The Impact of Location of Labor Migration on Rural Households’ Income: Evidence from Jiangxi Province in China

Abstract

With the increasing occurrence of labor migration (LM), off-farm employment has emerged as a crucial means to augment the income of agricultural households, bridge the urban-rural divide, and achieve rural regeneration. This study utilized a multiple linear regression model and quantile regression model to examine the effect of LM location on rural households’ income. The analysis is based on research data from Jiangxi Province in 2018. The outcomes reveal that both intra-country LM and outside-of-county LM could make a substantial contribution to the increase of overall household income. However, the coefficient of impact for outside-of-county LM is greater. The findings of this study successfully passed the rigorous tests for robustness and endogeneity. Furthermore, the quantile regression analysis indicates that the greatest income-generating impact of intra-county LM occurred at the 90% quantile, whereas the highest income-generating impact of outside-of-county LM appeared at the 75% quantile. The study aims to determine if there is a variation in the income impact of LM in samples with distinct features. Specifically, it investigated the scale of forestland management and the LM of the household head. The results show that the promotion effect of intra-county LM on the total income of rural households was only observed in the sample group with a forestland area larger than 50 mu. Additionally, outside-of-county LM could only promote the growth of the total income of rural households in the sample group in which the head of household has not experienced labor migration. Hence, to enhance the growth of income for rural households amidst China’s urbanization, policymakers should facilitate the controlled migration of labor from rural areas to urban areas while also encouraging the migration of labor within rural areas.

1. Introduction

The supply of rural labor is crucial for the progress of the development of the agricultural industry and rural economy. Nevertheless, as the social economy continues to progress, an increasing number of farmers are migrating to non-farm industries. A significant number of scholars primarily concentrated on examining the phenomenon of labor migration (LM) in developed countries like the United States. They proposed theories such as the Lewis model, Ranis–Fei model, Jorgenson Model, and Todaro Migration Model [1,2,3,4]. Recently, there has been a growing scholarly interest in the issue of LM in developing countries [5,6]. China, being a typical developing country with the largest rural population globally, has witnessed the phenomenon of rural LM since the 1980s. The amount of LM in China has been consistently increasing since the implementation of the reform and opening-up policy. China’s National Bureau of Statistics reported that as of 2023, the rural population in China amounted to 477 million individuals. Among them, there were 297.53 million migrant laborers, making up 62.38% of the rural population. Out of these migrant laborers, 120.95 million were local migrants, while 176.58 million were migrants leaving their home province. Among the outbound migrant laborers, 67.51 million were migrating between provinces, while 109.07 million were migrating within the same province. To increase household income, expand sources of income, and improve family welfare, many rural households have opted to migrate from the agricultural industry to the non-farm industry [7]. Nevertheless, with the Chinese government increasingly prioritizing rural, agricultural, and farmer concerns and providing greater policy assistance, the potential loss of LM for rural households is progressively increasing [8]. Furthermore, the income effect of LM is attained by securing employment in regions with higher wage rates. Consequently, is there variation in the income effect among different labor migration locations (LML)? Providing a more comprehensive response to these concerns would enhance the LM mechanism and offer valuable policy insights for the formulation of differentiated LM policies.

A lot of attention from both national and international scholars has been directed toward the subject of LM. Several scholars have examined the underlying reasons behind LM. According to foreign researchers, LM is greatly affected by policies, particularly health insurance policies and agricultural development policies, which are regarded to be significant factors driving LM [9,10]. Chinese scholars have included policy considerations in the analytical framework of LM incentives. Lyu et al. (2020) discovered that in Jiangsu Province, the surplus of rural labor and the decrease in employment opportunities for those farmers are significant factors that encourage LM [11]. They also found that by enhancing agricultural subsidies, it is possible to mitigate the trend of LM. Agricultural subsidies primarily result in a rise in income for rural households, which in turn can significantly influence the LM decisions made by these households. Hence, Ramsey et al. (2023) conducted a rigorous analysis of empirical research data from both the United States and Japan, revealing that LM takes place when the benefits of migrating outweigh the associated costs according to the Marshallian framework [12]. Subsequently, other scholars investigated the impact of LM on income. Nevertheless, the question of whether this income effect is beneficial or detrimental is still a subject of ongoing dispute. Some scholars believe that LM is believed to have a favorable impact on both agricultural output and rural households’ total income (FTI) [13]. An alternative perspective argues that rural households’ agricultural income (RAI) remains the primary component of FTI. It posits that LM will diminish the RAI and, hence, reduce the FTI. LM will result in the farmer from agricultural industries to off-farm industries. This will cause a shortage of labor in the forestry production process, which is known as a negative labor loss effect [14]. Conversely, when a significant number of young and strong laborers migrate elsewhere, it results in the depletion of rural areas. Consequently, the surplus labor in rural areas becomes increasingly older and dominated by women. This, in turn, contributes to the growing issue of land abandonment [15,16]. The imbalance between labor supply and demand in agricultural production has resulted in a decline in agricultural output [17]. Nevertheless, all the aforementioned assessments assume that the FAI and rural households’ off-farm income (ROI) are mutually exclusive. However, it is important to note that the two factors might indeed have an impact on each other as a result of limited resources. As Anan et al. (2020) discovered, the ROI positively influences the growth of RAI. This adds complexity to the examination of fluctuations in the RTI during the process of LM [18].

