Estimating Livelihood Vulnerability and Its Impact on Adaptation Strategies in the Context of Disaster Avoidance Resettlement in Southern Shaanxi, China

Abstract

In order to alleviate ecological environmental degradation and to enhance sustainable rural household livelihoods, the Shaanxi government of China launched a disaster mitigation program: the disaster avoidance resettlement. Measuring household livelihood vulnerability and further assessing its influence, hold the key to strengthening livelihood adaptation in the context of disaster resettlement. Taking Ankang—in southern Shaanxi Province, China—as an example, this article explores the impact mechanism of household livelihood vulnerability on adaptation strategies through a multinominal logistic regression model in which 657 rural questionnaires were employed. In order to provide more integrated empirical evidence, we draw lessons from the livelihood of the previously proposed vulnerability framework, which has three dimensions: the degree of sensitivity, exposure, and adaptive capacity. The adaptive strategies were divided into pure farming, non-agricultural, and diversified adaptation types according to the types of income sources. The results indicated that livelihood vulnerability varies with different resettlement characteristics. In terms of adaptive strategy types, the vulnerability of pure farming households was the largest. This article found that the relocated households who had a lower sensitivity preferred the non-farming livelihood adaptation strategy. Local households with a high adaptive ability preferred to implement diversified adaptation strategies. The weaker the social support network of the relocated households, the more likely they were to choose off-farm adaptation strategies. Our research results are robust and have broader implications in terms of promoting rural household diversifications of adaptation strategies and reducing livelihood vulnerability.

1. Introduction

In 2011, in order to improve the sustainable livelihood level of rural households and to alleviate the deterioration of the ecological environment, the Shaanxi government initiated a disaster mitigation and preparedness program named the disaster avoidance resettlement (DAR). Distance to the city is a limiting factor for rural households’ ability to obtain off-farm employment [1] and reduce the likelihood of chronic poverty [2]. DAR, which reduces the distance between farmers and cities [3], weakens this factor to a large extent. Therefore, DAR is a “win-win” project that not only contributes to the sustainability of an ecosystem, but also enables rural households to reduce their livelihood vulnerability [4]. Most of the existing studies on the DAR focused on the reasons for resettlement [5,6], policy effects [7,8], and suggestions [9]. For instance, Liu et al. (2020) found that the resilience of families that were relocated due to disasters was higher than that of families relocated for reasons of poverty reduction [10]. Xu et al. (2022) argued that resettlement policies had a significant negative impact on household adaptive capacity as they would reduce the adaptive capacity of rural households [11]. Livelihood, as a means for rural households to make a living, is the key element for rural sustainable development [12,13,14,15]. Given that there are still problems in livelihood deterioration or re-poverty after resettlement [16], it is particularly important to analyze the relocated people’s livelihood vulnerability.

Vulnerability is “a state of being susceptible due to exposure to stress related to social and environmental change and because of the absence of adaptive capacity” [17]. Vulnerability research has become a hot spot and a vital analysis tool in global environmental change and sustainability research [18]. In the new era of rural revitalization, how to reduce vulnerability and make farmers better adapted to environmental changes is a topic of interest to Chinese policymakers [19]. Vulnerability refers to a state of susceptibility due to low adaptive capacity and exposure to stresses related to social and environmental change [20]. Vulnerability evaluations enable people to discover the elements and segments most vulnerable to the adverse influences of environmental change [21,22]. Vulnerability evaluation to hazards takes into account not only disaster-specific factors such as losses and damage [23], but also social factors in (rural) communities, such as health conditions and social networks [24]. Most studies focus on the evaluation criteria on a global scale [25,26,27], national scale [28], and regional scale [29]. Despite there existing many studies on vulnerability to disaster resettlement [30,31,32], few studies have measured vulnerability at the rural-household scale. In fact, rural households play a major role in climate change adaptation decisions [33,34]. For example, Khan et al. (2022) proposed that the assessment of livelihood vulnerability indicators can help to formulate effective adaptation measures and thus can reduce the livelihood risks of rural households [35]. There are three types of vulnerability research methods: qualitative analysis [36], quantitative analysis [37], and methods that combine qualitative and quantitative analysis [38]. Many scholars have adopted the Intergovernmental Panel on Climate Change framework to characterize vulnerability in terms of three aspects: exposure, sensitivity, and the ability to adapt [39,40]. This article used quantitative analysis to measure rural household livelihood vulnerability.

