Characteristics and Determinants of Livelihood Diversification of Different Household Types in Far Northwestern China

Abstract

Livelihood diversification is beneficial to mitigate economic and environmental risks and to improve livelihood sustainability and regional sustainable development. Unsettled herder households (UHH), settled herder households (SHH) and farmer households (FH) are different household types in far northwestern China whose livelihood diversification has not been fully explored. By applying a framework of livelihood diversification, this paper presents a comparative analysis of the characteristics and determinants of the diversification of the three household types. The results show that livelihood assets have been unequally distributed, with FH possessing the least assets; however, FH are better than UHH and SHH in the diversification of livelihood activities. Agriculture remains the most important livelihood source. The high-income groups of the three household types have a higher number of livelihood activities but do not necessarily hold an advantage in equality of livelihood activities. Labor capacity and income are positively related to the number of livelihood activities for the three household types. Livestock size is negatively associated with the number of activities of herders. Moreover, age and subsidy have negative impacts on the number of activities for UHH. Based on the findings, we provide policy suggestions on livelihood enhancement and sustainable and effective development of pastoral regions.

1. Introduction

Livelihoods are fundamentally interwoven with the environment [1,2,3]. The excessive dependence of human livelihoods on natural resources, which is apparent in the overgrazing, overfishing and overuse of marginal land, is a major cause of environmental degradation that restricts regional and/or international sustainability; the loss of ecosystem services also impairs human well-being and compels households to seek alternative livelihoods [4,5,6]. Effective ecological restoration and sustainable development initiatives and policies need to consider the livelihoods of local households and encourage their participation [7,8]. The concept of Sustainable livelihoods, which links socioeconomic and environmental concerns in order to achieve environmental sustainability and social sustainability [9], resonates with the UN Sustainable Development Goals [10]. Productive livelihoods are key to achieving sustainable development [8,10]. One important pathway towards sustainable livelihoods for the inhabitants of marginal environments involves the avoidance of long-term dependency on only one income source [11].

Livelihood diversification (LD), which is defined as “the process by which rural families construct a diverse portfolio of activities and social support capabilities in order to survive and to improve their standards of living” [12], has attracted widespread attention from scholars and policymakers in developing countries [13,14,15]. Most studies report that LD is beneficial for relieving pressure on the environment [16,17], reducing livelihood risks and vulnerabilities [18,19,20], improving livelihood resilience and sustainability [21,22] and alleviating poverty [23,24]. Exploring the characteristics and determinants of LD are crucial issues in ecologically fragile regions to inform future appropriate policies and interventions for livelihood enhancement, ecological conservation and regional sustainable development.

The literature on LD has focused on two general types of households: farmer households who live in sedentary rural societies and herder households who have traditionally relied on pastoralism. These two types of households are often discussed independently of one another because they are typically separated by larger geographical, ethnic and cultural divides [25]. The studies on the LD of farmer households have tended to examine crop diversification and engagement in non-agricultural activities [26,27,28,29,30]. Herders have been diversifying into sedentary agriculture and non-agricultural activities to cope with socio-environmental risks, reduce rangeland pressure and increase income [18,20,31,32]. In terms of diversification motives, two broad perspectives have been distinguished: push (necessary) factors and pull (choice) factors [33,34]. Although a number of studies have argued regarding these two motives [26,35], addressing this issue by considering them in isolation is insufficient. The causes of diversification are multi-faceted [20], and push and pull factors can affect LD together. The impact of individual characteristics, family characteristics, and location on LD has been broadly discussed [25,36,37,38]. Based on the prevalent Sustainable Livelihood Framework (SLF) proposed by the Department for International Development (DFID), some researchers have analyzed the impact of livelihood assets that include human capital, social capital, natural capital, physical capital and financial capital on diversification strategies [23,39,40]. Disproportionate access to assets is often related to differing levels of LD.

Diversification is an infinitely heterogeneous social and economic process, and the research about this topic is supposed to emphasize the importance of the local context to therefore suggest policies tailored according to local circumstances [12,41,42]. Northern Xinjiang is a typical ecologically fragile region in China, where seasonal migration has been a traditional livelihood strategy for dwellers for thousands of years. Nevertheless, due to the fragile ecological conditions, grassland degradation and livelihood vulnerability of communities, development goals to achieve settlement or semi-settlement were established after the Northern Xinjiang Pastoralist Sedentarization Conference held in 1986. Meanwhile, the Household Responsibility System spread to these remote areas of China. Afterwards, the residents in this area were divided into two groups: herders and farmers. The difference is that the farmers do not have the right to use grassland but possess larger farmland plots to sustain a sedentary lifestyle. The Pastoralist Sedentarization Policy (PSP) has been implemented for over 30 years, and many ecological conservation and pastoral economic and social development initiatives have been implemented to alleviate the pressure of smallholder livelihoods on the pasture and promote sustainable development of the region. Currently, herder households can be divided into two distinct household types: unsettled herder households (UHH), who still graze for at least one season, and settled herder households (SHH), who do not practice mobility. Adaptive settlement outcomes of herders weigh policy intervention, grazing risks and settlement risks. Diversification has been recognized as a core livelihood strategy among many rural and pastoral regions [43,44,45,46]. Previous studies in northern Xinjiang have focused on research into the LD of the herders [18,47,48], whereas the livelihoods of farmer households (FH) has not been fully considered, but they are an important part of the local population whose livelihoods are also closely linked to the local environment and regional sustainable development.

Northern Xinjiang is an ideal study area to compare the characteristics of the LD of UHH, SHH and FH and to explore LD options while allowing these groups to maintain their current adaptive settlement status. Therefore, using Fuyun County, northern Xinjiang, as a case study, the objectives of this paper are as follows: (1) to suggest a conceptual framework for LD with reference to the current literature; (2) to present a case study that compares the characteristics of the LD of UHH, SHH and FH; (3) to explore and compare the determinants of LD for each household type; and (4) to provide policy recommendations towards promoting livelihood enhancement and sustainable and effective development of pastoral regions.