It has been argued that there is a negative labor loss and a positive income effect of LM on RTIs [19]. However, there is no consensus on which effect has a more significant impact on the RTI. The current framework for studying the effect of LM on RAIs offers a solid theoretical foundation for this study. After thoroughly examining the literature, we have identified certain deficiencies in the current study. Specifically, previous studies had only investigated the presence of LM in agricultural households without delving into the intricacies of this variable [20]. As theories on LM have advanced, numerous studies have started to examine the redistribution of labor factors resulting from LM. This redistribution is typically quantified by the ratio of LM to the total number of laborers in a household [15,21]. In the past, LM primarily occurred in economically advanced regions, with fewer instances of LM within counties. Consequently, studies on the spatial patterns of LM was limited. Nevertheless, due to the implementation of rural revitalization policies and the significant economic growth in the county, a portion of the LM is still inclined to seek employment opportunities outside of agriculture. Therefore, it is crucial to consider the LML. Furthermore, while examining the influence of LM on the revenue of rural households, several studies concentrate solely on one aspect of farmers’ income, such as rural households’ forestry income (RFI) [22] or FOI [23,24]. A mere alteration in income alone does not automatically yield a comprehension of the complete effect of LM on the revenue of rural households. This paper examines the effects of intra-county labor migration (${Labor}_{in}$) and outside-of-county labor migration (${Labor}_{out}$) on the RTI, RFI, and ROI at the LML level. The analysis was based on data from 506 rural households conducted in Jiangxi Province. Multiple linear regression and quantile regression methods were used for the analysis. This study primarily focuses on the following minor contributions: The study examined the impact of ${Labor}_{in}$ versus ${Labor}_{out}$ on the income of agricultural households, regardless of the LML. The majority of past studies have mostly examined the proportion of LM, while there have been fewer analyses conducted on LM across various regions. Thus, this study serves to address the research gap in this area and adds to the achievement of rural households’ income. Furthermore, the study method involves the use of both the multiple linear regression model and the quantile regression model. This allows for a comprehensive examination of the overall influence of LM on the income of rural households, as well as a detailed analysis of the varying effects within different value domains. The empirical findings confirm that LM had a positive impact on the RTI. Specifically, the income-generating effect of ${Labor}_{in}$ is highest at 90%, whereas the income-generating effect of ${Labor}_{out}$ is highest at 75%. This suggests the existence of an optimal ratio for LM, which offers theoretical justification for enhancing the LM system. This paper aims to assess the varying effects of LM on the income of rural households, specifically focusing on different forestland management scales and the LM of household heads. By doing so, it provides valuable insights for policymakers to identify the differential impact of LML on the income levels of rural households. This knowledge can help improve the efficiency of LM policies and ensure the sustainable growth of rural households’ income.

2. Research Hypotheses

2.1. The Impact of Intra-Country Labor Migration on Income of Rural Households

Since the initiation of China’s reform and opening-up policy in 1978, there has been a notable disparity in income between urban and rural areas in China. The fluctuation in agricultural commodity prices has significantly affected farmers’ earnings. To enhance their living conditions and access more social resources and public services, farmers have been inclined to reallocate their family’s labor resources and generate additional income by LM to off-farm sectors. The growth of secondary and tertiary sectors has created greater opportunities for employment for rural households, and these opportunities for off-farm employment typically offer higher wages compared to farming activities. By optimizing the industrial structure of the county, the sample rural households whose LML is within the county can also enhance their RTI through off-farm employment. Forestry is distinguished by the lack of uniformity in labor time and production time, with forest products requiring a lengthy production period. Farmers can effectively utilize this portion of their leisure time for off-farm employment. Income diversification can help mitigate the risks associated with nature and markets in forestry management [25]. Simultaneously, the increasing ROI offers a solid foundation for investing in forestland [26]. To enhance the efficiency of forestry management, farmers opt to allocate a portion of their ROI towards acquiring machinery. This allows them to reduce reliance on labor, which becomes increasingly costly with time [27,28]. To summarize, ${Labor}_{in}$ can enhance the RTI by diversifying revenue sources and using mechanization in the forestry sector. Consequently, this study presents the following hypotheses:

Hypotheses 1: 

Intra-county labor migration can significantly contribute to the growth of RTI.

2.2. The Impact of Outside-of-County Labor Migration on Income of Rural Households

Rural households’ labor that migrates outside the county is involved in secondary and tertiary industries that offer higher wages and quicker returns. Farmers who migrate their labor might enhance the accumulation of material capital by receiving greater wages, so effectively augmenting the property income of their family. Simultaneously, when rural households relocate to economically advanced areas for off-farm employment, they are incentivized to actively acquire professional skills and engage in professional training to enhance their competitiveness in the labor market [29]. Enhancing the farmers’ skill levels to enable them to access improved employment opportunities and attain steady growth in their income from off-farm employment [30].

However, in contrast to ${Labor}_{in}$, ${Labor}_{out}$ typically incurs more transportation expenses [31]. This creates a situation where laborers are unable to participate in off-farm sectors while also working part-time. The primary demographic of individuals who migrate are young people, whereas the individuals who stay in rural areas to engage in forestry production consist primarily of traditional elder laborers and a few female laborers. Consequently, a significant number of skilled workers proficient in contemporary forestry practices would be lost in rural areas, leading to a reduction in forestry production [32]. When a significant percentage of a household’s income comes from sources other than farming, farmers may opt to either lease out their forestland to someone else or let it remain unused [33]. When there is weak security of tenure for forestland, farmers may opt to let the land lie uncultivated rather than leasing ownership. This is due to their aversion to the risk of losing the forestland. In other words, during this period, off-farm employment will lead to the decision to abandon the forestland [15,34]. The implementation of forest rights system reform has greatly enhanced the security of farmers’ tenure. Farmers opt for the leasing out of forestland to secure a stable rental income, thus ensuring the attainment of a minimum income threshold [35]. However, typically, the rental revenue is lower than the money generated by the person involved in forestry production through LM. This results in a decline in their RFI.

To summarize, the analysis demonstrates that individuals who relocate to economically developed regions can earn a higher ROI through ${Labor}_{out}$ compared with being employed in the forestry industry or off-farm employment within the same county. Nevertheless, as a result of the increased transportation expenses, farmers are unable to participate in part-time work, resulting in a decrease in ROI (Figure 1). Thus, this paper positions the subsequent hypothesis:

Hypotheses 2: 

Outside-of-county labor migration will reduce RFI, but ROI will increase.

3. Research Design

3.1. Data Sources

The study locations chosen were regarded as collective forest areas and large labor migration provinces. Jiangxi Province is situated in the southeastern part of China, benefiting from its advantageous geographical position and a significant population. The scale of forestland in Jiangxi Province is 10.8 million hectares, accounting for 64.7% of the province’s total land base, exceeding the national average of 20% [36,37]. Meanwhile, the population of Jiangxi Province reached 45,150,100 individuals by the end of 2023. Among them, 27,400,600 people fell between the age range of 16–59, constituting 60.70% of the province’s total population. In other words, the majority of the population consisted of individuals who had reached working age. Jiangxi Province, a significant source of labor in China, is predicted to have a labor of 35.524 million in 2022. Nevertheless, Jiangxi Province is mostly known for its primary sector of industry. However, it still lags behind industrialized provinces like Beijing and Shenzhen in terms of industrial development. Statistics indicate that the industry’s higher comparative returns will draw a significant influx of LM. By the end of 2022, in Jiangxi Province, the number of rural migrant workers (referring to local individuals who have been employed in off-farm sectors outside of rural areas for a duration of six months or more) is projected to reach nearly 12 million people. It is highly important to consider if LM can effectively enhance the income of rural household in Jiangxi Province. The choice of study sites was based on the abundance of forestry resources and the income level of rural households, taking into account the importance of these areas in income-generating efforts to achieve shared prosperity to ensure a representative sample. In accordance with this, during July and August 2019, the research team carried out local investigations in 10 specifically chosen counties and districts within Jiangxi Province (Figure 2). In order to achieve thorough data collection, the research team collaborated with academics to create the questionnaires and chose the individuals who would participate in the study. The study locations chosen were regarded as traditional agricultural regions and important places for reducing poverty in Jiangxi Province. As a result, these sites were selected as representative areas for our study.