Rural households vary in their adaptation strategies, and a large part of the reason for it is the differences between respective vulnerabilities. Livelihood adaptation refers to the dynamic adaptation and maintenance of the appropriate status of rural households in using resources (assets) that they are close to or can retain [41]. The selection of adaptive strategies is a vital tool through which to improve the livelihoods of vulnerable groups, and it can help them survive in the presence of drastic changes in the environment [42,43,44]. Previous studies have examined the correlation between adaptation strategies and livelihood resilience. Specifically, Pagnani et al. (2021) found that adaptation strategies can improve the livelihood resilience of rural households [45]. Further, Zhou et al. (2021) proposed that the stronger the livelihood resilience, the more inclined rural households are to non-agricultural adaptation strategies [46]. Certain other scholars have explored the influence of different adaptation strategies on the adaptive capacity of rural households [47]. For example, Savari et al. (2022) developed 12 appropriate strategies through which to improve rural household drought adaptive capacity; in addition, they explored the two most efficient methods [48]. Scholars classify adaptive strategies according to adaptative behavior characteristics and income sources [49]. For example, Yin Sha et al. (2020) divided adaptation strategies into initiative and passive adaptation according to their adaptation features [50]. Drawing on previous studies [46], this paper classifies—according to income sources—adaptation strategies into the pure agriculture adaptation type, the non-agriculture adaptation type, and the diversified adaptation type.

As mentioned above, scholars have brought certain contributions to the study of livelihood vulnerability and adaptation strategies but have not yet explored this issue from the perspective of integration of the two. Instead, this paper teases out the associations between disaster resettlement, rural households’ livelihood vulnerability, and their adaptation strategies, which may contribute to sustainable development in local decision-making. Most past studies have used livelihood assets as important factors affecting livelihood vulnerability [51] and adaptation strategies [52]. However, this paper focuses more on the capacity of rural households to choose optimal strategies under resource endowment constraints, rather than using their capital stock, in order to achieve sustainable livelihoods under new environmental conditions [53,54]. Firstly, through drawing on the vulnerability index system of IPCC, we constructed a livelihood vulnerability assessment indicator system for the resettlement of households. This system has three aspects: sensitivity, exposure, and the ability to adapt. Secondly, principal component analysis was adopted to obtain the index weight. Finally, a multinomial logistic model was used to analyze the mechanisms through which the vulnerability characteristics of rural households may lead to potentially permanent changes in adaptation strategies. The rest of this article is arranged as follows: Section 2 introduces the data sources and research methods. Section 3 shows the results. Section 4 outlines the discussion. Section 5 provides the conclusions.

2. Data Sources and Research

2.1. Data Sources

Our data were obtained from Ankang Prefecture, China, which is one of the three states conducting disaster resettlement. Ankang city is located in the southeast of Shaanxi province, and it is surrounded by the Qinling Mountains in the north and the Bashan Mountains to the south (see Figure 1). It is one of China’s poorest regions. During 2011–2020, the program in Ankang built a total of 1364 resettlement communities, which involved 268 thousand households (a rural population of 945 thousand), thus accounting for 31% of the total population (3.03 million) of the city. Ankang, as the birthplace of disaster resettlement in Shaanxi Province, is a representative city used to study the livelihood of rural households under the background of rural revitalization.

The data in this paper are derived from a survey on the livelihood of farmers in Ankang Prefecture, southern Shaanxi Province (see Li et al. (2021)) [4]. This survey is mainly based on structured household questionnaires and community questionnaires, supplemented by semi-structured interviews. Considering the representativity of the sample selection and the feasibility of the survey plan, the research group randomly selected three centralized resettlement communities from two townships in Ziyang County, four townships in Hanbin District, and two townships in Ningshan County (see Figure 1) for investigation. The investigators were composed of teachers, postgraduates, and undergraduates. For certain villages and communities, convenience sampling is adopted in this survey to randomly survey the households at home on that day. Households, or spouses aged 18–65, were interviewed. The survey includes the basic information of households’ social and demographic features: exposure, sensitivity, and adaptive capacity. Ultimately, the survey obtained 657 valid questionnaires, of which 69.9% were relocated households and 30.1% were non-relocated households.