2. Conceptual Framework of Livelihood Diversification in Pastoral Regions of China

In terms of the measurement of the extent of LD, although a livelihood includes not only income (cash and in kind) but also social institutions, access to social and public services, etc. [12], most studies equate the extent of LD with income diversification as the other aspects of a livelihood are difficult to quantify [14,26,39]. Nevertheless, regarding the classification of income, little consensus exists. Barrett et al. [33] depicted three ways to classify household income: (1) Sector (e.g., farm vs. non-farm); (2) function (wage vs. self-employment); and (3) space (local vs. migratory). Tsegaye et al. [49] identified five income sources: Livestock, crops, non-farm non-pastoral income, relief aid and remittances. Liao et al. [47] listed six sources of income: Livestock, crops, wages, herding fees, government subsidy, and small business. For heuristic purposes and for future comparisons of the evidence across surveys, this study focuses on the definition of LD by Ellis and refers to previous LD empirical studies, as this definition has been widely used and is consistent with the actual situation in the pastoral regions of China. That is, to survive and improve their living standard, smallholders not only participate in diverse livelihood activities but also receive subsidies from the government that are an important part of their income and livelihood. We use income diversification to depict LD and divide income sources into 4 types according to the pastoral situations in China, including agricultural activities, non-agricultural activities, land rent and subsidies. Agricultural activities are practices that relate to agriculture on the householders’ own land and/or on the land that they rent, including raising their own livestock, crop production, and herding for other households. Non-agricultural activities are subdivided into casual labor, wage labor and small business. Land rent is a capital income that benefits from the policy of promoting land transfer for scale management. Subsidies are payments from the government, and these include allowances for elderly and poor families and an ecological subsidy that arises from the Grassland Ecological Protection Award Policy (GEPAP), which targets the protection of grassland, guarantees the supply of livestock products and improves household income.

Evidently, smallholders combine different options for agricultural activities, non-agricultural activities and land transfer while having different access to subsidies. This income classification approach allows for an in-depth analysis of the different dimensions of LD, for example, to analyze the characteristics of overall LD, diversification of actual livelihood activities, diversification in agricultural activities, diversification in non-agricultural activities, and the dependence of livelihoods on land rent and subsidies. A detailed description of the measurement methods and indexes of LD is presented in Section 3.3.1.

This study also aims to explore the factors that determine LD to explain smallholders’ behavior and inform future diversification policies and interventions. Based on the prevalent SLF proposed by the DFID, we propose a conceptual framework of LD in pastoral regions of China (Figure 1), which focuses on the effect of livelihood assets, the vulnerability context and the transforming structures and processes on LD. First, livelihood assets under smallholders’ control, are considered to be critical factors in the determination of LD for households in the same region [23,39,50]. In pastoral regions, labor, land, livestock and wealth constitute the main livelihood assets [47,49]; therefore, exploring the role of these assets in LD could be targeted. In the context of pastoral vulnerability, which is outside a person’s control, such as drought, snowstorm, population growth, land degradation, uncertainty in climate, agriculture expansion, urbanization, industrialization, and the seasonality of price, production and employment, smallholders seek LD to alleviate risks and uncertainty. Transforming structures and processes can create direct feedback for the vulnerability context, influence access to assets and affect smallholders’ extent of LD. For instance, settlement policy and the GEPAP aim to reduce grassland degradation (vulnerability context) and simultaneously increase herders’ living standard (livelihood assets), facilitate sedentarization and increase the extent of LD. In this study, we focus on the role of livelihood assets and subsidy policy on the LD of each household type, and we qualitatively discuss the impact of the vulnerability context and settlement policy on the LD of different household types.

3. Materials and Methods

3.1. Study Area

This study was conducted in Fuyun County (45°00′–48°03′ N, 88°10′–90°31′ E), which is located in far northwestern China and is under the administration of Altay Prefecture. It extends from the southern base of the Altai Mountains to the area north of the Junggar Basin, with elevation descending from north to south (Figure 2). Fuyun County lies within semi-arid and arid climatic belts of the temperate climate zone and is fed by the Irtysh River and the Ulungur River, which originate in the Altai Mountains. The mean annual temperature is 4.60 °C (2007–2016), and the mean annual precipitation is 208.41 mm (2007–2016) [53].

Fuyun County has an area of approximately 33,200 km², with a human population of 97,113 (2015). Kazakh and Han are the two main ethnic groups, which account for 73.51% and 21.41% of the population, respectively. The county contains 10 towns and 76 villages, and the villages are characterized as three types: farm village, pasture village and newly settled village. Farm villages mainly comprise FH, and only a very small number of herder households have settled here. Some farm villages are distributed in the piedmont plain and basins, and some are distributed along the banks of the Ulungur River. Pasture villages and newly settled villages comprise herder households except for a few young FH who bought houses in newly settled villages and now live there mainly along the banks of the Ulungur River. The newly settled villages, with unified housing standards, have been successively built by the government since 2006 to encourage herders to settle, improve their living conditions and reduce the pressure on the grassland. The herder households who moved to the newly settled villages can buy the houses at a subsidized rate to pay approximately half of the cost, and they receive 3.3 ha of farmland, on average, as a bonus to plant forage.

3.2. Data Collection

The data were collected from household surveys, including questionnaire survey and semi-structured interviews. Before conducting the household surveys, in 2015 and 2016, we conducted discussions with government officials of Altay Prefecture and Fuyun County on pastoral livelihoods, ecological conservation, the implementation of Pastoralist Sedentarization Policy and other major issues of regional development; we also conducted preliminary surveys to investigate the local livelihoods. The household surveys were conducted from July to August and October to November in 2017 and in June 2018. We investigated nine townships in succession, other than the township that is the seat of Fuyun County, that are mainly occupied by an urban population. The surveyed villages were selected after a discussion with town leaders according to the village type and geographical and economic conditions. In each village, we interviewed village leaders to obtain a comprehensive understanding of the household livelihoods in the village and the leaders’ opinions about the determinants of LD. If the village leaders were absent, we consulted one to two knowledgeable senior residents. Afterwards, we selected the respondent households through random sampling.

The heads of household were the primary representative subjects for the survey, which took one to two hours. In case of the absence of the head of household, we interviewed the second decision maker in the household. The questionnaire includes basic information on the family members, the livestock, grassland and farmland, other income-generating activities, and the income from each source. In the follow-up interviews, we discussed human welfare, land tenure, Pastoralist Sedentarization Policy and grassland policy. Since almost all of the farmers and herders were Kazakhs, three Kazakh students who are fluent in Mandarin and Kazak were recruited as language interpreters and were trained by the first author to become familiar with the questionnaire.

Based on this approach, we investigated 19 farm villages, 10 pasture villages and 11 newly settled villages, and collected 207, 111, and 117 valid questionnaires, respectively. The majority of household heads and interviewees were male, which accounted for 92% and 63% of the total sample, respectively. Through further sorting, these questionnaires represented 90 UHH, 159 SHH and 186 FH.

