I would like to express my profound gratitude to my research assistant who helped me find and gather various census yearbooks, and with whom I also had the opportunity to discuss much of the findings. However, given current circumstances, they prefer to remain anonymous. Thanks also to the anonymous peer reviewers, to Geoff Childs for his generous comradery in discussing all matters relating to Tibetan demographics, and to Tashi Rabgey, for her insights into how all of this is relevant in the broader context of Tibet and China. And most of all, to Katia Buffetrille for her endless patience and encouragement, and the other editors of EMSCAT for their support in bringing this to fruition.

4The question is whether similar experiences occurred in Amdo and in the parts of Kham outside of the TAR, or whether the tendency of outmigration continued to prevail into the 2010s in these areas that were less prioritised by the central government. Various development projects from the late-1990s onwards in Qinghai, for instance, particularly in Tsonub TMAP, which is part of Amdo, raised concerns that these other parts of Tibet were following the way of the TAR. More recently, several changes in the central government’s governance approach towards these eastern Tibetan areas might also have caused migration dynamics to increasingly resemble those of the TAR. These include the addition of the Tibetan areas outside of the TAR into a broader conceptualisation of “Tibet” by the Fifth Tibet Work Forum in 2010; the increasing subsidisation of Qinghai Province (see Fischer 2015); the spectacular economic growth of Sichuan Province centred around Chengdu, the capital city; and an increasing focus on “public security” from the late 2000s onwards in these Tibetan areas. However, any questions of aggregate population change could only be answered with the results of the 2020 census.

5An initial study of Han and “minority” nationality (Ch. minzu) shares throughout all Tibetan areas in China, based on the release of the first 2020 census communiqués in 2021, confirmed that the prevailing trends of Han outmigration rather than in-migration not only continued but deepened across eastern Tibet (see Fischer 2021a and 2021b). However, at that time, prefecture and county-level data with detailed minzu decompositions were not yet available. This study extends the analysis with the county-level data that were subsequently published in 2022, with a focus on the Amdo region and, for the sake of comparative interest, Yulshul (Ch. Yushu) in Qinghai, technically part of Kham. For Qinghai and Gansu, the analysis is based on the detailed population count of all 56 minzu in China at the county level, available in the detailed provincial census tabulations, which we were only able to find and access in 2023 and early 2024 (table 1-4 in GPCY 2022, QPCY 2022 and SPCY 2022 respectively). For Sichuan, the detailed minzu disaggregation is only provided down to the prefecture level in the provincial census tabulations, which limits our analysis of Ngawa (Ch. Aba) Prefecture. This was nonetheless complemented by the national county tabulation (CPCC 2022), which provides county-level data aggregated according to a Han and minority binary. Both sources also provide other valuable information such as age structures, residency status, education levels (measured in terms of levels of schooling obtained), illiteracy rates, employment, among other measures, although these latter tabulations are not generally broken down by minzu status in the available public sources, requiring some inference and extrapolation to analyse along minzu lines.

6These detailed county-level data confirm the overall trends presented by Fischer (2021a, 2021b), but also provide more nuanced insights into local variations, including some cases of rapid population changes over the short intercensal period of a decade. These nuanced local insights expose and highlight a host of research questions that cannot be answered by these aggregate data and require further research. The insights are nonetheless an indispensable baseline to consider in any contemporary social research on these areas.

7The general trend in the Amdo region, with some exceptions, was one of falling – in some cases sharply – Han populations and population shares. Indeed, the main difference with the 2000s was the sharp Han depopulation in many counties in the 2010s, whereas this was less apparent in the 2000s, as noted above (except in the Tibetan areas of Gansu, where the Han population also fell significantly in the 2000s). Moreover, the Han depopulation included Tsonub Prefecture, where the Han had previously increased rapidly in the 1990s and 2000s. Tibetan outmigration was also evident, mostly notably in Ngawa Prefecture in Sichuan, where there was also Tibetan depopulation, although much slower than the Han. As in Qinghai and Gansu, Tibetan (and Qiang) depopulation was mostly observed in the eastern counties that border Han areas and have higher Han population shares.

10In contrast to the Han outmigration and depopulation, the census data also confirm earlier insights (e.g. Fischer 2005, 2008b) that the dominant inter-group demographic pressure in Amdo for the last two decades, at least in Qinghai, has not been between Tibetans and Han, but between Tibetans and Muslims. In Qinghai, the population of Muslim groups indigenous to the broader region (principally Hui, and smaller numbers of Salar and Dongxiang) in most cases increased faster than Tibetans even in Tibetan areas, although mostly from very small numbers in these areas, and this was evidently driven by net in-migration. The one extreme case was Tsonub Prefecture, where Muslims surpassed the population of Tibetans, Mongols and Monguors, and partially substituted the net outmigration of Han from this prefecture, although such extremes are not observed elsewhere. Indeed, the inverse occurred in Ngawa Prefecture, with strong Hui population decline comparable to the Han, whereas in Ganlho the results in this regard were mixed. Nonetheless, in Qinghai, with the largest number of Amdo Tibetans, it is clear that urbanisation has put these Tibetans in increasing interaction with these Muslims groups.

11More generally, these insights also confirm earlier arguments (e.g. Fischer 2004, 2008a and 2014; also cf. Childs 2008 and Ma 2011) that, as stated above, the dominant structural trend facing these relatively poor peripheral areas is net outmigration, not net in-migration, which is stronger among the Han than among locally indigenous nationalities. This has generally led to a tendency for rising Tibetan (and Muslim) population shares that will likely continue for several more decades given the higher natural increase rates among Tibetans (and Muslims) in this region and the greater propensity of non-indigenous Han to leave the region altogether. Such tendencies are only counteracted by extremely high levels of subsidisation, as observed in the TAR, which then stimulates in-migration. However, given that Amdo falls entirely outside the TAR, is incorporated and subordinated within provincial governance structures and hence far less subsidised than the TAR, and is strategically less prioritised than the TAR, it logically succumbs to the norm in China of net outmigration and, more recently, even depopulation in more remote and rural settings.