In order to ensure the accuracy and completeness of data collection, the project team adopted a combination of questionnaires and structured interviews. Before the research, the project team conducted a five-day training session for the researchers, covering interviewing skills, questionnaire filling specifications, and other aspects. Prior to the survey, the project leader organized a five-day interview training session for the interviewers. Afterward, the project team carried out a preliminary survey. The advance team arrived in the research area 3–4 days prior to the official survey participants’ arrival date. A single village was randomly chosen from each county as a representative sample to gather preliminary information on the farmers’ situation and to make any necessary adjustments to the questionnaire on the basis of the pre-survey to ensure that the questions were reasonably set and clearly expressed. The sample villages in the formal study procedure were chosen by a stratified sampling approach that involved a combination of point and face sampling approaches. In order to ensure that the sample was randomized and unbiased, the following steps were used in this paper for the research design: The project team chose a total of 10 county districts based on the distribution of forestry resources and the income level of farm households in each county; within each selected county, the project team also randomly selected four to six village as a research target. This step ensured natural coverage of different types of households (with migration, without migration, and partially migrated) during the research; within each research village, the project team further used random sampling to randomly select 10 households from the list of households in the village for the research, ensuring a natural distribution of the sample. During the formal research process, the project team conducted household interviews that were strictly in accordance with the randomly selected sample household list and on the one hand, collected detailed data on the labor force status of the farmers, sources of income, basic characteristics of the household, and characteristics of the village and other aspects. Secondly, the village accounts data were obtained from village cadres. The village accounts data provide detailed information on key variables such as the village income, population, and the proportion of labor migration for the study. Upon collecting the surveys, the research team implemented a three-step quality assurance process. This approach included daily self-checking by the research members, mutual checking among team members, and checking by the lead teacher. The overall number of respondents was 510, with 506 valid questionnaires acquired after removing four surveys that did not include crucial characteristics. The result yielded a sample validity percentage of 99.22%.

3.2. Variable Selection

3.2.1. Dependent Variables

The data for this study came from a sample of rural households in collective forest areas in Jiangxi Province. The focus of this study is to examine the rural households’ income as the dependent variable. Based on the multiplicity of rural households’ income structure, this study selected RTI, RFI, and ROI as the main dependent variables to be analyzed. RTI is measured by the RFI, ROI, RAI, and other income. Among them, referring to the ROI includes off-farm incomes in this study, which is defined as off-farm income, wages, salaries, and remittances of household members [38]. RFI includes incomes from timber forests, bamboo forests, economic forests, under-forest economy, income from forest-related part-time employment, and other income. Income from timber forests includes income from timber and fuelwood; income from bamboo forests includes income from bamboo timber and bamboo shoots.

3.2.2. Independent Variables

As previously stated, due to variations in the LML, the effect on rural households’ income may not be fully uniform. This paper examined the studies conducted by Liu et al. (2014) and Xiao et al. (2020) to identify four main aspects of LML [39,40]. These aspects include labor hometown, non-hometown within the same county, non-hometown within the same province, and employment in foreign provinces. Since labor hometown and non-hometown within the same province make up a smaller proportion of the four types, the data are categorized into ${Labor}_{in}$ (which includes hometown and non-hometown within the same county) and ${Labor}_{out}$ (which includes non-hometown within the same province and out-of-province labor) to simplify data analysis. This research categorizes LM into two types based on the variations in geographical distance: ${Labor}_{in}$ and ${Labor}_{out}$.

3.2.3. Control Variables

To accurately assess the influence of LML on the income of rural households and ensure the reliability of the regression results, this study incorporated several control variables that could potentially affect LM. This was conducted to minimize any bias in the estimation caused by omitted variables and enhance the stability of the findings. Based on relevant research, control variables were included to account for the three aspects of personal characteristics of the head of home, family characteristics, and village characteristics. The head of household is the main decision maker in the household, so this paper referred to Chen et al. (2021), introducing the individual-level effects of age, educational level, and village cadres [41]. Family characteristics are often also one of the factors with significant impacts on the process of LM, so this paper referred to Minale (2018), who selected the number of family members, number of laborers, number of LMs, and number of ${Labor}_{out}$to control for the impacts at this level [7]. In addition to the above two types of characteristics, village characteristics are also an important control variable that affects the income of the farming households. Referring to Yang et al. (2020) and Liu et al. (2016), this paper incorporated the distance from the village to the town center, the village population, the village income, and the percentage of village LM in the village into the analysis framework [42,43].

3.3. Descriptive Analysis

Based on the research data, this study found that of the 506 rural households that participated in the research, a total of 324 households, or 64.0% of the total sample, had LM. In addition, this study found that 50.4% of rural households opted for intra-county labor migration; those choosing to migrate outside counties and within provinces accounted for 19.3%, while those choosing to migrate outside provinces accounted for 48.7%. There were a total of 1438 laborers in the 506 farming households. Of these, intra-county migration occurred in 22.3% of the total household labor per household, and outside-of-county migration occurred in 31.0% of the total household per household. In other words, approximately 321 intra-county labor migration and approximately 446 outside-of-county labor migration. We further derived long-term and short-term labor migration; the number of those who carried out short-term LM was 838 (accounting for 58.4% of the migrated households), and the number of those who carried out long-term LM was 597 (accounting for 41.6% of the migrated households) (

Table 1. Variable definitions.