3.3. Data Analysis

3.3.1. Measures of Livelihood Diversification

The measurements of LD draw on biodiversity indicators, and two approaches prevailed. One is a one-dimensional index determined by counting the number of income-generating activities [26,39,54]. The other is a two-dimensional approach that considers the number of income-generating activities and the proportion of each income, such as the Inversed Herfindahl-Hirschman Index [14], Shannon-Weiner Diversity Index [47] and Simpson Index [55]. In this paper, we use the simple Count index to indicate the richness of LD and the Inversed Herfindahl-Hirschman Index (IHHI) to indicate the evenness of LD because these two indexes have the same range of values, which is suitable for comparison.

As mentioned in Section 2, 8 income sources exist: livestock, crops, herding fees, casual labor, wages, small business, land rent and subsidies. We used the 5-year average sales rate of livestock in Fuyun County from 2011 to 2015 to calculate the number of livestock sold (The 5-year average sales rate of livestock in Fuyun County from 2011 to 2015 for goats (39%), sheep (42%), cattle (27%), horses (11%), and camels (10%) (Fuyun Statistical Yearbooks 2011–2015)). The price of livestock was reported from the interviewee. Crop income consisted of cash income and subsistence. The herding fee was calculated according to the number of livestock herded for other people, the length of herding time, and the herding price for each type of livestock. Income from other sources was directly reported from the interviewee. If a household had income from livestock, crops and casual labor, the richness of LD is 3. The IHHI was calculated by using the following equation:

$$\mathrm{IHHI}={\left({\displaystyle \sum }_{i=1}^{n}I{P}_i{}^2\right)}^{-1}$$

(1)

where $I{P}_i$ is the proportion of income source i to total income and n is the number of income sources for a household.

We calculated the overall LD (LD_o), overall LD without considering the subsidy (LD_o-s), diversification of livelihood activities (LD_a), agricultural LD (LD_ag), and non-agricultural LD (LD_nag) to analyze the extent of LD for the different aspects. The differences lie in the number of income sources considered. LD_o includes all of the 8 income sources, whereas LD_o-s includes 7 income sources but not subsidies. LD_a includes income obtained from the activities that require input of labor and time (i.e., not considering subsidies and land rent). LD_ag includes three agricultural income sources: livestock, crops, and herding fees. LD_nag includes the three non-agricultural income sources: casual labor, wage labor, and small business.

3.3.2. Kruskal-Wallis H Test and Chi-Square Test

The Kruskal-Wallis H test is a non-parametric method for testing whether two or more populations originate from the same distribution. It is similar to the one-way analysis of variance (ANOVA) but does not apply the ANOVA normality assumption [56]. This method was used to test whether significant differences exist in the household portfolios and LD (except for the variables of grassland tenure, LD in land rent, and LD in subsidies) among UHH, SHH and FH. The main procedures are as follows: (1) Rank all the data of a variable from all the groups and (2) calculate the Kruskal-Wallis H test statistic:

$$\mathrm{H}=\frac{12}{N\left(N+1\right)}{\displaystyle \sum }_{i=1}^k\frac{R_i{}^2}{n_i}-3\left(N+1\right)$$

(2)

where N is the total sample of a variable across all groups, k is the number of groups, $R_i$ is the sum of the ranks for group i, and $n_i$ (i = 1, 2, …, k) represents the number of samples in group i.

The distribution of H can be approximated by a chi-squared distribution with k–1 degrees of freedom. If the test statistic is in the critical region, the null hypothesis that the groups are of approximately the same form can be rejected, and the Kruskal-Wallis H post hoc test can then be used to conduct multiple comparisons.

The chi-square test was used to examine whether significant differences exist in the categorical data (i.e., grassland tenure, LD in land rent and LD in subsidies) among UHH, SHH and FH.

3.3.3. Regression Models for Livelihood Diversification

Multiple linear regression models have been widely used to quantitatively analyze the determinants of LD [25,39], and we used this model for each household type for this purpose. We calculated the extent of LD for the different aspects and selected the richness of LD_a as the dependent variable, which will be explained in Section 4.4. Based on previous studies [37,39,40], specific conditions in the study area and the availability of data, 6 independent variables were selected and are presented in

Table 1. Descriptions of the independent variables.

	Variables	Value Assignment
Labor	Age/x1	Age of household head
	Labor capacity/x2	The weighted average of the normalized values of the number of the labor force (0.5), average education level of the labor force (0.25), and average level of the Mandarin proficiency of the labor force (0.25). (1) The number of the labor force is the sum of the labor force value of each family member according to age and health condition. Assign 0 for infants, students, disabled individuals and members with serious illness, 1 for members aged 18–60 years, and 0.5 for people aged less than 18 years or older than 60 years. (2) The average education level of the labor force is as follows: 1 = illiteracy; 2 = primary school; 3 = middle school; 4 = high school; and 5 = college and above. (3) The average level of the Mandarin proficiency of the labor force is as follows: 0 = cannot speak Mandarin; 1 = can speak Mandarin for simple communication; and 2 = can speak Mandarin fluently
Land	Farmland/x3	Actual planting area (mu)
Livestock	Livestock size/x4	Index is in sheep units. Goat, cattle, horse and camel are equivalent to 0.8, 6, 6, and 7.5 sheep units, respectively (The sheep unit calculation was performed according to the Grass and Animal Balance Implementation Plan in Xinjiang).
Wealth	Total income/x5	Equivalent to the total income mentioned in Section 3.3.1 (thousand yuan)
Policy	Subsidy 1/ x6	Whether the household has subsidy income. 0 = no; 1 = yes

¹ For UHH and SHH, subsidy refers to the ecological subsidy, whereas for FH, subsidy refers to government allowances for poor and elderly families.

. Labor, land, livestock, and wealth are crucial livelihood assets for households. Subsidy policy has an uncertain impact on the diversification of livelihood activities: on the one hand, it increases the financial capital required to engage in livelihood activities; on the other hand, it may reduce incentives to adopt autonomous strategies to increase income. Exploring the impact of subsidy on the richness of LD_a is beneficial for understanding the relationship between policy and smallholders’ autonomous strategies and developing policies towards improving smallholders’ development abilities.

We used the variance inflation factor (VIF) to test for multicollinearity among the independent variables. The VIFs are less than 4, which suggests that no multicollinearity exists between the independent variables, and regression estimation can be performed.

The form of the model can be represented as follows:

$$\mathrm{Y}={\beta }_0+{\displaystyle \sum }_i{\beta }_i{x}_i+\epsilon$$

(3)

where Y refers to the richness of LD_a, ${\beta }_0$ is the intercept, ${\beta }_i$ is the coefficients of the i-th independent variable, $x_i$ is the value of the i-th independent variable, and $\epsilon$ is the error. Ordinary least squares (OLS) was used to estimate the unknown parameters. Then, we used the White test to examine heteroscedasticity in the models and found that there was heteroscedasticity in the models of SHH and FH. Therefore, robust standard error was used to eliminate heteroscedasticity in the models of SHH and FH. Poisson regression models were also used to estimate, and the results in this case were not meaningfully different from the results of the OLS model of UHH and the robust OLS models of SHH and FH (Table 5 and Table 6). Data analysis was conducted with the Stata 13 software.