12These insights are discussed in the following three sections. The first clarifies some background and technical issues regarding the census data. The second provides a brief overview of the demographic profile of Amdo Tibetans and a summary of the provincial minzu population dynamics. The third analyses the detailed prefecture and county-level data for all the counties of Amdo as well as Yulshul according to four broad patterns observed in the data, as introduced above. The conclusion discusses some of the development implications of these rapid population changes.

The census

13The most detailed census data analysed here are from the 2020 province census yearbooks (Sheng renkou pucha nianjian 省人口普查年鉴－2020), which are generally presented in three volumes (hereafter referred to as PCY, e.g. GPCY, QPCY or SPCY, all published in 2022). Except for Sichuan, these include tabulated data for all 56 minzu populations at the county or district level, also broken down by city, town, and rural areas (tables 1-4, 1-4a, 1-4b, and 1-4c; the numbering is consistent across all census yearbooks). The Sichuan PCY only provides the detailed minzu data at the prefecture level, although this can be supplemented by the next most detailed source, the 2020 China Population Census by County (Zhongguo renkou pucha fenxian zi 中国人口普查分县资－2020), also published in 2022 (CPCC 2022). However, the CPCC, for both the 2010 and 2020 censuses, only presents the minzu data according to a binary of Han and minorities, and there is no differentiation between various minorities, such as Tibetans and Hui Muslims. This limitation is important in Amdo given questions regarding the degree to which “minorities” in Amdo reflect the net in-migration of Muslims (Hui, Salar and Dongxiang) from neighbouring areas, which is a sensitive issue in this region, or else the in-migration of assimilated minorities from other parts of China, who would be effectively perceived by Tibetans as Han Chinese. We can assess these questions for Qinghai and Gansu at the county level, but only at a prefecture level for Ngawa.

17People in the last category are considered migrants (even though they might have been living in a location for many years) and are referred to synonymously as the “floating population”. These migrants are divided between those whose household registration is elsewhere in the same province, and those who are registered in another province. At the national level in the 2020 census, about two thirds of such migrants (or “floating population”) were intra-provincial, and one-third inter-provincial.

18This means, of course, that migrants who reside in a location for less than six months are not counted in that location, but instead in their place of household registration (or wherever else they might have been previously located). Hence, the census would not include short-term seasonal migrants, such as those who travel to Lhasa or other locations for only the summer months to work in tourism or construction. Or else, as is the case of censuses everywhere, it does not include what Roche et al. (2023) refer to as “visitors” and “sojourners”. However, given that the census was conducted in November, most of these short-term seasonal migrants or visitors in Tibetan areas would have already departed in any case. The censuses would thereby underestimate the summertime migrant populations in some Tibetan areas, although they would give a fairly accurate assessment of more settled year-round populations (registered or not).

20The alternative source of data is from the official Public Security Department (PSD) registers, but these report household registration (Ch. gong’an huji), not residency. They therefore poorly reflect migration (both outward and inward). To give one example, whereas the 2020 census counted 189,410 Han residents in Kardze (Ch. Ganzi) Prefecture in Sichuan, the PSD household register only counted 134,209 as of 6 January 2021 (GzSY 2021, p. 150). This difference of over 55 thousand people was likely because many Han residents did not register. The total prefecture population in the register was closer to the census count (1.085 million versus 1.107 million in the census, or a difference of 22 thousand people), although this is because the register inversely overestimated the Tibetan population by over 35 thousand people (because emigrated Tibetans would have kept their registration in Kardze). This demonstrates how the PSD source is simply not reliable. However, the discrepancies between these different sources of data often lead to scepticism, especially when people compare non-comparable data sets, although this is usually founded on a misunderstanding of how population data are collected and calculated.

21The census data are also often discounted based on claims that official data are regularly falsified. However, this claim is counter-intuitive with the census evidence. First, the region where Han in-migration is most evident – the TAR – clearly shows a large and accelerating increase of Han residents, as discussed below. In particular, the 2020 census reveals that the population of the urban core of Lhasa is one third Han, which given the predominantly working age and economically active profile of this population, clearly demonstrates that employment was dominated by the Han in Lhasa. Conversely, the counties where we observe falling Han numbers are far from the spotlight to the extent that the public, even in China, might not even conceive of them as Tibetan areas. But more importantly, falsifying the census data in a way that is consistent across a range of different demographic variables would take a lot of technical demographic knowledge and skill that simply would not exist at the local level where these data are collected and assembled. Rather, falsification would tend to happen in the public headline presentations of data but can be quickly discerned by comparing to the original census data. This is also the reason for corroborating the data across several demographic variables (such as age structure), to see if any inconsistencies can be detected.

24A related question regards interethnic marriage, and how a couple and especially their children get classified if a Tibetan marries someone from another minzu group. However, it again appears that this was quite limited among Tibetans, at least up until 2010. A study by Jian (2017), based on comparing the 2000 and 2010 census data, found that Tibetans in China had the third lowest rate of intermarriage in China (after the Uyghur and Kazak). Nationally, almost 94 percent of households with a Tibetan head had a Tibetan spouse in 2010, and this share slightly increased since 2000. The highest non-Tibetan choice of spouse was Han, at 5 percent, which similarly had slightly fallen since 2000 (ibid., p. 12). Notably, almost 99 percent of households in the TAR were single ethnic group households in 2010, with no change from 2000. This demonstrates how interethnic marriage remained marginal in aggregate population terms among Tibetans, with no apparent increase up to 2010 despite government efforts to encourage it. The study by Jian (ibid.) needs to be updated to the 2020 census data, especially considering the encouragement of interethnic marriage under Xi Jinping (see Leibold 2015). However, the trends in the 2000s, despite public encouragement and rapid urbanisation in that decade, suggest that this factor had at most a marginal effect on the aggregate minzu population counts in Tibetan areas in the 2010s.