Variable	Variable Symbol	Definition	Mean	SD
Dependent variable
Total rural households’ income	RTI (ln)	RTI (including income from forestry activities, off-farm employment, agricultural activities, and others) (Yuan) in 2018 taken in logarithms	11.027	1.458
Rural households’ forestry income	RFI (ln)	RFI (including income from timber forests, bamboo forests, economic forests, under-forest economy, the income from forest-related part-time employment, and other income) (Yuan) in 2018 taken in logarithms	5.682	4.208
Rural households’ off-farmer income	ROI (ln)	ROI (including off-farm income in this study is defined as off-farm income, wages, salaries, and remittances of household members) (Yuan) in 2018 taken in logarithms	10.660	1.983
Independent variable
Intra-county labor migration	$L{abor}_{in}$	Number of intra-county labor migration/Number of family laborers (%)	0.223	0.342
Outside-of-county labor migration	${Labor}_{out}$	Number of outside-of-county labor migration/Number of family laborers (%)	0.310	0.360
Control variables
Personal characteristics of the head of household
Age	$age$	Age of head of household in the year of survey (actual years)	56.805	10.185
Educational level	$edu$	Educational level of the head of household (elementary school and below = 1; middle school = 2; middle or high school = 3; college or bachelor’s degree or higher = 4	1.899	0.823
Whether village cadres	$carde$	Whether the head of household is a village cadre (Yes = 1; No = 0)	0.392	0.500
Family characteristics
Number of family members	$number$	Number of family (persons)	4.824	2.228
Whether or not there is labor migration?	$labortmig$	Whether or not there is labor migration in the rural household (Yes = 1; No = 0)	0.640	0.480
Number of laborers	$numlabor$	Number of family laborers (persons)	2.842	1.427
Number of outgoing laborers	$numlaborout$	Number of outgoing laborers (persons)	1.239	1.260
Village Characteristics
Area of forestland	$Area$	Total area of forestland (hectares)	8.044	20.325
Average area of forestland	${Area}_{ave}$	Total area of forestland/number of forested plots (hectares)	1.670	3.066
Distance	$distance$	Distance from the village to the town center (kilometers)	7.242	6.058
Percentage of village labor migration	$vlabormig$	Number of village labor migration/total number of villages	0.497	0.204
Village income	$lnvincome$	Village income (including financial assistance income, resource exploitation and remittance income, management income, investment income, and other income) (Yuan) in 2018 taken in logarithms	8.618	0.311
Village population	$lnvnumber$	Village population(person) logarithmic for the year before the year of household research	7.119	0.699

).

3.4. Model Construction

This paper focuses on the effect of LM on rural households’ income, so it is necessary to clarify the causal relationship of different LMLs on rural households’ income. Firstly, this paper analyses the impact of $L{abor}_{in}$ and ${Labor}_{out}$ on rural households’ income through multiple linear regression models, which were constructed as follows.

$$Y_i={\partial }_0+{\partial }_1{X}_i+{\partial }_2{Control}_i+{\epsilon }_i$$

(1)

In Equation (1), $Y_i$ is the sample household income matrix for household i, consisting of RTI (ln), RFI (ln), and ROI (ln). $X_i$ is the LM of household i, both $L{abor}_{in}$ and ${Labor}_{out}$. ${Control}_i$ is a vector matrix consisting of control variables. ${\partial }_0$, ${\partial }_1$, and ${\partial }_2$ are variable coefficients. ${\epsilon }_i$ is the random error term.

Equation (1) is a typical regression model that only estimates the average effect of LM on rural households’ income. However, there may be variations in the impact of LM on rural households’ income at different ratio levels, which reduces the accuracy of depicting this impact. Furthermore, the residual sum of squares in Equation (1) may be affected by outliers, resulting in skewed outcomes. This work utilized the quantile regression method developed by Koenker and Bassett (1978) to address the aforementioned factors [44]. The specific model was designed in the following manner:

$${Quant}_{\theta }\left(Y_i|X_i\right)={\beta }_{0\theta }+{\beta }_{1\theta }{X}_i+{\beta }_{2\theta }{Control}_i+{\epsilon }_i$$

(2)

where $X_i$ and $Y_i$ are the independent and dependent variables of this paper. ${Quant}_{\theta }\left(Y_i|X_i\right)$ denotes the conditional quantile of $Y_i$ corresponding to quantile θ given $X_i$. ${\beta }_{0\theta }$, ${ \beta }_{1\theta }$, and ${\beta }_{2\theta }$ are the coefficient matrices of the variables corresponding to quantile θ. The other variable settings are consistent with those above.

This study examines the anticipated effect of LM on rural households’ income and assesses the varying impact of different value domains using Equations (1) and (2). However, as rural households’ income also influences the decision to LM, additional tests are required to account for endogeneity. This work employs both instrumental variable (IV) and control function (CF) methods to address the endogeneity issue and enhance the robustness of the Hausman test.

This paper examines the reciprocal relationship between LM and household income. To address this, this study referred to Xiao et al. (2021) and replaced the original independent variable with the average LM ratio of households in the village that are not part of the sample households [40]. The regression analysis was then repeated using this new variable. The rationality of the choosing of instrumental variables can be observed through two characteristics based on theoretical analysis: first, social psychology posits that human behavior is influenced by the individuals in their environment. For instance, the decision of sample farmers to LM may be influenced to some degree by the LM situation of other farmers in the village. This aligns with the requirement for instrumental variables to be correlated with endogenous variables. Secondly, the LM of other sample farmers typically has little impact on the income level of sample farmers. The precise configuration of the model was as follows:

$$Y_i={\vartheta }_0+{\vartheta }_1{Z}_i+{\vartheta }_2{Control}_i+{\epsilon }_i$$

(3)

$Z_i$ is the instrumental variable selected for this paper.${\vartheta }_0$, ${\vartheta }_1$, ${\vartheta }_2$ is the coefficient matrix of the corresponding variable. The other variables are set as above.

The CF method involves a two-step process: first, an instrumental variable is used to estimate the endogenous variable, specifically the LM, in the first-stage regression. Then, the anticipated residual term ${\mu }_i$ is calculated. In the second phase, the variable ${\mu }_i$ is incorporated into Equation (1) for regression analysis, and its statistical significance is examined. If the residual term ${\mu }_i$ surpasses the 1% significance level test, the alternative hypothesis that there is at least one endogenous variable is accepted based on the principle of the Hausman test. The particular model was configured in the following manner:

$$X_i={\gamma }_0+{\gamma }_1{Z}_i+{\gamma }_2{Control}_i+{\mu }_i$$

(4)

$$Y_i={\delta }_0+{\delta }_1{X}_i+{\delta }_2{{\mu }_i+{\delta }_{3}Control}_i+{\epsilon }_i$$

(5)

${\mu }_i$ is the residual term of the instrumental variable on the endogenous variable.${\gamma }_0$, ${\gamma }_1$, ${\gamma }_2$, ${\delta }_0$, ${\delta }_1$, ${\delta }_3$ is the coefficient matrix of the corresponding variable. The other variables are set as above.

