Rural Migrant Workers in Urban China: Does Rural Land Still Matter?

Abstract

The purpose of this study is to test the response of rural–urban migration to land endowment while recognizing the heterogeneity of land tenure security perceptions. Based on the survey data of 751 migrant workers in Nanjing City, the latent class model identifies the secured group and unsecured group, with a particular focus on how heterogeneous concepts of land tenure security mediate the above relationships. Empirical modeling explores how land endowment affects household labor distribution and individual migration decisions, taking the heterogeneity of tenure expectations into account. The key findings reveal the following: (1) an inverse U-shaped relationship between land endowment and household labor distribution, but not between land endowment and individual migrant decisions; (2) farm households who perceive land tenure as unsecured tend to send fewer household members as job-seeking migrants, even as their land endowment increases; (3) however, individual migration decisions under conditions of a secured land tenure reduce the risk of losing land and induce migrant workers to stay in cities. The findings contribute to advancing the institutional analysis of the impact of land endowment on household labor allocation and how land tenure security affects immigration decisions, providing empirical evidence for China’s rural revitalization policies, which would support reforms that secure land tenure, such as the second-round contracting renewal policy and the rural revitalization plan in China.

1. Introduction

As an emerging economy, China has been experiencing an unprecedented labor migration from rural to urban areas since the late 1980s. The outmigrant population surged from 30 million in 1989 [1] to 177 million in 2023 [2]. This has attracted attention, focus, and some related research. Centering on the topic of internal migration, the initial literature in China, while extensive, mainly focuses on constraints from the labor market, including the household registration system, i.e., the hukou system [3,4], and the labor market discrimination [5,6]. In fact, despite China’s implementation of hukou reform and the removal of labor market discrimination, the migrant population increased less than one percent in the three years preceding the COVID-19 pandemic and was negatively affected during the pandemic. In other words, when the institutional and non-institutional constraints from hukou were gradually eliminated, the outmigration from rural areas to urban areas did not increase as expected. Migration may not develop if constraints from other factor markets remain. This is supported by the evidence that the extension of rural land rights to peasants has driven urbanization more effectively than the relaxation of migration restrictions [7]. One of its root causes lies in the fact that, at the inception of the rural household contract responsibility system, land allocation was linked to the number of labor or farmer members. In addition, the rural contracted law does not allow contracted land to be abandoned. Therefore, at the end of the last century, rural labor flow was constrained by rural land institutions. The experience of most developing countries also shows that farm households are systematically exposed to market imperfections and constraints other than labor markets [8]. With the explosive growth of migrant workers in China, off-farm labor market activities are the main catalyst of rising land rental activity [9]. In the face of population outflows, China’s government has put forward a large number of policies to guide the transfer of farmland management rights [10]. In theory, if the land and labor markets work well, labor allocation decisions should be independent of land endowment [8,11]. Nevertheless, the manner in which rural land endowment affects the decisions of household labor allocation is an extremely complex process.

The literature has further provided insights into the relationship between household land endowment and rural–urban migration. Specifically, several studies found evidence of the U-shaped relationship between land endowment and the probability of migration [12,13,14], implying that the probability of migration decreases as the area of land operation increases but that eventually the probability increases as the area of land operation continues to increase. Conversely, other research reports an inverse U-shaped relationship [15,16,17]. The critiques of the U-shaped relationship may stem from the absence of a formal, rigorous procedure for testing the U-shaped relationship. Past research almost exclusively included a quadratic term in its standard regression models and relied on the sign and significance of the level term and the quadratic term parameters to establish the U-shaped relationship’s existence [18]. However, this criterion can be misleading when the estimated extremum is located at the end of the data range [18]. Indeed, for a farm household, the operated land includes the contracted land (i.e., initial endowment) and the rent-in/rent-out land (i.e., increased endowment). The above studies, whether supporting a U-shaped or an inverse U-shaped relationship, have failed to distinguish between the impact of these two types of endowments on labor migration. Whether a U-shaped or an inverse U-shaped relationship exists, fundamentally, is a matter of a spatial mismatch between labor mobility and farmland fixation. The mismatch, in turn, leads to the coupling imbalance of rural–urban migration and rural land-use transition [19].