Aggregate profile and regional divergence

25The 2020 census counted a total of 7,060,731 Tibetans in China, of which 98 percent were residing in the five provinces containing Tibetan autonomous areas, and only 2 percent or 140,731 people in other provinces of China. Calculating the precise number of Amdo Tibetans is not possible given the increasing number who reside in the capital cities of their respective provinces (Xining, Lanzhou, and Chengdu), where it is not possible to differentiate the original source of emigration with the published census data. In addition, a small but significant number of Amdo Tibetans and closely related Monguors and Mongols have historically resided in non-Tibetan autonomous prefectures and counties, particularly in Haidong City in Qinghai and in various prefectures in Gansu.

27Roughly half of this population was based in Qinghai. The Tibetan population of the five autonomous prefectures in question and Haidong City was 943,704 in 2020 and increased over the decade by 55 thousand people, or 6.2 percent. The increase rate was higher than this in the five prefectures but was compensated by a falling Tibetan population in Haidong. The higher rate of increase for Qinghai Amdo than for the Amdo total is also because the Tibetan population fell in Ngawa and stabilised in the Tibetan areas of Gansu. Outside Qinghai, the Tibetan population in Ganlho and Pari was 436,532 in 2020 and increased by a mere 1.4 percent over the decade. The Tibetan population of Ngawa was 476,059 in 2020 and shrank by 2.8 percent over the decade.

28The rates of increase (or decrease in the case of Ngawa and Haidong) are indicative of net outmigration because they are much lower than the likely rates of natural increase (RNI) among Tibetans in these Tibetan areas over this period. Note that in demography, “natural increase” (or decrease) is the change in population that results from births minus deaths in a given year, not including net migration. It is derived from crude birth and death rates, and is expressed as a rate, typically per thousand population. Population increase (RI), or what is often referred to as “population growth”, adds net migration to natural increase, as the only two ways populations can change. Accordingly, as explained in Fischer (2012), net outmigration can be inferred when the average rate of natural increase over a period is substantially greater than the annualised rate of population growth of that same period, because the population is not growing as fast as would be implied by the addition of births minus deaths to that population – in other words, net outmigration is removing people from the population even as natural increase is adding them. When natural increase is sufficiently high and migration not overly predominant, net outmigration does not usually outweigh natural increase and the population continues to increase, just not by as much as indicated by the rate of natural increase. However, when natural increase slows due to low fertility sustained over sufficient time, and when migration becomes more dominant, populations often start shrinking as more people are leaving the population than are being added to it through natural increase. At the extreme, populations shrink because births are less than deaths in a given year, as was estimated to have occurred in China overall in 2023, but the Tibetan areas are far from this situation so far.

29Population increase (or growth) is measured most accurately by comparing two censuses, for the reasons discussed in the previous section. Natural increase rates, however, are calculated from the crude birth and death rates (CBR and CDR) measured in annual surveys that are taken between the census years (or civil registries where these are well developed). Censuses provide the birth and death rate data for the year of the census but not in other years – for instance, the CPCC (2012, Chapter 3) reports these data per county from the 2010 census, although these data were not reported in CPCC (2022) for the 2020 census. Alternatively, annual crude birth and death rates are provided in the Chinese public statistical sources by province, but not at lower levels of aggregation and not decomposed by ethnicity. As a result, we cannot directly know with the publicly available data the rate of natural increase among Tibetans in a province, or even in a county that is predominantly Tibetan outside of the census years.

30The only way around this limitation is to look at the birth and death rate data for the TAR, where the registered population is mostly Tibetan. Moreover, because the Han population in the TAR is mostly in the adult working ages (e.g. see Ma 2011, p. 76), they would be under-represented in both births and deaths, which in turn would mostly reflect Tibetan natural increase. Based on other demographic data that are quite similar in Tibetan areas both inside and outside the TAR (such as population age structures, as discussed below), we can presume that the TAR natural increase rates are also quite similar to other Tibetan areas. We can also cross-check this presumption by comparing the TAR rates to the rates reported in the 2010 census in Tibetan areas outside of the TAR with high Tibetan shares (namely, Malho, Tsolho, Golog – Ch. Guoluo – and Yulshul TAPs in Qinghai). These data are shown in figure 2 for the period from 1975 to 2020, also with China for comparison.

31For China as a whole, birth rates start to fall from the mid-1980s onwards. This followed from the sharp decline in TFRs in the 1970s, down to almost replacement levels by the late 1970s due to the introduction of strict family planning from 1970 onwards (Peng 1991). Death rates had already hit a trough by the 1980s and even started to rise slightly in the two most recent decades due to population aging. As a result, the RNI was quickly approaching zero by 2020.

32In the TAR (and western minority areas more generally), family planning was later to start and was more lenient, with the result that TFRs only started to significantly decline from the mid-1980s onwards (Childs 2008, p. 202; cf. Spoorenberg 2019 for all Tibetans in China). Birth rates then only fell below 20 per thousand from 2000 onwards (although keeping in mind that these data also reflect an increasing composition of Han in the TAR that would have also brought down birth rates – the Tibetan-only rates would have been higher than this and for longer). Combined with marginal improvements in death rates (from a level between eight and nine per thousand right up until the mid-1990s, to under five by the late 2010s), the natural increase rate of the TAR diverged from that of China from the late-1980s onwards (it had been roughly equal since the mid-1970s). The higher rate in the TAR was sustained right up to 2020, at about 0.5 percentage points higher than the China RNI (although this difference widened to 0.7 in 2020). Natural increase rates in the TAR did start to fall from the late 1990s onwards, but with a lag of about 15 to 20 years relative to China as a whole (e.g. the TAR natural increase rate of almost 0.9 percent in 2020 was last observed in China in the late 1990s).

33The 2010 natural increase rates for the four prefectures in Qinghai with high Tibetan population shares were very close to that of the TAR in the same year, which was 10.25 per thousand, or an increase rate of just over 1 percent a year. It was lower in Yulshul, although this was due to the spike in death rates from the 2010 earthquake, whereas counties in Yulshul that were not impacted by the earthquake had natural increase rates in the range of 13 to 17 per thousand, which was substantially higher than in the TAR. The rates were lower in Ganlho in Gansu and Ngawa in Sichuan (not shown), quite similar to the national rate for China in 2010, although this also reflects a higher proportion of Han (and Qiang in Ngawa), and possibly more advanced family planning and fertility transitions in these provinces compared to Qinghai. Nonetheless, these results confirm the fact that natural increase rates in Tibetan areas outside of the TAR were quite likely very similar to those in the TAR, at least in 2010.