4. Results

4.1. Household Portfolios

Labor, land, livestock and wealth are the main livelihood assets for smallholders in pastoral regions. In terms of these assets, UHH, SHH and FH have statistically significant differences, with the exception of labor capacity (

Table 2. Livelihood assets among the different household types.

Variable		UHH	SHH	FH
		(N = 90)	(N = 159)	(N = 186)
Labor	Age	53 (14) a	47 (13) b	48 (13) b
	Labor capacity	0.35 (0.14)	0.35 (0.15)	0.36 (0.15)
	Number of labor force	2.8 (1.0) a	2.3 (0.9) b	2.4 (0.9) b
	Average education level of labor force	2.6 (0.5)	2.7 (0.7)	2.7 (0.7)
	Average level of Mandarin proficiency of labor force	0.31 (0.33) b	0.49 (0.44) a	0.48 (0.42) a
Land	Grassland tenure (0/1)	0.99 (0.11) a	0.31 (0.46) b	0.00 (0.00) c
	Farmland tenure area (mu)	28.0 (23.4) a,b	31.9 (21.5) a	27.2 (22.8) b
	Farmland use area (mu)	18.7 (20.7) a,b	17.2 (21.0) b	27.3 (34.1) a
Livestock	Livestock size	254 (177) a	139 (127) b	74 (72) c
Wealth	Total income (thousand yuan)	96.2 (60.9) a	76.9 (43.6) a	63.1 (59.5) b
	Livestock (%)	48.6 (22.5) a	33.1 (21.5) b	23.0 (17.0) c
	Crops (%)	14.3 (17.7) b	17.9 (20.7) b	36.4 (26.4) a
	Herding fees (%)	8.2 (18.1) a	0.0 (0.0) b	0.4 (5.1) b
	Casual labor (%)	3.2 (10.1) b	11.5 (19.2) a	12.5 (21.1) a
	Wage labor (%)	5.0 (12.1) b	11.2 (20.6) a,b	13.8 (23.6) a
	Small business (%)	2.4 (7.9) b	8.2 (18.8) a,b	7.6 (16.4) a
	Land rent (%)	2.0 (6.0) b	4.3 (8.4) a	2.1 (7.4) b
	Subsidies (%)	16.2 (15.4) a	13.8 (17.8) b	4.3 (8.4) c

Notes: Data in brackets refer to the standard deviation. Figures labelled with different superscript letters within rows indicate a significant difference (p < 0.1) between household categories, and the value corresponding to superscript a is significantly higher than that of superscript b and the value corresponding to superscript b is significantly higher than that of superscript c. UHH: unsettled herder households; SHH: settled herder households; FH: farmer households.

).

(1) Labor. The age of the household head of the UHH is significantly higher than that of the SHH, with an average age of 53 years, which indicates that older herders are more reluctant to change their traditional seasonal migration ideology and settle down completely, and young people are flexible in adapting to a new lifestyle. No significant differences exist among the household types in labor capacity, and all groups possess a low mean value. The average number of the labor force of UHH, SHH and FH are 2.8, 2.3, and 2.4, respectively. The average education level of the labor force for the three household types are all between primary school and middle school education. The average level of Mandarin proficiency of the labor force of the UHH, SHH and FH are 0.31, 0.49, and 0.48, respectively, which shows that most of the labor force in Fuyun County cannot speak Mandarin even for simple communication. Although FH and SHH have a relatively good command of Mandarin, they still have great potential to improve.

(2) Land. Almost all UHH have grassland tenure, except for a few households that rent grassland for grazing. Although SHH are no longer engaged in grazing, 31% of them have grassland tenure. As grassland was distributed to herders in 1984 and has never changed since then, FH do not have grassland tenure. Moreover, because of the implementation of the Pastoralist Settlement Project from 2009 to 2015, herder households included in the project received 3.3 ha of farmland on average for planting forage; therefore, FH no longer have an advantage of a farmland tenure area and are significantly lower than SHH. However, FH have strength in planting skills and are higher than UHH and SHH in farmland use area. Restricted by planting skills and the insufficient infrastructure of newly reclaimed land and influenced by the local government’s initiatives to encourage land transfer, UHH and SHH have transferred some farmland to other people.

(3) Livestock and wealth. In terms of the livestock size, UHH > SHH > FH, which also indicates the order of the livestock income share. The total income of UHH and SHH are significantly higher than that of FH. In decreasing order of importance, for UHH, the main sources of income are livestock, subsidies, crops and herding fees; for SHH, the main sources of income are livestock, crops, subsidies and casual labor; and for FH, the main income sources are crops, livestock, wages and casual labor.

4.2. Characteristics of Livelihood Diversification

LD has been an important adaptation strategy for smallholders in response to the uncertainty in the environment and market economy. The LD characteristics of the UHH, SHH and FH are listed in

Table 3. Characteristics of the livelihood diversification of the different household types.

LD	Index	UHH	SHH	FH
LDo	Count	3.30 (0.92)	3.47 (1.02)	3.33 (0.96)
	IHHI	2.36 (0.68)	2.53 (0.73)	2.37 (0.72)
LDo-s	Count	2.56 (0.90) b	2.85 (0.90) a	2.89 (0.87) a
	IHHI	1.89 (0.68) b	2.19 (0.66) a	2.22 (0.70) a
LDa	Count	2.42 (0.89) b	2.57 (0.84) b	2.77 (0.85) a
	IHHI	1.82 (0.66) b	2.02 (0.63) a	2.15 (0.70) a
LD ag	Count	1.91 (0.57) a	1.62 (0.51) b	1.78 (0.51) a
	IHHI	1.53 (0.46) a,b	1.41 (0.43) b	1.50 (0.46) a
LDnag	Count	0.51 (0.67) b	0.96 (0.70) a	0.99 (0.72) a
	IHHI	0.48 (0.61) b	0.88 (0.59) a	0.91 (0.61) a
Land rent	/	0.13 (0.34) b	0.28 (0.45) a	0.12 (0.32) b
Subsidies	/	0.74 (0.44) a	0.62 (0.49) b	0.45 (0.50) c

Notes: Data in brackets refer to the standard deviation. Figures labelled with different superscript letters within a row indicate a significant difference (p < 0.1) between household categories, and the value corresponding to superscript a is significantly higher than that of superscript b and the value corresponding to superscript b is significantly higher than that of superscript c. UHH: unsettled herder households; SHH: settled herder households; FH: farmer households.