To further improve the robustness of this study, an ordered logit model was set up as follows:

$$Y_i^{*}={\phi }_0+{\phi }_1{X}_i+{\phi }_2{\mathit{Control}}_i+{\epsilon }_i$$

(6)

Meanwhile, the relationship between $Y_i^{*}$ and $Y_i$ is as follows:

$$Y_i^{*}=k, if {\tau }_k<Y_i^{\prime }<{\tau }_{k+1}, k=1,\dots , K$$

(7)

$Y_i^{\prime }$ is rural households’ income without logarithms, including the RTI, RFI, and ROI. ${\tau }_k$ is the threshold for different individuals i. The other variables are set as above.

4. Empirical Analysis

4.1. Baseline Regression

The regression results of LM on household income are presented in

Table 2. The impact of intra-county labor migration.

Variables	Dependent Variable: RTI (ln)		Dependent Variable: RFI (ln)		Dependent Variable: ROI (ln)
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
$L{abor}_{in}$	0.255 ** (0.190)	0.285 ** (0.172)	0.888 (0.547)	0.539 (0.527)	0.266 (0.258)	0.140 (0.240)
$age$	—	−0.021 *** (0.006)	—	0.005 (0.019)	—	−0.027 *** (0.009)
$edu$	—	0.085 (0.079)	—	0.112 (0.243)	—	0.163 (0.111)
$carde$	—	0.154 (0.122)	—	−0.550 (0.376)	—	0.343 ** (0.171)
$number$	—	−0.016 (0.027)	—	−0.067 (0.083)	—	−0.010 (0.038)
$labortmig$	—	−0.057 (0.182)	—	0.237 (0.559)	—	0.184 (0.255)
$numlabor$	—	0.249 *** (0.057)	—	0.667 *** (0.174)	—	0.218 *** (0.079)
$numlaborout$	—	0.190 ** (0.083)	—	−0.699 *** (0.255)	—	0.292 ** (0.116)
$Area$	—	0.006 (0.006)	—	0.024 (0.017)	—	0.001 (0.008)
${Area}_{ave}$	—	0.032 (0.038)	—	0.148 (0.116)	—	−0.154 *** (0.053)
$distance$	—	−0.004 (0.011)	—	−0.053 (0.033)	—	0.000 (0.015)
$vlabormig$	—	−0.240 (0.334)	—	−2.711 *** (1.024)	—	0.183 (0.466)
$lnvincome$	—	0.743 *** (0.199)	—	−0.134 (0.610)	—	0.886 *** (0.277)
$lnvnumber$	—	0.355 *** (0.087)	—	−0.470 * (0.266)	—	0.378 *** (0.121)
Cons	10.970 *** (0.077)	2.300 (1.902)	5.484 *** (0.223)	10.230 * (5.839)	10.600 *** (0.105)	0.420 (2.657)
N	506	506	506	506	506	506
Adj  R2	0.002	0.237	0.003	0.126	0.000	0.197

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% statistical levels, respectively; values in parentheses are regression standard errors.

. Model 1, which passed the 5% significance level test, indicates that ${Labor}_{in}$ significantly promoted the growth of the RTI (ln). Model 2 demonstrates that, even after accounting for control variables, ${Labor}_{in}$ continues to significantly enhance the RTI (ln), with the results remaining significant at the 5% level. Model 3 and Model 4 indicate that ${Labor}_{in}$ had no significant impact on the RFI (ln). In a similar vein, Model 5 and Model 6 reveal that ${Labor}_{in}$ had no significant effect on the ROI (ln).

The effects of additional control factors on household income are also shown in Table 2. Model 2 reveals that, at the individual characteristic level, age had a negative impact on the RTI (ln). At the household characteristic level, an increase in the number of laborers and the number of outgoing laborers both significantly boost the RTI (ln). At the village characteristic level, both village income and village population had a positive effect on the RTI (ln). In Model 4, at the household characteristic level, the number of laborers significantly promoted the growth of the RFI (ln), whereas an increase in the number of outgoing laborers engaged in external employment was associated with a reduction in the RFI (ln). At the village characteristic level, the percentage of village LM and village population were negatively correlated with the RFI (ln). The ROI (ln) is greatly increased at the household characteristic level by both an increase in the number of laborers and the number of outgoing laborers. The average area of forestland had a negative effect on the ROI (ln) at the village characteristic level, but the RTI (ln) was positively impacted by village income and village population.

The influence of Labor_out on the income of rural households is seen in

Table 3. The impact of outside-of-county labor migration on income.

Variables	Dependent Variable: RTI (ln)		Dependent Variable: RFI (ln)		Dependent Variable: ROI (ln)
	Model 7	Model 8	Model 9	Model 10	Model 11	Model 12
$L{abor}_{out}$	0.357 ** (0.180)	0.345 ** (0.174)	−1.082 ** (0.518)	−0.547 ** (0.335)	1.082 ** (0.518)	0.052 (0.243)
Control	—	Yes	—	Yes	—	Yes
Cons	11.07 *** (0.086)	2.241 (1.897)	6.018 *** (0.246)	9.944 * (5.835)	6.018 *** (0.246)	0.364 (2.656)
N	506	506	506	506	506	506
Adj  R2	−0.001	0.239	0.007	0.126	0.007	0.196

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% statistical levels, respectively; values in parentheses are regression standard errors.

. Labor_out was shown to significantly boost the rise of the RTI (ln) in Model 7. Model 8 indicates that, even after accounting for the control variables, Labor_out continues to significantly promote the RTI (ln). This effect was significant at the 5% level. According to Model 9, the RFI (ln) decreased when the Labor_out increased. This finding was significant at the 5% level. Even after accounting for the control variables, Labor_out was still adversely connected with the RFI (ln), as shown by Model 10, which was similarly significant at the 5% level. The growth of the ROI (ln) was greatly aided by Labor_out, according to Model 11, which is significant at the 5% level. Nevertheless, Model 12 concludes that the effect of Labor_out on the ROI (ln) was not statistically significant when controlling for the control variables.

4.2. Quantile Regression

The baseline regression model focused on the impact of the LML on the conditional expectation of income, which is a form of mean regression. If the distribution of the variable is skewed or contains outliers, it may lead to biased model estimation results. Applying the quantile regression method can mitigate these issues and reveal the impact of independent variables on the dependent variable across the entire distribution of the dependent variable. Therefore, this study utilized the quantile regression analysis method to examine the effect of LM on income across different conditional distributions. The results are presented in

Table 4. Quantile regression analysis of the income effects of labor migration.