2.2. Research Methods

2.2.1. Indicator System for Measuring Livelihood Vulnerability

Here, the authors identify livelihood vulnerability at the household level from three sub-dimensions: sensitivity, exposure, and adaptive capacity [20]. Exposure has both spatial and temporal dimensions [55], and it can be measured by the extent to which farmers are exposed to agricultural, property, and livestock risk shocks, as well as the availability of credit. Sensitivity is defined as the elements that affect a system’s vulnerability [28], such as labor force shocks, income, food and energy dependence, and access to water. Both values are proportional to vulnerability. In other words, a higher score of both would exacerbate the livelihood vulnerability level of farmers. Adaptive capacity is the household’s capacity to anticipate, cope with, minimize, and recover from natural or human-induced disturbances [56,57]. Adaptive capacity was examined from the level of livelihood capital, which includes financial, natural, social, human, and physical capital [58,59]. The livelihood vulnerability index (LVI) is in contrast to the adaptive capacity index. That is to say, if the value of adaptive capacity becomes higher, the households can resist the interference and intimidation from the system, and beyond, more easily (

Table 1. The indexes of the livelihood vulnerability assessment of rural households.

		Relocated		Not Relocated
Variables	Definition (Unit)	M or %	SD	M or %	SD	p Value
Exposure
Agriculture shock	Annual actual agricultural losses (CNY)	21.36	239.70	186.82	1468.91	0.019
Property shock	Annual actual property losses (CNY)	149.61	3033.93	589.65	7143.73	0.268
Livestock shock	Actual annual livestock loss (CNY)	33.40	521.37	13.38	149.23	0.596
Loan availability	A value of 1 means very likely; 2 means possible; 3 is general; 4 means impossible; and 5 is highly unlikely	3.55	1.32	3.15	1.33	0.000
Sensitivity
Labor force shock	The proportion of medical expenditure in total family income <20% is 0.33; 20–50% is 0.67; and the ratio >50% is 1.00	0.33	0.33	0.39	0.36	0.041
Income dependence	The proportion of the combined income from agriculture, forestry, and animal husbandry in total household income	13.1%	0.21	25.2%	0.31	0.000
Food dependence	The subsistence food income divided by the total annual household food expenditure	0.8%	0.06	3.6%	0.11	0.000
Energy dependence	Collect firewood divided by the number of annual energy consumption spending	9.7%	0.23	36.3%	0.30	0.000
Access to water	Whether the home has running water (0 means no and 1 means yes)	0.03	0.16	0.13	0.34	0.000
Adaptive capacity
Age	Age of household head	50.72	11.79	50.93	11.64	0.830
Experience	The number of family members who were ever employed (number)	0.48	0.85	0.54	0.83	0.393
Housing structure	A value of 0.33 for civil structure; 0.67 for brick-wood structure; and 1 for brick–concrete structure	0.38	0.16	0.54	0.29	0.000
Farmland area	Per capita cultivated land area (mu)	1.16	2.90	0.69	0.83	0.024
Distance to the main	Time spent walking to the nearest market (1 for more than 6 h; 2 for 4–6 h; 3 for 2–4 h; 4 for 1–2 h; and 5 for less than 1 h)	0.98	0.12	0.96	0.16	0.095
Physical assets	Standardized values for the range of assets owned by farmers	0.34	0.12	0.37	0.13	0.001
Training	Whether family members have received training (0 means no and 1 means yes)	0.18	0.38	0.37	0.48	0.000
Social relationships	Number of relatives and friends serving as village cadres (persons)	0.37	1.10	0.79	2.00	0.001
Agriculture	Amount of annual agricultural income (CNY)	817.62	4774.60	2394.85	16,297.06	0.059
House value	The building’s market value (1 is the value for <CNY 10; 11–20 is 2; 21–30 is 3; and 4 is the value for >CNY 30)	2.74	0.98	2.41	1.15	0.000
Household size	Number of family members (number)	4.51	1.59	4.46	1.65	0.702
Non-agriculture	Amount of non-farm income for the year (CNY)	4189.10	6594.91	3979.56	6275.78	0.705
Monetary help	Number of rural households available for assistance (persons)	3.82	4.88	4.33	6.01	0.247
N		459		198

).