More importantly, existing studies of the U-shaped relationship in terms of land endowment oversimplify to provide an economic explanation of the identified relationships. On the one hand, the studies ignore the lack of well-defined land tenure in rural China, which can create considerable distortions in the relationship between land endowment and labor allocation. An unsecured land tenure increases the need for “guard labor”, which was previously principally devoted to agricultural production, to solidify informal land tenure [20,21]. Even considering the marketization of land leasing, land renting-in tends to increase the labor involved in agricultural production [22]. However, land renting-out may reduce the labor allocation toward agricultural activities, and in cases of unsecured land tenure, it may also consume some of the time of outmigrants to tackle leasing affairs. On the other hand, studies confuse the perspectives of individual decisions and joint decision-making in households. While the early migration models emphasized the migration decisions made by an individual household member [23], more recent research has focused on household labor allocation [12,15,17]. Given the fundamental distinction between the household and individual agency, it is worthwhile to distinguish between those decisions. Against this backdrop, the purpose of this paper is to focus on the relationship between land endowment and rural labor migration in the case of unsured land tenure. By addressing the institutional complexities of land rights and differentiating between individual and household decision frameworks, this research seeks to provide a more comprehensive explanation of the observed U-shaped relationship.

This paper examines how rural land endowment shapes household and individual labor allocation decisions in an attempt to provide an explanation for the pervasive, incomplete rural-to-urban migration. To do so, this paper uses survey data collection through a specifically designed questionnaire administered to migrant rural workers with urban jobs. The latent class model identifies distinct groups of migrants with different perceptions of land tenure, specified as a “secured group” (those confident in their land rights) and an “unsecured group” (migrants unsure of the security of their land rights). The results show an inverse U-shaped relationship between land endowment and migrant workers’ household labor distribution in the secured group. For households in the unsecured group, the larger the land endowment is, the less likely it is that a household sends members in search of urban jobs. Further, the results show land that endowment has a limited impact on a rural individuals’ migration plan, suggesting that land endowment affects household labor distribution rather than individual migration plans. Additionally, land tenure insecurity hinders rural labor outmigration at the household and individual level. The findings are robust under alternative models and variable specifications. Complementary evidence from the survey sample’s demographic profile strengthens the analysis: the average age was approximately 43 years, projected to rise to 57 by 2025. As a result, the competitiveness of this group above the average age in the urban labor market has declined, while fostering a general return-to-rural trend. Significantly, most of the participants went back to engage in farming, reflecting their relatively stable attachment to land. Their perceptions of land tenure security thus continue to shape rural labor allocation dynamics at both the household and individual level, underscoring the institutional and demographic complexities influencing migration patterns.

This study complements previous studies on labor allocation across sectors [24,25,26,27] through two distinct contributions. Firstly, it may extend the research scope of labor allocation by focusing on migrant workers and their cross-market constraints from land endowment. Unlike prior studies, this paper systematically differentiates the effects of land endowment between two groups of migrants with different perceptions of land tenure security, thereby uncovering nuanced institutional influences on migration behavior. Secondly, methodologically, the study helps to identify heterogeneous migrant subgroups by employing a latent class cluster model, and helps to test for U-shaped relationships by implementing a robust analytical approach. These methodologies enhance the rigor of our labor allocation dynamics analysis, addressing limitations in earlier research. Collectively, this work will help to deepen our understanding of how land endowment, moderated by tenure security perceptions, shapes rural–urban migration and cross-sectoral allocation.

2. Theoretical Framework

2.1. Labor Migration and Land Tenure

The early literature on development economics documented that the growth path of most advanced economies was accompanied by the process of structural transformation involving the reallocation of the labor force across sectors and/or regions [23,28]. While rural households in China are able to migrate, the household registration system (i.e., the hukou system in China) [3,4], labor market discrimination [5,6], and other institutional and non-institutional constraints result in the phenomena of temporary and circular migration [29,30,31]. Temporary migration leaves migrants largely excluded from public services such as education, health care, and other forms of social support in cities [4,32].