34Despite the lack of time series data for calculating the natural increase rates of Tibetan prefectures or counties in Amdo over time, we can also infer from the population age structure of various parts of Amdo that the trends would have been similar to the TAR, albeit with some surprising results. In this regard, QPCY (2022, table 2-2) provides the 2020 census population data for all Qinghai by 5-year age cohorts broken down by each minzu, although only at the provincial level. Alternatively, the CPCC (2022: Chapter 2) provides the age structure per county, but not by minzu, although this can nonetheless provide some very interesting fine-grained insights into counties with high Tibetan population shares, such as in Yulshul or Golog. The age structures for both Tibetans and Han in Qinghai are presented in figures 3 and 4.

35The Han population structure in Qinghai is typical to elsewhere in China proper. The largest cohort was in the 50- to 54-year-old age range, corresponding to the cohort that was born from 1966 to 1970, just before the beginning of the sharp reduction of fertility rates in China. As Han women of this generation started having children, we observe an echo effect 20 years later, in the 30-34 age range, born in the late 1980s. As is standard with such echo effects in the context of falling or low fertility, this cohort of the late 1980s is smaller than the cohort of the late 1960s or early 1970s. The structure then tapers off in the younger cohorts, which is consistent with low fertility sustained over several decades, although a very slight echo effect can be observed in the 5-9 age range, as Han millennials in Qinghai started having children (albeit at older ages).

36In contrast, the Tibetan age structure in Qinghai reveals a surprising result – a veritable baby boom from about 2005 to 2015, with the largest ever cohort in the 5-9 age range, born from 2011 to 2015 (and, as an extra surprise, the age structure resembles a stupa). Because this is province-level data, it is unlikely that the structure is explained by the net outmigration of Tibetan adults given that Tibetan migration out of the province would have been marginal. Some of the structure might be explained by the echo effect of millennial women starting to have children, except that as noted above, echo effects are generally smaller, not larger than the preceding generation. The fact that the two cohorts are larger can only be explained by a short resurgence of fertility rates, after they had fallen in the 1980s and 1990s (like the baby booms in the US and Europe following WWII). As noted in the previous section, we do not have the TFR data to confirm this, and the alternative census data on live births per women aged 15-64 (which we do not deem reliable) do not give any indication of this baby boom in 2010, even though the 2010 census data show relatively rapid natural increase rates in the prefectures of Qinghai with high Tibetan shares, as discussed above. Nonetheless, according to the age structure, fertility rates would have fallen again after 2015 given the sharp indent in the population structure in the youngest cohort born between 2016 and 2020, which would be a consequence of a sharp drop in fertility (or at least births).

38One possible counter explanation for the baby boom from 2005 to 2015 is that it reflects a compositional shift. In other words, the more remote southern Tibetan populations in Golog and Yulshul, which were growing more rapidly and maintaining higher fertility rates, started to become more prominent in the population than the erstwhile more dominant Tibetan farming areas to the north. However, if we examine the population structure of Darlag (Ch. Dari) County in Golog Prefecture (fig. 2c), which, as analysed below, had the fastest growing Tibetan population in all Amdo and remained over 94 percent Tibetan in 2020, we actually observe similar dynamics, although accentuated by the likely net outmigration of adults.

39We can also observe similar structures elsewhere, such as in the western counties of Ngawa Prefecture in Sichuan, where Tibetans account for more than 90 percent of the population, and where the largest-ever cohorts in the 2020 census were also in the 5-9 age range. This contrasted with the prefecture overall, which had a structure closer to the Han in Qinghai, probably due to the weight of Han and Qiang in the eastern counties of the prefecture. However, because outmigration might be prominent in these counties, we cannot draw any definitive conclusions from these county-level data without further research.

40Similar dynamics can also be observed among the Hui Muslims in Qinghai (not shown here). They apparently had a sharper fertility decline in the late 1970s and 1980s compared to Tibetans, but with a similar surge from 2005-15, with the largest ever cohort in the 5-9 age range. This suggests that something particular was happening among minorities in the region, although at the moment we do not have an explanation for this.

Macro-regional divergence in Han migration

42In contrast, the Han population share declined in the three provinces that account for all of Amdo and most of the parts of Kham outside of the TAR (Gansu, Qinghai, and Sichuan), while it remained stable in Yunnan Province (the Tibetan areas in these four provinces account for over half of the Tibetan population in China and about half of the areas). The decline was sharpest in Qinghai, where the share also fell faster in the 2010s than in the 2000s, dropping from 54 percent in 2000, to 53 percent in 2010, and to 50.5 percent in 2020. In Gansu, the declining Han share was also the result of Han depopulation, that is, falling numbers of Han people, which would have been due to strong outmigration.

43Differences in population growth underlie these trends, as shown in table 1. The TAR population, like that of Xinjiang further west, grew rapidly during the 2010s at close to 2 percent a year (a rate at which the population would double in roughly 36 years). This growth came through a combination of net in-migration and relatively high natural increase rates (the highest provincial rates in China).

45The stabilisation or decrease in Gansu, Qinghai and Sichuan was driven by the Han population changes (as analysed in further detail below). This would have been due to strong outmigration among the Han, combined with lower natural increase rates compared to minorities, as discussed above. In contrast, minority populations grew at rates significantly faster than the provincial RNIs, in this case reflecting higher birth and natural increase rates among these minorities than the provincial averages, and probably higher fertility rates, as also discussed above.