2.2.2. The Livelihood Vulnerability Assessment Model

The main difficulty is to aggregate the indicators into composite indicators because the indicators are expressed in different units. Thus, the authors utilized the means of deviation normalization to avoid problems. By means of standardized data processing, the resulting values are within (0 to 1). The following can be used as a suitable equation:

$${\mathrm{X}}_{\mathrm{ij}}^{\prime }=\frac{{\mathrm{X}}_{\mathrm{ij}}-{\mathrm{X}}_{\mathrm{jmin}}}{{\mathrm{X}}_{\mathrm{jmax}}-{\mathrm{X}}_{\mathrm{jmin}}}$$

(1)

where X_ij represents the initial value of the households’ indicator; X_jmin is the minimum of each indicator; and X_jmax is maximum of each indicator.

In this paper, the principal component analysis (PCA) method is used to determine the weight to ensure the scientific and objective research process of evaluation. The Kaiser–Meyer–Olkin (KMO) value is 0.657, indicating that the sampling adequacy of this paper is reasonable. The approximate chi-square value of Bartlett’s test of sphericity was 1416.833, and the p-value was 0.000—thus, this rejected the null hypothesis of no correlation between indicators. Therefore, it is feasible to use PCA methods to deal with the variables in statistical significance. Finally, eight principal components were extracted according to the criterion that the eigenvalue is higher than 1, and that the proportion was the largest.

Table 2. The result of principal component analysis.

Principal Component	Factor Loading	Factor Loading	Factor Loading	Eigenvalue	Variance Proportion	Cumulative Proportion
1	Access to loans (−0.520)	Physical assets (0.673)	House value (0.573)	2.413	0.110	0.110
2	Energy dependence (0.689)	Housing structure (0.608)		2.170	0.099	0.208
3	Age (0.501)	Household size (0.527)		1.494	0.068	0.276
4	Agriculture income (0.590)			1.423	0.065	0.341
5	Monetary help (−0.532)			1.228	0.056	0.397
6	Experience (0.507)			1.042	0.047	0.494
7	Agriculture loss (0.502)			1.031	0.047	0.541

The factor loading coefficients are in parentheses, and only those with an absolute value greater than 0.5 are listed here.

shows the analysis results of the principal components.

In Table 2, the factor loading coefficient represents the correlation coefficient between principal components and key variables. Energy dependence and physical assets are the pivotal indicators through which to measure the sensitivity of the household. Access to loans and agriculture loss are crucial factors affecting the exposure levels of the farmers. Monetary help and agricultural income help to increase adaptive capacity; in addition, house value, house structure, household size, age, and experience are the crucial indicators through which to calculate adaptive ability. These factors all play a crucial part in the construction of livelihood vulnerability.

Based on the “exposition-sensitivities—adaptive capacity” analysis framework of IPCC [60], this paper conducted a livelihood vulnerability measurement index for disaster resettlement farmers in the intolerant areas of southern Shaanxi. This model can effectively assess the vulnerability of rural households and has become a popular vulnerability assessment method in the world as it has the advantages of simple operation and is easy to understand. The LVI was constructed as per the following formula:

$$\mathrm{LVI}=\left(\mathrm{E}+\mathrm{S}\right)-\mathrm{A}$$

(2)

where LVI is the livelihood vulnerability index, and a higher value means that the household is more vulnerable to the adverse impacts that are as a result of the deterioration of the environment; E is the exposure; S is the sensitivity; and A is the adaptive capacity. The LVI is proportional to the sensitivity and exposure index, which is in contrast with the adaptive capacity index.

