacement far more closely than the severity of destruction.

July 19, 2026 · Official YSK and TÜİK data · province/district level

On the morning of February 6, 2023, two major earthquakes nine hours apart struck eleven provinces; the first official assessments in March 2023 put the number of heavily damaged buildings at 179 thousand, a figure that rose above 200 thousand as inspections were completed, and millions of people left their homes. Three months later, on May 14, ballot boxes were set up in all of these provinces. What happened to turnout in the region that day? This article looks for the answer. It should be said at the outset that the search did not go as planned: publicly available damage data ran out at a certain point, and a more robust measure had to take its place. Because that process also shaped the conclusion, we describe at the end of the article what we examined and what we chose not to use.

A first look at the province level: a gap, but not proof

Let us start with the crudest comparison. From 2018 to 2023, the turnout rate fell by an average of 1.4 points in the 11 provinces declared part of the earthquake zone, while it rose by 0.4 points in the remaining 70. That is a gap of roughly 1.8 points, and the top of the list is made up of earthquake provinces: Malatya lost 5.9 points, Hatay 5.6, Adıyaman 4.4 and Kahramanmaraş 4.0.

Yet this table alone does not justify the claim that the earthquake lowered turnout. Near the top of the same list sit provinces that experienced no earthquake at all: Ağrı fell by 5.6 points, Van and Muş by 4.2 each. The eastern provinces clearly saw a turnout decline unrelated to the earthquake, and anyone looking only at the province level cannot separate the earthquake's effect from that broader pattern. The variation within the 11 provinces is also wide: while Malatya lost 5.9 points, turnout in Kilis and Osmaniye — both on the official list — actually rose by 1.6 points. The "earthquake zone" is, in the end, an administrative list; to see the real distribution of destruction one has to go below the province line, down to the districts.

Gaziantep: a comparison within a single province

At the district level the picture sharpens, and a single province offers an instructive comparison. Two of Gaziantep's districts — Nurdağı and İslahiye, sitting on the fault line — were devastated. The province's central districts, Şehitkâmil and Şahinbey, and Nizip to the east remained relatively intact. These districts of course differ in size, income and voter profile; but they share the common conditions of the same province, and the sharpest difference between them is what they lived through on February 6. The result:

Across the earthquake zone as a whole, the steepest declines cluster in the places known to have suffered the worst destruction: Antakya alone lost 13.0 points — a collapse seen in no other district in Türkiye at this election. It is followed by Hatay's Kırıkhan (−8.8) and Defne (−8.6), Malatya's central district Yeşilyurt (−8.2), central Adıyaman (−7.8) and Elbistan (−7.3), the epicenter of the second earthquake. The epicenters of both earthquakes appear on the list.

Chasing the damage data: what we found and what we could not

The natural question at this point: does the loss of turnout line up, district by district, with the severity of destruction? Measuring that required official damage-assessment data, and the search proved instructive. Neighborhood-level damage data was never published; the Ministry announced its final assessments only "on site", through posted lists. Final district-level figures could be obtained for Hatay alone — and even those came not from a data portal but from news records of the period, verified against two independent sources. The figures circulating as district breakdowns for the other provinces are the incomplete first assessments of early February 2023: they give 8.6 thousand for Kahramanmaraş, where the final count exceeds 35 thousand. Building an analysis on a count missing roughly three quarters of the total would have been misleading; we did not use it.

The calculation we ran on Hatay's 15 districts confirmed the relationship: as the number of collapsed buildings per thousand voters rises, so does the loss of turnout. But the strength of that relationship rests largely on a single observation, Antakya, and a sample of fifteen cannot carry an article on its own. Counting buildings also has a structural flaw: a two-story village house and a six-story apartment block in a city center are each "one building", yet they do not displace the same number of households. The dwelling-unit data that would correct this distortion was never published at the district level. In short, the damage data took us part of the way and then ran out.

From counting damage to counting people

So we turned the question around. The earthquake affects an election not because buildings collapse, but because the people living in them leave. Rather than buildings, then, let us measure the movement of people: TÜİK's Address-Based Population Registration System gives each district's population at the end of 2022 and 2023. Let us say at the outset that this is not a perfect measure of displacement: it is taken at year's end, so it also covers the seven months after the election, and it includes births, deaths and migration unrelated to the earthquake. It is nonetheless the most comprehensive and most reliable indicator of human movement available. We joined this data with our own ballot-box records and, for each of 972 districts, placed two numbers side by side: by what percentage did the population change in the earthquake year, and by how many points did turnout change? (TÜİK lists 973 districts; Derecik in Hakkâri, which did not yet exist in 2018, cannot be compared across the two elections and is excluded.)

The result can be summarized in two sentences. Across Türkiye as a whole there is no relationship between these two numbers — in an ordinary district losing population, turnout does not fall. The relationship appears only, and sharply, in the earthquake zone: there, the larger the population loss, the larger the loss of turnout.

