A Century of Numbers

Every population figure on this site comes from a count that somebody actually conducted. They are two imperial and early Soviet enumerations, six all-Union Soviet censuses, and two censuses run by the unrecognized republic itself. None of them is beyond criticism, and this essay states the criticisms before anyone else can. The argument is simple. A record that admits what it cannot show earns the right to be believed on what it can.

Two baselines#

The first modern count is the Russian imperial census of 1897. It recorded the Shushi uezd of Elisavetpol Governorate, the administrative unit covering most of mountainous Karabakh, at 138,771 inhabitants, of whom 73,953 gave Armenian as their native language and 62,868 the local Turkic language.¹ Two cautions attach to this number, and both matter. The census asked about native language, not ethnicity, so its categories are not the same variable as the nationality question of the later Soviet censuses and should not be charted as if they were. And the uezd is not the later autonomous oblast; its boundaries take in lowland territory the oblast never included, so the 1897 aggregate must never be set directly against the NKAO figures.

Within that count, one city can be read on its own. Shushi held 25,881 people, of whom 14,420 were Armenian-speakers and 10,778 fell in the column the census volume calls 'tatarskiy'.² Hold that figure. In 1926 the first Soviet census found 93 Armenians in Shushi.³ Between the two counts lies the destruction of the city's Armenian quarter in March 1920. The census tables do not narrate that event; they record its result, which is a different and in some ways harder kind of testimony. A fall from 14,420 Armenian-speakers to 93 Armenians needs no adjectives.

The second baseline is the enumeration the Baku SSR conducted across Karabakh in 1921, the only settlement-by-settlement count that survives in accessible form for this period. It is a limited instrument. It records only two nationality columns, Armenian and Turkic, and it is not one of the all-Union census years. And it survives through a republication by an Armenian documentary site, read via the Internet Archive; that is an Armenian source, and this dataset says so, which is why every figure taken from it was re-verified by hand against the archived tables, down to checking that the male and female columns sum to the printed totals.⁴ What it shows is a countryside of Armenian villages, Hadrut, Togh, Talish, Chartar, Avetaranots and dozens more recorded as entirely or almost entirely Armenian, and a village of 1,379 people, 981 of them Armenian, listed as 'Stepanakert (s. Khankendy)', which two years later would be the oblast capital.⁴ Where this dataset had to combine two printed rows, as with Chartar's upper and lower villages, 1,319 plus 509, the addition is disclosed as ours so a reader can undo it.⁵ The 1921 table for the Shushi district, notably, covers only the Turkic villages of one volost and omits the ruined city altogether. The likeliest reason is the obvious one, but the table does not say so, and the dataset records a gap rather than an inference.⁶

The boundary of 1923#

On 7 July 1923 the Central Executive Committee in Soviet Baku decreed the formation of the Nagorno-Karabakh Autonomous Oblast.⁷ The population inside the new line was about 90 percent Armenian. The line itself was drawn so that this Armenian oblast shared no boundary with Soviet Armenia; a narrow strip of intervening territory separated the two, at what would later be called the Berdzor corridor.⁸ That is a cartographic datum, not a demographic one, and it is documented in the decree record rather than deduced. Everything the census series later shows happened inside a border designed, from the first day, to keep the region administratively alone.

The Soviet series#

Six all-Union censuses counted the NKAO, and the aggregates are the most solidly established numbers in this entire record. In 1926 the oblast held 125,300 people, 111,694 of them Armenians, 89.1 percent, and 12,592 recorded under the census term of the day, 'Turks', 10.0 percent.⁹ In 1939, 150,837 people, 88.0 percent Armenian.¹⁰ Then comes the largest break in the whole series. By 1959 the total had fallen to 130,406, with war losses and out-migration both bearing on the fall, and the Armenian share stood at 84.4 percent while the Azerbaijani share had risen to 13.8.¹¹ In 1970, 150,313 people, 80.5 percent Armenian, 18.1 percent Azerbaijani.¹² In 1979, 162,181 people, of whom 123,076 were Armenians, 75.9 percent, and 37,264 Azerbaijanis, 23.0 percent.¹³

That is the arc usually compressed into a sentence. Between 1926 and 1979 the Armenian share of the NKAO fell from 89.1 to 75.9 percent, while the Azerbaijani population roughly tripled, from 12,592 to 37,264. Armenians in the oblast attributed the shift to Baku's economic and settlement policies, and scholars who studied the region identified that grievance as a core driver of the 1988 movement.¹⁴ Armenian sources report that the appeals sent to Moscow in 1977 already argued from these numbers; that report rests on Armenian sources alone, and is labelled accordingly.