.

Regarding the richness and evenness of LD_o, no significant difference exists among these three types of households. The richness of LD_o in Fuyun County is between 3 and 4, and the evenness of LD_o is between 2 and 3. If the subsidy income is not considered in the calculation of indexes of LD_o, the richness and evenness of LD_o-s of SHH and FH are significantly higher than that of UHH, which shows that UHH are more dependent on subsidy income. That is, 74% of UHH have subsidies, of which 95.5% have a grassland ecological subsidy that varies greatly among herder households, with a maximum of 92,000 yuan, a minimum of 1500 yuan, a median of 18,000 yuan, and a mean of 20,730 yuan. Of the SHH, 62% have subsidy income, of which 75.8% have a grassland ecological subsidy. Of the FH, 45% have subsidy income, which mainly consists of allowances for elderly and poor families and is far below the grassland ecological subsidy, with a mean of 3759 yuan.

In terms of their livelihood activities, i.e., not considering subsidies and land rent, the richness of LD_a is significantly higher for FH than for UHH and SHH, whereas FH and SHH exhibit a significantly higher evenness of LD_a than UHH. Generally, FH participate in the most livelihood activities, and UHH are involved in the least livelihood activities.

Agricultural activities, including raising livestock, crop production, and herding for other people, are the main livelihood sources for smallholders. In terms of the richness of LD_ag, the values for UHH and FH are significantly higher than that for SHH. For the evenness of LD_ag, the mean is higher for UHH and FH than for SHH; however, a statistically significant difference exists only between FH and SHH.

All UHH raise livestock, and 25.6% of them herd for some SHH and FH. In total, 64% of UHH plant forage crops, such as alfalfa or silage maize, to supplement the shortage of forage in winter (Figure 4A). UHH possess the highest livestock diversity, with 67% raising goats and/or sheep, cattle, camels and/or horses and 88% keeping at least two livestock types (Figure 3A).

SHH have the highest proportion that have transferred land (Table 3) because 48% of those in the sample dwell in newly settled villages, where the local government has encouraged villagers to transfer their farmland to skilled farmers from other regions who can manage the newly reclaimed farmland at scale and install irrigation facilities. However, the government officials of Altay Prefecture and Fuyun County and some households have complained that the new land managers have adopted predatory approaches to land management by planting commercial crops at a large scale, which can generate good economic benefits but imposes a serious burden on the land and leads to a decline in soil fertility. When the land lease expires, the land is no longer suitable for cultivating.

The livestock diversity is lower for SHH than for UHH, with an apparent decline in the percent of horses/camels from 68% for UHH to 39% for SHH (Figure 3A,B). Regarding the breeding model, only 21.4% of SHH raise livestock in pens or near the settlements throughout the year; 54.7% of them pay herders to herd their livestock between April/May to October. Furthermore, the livestock from 23.9% of SHH freely graze on the grassland of relatives. The cropping structure between SHH and UHH are similar, with mainly forage crops being planted (Figure 4A,B).

FH have a relatively balanced income between livestock and crops; thus, they possess a strength in LD_ag (Table 2). FH have the lowest livestock diversity and the highest cropping diversity. The most common combination of livestock types for FH is goats/sheep and cattle, which accounts for 51% of the sample (Figure 3C), and only 19% of FH keep camels/horses. In terms of the breeding model, 47.9% of FH raise livestock in pens or near the settlements throughout the year; the proportion who pay for herding and free herding by relatives account for 46.8% and 5.4%, respectively. The main crop type of FH is still forage, whereas the percent of grain and commercial crops show obvious increases compared to UHH and SHH (Figure 4).

The proportions of farmland abandonment for UHH, SHH and FH are 12.2%, 12.6%, and 8.6%, respectively. The three main reasons are water shortages (28.9%), poor soil fertility (28.9%), and the flooding of riverside land (17.8%). Approximately one-third of the households (31.5%) stated that water shortages are a major agricultural production pressure.

LD_nag is notably lower than LD_ag. Concerning the richness and evenness of LD_nag, the values for SHH and FH are significantly higher than that for UHH. As the primary labor force of UHH are engaged in grazing and lack non-agricultural employment skills and information, 58.9% of UHH do not participate in non-agricultural activities, whereas the proportions of SHH and FH without non-agricultural activities are 24.5% and 24.7%, respectively. Most SHH and FH participated in at least one non-agricultural activity. In decreasing order of participation, the main non-agricultural activities for each household type are casual employment, wage employment and small business.

4.3. The Extent of Livelihood Diversification of Different Income Levels

Trisection, quartiles or quintiles have been widely used to divide income levels to examine livelihood diversification variation over income strata [14,36,38]. Based on previous studies and our sample size, each household type was subdivided into three income groups, with each group having the same sample size. These groups are a low-income group, medium-income group, and high-income group. As shown in

Table 4. Comparison of the extent of the LD at different income levels.

Household Types	LD	LD Indexes	Low-Income Group	Medium-Income Group	High-Income Group
UHH	LDo	Count	3.00 (0.95) b	3.33 (0.76) a,b	3.57 (0.97) a
		IHHI	2.33 (0.80)	2.35 (0.67)	2.39 (0.58)
	LDa	Count	2.20 (0.89)	2.47 (0.78)	2.60 (0.97)
		IHHI	1.84 (0.73)	1.81 (0.66)	1.81 (0.60)
SHH	LDo	Count	3.19 (0.96) b	3.43 (0.97) a,b	3.79 (1.06) a
		IHHI	2.43 (0.76)	2.55 (0.69)	2.61 (0.75)
	LDa	Count	2.34 (0.83) b	2.51 (0.80) b	2.87 (0.81) a
		IHHI	1.95 (0.69)	2.03 (0.61)	2.09 (0.60)
FH	LDo	Count	2.94 (0.88) c	3.32 (0.76) b	3.76 (1.05) a
		IHHI	2.23 (0.71)	2.42 (0.63)	2.46 (0.80)
	LDa	Count	2.35 (0.85) c	2.79 (0.73) b	3.16 (0.77) a
		IHHI	1.95 (0.76) b	2.21 (0.61) b	2.30 (0.67) a

Notes: Data in brackets refer to the standard deviation. Figures labelled with different superscript letters within a row indicate a significant difference (p < 0.1) between household groups, and the value corresponding to superscript a is significantly higher than that of superscript b and the value corresponding to superscript b is significantly higher than that of superscript c. UHH: unsettled herder households; SHH: settled herder households; FH: farmer households.