2.2.3. The Multinomial Logistic Regression Analysis Model

To analyze the influence of rural household livelihood vulnerabilities on adaptation strategies, a multinomial logistic regression analysis model was adopted, which can determine the role and strength of independent variables in predicting the probability of occurrence of classified dependent variables. The indicators of the three dimensions of livelihood vulnerability are independent variables. The dependent variables are the types of adaptive strategies, which contain the pure agricultural type, non-farming type, and diversified adaptive type. Among them, the pure agricultural type refers to rural households who are neither engaged in labor nor engaged in non-agricultural activities. Non-farming type refers to households with no income from agroforestry or from raising livestock. For those rural households whose income sources include three or four of the above, their adaptation strategies are diversified [39,43]. The equation is as follows:

$$\mathrm{logit}\left(\mathrm{Y}\right)=\propto +{\mathsf{\beta }}_1{\mathrm{X}}_1+{\mathsf{\beta }}_2{\mathrm{X}}_2+{\mathsf{\beta }}_{\mathrm{i}}{\mathrm{X}}_{\mathrm{i}}+\mathsf{\epsilon }$$

(3)

where X_i refers to the indicators of sensitivity, exposure, and adaptive capacity; Y is the category of the adaptive strategies of resettled households; β_i and α refer to the assessed parameters of the model; and ε is the model residual.

3. Results

3.1. Descriptive Statistical Analysis of Livelihood Vulnerability

A t-test was used to analyze the differences in vulnerability and its three sub-dimensions between the resettled households and local households (

Table 3. The sensitivity, exposure, adaptive capacity, and livelihood vulnerability of the different households.

	Relocated		Not Relocated		Difference
	Mean	SD	Mean	SD	Mean	T-Value
Exposure	0.042	0.022	0.036	0.025	0.005	−2.736 ***
Sensitivity	0.031	0.027	0.059	0.035	−0.028	10.944 ***
Adaptive Capacity	0.116	0.023	0.127	0.027	−0.011	5.592 ***
Vulnerability	−0.043	0.040	−0.032	0.055	−0.011	2.870 ***

Note: *** p < 0.01.

). The results indicated that the relocated households’ exposure (0.042) was significantly higher than that of local households (0.036). The relocated households’ adaptive capacity (0.116) was significantly lower than the local households’ (0.127). That is to say, the resettled households were more susceptible to the interference of external forces than the local households. One possible reason is that disaster resettlement is not an overnight process, resettled households will go through a series of complex economic recovery and reconstruction processes in the new environment.

However, the relocated households’ livelihood vulnerability (−0.043) was significantly lower than that of local households (−0.032). Additionally, the sensitivity of relocated households (0.031) was noticeably lower than that of local households (0.059). Generally speaking, the lower the vulnerability and sensitivity of households, the less susceptible they are to the disturbance of external forces. This shows that the disaster resettlement policy is still an effective and positive intervention through which to deal with the vicious circle of poverty.

3.2. Analysis of the Households’ Livelihood Vulnerability

Vulnerability is often thought of as a function of sensitivity, exposure, and adaptability [60]. Generally speaking, sensitivity and exposure are proportional to the vulnerability of rural households; thus, for adaptable farmers, the livelihood vulnerability is lower [59].

Table 4. Values of adaptive capacity, exposure, sensitivity, and vulnerability of different types of households.

Type of Household		Exposure	Sensitivity	Adaptive Capacity	Vulnerability
Livelihood type	Pure farming type	0.042	0.065	0.117	−0.010
	Non-farming type	0.043	0.023	0.115	−0.050
	Diversified livelihood type	0.037	0.045	0.124	−0.042
Livelihood diversification	Single-livelihood households	0.043	0.027	0.115	−0.045
	Double-livelihood households	0.038	0.043	0.119	−0.038
	Diversified-livelihood households	0.036	0.054	0.129	−0.039
Resettlement reason	Ecological restoration	0.038	0.019	0.114	−0.056
	Project-induced	0.046	0.046	0.126	−0.034
	Disaster-related	0.044	0.027	0.115	−0.044
	Poverty reduction	0.039	0.032	0.112	−0.041
	Other reasons	0.034	0.039	0.119	−0.046
Type of resettlement	Centralized	0.043	0.029	0.114	−0.042
	Scattered	0.042	0.042	0.121	−0.037
	Self-determined	0.037	0.035	0.120	−0.049
	Other resettlement	0.024	0.039	0.120	−0.057
Resettlement time	Less than 3 years	0.043	0.029	0.112	−0.041
	3–5 years	0.040	0.028	0.120	−0.052
	More than 5 years	0.041	0.038	0.117	−0.039

analyses the sensitivity, exposure, adaptability, and vulnerability of each type of household.