Variables	Dependent Variable: RTI (ln)				Dependent Variable: RFI (ln)
	25%	50%	75%	90%	25%	50%	75%	90%
	Model 13	Model 14	Model 15	Model 16	Model 17	Model 18	Model 19	Model 20
$L{abor}_{in}$	0.019 (0.140)	0.162 (0.130)	0.229 * (0.181)	0.341 ** (0.181)	0.911 (1.327)	0.866 (0.719)	−0.173 (0.399)	−0.131 (0.507)
$L{abor}_{out}$	0.003 (0.168)	0.250 * (0.161)	0.980 ** (0.452)	0.461 ** (0.228)	0.061 (1.028)	0.002 (0.748)	−0.601 (0.467)	−0.898 ** (0.424)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Cons	5.132 *** (1.437)	5.280 *** (1.447)	4.373 ** (1.854)	4.693 (3.527)	16.130 (12.770)	9.025 (9.098)	5.143 (4.630)	−0.248 (5.112)
N	506	506	506	506	506	506	506	506
R2	0.242	0.214	0.201	0.220	0.044	0.093	0.108	0.137

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% statistical levels, respectively; values in parentheses are regression standard errors.

. Models 13 to 16 represent the results for the RTI (ln) at the 25%, 50%, 75%, and 90% quantiles. Models 17 to 20 show the results for the RFI (ln) at the 25%, 50%, 75%, and 90% quantiles.

After controlling for the factors, the effect of LM on income is shown in Models 13- Model 16. The results indicate that the influence of ${Labor}_{in}$ on the RTI (ln) increases as the quantiles increase, and this effect is statistically significant only at the 75% and 90% levels. The impact of ${Labor}_{out}$ on the RTI (ln) was not significant at quantiles below 25%. From the 50% to the 75%, the income effect of ${Labor}_{out}$ showed an upward trend. Beyond 90%, this income enhancement effect remained significant, but the magnitude of the impact coefficient decreased to some extent.

After controlling for the factors, Models 17–Model 20 show how LM affects rural households’ income. The findings show that, for all quantiles, the impact of ${Labor}_{in}$ on the RFI (ln) was not statistically significant. This implies that an unequal distribution of income levels was not the cause of the overall mean finding, which indicated no discernible effect of ${Labor}_{in}$ on the RFI (ln). Additionally, the impact of ${Labor}_{out}$ on the RFI (ln) was not significant at quantiles below 90%. However, at 90% and above, each unit increase in the ${Labor}_{out}$ results in a decrease in the RFI (ln) by 0.898 units.

4.3. Robustness Test

In the quantile regression analysis, this study accounted for the potential bias caused by extreme values. However, outliers in the data could also impact the conclusions. To address this, the study applied a 1% bilateral shrinkage to both the LM and income. Furthermore, an ordered logit model was used for the analysis in this study in place of the multiple linear regression model, further strengthening the credibility of the previous discoveries. In this study, the RFI was divided into five groups, namely 0–1000, 1001–4000, 4001–7000, 7001–12,000 and above; the RTI and ROI were divided into five groups, namely 0–150,000, 150,001–250,000, 250,001–400,000, 400,000–550,000, and 550,000 above. The results of the robustness test are displayed in

Table 5. Robustness analysis of the income effects of labor migration.

Variables	Dependent Variable: RTI (ln)	Dependent Variable: RFI (ln)	Dependent Variable: ROI (ln)	Dependent Variable: RTI (ln)	Dependent Variable: RFI (ln)	Dependent Variable: ROI (ln)
	Model 21	Model 22	Model 23	Model 24	Model 25	Model 26
$L{abor}_{in}$	0.364 ** (0.159)	0.427 (0.583)	0.153 (0.265)	0.314 * (0.216)	0.272 (0.286)	−0.152 (0.297)
$L{abor}_{out}$	0.282 ** (0.132)	−0.596 ** (0.461)	0.145 (0.271)	0.327 * (0.157)	−0.433 * (0.304)	0.296 (0.310)
Control	Yes	Yes	Yes	Yes	Yes	Yes
Cons	4.632 *** (1.302)	10.830 * (5.892)	−0.364 (2.681)	13.040 *** (3.097)	0.919 (2.864)	9.201 *** (3.038)
N	506	506	506	506	506	506
Adj  R2	0.293	0.178	0.178	0.199	0.200	0.199

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% statistical levels, respectively; values in parentheses are regression standard errors.

. The regression results using the ordered logit model are presented in Models 24, 25, and 26, whereas the regression results following shrinkage are displayed in Models 21, 22, and 23. According to Models 21 and 24, ${Labor}_{in}$ and ${Labor}_{out}$ both contributed to the rise in RTIs (ln). It is demonstrated that ${Labor}_{out}$ impeded the increase of RFIs (ln) in Models 22 and 25. The robustness of the study’s conclusions is supported by the results’ concordance with earlier findings and comparable significance levels.

4.4. Heterogeneity Analysis

There is a bidirectional causal relationship between the scale of forestland and income. When households achieve a certain size, they can realize economies of scale through intensive management, which in turn affects income. Conversely, as income levels increase, they have more capital available to expand the size of forestland management, thereby pursuing higher marginal returns. Thus, this study adds the scale of forestland as a grouping variable to the analysis framework in order to examine whether the effect of LM on income differs across samples. The groupings are as follows: households with less than or equal to 50 mu of the scale are assigned $Scale$ = 1, households with more than 50 mu of the scale but less than or equal to 100 mu are given $Scale$ = 2, and households with more than 100 mu of the scale are assigned $Scale$ = 3. The grouped regression results are presented in

Table 6. Heterogeneity analysis based on the scale of forestland management.

Variables	Dependent Variable: RTI (ln)			Dependent Variable: RFI (ln)
	Scale = 1	Scale = 2	Scale = 3	Scale = 1	Scale = 2	Scale = 3
	Model 27	Model 28	Model 29	Model 30	Model 31	Model 32
Laborin	0.433 (0.308)	0.473 ** (0.382)	0.342 ** (0.198)	−0.414 (1.027)	1.309 (0.957)	−1.008 (1.070)
Laborout	0.997 ** (0.571)	0.581 * (0.411)	0.532 ** (0.359)	0.263 (0.903)	1.372 (1.030)	−2.497 ** (1.123)
Control	Yes	Yes	Yes	Yes	Yes	Yes
Cons	−2.436 (3.889)	−0.726 (4.355)	6.600 ** (2.818)	10.100 (12.980)	−9.495 (10.910)	15.380 * (8.815)
N	166	171	169	166	171	169
R2	0.307	0.150	0.155	0.162	0.175	0.173

Note: * and ** indicate significance at the 10% and 5% statistical levels, respectively; values in parentheses are regression standard errors.