Then the series does something a partisan dataset would be tempted to hide. In January 1989, the last Soviet census, taken after the movement began but before the mass violence and population transfers, counted 189,085 people in the NKAO, of whom 145,450, or 76.9 percent, were Armenians.¹⁵ The Armenian share went up between 1979 and 1989, from 75.9 to 76.9 percent. The record itself contradicts any account of smooth, continuous Armenian decline across the Soviet period. The honest shape of the curve is a steep fall to 1979 and a slight recovery after it, and this site draws it that way.

What the numbers cannot prove#

The census series proves the direction and the size of the demographic shift. It does not, by itself, prove intent. A nationality table has no column for motive; the argument that the decline was engineered rests on the boundary decision of 1923, on the pattern of investment and settlement that Armenians protested at the time, and on the scholarship that took those protests seriously.¹⁴ This dataset keeps the two kinds of claim apart. The percentages are census facts; the explanation is an attribution, labelled as one.

The record also has holes, and they are stated rather than papered over. No settlement-level Soviet census data exists in published form for 1926 through 1989; the accessible tables stop at districts, cities and urban-type settlements, which is why village series in this dataset jump from 1921 to 2005. For 1989 the situation is worse. Only the oblast aggregate survives, with no city or district breakdown found anywhere, so the census closest to the war is also the thinnest. And the corroboration of the aggregates has a stated limit. Two independent publishers, the Demoscope archive in Moscow and a Caucasus census compilation, agree digit for digit at every census year, but for 1959 through 1989 both transcribe the same underlying archive, RGAE fond 1562. That is confirmation of the record, not a second enumeration. No second enumeration exists, because the Soviet census is the enumeration.¹⁶

Reported because it is true#

The discipline runs in both directions. The territories around the oblast were counted too, and this dataset reports those numbers even though no Armenian case is served by them. The district the Soviet tables call 'Lachin', around the town Armenians know as Berdzor, was overwhelmingly Azerbaijani and Kurdish at every Soviet census; in 1979 it contained 34 Armenians. The district around Karvachar counted 40,516 people in 1979, of whom 49 were Armenians.¹⁷ Those figures appear here because they are true, and because a record that concealed them would deserve to be disbelieved about everything else. The same tables show the Shahumyan district, north of the oblast and outside it, Armenian-majority at every Soviet census, 85.9 percent in 1939 and still Armenian-majority in 1979.¹⁷ The numbers cut in every direction; that is what makes them numbers.

The last counts#

After the Soviet collapse, the counting passed to the unrecognized republic. Its census of 18 October 2005 recorded a permanent population of 137,737, of whom 137,380 were Armenians; its census of October 2015 recorded 145,053.¹⁸ Both were conducted and published by the statistical service of a de facto state, without international supervision. That does not make them false, and this site says so plainly rather than hiding the provenance. The 2015 figure also needs guarding. The widely repeated 150,932 is a preliminary figure announced in 2016 that appears in no published census table, and a number near 148,900 is a yearbook estimate, never a census count. The counted, final, de jure figure is 145,053.¹⁹

No census will follow it. Between 24 September and about 1 October 2023, after the blockade, the population left. The figure this site uses is the UN's own, not an Armenian count. More than 100,000 refugees arrived in Armenia, per UNHCR, whose staff had registered over 88,000 in the first week alone, and the first UN mission to the region in thirty years was told that between 50 and 1,000 ethnic Armenians remained in the whole of Karabakh.²⁰ A century of numbers ends at approximately zero, and the record of how the land filled and emptied is exactly as strong as the discipline with which it was kept. That is why every concession in this essay is there.