Access to land and land ownership is a major household asset in most developing countries, shaping the socioeconomic stratification within rural communities. Land area affects the labor (or working time) distribution of rural households, as well as that of household members who have already migrated in search of urban jobs. For example, when farming labor runs short, households operating with sizable land areas do not necessarily recall household members back from migration because they can lease out their land in order to lower labor requirements or hire a worker to substitute for their migrating household members. However, farm households in most developing countries are systematically exposed to various market imperfections and constraints [8]. Market imperfections arise from high transaction costs and limited labor, including the nonexistence of local labor markets [8]. In China, local labor markets and land leasing markets have been underdeveloped [33,34]. The exception to this is that the relationship between suburban and rural areas and land transfer is much stronger [35]. Therefore, the understanding of how migrant workers allocate labor in the context of market failures should consider the inseparability between land endowment and the labor market. The migration literature, while extensive, provides little insight into the effects of land endowment on migrant workers’ allocation of labor between farms and urban jobs. Therefore, it is important to rebuild the rural household labor allocation model to consider land endowment.

2.2. Rural Household Model for Labor Allocation

Consider a rural household that has two available resources: labor and farmland. The household can allocate labor to farming and urban jobs. Let the household have a fixed total labor resource normalized to size 1 and allocate l of the labor to rural-to-urban migration and 1−l to farming activities. The household has a land endowment of A, which is exogenously determined. Farmland can generate income by raising crops, be left idle, or be underutilized. Land rental activities are ignored in this study because the prevalence of the land rental market in rural China remains low [36]. In other words, the land leasing market is not active. If farm households perceive their land rights to be secure, they will invest production factors, such as labor and fertilizer, into their farmland in expectation of additional returns. Otherwise, farmers underutilize their land to avoid the risk of a loss of investment under insecure tenure.

Let the development of land tenure be given by $\lambda \in \left[\mathrm{0,1}\right]$. As land tenure is viewed as more secure, the value of $\lambda$ increases. There are two extremes of the land tenure system: $\lambda =0$ indicates the “public” and unprotected land tenure regime, whereas $\lambda =1$ represents “private” and protected land tenure. Any situation in between implies imperfect land tenure. In the absence of the land market, $\lambda$ may be considered as land adjustment or reclamation. The Rural Land Contracted Law in 2002 stipulated that farm households’ contracted land should be withdrawn if they migrated and registered in urban areas (see Article 26). The law was amended in 2018, as well as the withdrawal article was abolished. If $A^f$ is the amount of land available to the amount of household labor left to farm and $A^u$ is the underutilized farmland distributed to household farm labor, we assume the following:

$$A^f=A^f\left(l,\lambda \right)$$

(1)

$$A^u=A^u\left(l,\mathsf{\lambda }\right),$$

(2)

where $A_{\lambda }^f\left(l,\lambda \right)>0$, $A^f\left(1,0\right)=0$, $A_{\lambda }^u\left(l,\lambda \right)<0$, $A^u\left(l,1\right)=0$, and $A_{\lambda }^f\left(l,\lambda \right)$ and $A_{\lambda }^u\left(l,\lambda \right)$ are the derivatives of $A^f\left(l,\lambda \right)$ and $A^u\left(l,\lambda \right)$ with respect to $\lambda$, respectively. The positive sign indicates that, as land tenure becomes secure, more land will be farmed than left underutilized. We assume two simplified expressions for land distribution:

$$A^f\left(l,\lambda \right)={\displaystyle \frac{\lambda (1-l)}{l}}A$$

(3)

$$A^u\left(l,\lambda \right)={\displaystyle \frac{\left(1-\lambda \right)l}{1-l}}A$$

(4)

where ${\displaystyle \frac{\lambda (1-l)}{l}}\in \left[\mathrm{0,1}\right]$, and ${\displaystyle \frac{\left(1-\lambda \right)l}{1-l}}\in \left[\mathrm{0,1}\right]$, because this study assumes that land rental activities are ignored. If the land rental market works, the results of ${\displaystyle \frac{\lambda (1-l)}{l}}$ and ${\displaystyle \frac{\left(1-\lambda \right)l}{1-l}}$ will be greater than 1.

Household labor can generate income from off-farm activities and/or farming. Each activity is characterized by a twice-differentiable production function, with$P_a$ denoting the farming output and $P_{na}$ representing the migration output:

$$P_a=f\left[l,A^f\left(l,\lambda \right)\right] \mathrm{and}$$

(5)

$$P_{na}=g\left[l,A^u\left(l,\lambda \right)\right],$$

(6)

where $f^{\prime }\left(l\right)<0$, $f^{″}\left(l\right)>0$, $g^{\prime }\left(l\right)>0$, and $g^{″}\left(l\right)<0$. The derivatives show that the farming output is a continuous and convex function in terms of migrant labor, while the output function of migration is continuous and concave in terms of labor for off-farm activities. Meanwhile, as land area increases, households increase farming output; that is, $f^{\prime }\left(A^f\right)>0$. If households leave farmland idle or underutilized, the risk of loss of farmland exists when household members outmigrate; thus, $g^{\prime }\left(A^u\right)<0$.