. Models 27, 28, and 29 indicate that the promotion effect of ${Labor}_{in}$ on the RTI (ln) occurred only in the sample groups with a scale greater than 50 mu. On the other hand, the positive impact of ${Labor}_{out}$ on the RTI (ln) did not differ across the sample groups with varying scales. The findings of Models 30, 31, and 32 indicate that a scale higher than 100 mu especially experienced a negative impact on the RFI (ln) due to ${Labor}_{out}$.

The number of labor and the level of family caregiving can be influenced by the LM of the head of the household, which can impact income. Thus, this study found out if sample groups with different situations of household head LM have different income effects from LML. This study assigned a value of $headlabortmig$ = 1 to the sample groups that the household head had LM behavior, and $headlabortmig$ = 0 to those household head that had not.

Table 7. Heterogeneity analysis based on household head’s labor migration behavior.

Variables	Dependent Variable: RTI (ln)		Dependent Variable: RFI (ln)
	$\mathit{h}\mathit{e}\mathit{a}\mathit{d}\mathit{l}\mathit{a}\mathit{b}\mathit{o}\mathit{r}\mathit{t}\mathit{m}\mathit{i}\mathit{g}$ = 0	$\mathit{h}\mathit{e}\mathit{a}\mathit{d}\mathit{l}\mathit{a}\mathit{b}\mathit{o}\mathit{r}\mathit{t}\mathit{m}\mathit{i}\mathit{g}$ = 1	$\mathit{h}\mathit{e}\mathit{a}\mathit{d}\mathit{l}\mathit{a}\mathit{b}\mathit{o}\mathit{r}\mathit{t}\mathit{m}\mathit{i}\mathit{g}$ = 0	$\mathit{h}\mathit{e}\mathit{a}\mathit{d}\mathit{l}\mathit{a}\mathit{b}\mathit{o}\mathit{r}\mathit{t}\mathit{m}\mathit{i}\mathit{g}$ = 1
	Model 33	Model 34	Model 35	Model 36
$L{abor}_{in}$	0.228 ** (0.121)	0.354 ** (0.147)	0.002 (0.695)	1.018 (1.187)
$L{abor}_{out}$	0.431 ** (0.234)	0.126 (0.335)	−0.494 (0.712)	−1.184 ** (0.725)
Control	Yes	Yes	Yes	Yes
Cons	3.306 (2.301)	−2.515 (3.576)	7.676 (7.012)	2.160 (12.00)
N	364	142	364	142
R2	0.172	0.184	0.198	0.093

Note: ** indicates significance at the 5% statistical levels; values in parentheses are regression standard errors.

presents the results. Models 33 and 34 indicate that, regardless of whether the household head had LM behavior, ${Labor}_{in}$ contributed to the growth of RTIs (ln). However, only in sample groups where the household head had no LM did ${Labor}_{out}$ significantly promote the growth of RTIs (ln). Models 35 and 36 show that ${Labor}_{in}$ had no discernible effect on the RFI (ln), irrespective of LM. However, only households with a head that exhibited LM behavior experienced a substantial negative effect of ${Labor}_{out}$ on the RFI (ln).

4.5. Endogeneity Test

The baseline regression analysis has established that LM significantly impacts income. Conversely, changes in income may also influence LM, either encouraging or discouraging LM and affecting the choice of LML. To address this bidirectional causation, this article employs the CF method and IV method, respectively. It used the average LM ratio of other rural households in the village, excluding the sample rural households, as instrumental variables. Let ${{Labor}_{in}}^{\prime }$ represent the average ${Labor}_{in}$ of other rural households, and let ${{Labor}_{out}}^{\prime }$ represent the average ${Labor}_{out}$ of other rural households. The regression results are shown in

Table 8. Endogeneity test results.

Variables	Dependent Variable: RTI (ln)	Dependent Variable: RFI (ln)	Dependent Variable: ROI (ln)	Dependent Variable: RTI (ln)	Dependent Variable: RFI (ln)	Dependent Variable: ROI (ln)
	Model 37	Model 38	Model 39	Model 40	Model 41	Model 42
${L{abor}_{in}}^{\prime }$	0.327 ** (0.192)	0.371 (0.591)	0.205 (0.269)	0.267 * (0.171)	0.135 (0.434)	0.153 (0.123)
${L{abor}_{out}}^{\prime }$	0.662 ** (0.310)	0.742 (0.150)	8.842 (15.540)	0.432 ** (0.249)	0.120 (0.978)	0.343 (0.278)
Control	Yes	Yes	Yes	Yes	Yes	Yes
${\mu }_i$	−9.929 (11.110)	−1.121 (34.170)	−8.704 (15.550)	—	—	—
Cons	−0.613 (3.701)	9.803 (11.390)	−2.026 (5.180)	5.045 (5.172)	5.394 (289.700)	0.142 (0.826)
N	506	506	506	506	506	506
R2	0.126	0.121	0.107	0.093	0.086	0.042

Note: * and ** indicate significance at the 10% and 5% statistical levels, respectively; values in parentheses are regression standard errors.

. The CF method was used for Model 37, Model 38, and Model 39, while the IV method was used for Model 40, Model 41, and Model 42.

The residual term ${\mu }_i$ was found to fail the significance level test in Model 37, Model 38, and Model 39, i.e., the endogeneity assumption above is not valid. Additionally, both Models 37 and 40 passed the significance level tests, indicating that ${Labor}_{in}$ and ${Labor}_{out}$ both significantly promote the growth of the RTI (ln). These findings are consistent with the baseline regression results, further reinforcing the robustness of the study’s conclusions.

5. Discussion

This paper employs a multiple linear regression model to examine the effects of${Labor}_{in}$ and ${Labor}_{out}$ on the income levels of a sample of rural households. The empirical findings demonstrate that LM has a substantial positive impact on the RTI. This outcome validates hypothesis 1 and aligns with the conclusions gained from prior research [45]. Nevertheless, this study fails to address the further attributes of LM, leaving several areas untouched. Studies indicated that individuals of different genders may contribute varying levels of labor quality, which might result in unequal remuneration [46,47]. Furthermore, Ma et al. (2023) discovered that those who work as part-time peasants and those who belong to non-agricultural households exhibit a greater inclination to lease out their forestland [48]. Multiple studies have demonstrated that the lease of forestland by rural households has a significant impact on labor reallocation [21,49]. Does the diversification of rural households impact both LM and income? Alternatively, do different types of rural households result in variations in the income effect of LM? However, the issue of household registration discrimination continues to persist in China’s labor market. According to Meng and Zhang (2001), there is a notable disparity in wages between urban residents and rural migrant workers when examining the Shanghai labor market [50]. The aforementioned elements will have an impact on the income effect of LM. However, this research does not incorporate them into the analysis framework. On the other hand, this paper thoroughly examined the income changes in rural households by taking into account their RTI, RFI, and ROI. This approach allows for a better understanding of how the LM affects the overall income structure. However, this solely considers that the absolute income level is insufficient to remove the influence of macro factors like inflation. Consequently, some scholars have examined the effect of LM on the income disparity among rural households. The findings indicated that LM can enhance the income distribution of rural households by the remittance effect, thereby reducing the income gap between urban and rural regions [51]. Given the current rapid development of the rural economy, it is unclear if various LMLs can have varying effects on the income disparity between urban and rural areas. Hence, our team will persist in conducting further research to augment the empirical assessments in these areas of deficiency.