46At a provincial level, these higher minority natural increase rates compensated for the outmigration and lower natural increase among the Han. Moreover, as noted above, net Tibetan outmigration from their respective provinces was probably very marginal, even though Tibetans were clearly migrating out of their counties and prefectures. For instance, the annualised growth rate of Tibetans in Qinghai during the 2010s was quite close to the RNIs in 2010 for the Qinghai prefectures shown in figure 2 above (e.g. the annualised growth rate was 0.94 percent, versus natural increase rates in 2010 of 1.02 percent for Golog, 1.12 percent for Tsolho, and 1.10 percent for Malho). If subsequent natural increase rates among Tibetans in Qinghai were similar to the TAR and remained at this level or slightly decreased throughout the decade, which is likely, then their net outmigration from Qinghai would have been very minimal.

48One problem with these provincial aggregates, however, is that the Han majorities of these provinces mostly reside outside of the Tibetan areas. For instance, it could be plausible that Han shares increased in Tibetan prefectures and counties even while they were falling at the provincial level. This could occur because of churning population movements within provinces, not only of Han migrants coming to Tibetan areas but also Tibetans moving into Han areas, such as large numbers of minorities migrating to the capital city of their province, as has become increasingly common (Yeh & Henderson 2008; Yeah & Mackley 2019). Indeed, encouraging Tibetans to migrate outside of Tibetan areas could be considered a strategy to encourage minority integration for national development, as claimed by Roche et al. (2023; cf. Ma 2022), although it is not clear that this is necessarily a significant government policy. The aim of residential schools, for instance, is to keep children and their families within counties and prefectures, not to leave them. The county-level data is therefore crucial to examine these types of questions.

Prefecture- and county-level data from Amdo (and Yulshul)

49The prefecture- and county-level data for the Tibetan areas of Gansu, Qinghai and Sichuan roughly correspond to these provincial observations, although with significant variations. These can be synthesised according to three dominant patterns and one qualification. The first is the strong Tibetan population growth in the remote counties of Amdo, as well as in neighbouring Yulshul, where Tibetans also constitute most of the population (i.e. over 90 percent). In Qinghai specifically, this is complemented by slight increases in Han numbers and shares in these remote counties, but from small numbers (similar in this sense to the more remote prefectures in the TAR). The second pattern is the otherwise sharp falls in Han numbers and shares in many of the more central prefectures and counties of Amdo that border Han areas and/or that already had higher Han shares. The third is the remarkable increase in the resident Tibetan populations in the provincial capital cities. The qualification is the increase of Muslim populations in most Tibetan prefectures and counties of Qinghai, although not in the other parts of Amdo.

Strong Tibetan population growth in remote counties

52Moreover, similar dynamics occurred in Yulshul, with larger populations, and can also be inferred from the census data for the more remote western counties of Ngawa Prefecture, although less so in Gannan. For instance, the minority (mostly Tibetan) population of Dzamthang in Ngawa Prefecture grew 15 percent over the decade, with a youth share of almost 31 percent and a minority share of the total population at almost 95 percent in 2020. Neighbouring counties in Kardze Prefecture such as Serthar (which is considered to be part of Amdo) or Sershul (Ch. Shiqu) County ­(tucked in between Yulshul and Golog) had comparable outcomes. Sershul in particular grew 26 percent and had a minority share of 96 percent in 2020, which is also significant because it is a more populous county, with a minority population reaching almost one hundred thousand by 2020. Given its youth ratio of 33 percent in 2020, we can presume that the addition of over twenty thousand minority people during the decade was mostly Tibetans being born rather than adults in-migrating.

53In Qinghai specifically, these population dynamics in the more remote prefectures and counties were complemented by slight increases in Han numbers and shares in these locations, but from small numbers (similar to the more remote prefectures in the TAR). In other words, small increases to very small numbers of Han were sufficient to raise the Han population shares. This occurred in Golog, Yulshul and Malho (Ch. Huangnan Prefectures), although with a few county exceptions within these prefectures. As an example, an increase of 3,307 Han in Golog Prefecture was sufficient to increase the Han share by half a percentage point, to 7.1 percent of the prefecture population, despite strong growth in the Tibetan population by 29,240 Tibetans over the decade. Notably, over half of this Han population in 2020 was concentrated in Machen County (Ch. Machin), the prefecture seat.

54In contrast, however, there was an increase in Tibetan shares in most of the counties with very high Tibetan shares in Ganlho Prefecture (Gannan) or in western Ngawa Prefecture because the Han population shrank in these counties, unlike in Golog, Yulshul or Malho in Qinghai. The only county where the Han population grew in western Ngawa was in Ngawa County (by 1,752 people), and the Han population grew by 122 people in Machu County in Ganlho Prefecture, where Tibetans accounted for 90 percent of the population in 2020. These were the only two counties outside of Qinghai that had marginal increases in Han population similar to Golog, Yulshul and Malho in Qinghai. However, this was set against the relatively rapid Tibetan population growth across all these more remote areas.

Han (and sometimes Tibetan) depopulation in more central counties

56The large reductions in Han population in these cases can only be explained by strong net outmigration. The most extreme cases were Tsonub Prefecture in Qinghai, where the Han population fell by 14 percent (as discussed further below), or else Ngawa Prefecture in Sichuan, where the Han population fell by almost 18 percent. However, similar dynamics occurred in Tsolho (Ch. Hainan) and Tsojang (Ch. Haibei) Prefectures in Qinghai, as well as in Haidong (not a Tibetan autonomous prefecture but with a substantial population of Tibetans in several counties), and in Ganlho (Ch. Gannan) Prefecture and Pari (Ch. Tianzhu) County in Gansu. The same dynamics are observed outside Amdo, in Kardze (Ch. Ganzi) Prefecture in Sichuan, and Dechen Prefecture in Yunnan.

57The county level reveals even more extreme examples. In Tsolho (Ch. Hainan) Prefecture, for instance, overall the Han population fell by just over 9 percent, but this was concentrated in three of the more remote counties, namely in Hwal (Ch. Tongde), where the Han population fell by 36 percent, Dragkar (Ch. Xinghai) by 26 percent, and Mangra (Ch. Guinan) by 40 percent, while there was a slight increase in the number of Han in the prefecture seat, by almost 3 percent. Moreover, the Han population share in these counties was not insignificant either – in Mangra, it was just over 16 percent in 2010 and fell by almost 6 percentage points in the decade (or by almost five thousand people) to just over 10 percent of the county population.