2.3. Locally Stable Equilibrium and Conditions

Farm households allocate labor between urban and rural areas. Their family members engage in agricultural or non-agricultural activities in pursuit of more substantial income. Household members capable of working face the choice of migrating or farming the land. The equilibrium state arises when household working members lack the incentive to switch between migration and farming:

$$p{P}_a=w{P}_{na},$$

(7)

where p and w are the exogenous prices for agricultural output and migrant labor, respectively. Two inequalities rule out the corner solution to labor distribution (or specialization in production): “$l=1$” and “$l=0$”. Thus, $pf\left(0,A^f\right)<wg\left(1,A^u\right)$ implies that if all household working members remain farming, the income from farming will be lower than the income from an urban job. There is an incentive for household members to migrate. This inequality establishes the first condition for a locally stable equilibrium.

Condition 1: $pf\left(0,A^f\right)<wg\left(1,A^u\right)$.

Another requirement for a locally stable equilibrium is that any transfer of household working members to urban jobs should lead to a smaller marginal income from off-farm activities than from farming. Otherwise, the transfer will persist until all the rural household labor force migrates to urban areas. The condition is as follows:

Condition 2: $p{\displaystyle \frac{\partial P_a}{\partial l}}>w{\displaystyle \frac{\partial P_{na}}{\partial l}}$ or $p{f}_l+p{f}_{A^f}{\displaystyle \frac{\lambda }{l^2}}A>w{g}_l+w{g}_{A^u}{\displaystyle \frac{1-\lambda }{{\left(1-l\right)}^2}}A$.

However, both conditions do not necessarily map onto a unique solution to a locally stable equilibrium. According to Equation (7), both migration to urban jobs and farming are functions of land tenure security perception, $\mathsf{\lambda }$. Figure 1 shows a locally stable equilibrium for two scenarios: homogeneity and heterogeneity in land tenure perceptions. Under the assumption of the homogeneity of land tenure perceptions, there exists a unique, locally stable equilibrium (Figure 1a).

In contrast, relaxing the assumption allows for a heterogeneity of land tenure perceptions. There are two locally stable equilibria and one unstable equilibrium, shown in Figure 1b. In Figure 1b, the upper interaction point represents a locally stable equilibrium under the expectations of secured land tenure, while the lower interaction point is a locally stable equilibrium with the expectation of unsecured land tenure. The interaction shown in the middle of Figure 1b is an unstable equilibrium since the slope of migration income is greater than the agricultural income’s slope. Thus, under these conditions, higher migration will result in additional income from migration, which violates the second condition of a stable equilibrium.

2.4. Land Endowment and Labor Supply: Heterogeneity in Land Tenure Security

When the equilibrium state represented by Equation (7) occurs for a farmer family, its land endowment will also correspond to the result of an equilibrium state$\overline{A}$. Obviously, it will also be determined by l and $\lambda$, i.e., $\overline{A}=\overline{A}(\mathrm{l},\lambda )$. In this way, $\overline{A}$ is an exogenous variable. Thus, the equilibrium Equation (7) can be written as an identity:

$$F=p\ast f\left\{l,\overline{A}(l,\lambda )\right\}-w\ast g\left\{l,\overline{A}(l,\lambda )\right\}\equiv 0$$

(8)

The critical causal effect in the current study is the impact of land endowment on rural labor migration to urban jobs. First consider the overall impact, assuming that rural laborers are homogeneous in terms of their perceived land tenure security. That is, i.e., how land endowment affects rural household labor distribution while holding other factors constant. According to the implicit function theorem [37], using Equations (3)–(8) and two conditions, the locally stable equilibrium is as follows:

$${\displaystyle \frac{dl}{d\overline{A}}}=-{\displaystyle \frac{\partial \mathrm{F}/\partial \overline{\mathrm{A}}}{\partial \mathrm{F}/\partial \mathrm{l}}}={\displaystyle \frac{w{g}_{\overline{A}}\frac{\lambda (1-l)}{l}-p{f}_{\overline{A}}\frac{\left(1-\lambda \right)l}{1-l}}{p{f}_l+p{f}_{\overline{A}}\frac{\lambda }{l^2}\overline{A}-w{g}_l-w{g}_{\overline{A}}\frac{1-\lambda }{{\left(1-l\right)}^2}\overline{A}}}$$

(9)

therefore, the overall sign of the deviation of household migrant labor over land endowment is negative.