In terms of the type of adaptation strategy, pure farming households have the highest vulnerability (−0.010), with strong sensitivity and high exposure. Through considering differences in the degree of livelihood diversification, the double-livelihood households were found to be the most vulnerable at −0.038, and this was due to their high exposure level and low adaptive capacity. With respect to the types of resettlement households, the vulnerability of project-induced resettlement households was the largest (−0.034), and this was due to high sensitivity and exposure. With respect to the type of resettlement, the scattered resettlement households had the highest vulnerability (−0.037), and this was due to higher levels of exposure and sensitivity. Regarding the differing resettlement times, the vulnerability of households who had already been resettled for more than 5 years was the highest (−0.039) due to strong exposure and high sensitivity. Therefore, we can draw a conclusion that farmers with high exposure and sensitivity are more vulnerable to interference from the system; on the contrary, farmers with strong adaptive capacity are more likely to resist risks from the system. Fortunately, the research of this paper confirms the previous research of Gupta et al. (2020) [61].

3.3. The Regression Results of the Adaptive Strategies of Relocated Households

This paper divides adaptation strategies into three types: pure farmers, non-farmers, and diversified livelihoods [44,46].

Table 5. The regression results of the adaptive strategies of relocated households in Ankang Prefecture for disaster resettlement.

Variables	Relocated		Not Relocated		Total Sample
	Model 1 (Non-Agricultural Adaptation Type)	Model 2 (Diversified Adaptation Type)	Model 3 (Non-Agricultural Adaptation Type)	Model 4 (Diversified Adaptation Type)	Model 5 (Non-Agricultural Adaptation Type)	Model 6 (Diversified Adaptation Type)
Exposure	−0.499	−0.481	0.004	0.002	−0.165	−0.184
Sensitivity	−1.160 ***	−0.977 ***	−1.622 ***	−1.413 ***	−1.397 ***	−1.217 ***
Adaptive capacity	0.678	1.321	0.135	2.144 *	0.584	1.337 *
Average education (in years)	0.038	−0.000	0.081	0.279 ***	0.119 **	0.123 **
Whether participation was had in the sloping land conversion program	−1.309 ***	−0.569	−2.418 ***	−1.492 **	−1.542 ***	−0.825 **
Phone charge	0.003 *	0.002	0.002	0.001	0.002 **	0.002
Social support network	−0.005 *	0.000	−0.001	−0.003	−0.004 **	−0.000
Frequency of participation in collective affairs	0.377 ***	0.128	−0.021	0.151	0.278 ***	0.086
Ningshan County	1.574 *	0.076	0.868	−0.285	1.269 ***	−0.262
Ziyang County	1.324 *	0.074	−0.797	−1.188 *	1.527 ***	−0.108
LR chi2	181.18		76.00		302.82
Prob > chi2	0.0000		0.0000		0.0000
Pseudo R2	0.2282		0.2174		0.2478
N	459		198		657

Note: * p < 0.1, ** p < 0.05, and *** p < 0.01.

shows the results of the multinomial logistic regression model, which was used to evaluate the influence mechanism of household vulnerabilities on adaptation strategies. The regression idea is to incorporate exposure, sensitivity, and adaptive capacity as explanatory variables into the regression model. In turn, model 1–model 2, model 3–model 4, and model 5–model 6 are the regression results of relocated households, local households, and the total samples, respectively.

Table 5 demonstrates that sensitivity has a negative, noticeable impact on the choice of non-agricultural adaptation strategies and diversified livelihood strategies by households at the 1% statistical level. Adaptive capacity has a positive and noticeable impact on local households at the 10% statistical level, but it has no significant positive influence on relocated households. The average years of education have a positive, noticeable influence on the diversified adaptation strategies of local households, but the positive impact on relocated households was not significant. Participation in the sloping land conversion program (SLCP) had a noticeable, negative impact on the choice of off-farm adaptive strategies in households at the 1% statistical level. The communication cost one month before the household survey had a positive and noticeable influence on the non-farm adaptation type of the relocated and the whole sample at the 10% and the 5% statistical levels, respectively.