59There were also strong declines in the Han populations of the more remote western (and mostly Tibetan) half of Ngawa Prefecture, unlike in the remote parts of Qinghai. For instance, the Han population fell by almost 20 percent in Dzögé (Ch. Zoige) County, or by 37 percent in Kakhog (Ch. Hongyuan) County. The Han population even fell in the prefecture seat of Barkham (Ch. Malkang) by just over 11 percent. As mentioned above, the only county of Ngawa Prefecture that had an increase in Han population was in Ngawa County, but from a very small number.

60The shrinking of the Han population in Ganlho was smaller but still significant. Overall, it shrank by 4.4 percent, and large falls occurred in some of the eastern counties where the Han were already a majority, such as in Hwatsé (Ch. Lintan) by 15 percent, although smaller falls also occurred in the more remote counties, such as Sachu (Ch. Xiahe) by over 6 percent. These falls were compensated by a strong increase of the Han population in the prefecture seat of Tsö (Ch. Hezuo) by 35 percent, leading to an increase in the Han population share by almost 3 percentage points to 34.6 percent in this county. However, the Han population decline in the eastern counties, especially Hwatsé, exceeded the Han increase in Tsö – the Han population of Tsö increased by 10,125 people, but fell in Hwatsé alone by 14,600 people – meaning that there was net Han outmigration from the prefecture despite their increase in Tsö.

The case of Tsonub (Haixi)

61The outmigration of Han from Tsonub Mongol and Tibetan Autonomous Prefecture in Qinghai, part of Amdo, provides a telling example about how, despite past predictions of swamping, the government has struggled to retain Han migrants in many of these remote Tibetan areas outside of the TAR. This arid and sparsely populated prefecture came under the international spotlight in the late 1990s over allegations of a World Bank-funded government project encouraging population transfer and ethnic dilution in this prefecture (the “Qinghai Project”; see Fischer 2021a). The project was related to the expanding petroleum and mining exploitation in the Tsaidam (Ch. Chaidamu) Basin, as part of the central government’s broader western development strategies that started in the mid-1990s. The prefecture was also the locus of large development projects in the 2000s given that Golmud (Tib. Nagormo; Ch. Ge’ermu), the second-largest city in Qinghai and the centre of the prefecture’s mining industry, served as the logistical hub for the construction of the first railway to Lhasa in the TAR. Together with increasing mining, petroleum, and natural gas exploitation, population swamping seemed fait accompli. Despite its exceptional circumstances, it was taken to be a precursor of things to come in other Tibetan areas.

62Tsonub was indeed the strongest case of Han in-migration across all the Tibetan areas up until the 2000s. The prefecture population grew from just over 20,000 people in 1953, when it was mostly composed of Tibetans and Mongols, to 489,338 in 2010, of which 322,997 (66 percent) were Han. The Hui, who were not originally indigenous to Tsonub, even though from the broader region, accounted for an additional 13 percent of the population in 2010 (up from 12 percent 2000), many of whom had arrived from Haidong Prefecture in the 1990s through the Qinghai Project.

64Yet as Han left, the minority population increased by almost 15 percent, or by 24,542 people, reaching 190,868 people, resulting in a sharp increase in the minority share by almost 7 percentage points, reaching 41 percent of the population in 2020. Nevertheless, there was a rapid shift in the composition of minorities, as the Muslim groups (mostly Hui, but also Salar and Dongxiang) overtook the combined population of Tibetans, Mongols and Monguors, as discussed further below. However, this rapid shift in the composition of minorities from Tibetans to Muslims did not occur in other parts of Qinghai, even while falling Chinese shares did.

Cases of Tibetan depopulation

65In some of these prefectures and counties, Tibetan populations also fell, but by a slower rate than the Han, hence leading to an increasing Tibetan (or minority share) in the context of overall depopulation. A good example of this is in Ngawa Prefecture, where the Tibetan population shrank by almost 3 percent (as noted above, the Qiang population shrank by 13 percent). As a result of this differential minzu deceleration, the Tibetan share increased by 3.4 percentage points to 57.9 percent, and the combined Tibetan and Qiang share increased by 2.5 percentage points, to 74.6 percent. Another exception was in Tewo (Ch. Diebu) County in Ganlho, where the Tibetan population shrank and the Han grew, leading to a drop in the Tibetan share by almost 3 percentage points, although the Tibetan share remained at almost 84 percent in 2020. Most of the county examples of shrinking Tibetan populations were nonetheless outside Qinghai, besides a few cases that might have been related to resettlements (one in Tsolho Prefecture and three in Tsonub Prefecture).

66There was also strong outmigration of Tibetans (and Han) from the two Muslim counties of Haidong City where Tibetans have traditionally been a significant minority (Hualong Hui Autonomous County and Xunhua Salar Autonomous County). Tibetans also outmigrated from several of the outlying counties of Xining City. This was contrasted, however, by a strong growth in the Tibetan population of Ping’an District (the birthplace of the current Dalai Lama), also in Haidong City and close to Xining.

Tibetan migration to provincial capital cities

67Related to this last point, the third dominant pattern was very strong Tibetan and other minority migration to the provincial capitals of these provinces (Xining, Lanzhou, and especially Chengdu). The migration was also focused on the built-up urban districts of these cities, which appear to have been pulling in migrants from both nearby peripheries and from further afield across the Tibetan and Muslim areas. Contrary to the claims made by Roche et al. (2023), who argue that Tibetans experience segregation in these cities and barriers to settle in them, the rapid increase in resident (i.e. settled) Tibetan populations in Xining and Lanzhou (for which we have district-level data) also indicate an increasingly spread-out presence of Tibetans in all the core urban districts, not only those traditionally associated as Tibetan neighbourhoods. Our interviews with Tibetan residents in Chengdu confirm the same.