, the groups with different income levels of each household type have significant differences in the richness of LD_o. Specifically, for UHH and SHH, the richness of LD_o is significantly higher for the high-income group than for the low-income group. For FH, significant differences exist between every two groups, with the high-income group receiving the highest Count value and the low-income group receiving the lowest Count value. With respect to the evenness of LD_o, although the IHHI value increases as the income level increases, no significant differences were observed among the groups for each household type.

In terms of LD_a, no significant difference exists between the different income groups of UHH, but the Count value of LD_a increases as the income level increases. For SHH, the richness of LD_a is significantly higher for the high-income group than for the low- and medium-income groups. For FH, the richness and evenness of LD_a are both significantly higher for the high-income group than for the low- and medium-income groups. Taken together, the Count indicator is more representative of the difference in LD between different income groups. The high-income group is more likely to seek diversification, especially the people with high income from SHH and FH.

4.4. Determinants of Livelihood Diversification

Subsidies and land rent represent income from government transfers and capital revenue, respectively, which improve the extent of LD_o. As shown in Table 3, subsidies provide an income source for most households, but it is out of the smallholders’ control; the number of households that have transferred land is low. Moreover, subsidies and land rent do not require an input of labor and time. Therefore, exploring the role of livelihood assets and subsidies on the diversification of livelihood activities (LD_a) that are under household control is more effective than considering the LD_o as the dependent variable. According to Section 4.2 and Section 4.3, the richness indicator is more representative than the evenness indicator; thus, we take the richness of LD_a as the dependent variable. A multiple linear regression model and a Poisson regression model were conducted for each household type, and the results of the models are presented in

Table 5. Estimation of the determinants of the richness of LD_a with multiple linear regression model.

Variables	UHH	SHH	FH
	Coef.	Coef.	Coef.
Age	−0.015 ** (0.006)	0.003 (0.005)	−0.0001 (0.005)
Labor capacity	1.756 *** (0.637)	1.076 ** (0.464)	1.130 ** (0.474)
Farmland	0.006 (0.004)	0.005 (0.006)	−0.005 (0.004)
Livestock size	−0.002 ***(0.001)	−0.001 * (0.001)	−0.0003 (0.001)
Total income	0.008 *** (0.002)	0.005 ** (0.002)	0.006 ** (0.003)
Subsidy 1	−0.358 * (0.198)	−0.010 (0.134)	−0.119 (0.128)
Constant	2.469 *** (0.322)	1.771 *** (0.236)	2.218 *** (0.260)
Adjusted R2	0.334	0.136	0.145
F-statistic	8.45	4.15	7.37
Prob > F	<0.001	<0.001	<0.001

Note: ¹ For UHH and SHH, subsidy refers to the ecological subsidy, whereas for FH, subsidy refers to government allowances for poor and elderly families. For UHH, standard errors are given in parentheses; for SHH and FH, robust standard errors are given in parentheses. *** p < 0.01; ** p < 0.05; * p < 0.1. UHH: unsettled herder households; SHH: settled herder households; FH: farmer households.

and

Table 6. Estimation of the determinants of the richness of LD_a with Poisson regression model and average marginal effects.

Variables	UHH		SHH		FH
	Coef.	dy/dx	Coef.	dy/dx	Coef.	dy/dx
Age	−0.006 ***	−0.015 ***	0.001	0.003	−0.0001	−0.0003
	(0.002)	(0.005)	(0.002)	(0.004)	(0.002)	(0.005)
Labor capacity	0.731 ***	1.770 ***	0.420 **	1.080 **	0.425 ***	1.176 ***
	(0.227)	(0.548)	(0.171)	(0.439)	(0.161)	(0.444)
Farmland	0.002	0.006	0.002	0.004	−0.002	−0.004
	(0.002)	(0.004)	(0.002)	(0.005)	(0.001)	(0.003)
Livestock size	−0.001 ***	−0.002 ***	−0.0004 *	−0.001 *	−0.0001	−0.0002
	(0.0003)	(0.001)	(0.0003)	(0.001)	(0.0003)	(0.001)
Total income	0.00 ***	0.007 ***	0.002 **	0.004 **	0.002 **	0.005 **
	(0.001)	(0.002)	(0.001)	(0.002)	(0.001)	(0.002)
Subsidy 1	−0.148 *	−0.359 *	−0.002	−0.005	−0.048	−0.133
	(0.083)	(0.199)	(0.051)	(0.131)	(0.045)	(0.125)
Constant	0.911 *** (0.129)	/	0.636 *** (0.091)	/	0.823 *** (0.092)	/
Pseudo R2	0.037		0.014		0.013
Wald chi2 statistic	55.79		27.80		46.68
Prob > chi2	<0.001		<0.001		<0.001

Note: ¹ For UHH and SHH, subsidy refers to the ecological subsidy, whereas for FH, subsidy refers to government allowances for poor and elderly families. Robust standard errors are given in parentheses of corresponding columns of Coef.; Delta-method standard errors are given in parentheses of corresponding columns of dy/dx. *** p < 0.01; ** p < 0.05; * p < 0.1. UHH: unsettled herder households; SHH: settled herder households; FH: farmer households. Average marginal effects (dy/dx) are calculated for comparison with the results of multiple linear regression models.

.

The common factors that are significantly correlated with the richness of LD_a are labor capacity and total income. The higher the labor capacity is, the greater the opportunity for livelihood activities. The total income is positively related to the richness of LD_a, which further confirms that the households with high income in each household type are more likely to seek diversification. Livestock size is significantly and negatively related to the richness of LD_a of UHH and SHH. When more livestock is owned, the need for more labor force and time from the households for livestock raising is greater, which leaves fewer opportunities to participate in other activities.

Age is significantly and negatively related to the richness of LD_a of UHH, which indicates that young UHH are more likely to engage in diversified livelihood activities. Interestingly, the subsidy that increases the extent of the LD_o of UHH is significantly and negatively correlated with their richness of LD_a. The UHH without a subsidy tend to have a higher number of livelihood activities.

5. Discussion

Previous studies about LD in the pastoral context have focused on herder households that have access to grassland or whose elder generation has access to grassland. This study, which used Fuyun County located in northern Xinjiang as an example, considered the FH that do not have grassland tenure due to the historical land contracting system but also have negative externalities to grassland. Furthermore, this paper compared the characteristics and determinants of the LD of UHH, SHH and FH to provide tailored development policies for differentiated households and to promote sustainable and effective development of pastoral regions.