The clearest evidence for this distinction comes from districts that lost population without an earthquake. Aydıncık in Yozgat lost 11.6 percent of its population in 2023 and Bozkurt in Denizli 10.8 percent — roughly half of Antakya's loss. Yet their turnout barely moved. In earthquake districts losing people at the same rate, turnout collapsed by 7–9 points:

The mechanism best suited to explaining this difference is that ordinary migration and disaster migration leave different traces in the voter register. A person who moves for work or education takes their address record along; the register follows them, and the turnout rate is undisturbed. A person who loses their home overnight in an earthquake, by contrast, needs both a reason and an opportunity to update their record. The YSK gave earthquake-affected voters the chance to update their address until April 2, 2023; no public figure exists for how widely it was used. The aggregate data, however, is consistent with limited uptake: in Hatay, the province's population fell by 141 thousand in 2023 while the number of registered voters rose. Every voter who no longer actually lived in the district where the register placed them faced a choice on May 14: return to their assigned ballot box, or not vote — and this, by inflating the denominator of the turnout rate, exerted downward pressure on the rate in every district whose people had left. District-level data cannot prove this mechanism at the individual level; but it is the explanation most consistent with the observed pattern.

Rural destruction, urban destruction

Splitting the data by district size adds another layer. Dividing the 124 earthquake-zone districts into three by population: in the 41 large districts above 100 thousand, the population–turnout relationship is very strong (r=0.83) and the average turnout loss is 1.7 points. In the 39 small, largely rural districts below 20 thousand, the average turnout change is zero — despite sitting in the middle of the earthquake zone.

This difference is consistent with field observations after the earthquake. In rural areas, families whose homes were damaged mostly continued to shelter near their villages — in tents, in containers, with relatives — and when election day came, the ballot box where they were registered was still in place. In city centers, every collapsed apartment block left dozens of households without shelter at once, and neighborhoods emptied wholesale. Housing could usually be resolved locally at village scale; at urban scale it could only be resolved by moving to another city. The data shows that the loss of turnout moved with the scale of that relocation far more than with the destruction itself.

Why does this matter?

First, registration. Türkiye's voter register is address-based, and in ordinary times the system works well. February 6 showed how it responds to an extraordinary one: hundreds of thousands of people actually moved, while the register stayed put. The consequence is not merely statistical — every voter unable to return to the district where they were registered was, in practice, unable to exercise their right to vote. To give a sense of scale: had turnout in Antakya matched its 2018 level, roughly 28 thousand more votes would have been cast — in a single district.

Second, measurement. Every turnout and vote share read in the earthquake zone in the elections to come should be read in the shadow of this gap between register and residents. In a district whose population fell by a quarter but whose register shrank by a tenth, the denominator of every rate now includes people who no longer live there. This has nothing to do with any party; it is a technical matter of registration, but it directly affects how election results should be interpreted.

How we calculated — and what we chose not to use

This section describes the data used in the analysis, the verification steps applied, and the sources that were rejected.

What we used. Turnout: YSK ballot-box data for the 2018 and 2023 presidential elections (first round), aggregated to 972 districts. Population: TÜİK's Address-Based Population Registration System, 2022 and 2023, district level (TÜİK MEDAS). Before using the population data we ran a consistency test: the district sums of all 81 provinces matched TÜİK's official province totals exactly, in all 81 cases. Hatay's district damage breakdown: news records of the period, cross-checked against two independent sources.

What we rejected, and why. District-level damage counts (outside Hatay): only the incomplete first assessments of February 2023 exist, at roughly a quarter of the final count — rejected. Neighborhood-level damage: never published. An online encyclopedia's district population tables: the district sums did not match the official province totals (136 thousand short in 2022) and even showed Hatay gaining population — rejected. A rural/urban split at the neighborhood level: Law 6360, passed in 2012 and in force with the 2014 local elections, legally converted villages into neighborhoods in metropolitan provinces, so the distinction cannot be derived from the records; since every one of the worst-hit districts lies in a metropolitan province, the distinction is missing exactly where it is needed.

Robustness. The relationship in the earthquake zone does not rest on a single outlier: the correlation of r=+0.46 across the 124-district group persists at +0.38 when all of Hatay is removed and at +0.30 when Hatay and Malatya are removed together. The +0.83 coefficient in the large-district group likewise stays at +0.71 when the three most extreme observations (Antakya, Defne, Yeşilyurt) are all removed. In the middle group (20–100 thousand population, 44 districts) the linear relationship is weak (r=+0.22), but the average turnout loss is the same as in the large group: 1.7 points. The population thresholds (20 and 100 thousand) are round practical values; the results are not sensitive to small shifts in the cutoff. Because districts of the same province share common conditions — that is, they are not independent observations — we treat the province-removal tests as the primary measure of robustness.

Limits. ADNKS population is measured as of December 31; the election was on May 14. The 2023 figure therefore also covers the seven months after the election — the direction is reliable, the exact magnitude is not. ADNKS itself rests on address records; a person who does not move their registration continues to appear there for a while too, meaning true displacement may be larger than what we measure here. We also work with district totals: it cannot be shown at the individual level that the non-voters in a district losing both population and turnout are the same people who left. Finally, this is an observational analysis: what it shows is a strong and durable pattern of association, not laboratory-grade proof of cause and effect.

Data sources: YSK open data (ballot-box level turnout, 2018 and 2023 presidential elections) · TÜİK MEDAS (ADNKS district populations 2022–2023) · Hatay damage assessment: AA, April 8, 2023 · total damage: AFAD. You can explore every result on this site down to the ballot box.

This is the sixth article in the Analysis of the Week series. The data may be used with attribution.