68For instance, while the population of Xining grew by almost 12 percent over the decade, to almost 2.5 million people, the Han component only increased by just under 8 percent, accounting for just under half the increase of 259,257 people. Minorities increased by 23 percent, accounting for just over half of the increase, and resulting in an increase of share by 2.6 percentage points, to almost 29 percent of the city population. This share was concentrated in certain downtown districts, such as Chengdong District, where the minority share rose by almost 5 percentage points, to 43 percent of the district population in 2020, or the Chengbei District, where the minority share also rose almost 5 percentage points, reaching over 14 percent. This is despite rapid Han growth in these districts as well, such as in the Chengdong District, where the Han population increased by 56,233 people, from 222,107 to 278,340 people, while the minority population increased by 73,451 people, from 137,581 to 211,032 people. Indeed, this is consistent with the analysis of Grant (2022) about how parts of Xining are becoming minoritised.

69Among these minorities, Tibetans were the fastest growing population in Xining, although from a smaller base than the Hui. The Tibetan population grew by almost 32 percent, reaching 160,049 people, while the Muslim (mostly Hui) population grew by almost 20 percent, reaching 442,253 people. The Tibetan share was still small, increasing by 1 percentage point to 6.5 percent of the total population in 2020 (or 10 percent if including Mongols and Monguors), which was roughly 23 percent (or 35 percent) of the total minority population in the city.

70The increase in the Tibetan population was fastest in two of the urban districts; Chengdong District, on the eastern part of the city and where the Tibetan market and main transportation hub are located, and Chengbei District, where most of the universities are located. The Tibetan population in Chengdong grew by 131 percent (i.e. more than doubled), and in Chengbei it grew by 186 percent (almost tripling). Muslims were still the dominant minority in Chengdong, which was historically the main Muslim district and where more than half of the Hui population of the downtown area still resided in 2020. However, the Tibetan population surpassed Muslims in Chengbei by 2020. Tibetans also grew rapidly in the two other downtown city districts, and Mongols and Monguors had similar growth rates in these four city districts, although from much smaller numbers. In aggregate terms, the Tibetan population was fairly evenly spread throughout all the city districts, in this case including the outlying district of Huangzhong where, as mentioned above, the Tibetan population declined, although the district still remained the most populous Tibetan jurisdiction of the city in 2020. Combined with the two counties of the municipality that also had Tibetan depopulation, this reflects the strong centripetal influence that the city had on Tibetans, even within this core administrative unit of the province.

72It is also apparent in the census data that Tibetans, in migrating out of their areas, seem to show a preference (at least in the aggregate) for migrating to these large capital cities rather than to prefecture seats. For instance, the main prefecture seat counties of Qinghai, such as Chabcha, Rebgong, Machen or Jyekundo, did not have Tibetan population growth rates that were more than would be expected from their probable local rates of natural increase (as discussed above). This implies that net migration (e.g. from other parts of the respective prefectures) was not necessarily contributing to the Tibetan population growth in these counties. In Barkham (Ch. Malkang) City, seat of Ngawa Prefecture, the minority population only grew by 3 percent over the decade, which would imply net outmigration.

74Whereas Chengdu was clearly the recipient of net Han in-migration, it is interesting that Xining was not. In contrast to Tibetans and other minorities, the moderate growth of the Han population in Xining was not much more than would be expected from their natural increase rates, and hence it is likely that Han net migration contributed only marginally to the Han population growth of Xining. Instead, Han outmigration from across Qinghai (as discussed above) would have tended to leave the province altogether (in aggregate terms, i.e. Han from counties might have moved to Xining, but then Han from Xining might have left the province, etc.). Lanzhou was between these extremes of Xining and Chengdu, with the census data seeming to indicate moderate in-migration of Han – the Lanzhou Han population grew by 19 percent over the decade, in addition to the minority population that grew by 53 percent, and the Tibetan population by 117 percent.

Muslims in Tibetan areas

75As a qualification to these three dominant patterns, the increasing demographic presence of Muslims in the Tibetan areas of Qinghai stands out strongly. In Qinghai Province overall, the Tibetan population grew by 10 percent, while the Muslim (mostly Hui) population grew by 15 percent. Moreover, Muslim populations in most Tibetan prefectures and counties in Qinghai have been increasing at a faster pace than Tibetans in these prefectures and counties, although from very small numbers, and this was clearly driven by net in-migration given the speed of growth in most cases. These patterns are effectively the inverse of Xining, where Tibetans grew faster than Muslims in even the traditionally Muslim district of Chengdong, as discussed above.

77The two groups were nonetheless spatially distributed quite differently, with 64 percent of Muslims concentrated in Golmud (Ch. Geermu) City, where they were already larger than the combined TMM population in 2010. This also reflects their niche role in working for and/or servicing the mining economy of this region, and to a certain extent substituting the Han in these roles, especially in Golmud where the increase of Muslims exceeded the decline of Han. The TMM population, on the other hand, was spread evenly throughout most of the counties of Tsonub, and hence was still larger than the Muslim population outside of Golmud. Moreover, in a county like Themchen (Ch. Tianjun), where about one third of Tibetans in the prefecture were living in 2020, both the Han and Muslim populations fell sharply, by more than two thirds for the Han, and by almost half for the Muslims, although the latter from a very small number. However, key loci of pressured confluence were more evident in other economically strategic counties such as Dulan or Terlin (Ch. Delingha) City, the prefecture seat, where Tibetans were still the largest “minority” minzu, but Muslims were catching up in population weight.

78This rapid shift in the composition of minorities from Tibetans to Muslims, however, was not typical to the broader Amdo region. For instance, in Malho (Ch. Huangnan) Prefecture, the Tibetan population grew by 7.6 percent, and the combined TMM population by 6.2 percent, in total from 221,899 to 235,697 people, while the Muslim population of the prefecture grew by 14.4 percent, from 18,522 to 21,196 people, slightly increasing their share of the prefecture population. However, over half of this Muslim population was concentrated in Chentsa (Ch. Jianzha) County, which has traditionally had a significant Muslim minority of about one fifth of the population. In Rebgong (Ch. Tongren), the prefecture seat, the Muslim population grew by almost 12 percent, versus almost 10 percent for Tibetans. Hence, despite the faster Muslim population growth rates leading to marginal increases in their population shares, this did not alter much of the pre-existing ethnic composition of the prefecture. These dynamics were similar in the other Tibetan prefectures of Qinghai besides Tsonub – even in Tsojang TAP, where the Muslim population was larger than the Tibetan in both 2010 and 2020, more than three quarters of this Muslim population was based in the one Hui autonomous county of the prefecture (Menyuan HAC), whereas Tibetans were concentrated in Kangthsa (Ch. Gangcha) County, where the Hui were a small minority.