Compared to herder households, FH do not have grassland property, and their advantages in terms of farmland area were overtaken because of the implementation of the Pastoralist Settlement Project since 2009. Moreover, the GEPAP, which has been implemented since 2011, is closely related to grassland area; therefore, even with regard to subsidies, FH receive significantly less than the herder households. All of these factors lead to FH being the lowest income group in the region. The farmer respondents reported that herders are relatively wealthy because they have pasture to support more livestock, and they receive more subsidies from the government; in contrast, FH have to pay herders for grazing, and only a few of them have the relatively higher educational levels or skills required to possess stable non-agricultural jobs with a decent salary. UHH have the highest wealth and livestock size, which is consistent with previous studies in northern Xinjiang showing that pastoralists are the highest earners and that sedentarization has been accompanied by declines in household income and asset holdings [18,47].

Our results confirm that LD is the norm in pastoral regions, which is consistent with most previous studies [13,33], and the interdependence between different household types has contributed to the diversification of their livelihoods to some extent, with some differences existing with respect to the characteristics of LD. Agricultural livelihoods are still the most crucial source to sustain a living [41,49,57], and most households combine livestock raising and crop cultivation. UHH possess the highest livestock diversity, which was defined as one of the ecological indicators of LD [47] and has been found to be vital to livelihood resilience and rangeland sustainability as goats, sheep, cattle, camels and horses have different palatability for various plants and the ability to resist drought and pathogens [58,59]. Sedentarization has accompanied a decrease in livestock diversity, mainly in the holding of camels and horses. Although SHH and FH do not have access to pasture, most of them pay herders to graze their animals in the warm seasons. On the one hand, this mutually supportive behavior allows UHH to have an additional source of income, and their farmland can be planted by relatives during grazing. On the other hand, SHH and FH can raise a certain amount of livestock with limited forage and have more time to engage in other livelihood activities. In terms of cultivation, most households plant forage crops such as alfalfa or silage maize to feed livestock, and only approximately one-third of FH plant grain for subsistence and commercial crops to increase their incomes. The lack of crop diversification is probably due to the limitation of cultivating skills, climate-induced risk and the price fluctuation of commercial crops. Land transfer has promoted scale operations and land production efficiency, but predatory farming and the lack of supervision of land contractors lead to declining soil fertility and serious ecological problems [60]. The supervision of land transfers needs to be strengthened.

Moreover, water shortages and the flooding of riverside land coexist and are two of the main reasons for farmland abandonment. This is due to the backward infrastructure of the farmland system and the lack of water resources management. Fan et al. (2014, 2015) [61,62] debated about how engaging in sedentary agriculture may have succeeded in the short term, but this production style has increased the usage of water resources, reduced the efficiency of water use, and has led to decreasing river flow and more serious ecological problems. Water shortages and the destruction of the ecological environment along the Ulungur River have also been noticed [63]. However, due to population increase and grassland degradation, the rangeland is unable to support all the residents, and some households have adopted voluntary sedentarization [18,64]. Therefore, the key points might be to develop ecological sedentary agriculture through technology innovation and increase investment in farmland infrastructure.

Non-agricultural diversification accompanies sedentarization with complementary activities to agricultural livelihoods. The extent of the LD_nag is significantly higher for SHH and FH than for UHH. However, the extent of their non-agricultural livelihoods is also relatively low and has great potential to develop compared to the extent of non-agricultural livelihoods in other rural areas in China [65] and for some pastoralists in the Hindu Kush Himalayan region [66]. Fuyun County is a typical pastoral region in China. The livestock raised in mobility have better quality, but local herders and farmers often sell livestock directly. In the context of the rural revitalization of China, the Rural Revitalization Strategic Plan in the Xinjiang Uygur Autonomous Region (2018–2022) was issued in 2018, and the promotion of the integrative development of primary, secondary and tertiary industries in rural areas has been proposed in this plan. Implementation of the plan would offer an opportunity for non-agricultural livelihoods in Fuyun County to become closely linked to agricultural livelihoods through the extension of the agricultural industry and the promotion of industrial integration to build an agricultural product processing base with local characteristics (for example, milk and meat processing, tanning, and embroidery). Moreover, UHH could also add value to pastoralism with the orderly development of a creative industry of pastoral culture and ecotourism, which also requires investment and guidance from multiple agents. The development of secondary and tertiary industries based on indigenous specialties is conducive to further enhancements in the extent of the LD_nag for the three household types.

Our analysis of the extent of the LD of the different income levels suggests that high-income groups have higher richness of LD but do not necessarily have an advantage in terms of evenness. Asfaw et al. [29] employed three indexes to measure diversification in Niger that include the Count index, the Berger-Parker index and the Shannon-Weaver index, and the results showed that the mean value of the Count is greater than 3, whereas the mean values of the Berger-Parker index and the Shannon-Weaver index are slightly more than 1. Limited by the knowledge and capability that the households possess, the evenness of LD is difficult to improve, but involvement in diverse activities can contribute to income accumulation and the spreading of risks. The paths to successful livelihoods are multiple [67]. In terms of specialization, Deininger and Olinto [43] discussed the impact of specialization on household welfare and found significant gains from specialization but only for the households able to specialize. Therefore, in the context of pastoral vulnerability and facing not only barriers formed by a lack of education and skills for most households but also the imperfections in the markets, maintaining a certain level of specificity in one to two activities while having multiple sources of livelihood may be more conducive to livelihood enhancement and risk aversion.

The pastoral vulnerability context and the Pastoralist Sedentarization Policy (PSP) are crucial external factors that drive diversification, but the quantitative analysis required to understand these factors is beyond the scope of this paper. However, we attempt to qualitatively analyze the role of the vulnerability context and the PSP on LD. Shocks (such as drought and snowstorm) and grassland degradation make a single pastoral livelihood risky; therefore, even UHH have adopted diversified livelihoods [18,49]. With population growth and the uncertainty in climate, rangelands cannot support each household to maintain a minimum number in the herd; thus, sedentarization and diversification into other livelihoods becomes inevitable [68]. The PSP has encouraged herders to settle and transform their traditional single livelihood, and sedentarization is often associated with diversification and provides some income-earning opportunities [20,69]. Additionally, agriculture expansion with the sedentarization policy and urbanization and industrialization in China cause households to seek non-pastoral livelihoods. Due to the fluctuation of prices of agricultural products and the seasonality of agricultural production, many SHH and FH have diversified into non-agricultural livelihoods. Meanwhile, limited by the seasonality of employment opportunities, most smallholders cannot detach from agricultural livelihoods. All of these reasons make diversification the norm, but the extent and pattern of the LD differs. In our estimation, similarities and differences exist in the impact of livelihood capital and subsidy on the richness of LD_a for different types of households.