In the dimension of adaptive capacity, a social support network has a negative and noticeable effect on the non-farm adaptive strategies of the relocated households and the whole sample with the significance levels of 10% and 5%, respectively. In terms of social capital, the rural households’ willingness to participate in collective affairs has a positive, noticeable impact on the rural household non-agricultural adaptation strategies of the relocated households and the whole sample at a significance level of 1%.

3.4. Robustness Test

The robustness test results of the relocated households are shown in

Table 6. Robustness test for the relocated households.

Variables	Added Control Variables
	Model 7 (Non-Agricultural Adaptation Type)	Model 8 (Diversified Adaptation Type)	Model 9 (Non-Agricultural Adaptation Type)	Model 10 (Diversified Adaptation Type)	Model 11 (Non-Agricultural Adaptation Type)	Model 12 (Diversified Adaptation Type)
Exposure	−0.525	−0.360	−0.431	−0.426	−0.392	−0.292
Sensitivity	−1.168 ***	−0.979 ***	−1.150 ***	−0.949 ***	−1.189 ***	−1.001 ***
Adaptive capacity	0.126	1.044	0.916	1.576	0.118	0.906
Average education (in years)	0.038	0.005	0.047	0.007	0.046	0.004
Whether participation was had in the sloping land conversion program	−1.311 ***	−0.514	−1.392 ***	−0.628	−1.320 ***	−0.558
Phone charge	0.003 **	0.002	0.003 *	0.002	0.003 **	0.002 *
Social support network	−0.006 **	0.000	−0.005 *	0.000	−0.006 ***	−0.000
Frequency of participation in collective affairs	0.327 **	0.070	0.389 ***	0.128	0.349 ***	0.077
Ningshan County	1.406 *	0.075	1.476 *	−0.182	1.334 *	0.053
Ziyang County	1.112	−0.039	1.231	−0.150	0.834	−0.045
Control variables	Resettlement reason		Resettlement time		Type of resettlement
N	427		412		427
LR chi2	197.73		191.69		194.69
Prob > chi2	0.0000		0.0000		0.0000
Pseudo R2	0.2395		0.2419		0.2358

Note: * p < 0.1, ** p < 0.05, and *** p < 0.01.

. The robustness test usually examines whether the evaluation method and variables can maintain a relatively stable and consistent interpretation when some parameters of the model are changed. The above regression of rural household exposure, sensitivity, adaptive capacity, and livelihood adaptation strategy selection may have endogeneity problems due to the cross-sectional data used. In view of this, it is necessary to check whether the regression results are robust by controlling certain variables. The reason for resettlement, resettlement time, and type of resettlement are introduced into the regression model of relocated households as control variables, and these correspond to model 7–model 8, model 9–model 10, and model 11–model 12, respectively. In general, when different variables are controlled, the test results of the main indicators are similar to the regression results of model 7, and the coefficients of each indicator fluctuate slightly but have the same effect trend, which comprehensively indicates that the regression results are robust.

4. Discussion

Disaster resettlement is a vital approach in decreasing the livelihood vulnerability of rural households and diversifying their adaptation strategies. Previous studies revealed that rural household vulnerability is often influenced by region-specific factors, such as social, economic, political, environmental resource, and material factors [17]. Furthermore, this article revealed that the vulnerability of relocated households is significantly lower than that of local households. In essence, disaster resettlement allows farmers to live closer to the town or community, which helps them to rely more easily on public services (emergency response, evacuation, rescue, health, etc.) when disasters occur [3]. Therefore, resettlement is an effective measure that can reduce rural household vulnerability, as is confirmed in this paper. The study of Han et al. (2020) also indicates that resettlement can reduce the vulnerability of vulnerable groups, and it can realize sustainable development [62].