However, Equation (9) also indicates that the deviation of household migrant labor over land endowment is a function of land tenure security, $\mathsf{\lambda }$. Thus, the research question of this article is expressed as how farm households balance the allocation of land endowment and labor under the constraints of land tenure security/insecurity; that is, how much land endowment to allocate in the agricultural sector and how much labor to allocate in the non-agricultural and agricultural sectors. Therefore, the extreme cases, $\mathsf{\lambda }=1$ and $\mathsf{\lambda }=0$, provide a simple and intuitive illustration.

Scenario 1: λ = 1.

This case involves secured land tenure with well-defined, individualized land allocation and legislative protection. Equations (3)–(9) lead to the following equation:

$${\displaystyle \frac{dl}{d\overline{A}}}{|}_{\lambda =1}={\displaystyle \frac{w{g}_{\overline{A}}\frac{1-l}{l}}{p{f}_l+p{f}_{\overline{A}}\frac{1}{{\left(1-l\right)}^2}\overline{A}-w{g}_l}}>0$$

(10)

The denominator of Equation (10) is the difference in marginal incomes between farming and employment in urban areas. According to China’s National Bureau of Statistics, the average annual income per migrant worker was CNY 57,360, while the average annual income per farmer was CNY 21,691 in 2023 [2,38]. This significant difference implies a negative sign for the denominator. Because the sign of the numerator is negative, the overall sign of the deviation of migrant labor over land endowment under the expectations of secured land tenure is positive. That is, the increase in land endowment inducing migration allows us to state the first hypothesis in terms of the relationship between land endowment and migration decisions.

Hypothesis 1:

In a secured land tenure setting, a large land endowment induces migration from rural areas to urban areas.

Scenario 2: λ = 0.

This case indicates a “public” and unprotected land tenure system. From Equations (3)–(9), we obtain the equation below:

$${\displaystyle \frac{dl}{d\overline{A}}}{|}_{\lambda =0}={\displaystyle \frac{-p{f}_{\overline{A}}\frac{l}{1-l}}{p{f}_l-w{g}_l-w{g}_{\overline{A}}\frac{1}{{\left(1-l\right)}^2}\overline{A}}}<0$$

(11)

where the sign of the numerator on the right side of Equation (11) is negative, as mentioned in Section 2.2. The sign of the denominator is not easily determined. It contains the following: (1) $p{f}_l$, the marginal income from farming; (2) $w{g}_{\overline{A}}{\displaystyle \frac{1}{{\left(1-l\right)}^2}}\overline{A},$ the opportunity cost from the loss of land tenure because of migration to urban areas; (3) $w{g}_l$, the marginal income from migration. The China Migrants Dynamic Survey indicates that only 11.8% of migrants are willing to convert a rural hukou to an urban hukou if they are required to return the land they are allowed to farm. It appears that migrants place more value on the sum of income obtained from farming, $p{f}_l$, and the opportunity cost of the loss of land tenure because of migration, $w{g}_{\overline{A}}{\displaystyle \frac{1}{{\left(1-l\right)}^2}}\overline{A}$, than the benefits package associated with the urban hukou and the income from migration ($w{g}_l$). Thus, one can infer that the overall sign of Equation (11) is negative, leading to the second hypothesis:

Hypothesis 2:

If land tenure is unsecured, the greater land endowment, the more likely it is to hinder migration from rural to urban areas.

3. Data

3.1. Questionnaire Survey

The data used in the current paper were collected via a structured questionnaire survey in Nanjing, Jiangsu Province, in 2011. Nanjing is the capital city of Jiangsu Province and located in the Yangtze River Delta economic zone, which is one of the most developed regions in China. According to the Nanjing Statistics Bureau, Nanjing hosted more than 1.78 million migrant workers from different areas of the country by the end of 2013. Several locations frequented by migrants were identified in Nanjing, such as the Andemen job seeker market, Shenyang Village, dining halls in universities and restaurants where most of the servers were migrants, and dormitories at construction sites. To capture representative migrant perspectives, the survey team targeted these migrant locations. Survey data were collected during weekends and holidays from September to November 2011, employing a convenience sampling method. The enumerators randomly approached individuals at the listed locations and, following a customary greeting, asked screening questions. Upon agreeing, the respondent was presented with a consent form and asked to read it and sign it or withdraw from the survey. The interview took about 20 min, and the participants were rewarded with a token gift such as a towel, toothpaste, or a bar of soap upon the completion of the interview.