Two common factors are identified. Labor capacity, as a surrogate indicator of human capital, is significant for facilitating LD. This finding is consistent with most previous studies [67,70,71] that suggested that households with higher human capital tend to engage in multiple livelihood activities. On the one hand, as the number of laborers increases, more labor divisions can be made. On the other hand, a higher education and Mandarin language level increase the probability of diversification into non-agricultural livelihood activities. The second variable related to LD is total income. The relationship between income and diversification differs in the different regions, such as negative relationship, positive relationship, the U-shaped pattern, the inverted U-shaped pattern, or otherwise with no clear relationship [13]. Liao et al. [47] also analyzed the relationship between income levels and income diversification in northern Xinjiang and found a statistically significant negative association between these aspects, which showed that pastoralists with the highest income have the lowest diversity index. In our study, from the comparison among household types, UHH have the highest income and the lowest extent of LD, which indicates that there is a negative correlation between LD and income level, to some extent. This finding is consistent with the results of Liao et al. [43]. However, from a within-type perspective, the total income is positively correlated with diversification. Possible reasons for this are that in a pastoral context, animal husbandry is a superior industry, and cultivating is relatively inferior unless water resources and intensive management are ensured; UHH benefited from their natural advantage and were also favored by national policies. Even if FH engage in more livelihood activities, they will find it difficult to surpass the income of UHH. Within a type, however, wealthier households not only engage in agricultural activities but also have more opportunities to get involved in non-agricultural activities. This finding is in accordance with the literature, which indicated that income, as crucial financial capital, enables households to diversify their activities, promote non-agricultural livelihood strategies and/or pursue more lucrative activities that require higher non-labor and non-land inputs [19,22,36].

For UHH and SHH, livestock size is negatively related to the richness of LD_a; for FH, farmland area is negatively related to the richness of LD_a, although the estimation is insignificant. Livestock and farmland are important assets for herders and farmers, respectively. Herders with livestock of a greater size and farmers who possess larger farmland areas tend towards intensification and specialization, which might provide a decent income. However, these livelihoods are highly dependent on environmental resources and are under great risks. If natural shocks or disease affects an area, these livelihoods would be greatly affected. Dehghani Pour et al. [70] advocated that enhancing human assets should occur prior to or at least simultaneously with the enhancement of financial assets, especially among environmental resource harvesters in the area, to reduce the transformation of financial assets to the required resource-harvesting assets. This initiative is also suitable in our study area, where the education level is generally low, and the ecological environment is under great pressure. To reduce the dependence of farmers and herders on natural resources and improve their LD, the promotion of labor capacity, especially the education level and Mandarin proficiency, should be given top priority.

The results of this estimation suggest that two additional variables are relevant for the richness of LD_a for UHH. The first variable is age, which is significantly and negatively related to the number of activities for UHH. However, age tends to be positively correlated with the number of activities of SHH, although the estimation is insignificant. The result of the UHH is consistent with previous studies that found that young people have more possibilities concerning diversification, especially non-agricultural diversification [39,40]. The reason that the coefficient of age in the model of SHH is positive might be that some of the young households do not have grassland tenure and farmland tenure; therefore, although they have engaged in non-agricultural livelihoods, they are weaker in terms of LD_ag, and furthermore, they do not hold an advantage in LD_a. Moreover, because of settlement, some of the middle-aged SHH also have diversified into non-agricultural livelihoods. Second, the grassland ecological subsidy, as an important income source for UHH, has a negative impact on the richness of LD_a. For SHH and FH, the coefficients of the subsidy are also negative, although the estimation is insignificant. Simple direct compensation mechanisms may generate reduced incentives [29] rather than translate into the driving force behind the diversification of activities and the transformation of livelihoods. Development, not just monetary subsidies, is gaining increasingly more consensus [23,32].

LD is dynamic, and its determinants are multi-faceted [20]. This paper provides a snapshot of the evidence of LD in far northwestern China based on household differentiation, quantitatively explores the determinants from the perspective of major livelihood assets and subsidy and qualitatively analyses the role of the vulnerability context and the PSP in diversification. Future research needs to explore the dynamic mechanism of LD and could quantify the role of the vulnerability context and the transforming structures and processes according to the perception of the smallholder while also considering other variables, such as the risk preferences that have been found to be critical in household adaptation strategies [72]. Furthermore, we use two measurements of LD and find that income level is related more to a richness index than to an evenness index, which requires further examination in other study areas.

6. Conclusions and Policy Implications

Livelihood diversification is an important pathway towards environmental sustainability and social sustainability in pastoral regions. UHH, SHH, and FH are the three different household types that utilize natural resources in pastoral regions of China. By applying a framework of LD, this paper presents a comparative analysis of the characteristics and determinants of the LD of three household types in far northwestern China. The major conclusions are as follows.

(1) Livelihood assets have been unequally distributed among the three household types, with FH possessing no grassland tenure and the least farmland tenure area, livestock, subsidies and income.

(2) FH have more livelihood activities than herder households. Agriculture remains the most important source of livelihood for all household types. UHH show lower non-agricultural livelihoods than SHH and FH but have the highest livestock diversity.

(3) The high-income groups of the three household types have a higher richness of LD but do not necessarily have an advantage in terms of evenness. Maintaining a certain level of specificity while getting involved in multiple sources of livelihood may be more conducive to livelihood enhancement and risk aversion.

(4) Labor capacity and income are common variables for the three household types that are positively related to the number of activities. The increase in livestock size would lead to a decrease in the number of activities of herder households. Moreover, age and the grassland ecological subsidy have negative impacts on the number of activities for UHH.

The findings of this paper provide five policy implications for promoting livelihood diversification, enhancement, sustainable and effective development of pastoral regions in China and other developing countries with similar situations. First, a pilot reform of the grassland property rights system should be taken into consideration by policymakers to alleviate the unfairness in the utilization of natural resources and to safeguard farmers’ rights. Second, diversified and targeted subsidies rather than a single form of subsidy would be more effective in stimulating smallholders’ initiative for LD (for example, subsidies for superior crop varieties, education for the next generation, and training for skills with part of the subsidy being monetary). Third, initiatives that enhance labor capacity should be established prior to or at least simultaneously with enhanced income to avoid transforming income to the assets required for resource-harvesting. Fourth, the government should increase investment in the farmland infrastructure and in the research and development of eco-economic sustainable planting models. Fifth, non-agricultural industries have great potential for development by extending the agricultural industry chain and promoting industrial integration based on local characteristics, such as milk and meat processing, tanning, and embroidery. UHH can also add value to pastoralism by orderly developing a creative industry of pastoral culture and ecotourism.