The results indicated that households with a high sensitivity are more willing to choose pure agricultural adaptive strategies than non-agricultural or diversification strategies. The possible reason for this is that the relocated households will face higher social relationship risks in the new environment, which will prompt them to engage in out-migration activity [63]. Poudel et al. (2020) believe that the more sensitive the residents are, the greater the threat to their livelihood capitals [64]. The relocated households with a single agricultural income are more vulnerable, while non-agricultural income can enhance the livelihood resilience of rural households [65]. Moreover, adaptive capacity has a positive and significant impact on local households’ diversified adaptation strategies, but it has no significant effect on relocated households. The possible reason for this is that disaster resettlement, as a kind of policy intervention in the form of external impact, lacks local expert knowledge and experience and fails to achieve the expected policy effect [66]. Exposure has no significant effect on the adaptation strategy selection of relocated households. This may be because ecological restoration and compensation cases reduce exposure to hazards [67].

Education encourages rural households to diversify their livelihoods [68]. However, we found that it had no remarkable effect on the choice of diversified adaptation strategies for relocated households, which may be mainly due to the “human capital failure” [69]. Social capital is a pivotal element in rural households’ ability to adapt to disasters [70]. It was found that rural households with a high frequency of participation in collective activities were more likely to choose off-farm adaptation strategies because of their relative abundance of social capital. Social network support, as an important means to ensure rural household access to resources, can theoretically promote farmers to choose a non-agricultural adaptation strategy [59,71]. However, the empirical analysis results of this paper show quite the opposite. This may be because disaster resettlement often brings shocks to the original social networks and organizations of relocated farmers (who still need to go through the process of livelihood restoration and adaptation [13]). Participation in the SLCP has a noticeable, negative influence on the choice of off-farm adaptive strategies in households. Certain scholars have shown that farmers who participate in SLCP only take response measures passively; thus, their sensitivity reduction is temporary, and their adaptive capacity is not improved [72,73]. Therefore, non-participants are more inclined to off-farm adaptation strategies than participants. As noted here, this paper introduces the resettlement reason, resettlement time, and resettlement type as control variables into the resettlement household regression model, and the results of this paper were proven as robust.

5. Conclusions

This study constructed and measured a rural household livelihood vulnerability index; in addition, it explored the influence mechanism of livelihood vulnerability characteristics on adaptation strategy selection. Firstly, the results show that, although the relocated farmers’ vulnerability was significantly lower than that of local households, the farmer adaptive capacity was significantly lower than local households. Secondly, the livelihood vulnerabilities were different because of their resettlement characteristics. For instance, when considering the reasons for resettlement, the vulnerability of ecological restoration households was the lowest. Moreover, in terms of livelihood types, the vulnerability of pure farming households was the largest. In addition, the lower the sensitivity of the relocated households, the more likely they were to choose the off-farm or diversified adaptation strategies. The local households with stronger adaptive abilities were more partial to the diversified adaptation strategy. The higher the telephone cost and the higher the frequency of participating in collective affairs, the more likely the resettlement households were to choose a non-agricultural adaptation strategy as opposed to a pure agricultural adaptation strategy.

There are certain implications in this article. The authors of this study believe that the government should ensure the effective promotion for the restoration and reconstruction of relocated rural household livelihood capacities and that the government should strive to find ways through which to strengthen the adaptive capacity of farmers so as to ensure the relocated rural household’s stability and prosperity. Further, relying on the advantages of local resources and cultivating and developing the immigrant industry is an effective measure through which to improve the diversification of livelihoods and the adaptive capacity of the migrants. Specifically, the government should focus on strengthening vocational skills training for relocated households and help these affected rural households build social relations and network structures, thereby enriching their means of livelihood and sources of income. Additionally, a perfect follow-up guarantee mechanism should be established according to the production and living conditions of the different relocated rural households.

There still exist limitations to this article. Firstly, the multinominal logistic regression analysis method used in this paper is limited by the limited number of surveys and the fact that some variables are categorical. Therefore, the method used in this study can be further improved by using more advanced statistical models and weight assignments. Secondly, the factors affecting the adaptation strategies of rural households are complex. Although control variables are introduced in this paper, variables such as rural household risk perception were not included. Therefore, it is impossible to conduct a more comprehensive analysis on the factors of adaptation strategies. Finally, the investigation only covers Ankang city in southern Shaanxi, while the Shaanxi disaster avoidance resettlement program also covers parts of northern Shaanxi. Therefore, the generalization significance of the research results is limited.