Conclusion

81These insights into the changing minzu (or ethnic) population dynamics of Amdo raise a host of fascinating research questions that can only be answered through further fieldwork and related investigation into the local circumstances of each case. One of the more surprising findings with important implications is the apparent baby boom that took place for a decade from the mid-2000s to the mid-2010s among Tibetans in Qinghai and possibly in other parts of Amdo as well, which corroborates well with the relatively rapid population growth observed in many Amdo Tibetan counties in the 2010s. Against these unexpected Tibetan demographic dynamics, why have certain counties experienced such strong outmigration of Han? Has this been encouraged by or despite of government policy? Why have Muslims been migrating to Tibetan towns across Qinghai, but not to the other parts of Amdo, in Ganlho or Ngawa? In Qinghai specifically, how has this played out in terms of Tibetan-Muslim coexistence in the emerging urban economies, and why have the more explosive outbursts of Tibetan-Muslim tensions in the 1990s and 2000s, as Muslim migration into Tibetan towns was beginning to take hold, been contained over the past decade, despite the even stronger presence of Muslims in these towns? What has been driving the various patterns of Tibetan migration revealed by these aggregate data, and how have resettlements, boarding schools, or other policy initiatives shaped these patterns in specific localities? How has the extremely rapid growth of Tibetan communities in Xining, Lanzhou and Chengdu occurred and been experienced?

82Beyond the specificities of each local case, however, broader patterns can also be discerned. For instance, these census insights confirm earlier analyses (e.g. Fischer 2004, 2008a) that the dominant structural tendency facing these relatively poor peripheral areas is net outmigration towards more central locations, not net in-migration. With a few exceptions of slight net in-migration of Han to Tibetan areas in Qinghai (in Golog, Yulshul and Malho), outmigration was stronger among the Han than among Tibetans or local Muslims everywhere else in Amdo. Moreover, outmigration was strongest in counties with larger Han communities, established from earlier waves of in-migration, especially in the counties bordering Han Chinese areas along the eastern edge of Amdo (and Kham), or else in the mining areas of Tsonub in Qinghai. In contrast, the more remote Tibetan-dominated counties generally exhibited relatively rapid Tibetan population growth. The young population structures strongly suggest that this recent growth was not due to Tibetan in-migration but, instead, some of the highest levels of natural increase in China.

83These differences have driven the tendency for rising Tibetan and Muslim shares in the contemporary context of rapid development in this region. This tendency has been counteracted in the TAR by extremely high levels of subsidisation, but the less prioritised and more austere fiscal circumstances of Tibetan prefectures and counties outside of the TAR, which are subsumed and subordinated within their respective provinces, do not appear sufficient to produce the same effect. Hence, in the absence of such extreme subsidisation and strategic prioritisation, the trends across Amdo generally conform to the logical dynamics facing poorer peripheral regions, to such an extent that “minorities” will probably become the majority of Qinghai Province well before the next census in 2030, as discussed in the second section. In other words, minorities in Qinghai were already 49.5 percent of the population in 2020, up from 47 percent in 2010 and 46 percent in 2000. Hence, if their share continues to rise at the same pace as in the 2010s, we could expect it to reach 52 percent by 2030, with Tibetans as the largest non-Han minzu of the province.

84Indeed, this also sheds light on the future of Amdo Tibetans as increasingly urban, not just in their own towns or cities of their autonomous areas, but also in Xining, Lanzhou, and Chengdu. For instance, already in 2020 more than 10 percent of the Qinghai Tibetan population lived in Xining as residents, and in an increasingly spread-out pattern of residency across the five city districts. Such insights refute the claims of Roche et al. (2023), for instance, that Tibetans are obstructed from becoming residents in these cities, or else are segregated if they do, which does not appear to be the case from these census data or from our interviews with many of these Tibetan city residents. Amdo Tibetans might well still experience various forms of exclusion, subordination, or marginalisation within this rapidly urbanising context, especially in labour markets, as earlier argued in Fischer (2008a, 2014). However, the pace and patterns of change, as well as the possibilities of Tibetan agency to navigate this change and to appropriate the entirely new urban spaces as part of an Amdowan future, force us to rethink many of the conceptual frameworks that have been deployed up to the present to understand and interpret this region.

85The census data also speak strongly to the ongoing and emerging development, governance, and public policy challenges in Amdo Tibet, although space does not permit analysing these different dimensions here. Suffice to say that these rapid changes in population and migration over the 2010s have been occurring in a context of a rapid urbanisation and rural depopulation. And while the rapid changes over a short period of time contradict many of the conventional representations of population dynamics in eastern Tibet, they nonetheless accentuate many of the challenges that Amdo Tibetans face in the rapid socio-economic development of their region, particularly with respect to ongoing educational lags in a context of increasing integration into urban employment. Given that the Han and Muslim migrant populations are mostly concentrated in urban areas, the rapid urbanisation of Amdo Tibetans accentuates the confluence of these two streams of migrants – Tibetans from rural areas and Han and Muslims from the outside – and their competition over the limited and austere economic opportunities in these Tibetan towns. Moreover, these pressures can arguably persist even when Han population shares are declining if urban employment opportunities do not keep up with the needs of the rapidly urbanising and relatively young local Tibetan population. Indeed, the strong increase of Tibetan populations in the capital cities might itself be a response to such pressures. We need to have a better understanding of these subtilities to appreciate the grievances and responses that population changes might induce in such contexts.