This study categorizes LM into two types: ${Labor}_{in}$ and ${Labor}_{out}$. The examination of the income impact of ${Labor}_{in}$ reveals that ${Labor}_{in}$ only had a beneficial influence on the RTI. However, the effect on RFI and ROI was not statistically significant. Analysis of the economic effect of ${Labor}_{out}$ reveals that ${Labor}_{out}$ could stimulate the growth of ROIs and decrease RFIs. However, the RTI demonstrates an upward tendency. Research has indicated that the LM had a substantial negative impact on the growth of income derived from forestry activities [22]. Nevertheless, this conclusion holds true only when it comes to the ${Labor}_{out}$. It suggests that the migration effect and remittances effect resulting from ${Labor}_{in}$ is not significant. However, the remittance effect of ${Labor}_{out}$ is considerably greater than the migration effect.

This research utilized a quantile regression model to investigate the impact of LM on income enhancement across several value domains. By dividing the data into four quartiles at 25%, 50%, 75%, and 90%, it was observed that the highest income increase resulting from ${Labor}_{in}$ occurred at the 90% quartile. On the other hand, the largest impact on income resulting from ${Labor}_{out}$ was observed at the 75% quartile. Although studies on the topic are scarce, it has been suggested that there is an optimal migration ratio for the impact of LM on rural households’ income. Furthermore, it is possible that the optimal ratios for ${Labor}_{in}$ and ${Labor}_{out}$ could differ.

According to the theoretical analysis, this study suggests that there could be a reciprocal relationship between the LM and rural households’ income, meaning that there might be an issue of endogeneity. Thus, this work utilized the IV and CF methods to explore endogeneity. However, the results indicate that there is no presence of endogeneity, which diverges from both our hypothesis and previous research. This finding may be attributed to the following factors: (1) This study utilized the 2018 cross-sectional data. However, it is possible that the impact of income on LM is captured in the data from the previous era. (2) With the increasing frequency of LM, the rural labor market may have transitioned from being saturated to experiencing a shortage of labor. Farmers will opt to retain a portion of their labor instead of migrating it due to the demands of family care and forestland management. Additionally, the impact of revenue increase on LM diminishes over time.

6. Conclusions and Policy Implications

6.1. Conclusions

This study examined the effects of LM on the income of rural households in Jiangxi Province, taking into account the policy of promoting rural revitalization. It analyzed the differential impact of different LMLs using 506 research data. The study employed a multivariate linear regression model and a quantile regression model for analysis. Simultaneously, it investigated the impact of the fluctuation in the scale of forestland management and the situation of the family head to LM. Furthermore, this paper also performed rigorous tests to assess the resilience and potential confounding factors in the relationship between income and LM. The following findings are derived (Figure 3).

First of all, intra-county labor migration would stimulate the growth of the total income of rural households. Specifically, for each additional unit in the ${Labor}_{in}$, the RTI would increase by 0.255 units.

Secondly, outside-of-county labor migration played a substantial role in increasing the total income of rural households and the off-farm income of rural households but had a notable impact on reducing the forestry income of rural households. To clarify, a one-unit rise in outside-of-county labor migration results in a 0.357-unit gain in the total income of rural households, a 1.082-unit increase in off-farm income of rural households, and a 0.547-unit drop in the forestry income of rural households.

Third, the analysis of quantile regression demonstrates that when considering the 50%, a one-unit rise in intra-county labor migration is linked to a 0.250-unit increase in the total income of rural households. At 75%, an increase of one unit in intra-county labor migration was correlated with a 0.229-unit increase in the total income of rural households. As the quartile increases, the coefficient of income gain from outside-of-county labor migration reaches its highest value at 75%. This means that a one-unit increase in outside-of-county labor migration was connected with a 0.980-unit increase in the total income of rural households. At 90%, a one-unit increase in intra-county labor migration led to a maximum increase of 0.341 in the total income of rural households. Similarly, a one-unit increase in outside-of-county labor migration resulted in an increase of 0.461 in the total income of rural households. However, at this point, there was a decrease of 0.898 in the forestry income of rural households.

Last but not least, the study of heterogeneity in the scale of forestland management reveals that the positive impact of intra-county labor migration on the total income of rural households was only observed among those who possess more than 50 acres of forestland. The favorable effect of outside-of-county labor migration on the income of rural households did not vary among different sample groups with varying amounts of forestland. However, the negative impact on the total income of rural households was observed specifically among those with forestland areas larger than 100 mu. Furthermore, this study examined the diversity in LM by the head of the family, and the findings indicate that the intra-county labor migration helped the growth of the total income of rural households, regardless of whether the head of the household engages in LM or not. Outside-of-county labor migration could only contribute to the growth of the total income of rural households for the sample group when the head of the family does not LM. The impact of intra-county labor migration on the forestry income of rural households is not statistically significant, independent of the occurrence of LM. The adverse effect of outside-of-county labor migration on the forestry income of rural households is only observed in the subset of households where the head of the household undergoes LM.

6.2. Policy Implications

First of all, it is imperative to actively foster the county’s distinctive economy and encourage the process of rural urbanization. By leveraging the county’s resource endowment and economic foundation, it is possible to strategically develop the county’s economy in a way that capitalizes on its unique characteristics and competitive advantages. Secondly, the government should enhance the dissemination of policy information regarding LM and enhance the efficiency of the LM system. The government’s efforts to promote LM are beneficial in fostering greater uniformity, minimizing the risks and uncertainties associated with LM, and enhancing the protection of farmers’ rights and interests. Thirdly, it is imperative for the government to provide guidance to farmers in implementing LM in a measured manner while also discouraging farmers from engaging in indiscriminate ${Labor}_{out}$. Simultaneously, it persists in advancing the reform of the rural land system and expediting the conversion of forestland. Expediting the lease of forestland can facilitate the expansion, industrialization, and institutionalization of forestland management. This can enhance the efficiency of forestland assets, diversify income sources for farmers, and prevent a significant decrease in RFI following LM.