The five-part questionnaire included closed and partly open questions. The first part included questions about household labor distribution and respondents’ future migration plans. The second set of questions asked the respondents to share information about their household and individual characteristics, such as income, gender, age, and education. The third part of the questionnaire involved questions pertaining to land, such as land endowment, the land rental market, land contracts, and attitudes towards land adjustment. Finally, the fifth part probed the respondents about residing in urban areas as migrants, covering subjects such as labor contracts, job training, and integration into city life. This study used fully completed questionnaires obtained from 751 individuals.

3.2. Variable Explanation and Descriptive Analysis

Table 1. Selected descriptive statistics of variables used in estimation.

Variable Name	Definition	Mean	Std. Dev.	Min	Max
Migration variables
Share	Share of migrants in household	0.55	0.24	0.17	1
Plan	1 = stay in cities; 0 = return to villages	0.47	0.50	0	1
Land endowment
Acre	Land endowment in mu	5.39	3.99	0.6	23
Acreind	Land endowment per household laborer	1.96	1.67	0.23	10
Parcel	Land parcel number	3.94	2.81	1	15
Land tenure variables
Landcont	1 = land contract with collective	0.65	0.48	0	1
Landadjustatt	Attitude towards land adjustment a	2.29	1.32	1	5
Landprivate	Land ownership assessment b	2.09	1.40	1	4
Landrent	1 = land rental market	0.18	0.38	0	1
Control variables
Individual-level
Male	1 = male	0.82	0.38	0	1
Educ	Respondent education level c	2.72	0.85	1	5
Age	Age in years	42.66	9.86	16	72
Incomediff	Annual income difference between city and farming job(s), in 1000 CNY	34.97	20.74	7.2	138
Integration	1 = integrated into cities	0.67	0.47	0	1
Household-level
Hhedu	Household-head education c	2.67	0.83	1	5
Hhsizenl	Number of non-working household members	3.04	0.95	1	6
Hhincomediff	Household annual income difference between city and farming job(s), in 1000 CNY	54.15	40.36	−15	237.6
Instrumental variable
Moneygift	Monetary gift-giving in 1000 CNY	4.15	3.82	0	36

Note: N = 736. a 1 = strongly agree; 2 = agree; 3 = neither agree nor disagree; 4 = disagree; 5 = strongly disagree. b 1 = state ownership; 2 = city-owned; 3 = village-owned; 4 = personally owned. c 1 = no formal education; 2 = elementary school; 3 = middle school; 4 = high school; 5 = college or postgraduate education.

shows that 55% of the working individuals in households had jobs as migrant workers in urban areas. Among the migrants with jobs, 47% of the respondents wanted to stay there for the foreseeable future. In contrast to other studies, where the migration decision was investigated regarding the intent to migrate from rural areas to urban areas, this study asked respondents whether they wanted to stay in the city or return to their village.

The average area of land endowment was (Acre—see Table 1) 5.39 mu (0.88 acres) per surveyed household, and the land endowment per working household member (Acreind) was 1.96 mu (0.32 acres). To explain migration decisions, a variable that related a household’s land area to the number of laborers in the household was more appropriate than one that related land area to the number of household members. Both figures correspond to the national agricultural census results. The survey results showed that agricultural land is very fragmented in the households where the migrants come from, with an average household forming four parcels (Parcel). Each parcel averaged about 1 mu.

The survey included land variables that distinguished secured and unsecured land tenures. Instead of using a single proxy measuring land tenure setting, such as land readjustment risk [39], this study measured farmers’ understanding, perceptions, and expectations of land tenure security using four indicators. The first indicator was the land contractual relationship between farmers and their village collective (Landcont). Farmers in China are not the landowners of their contracted land. Instead, the village collective owns the land and signs land-use contracts with every household to authorize the cultivation, management, lease, and revenues from land use. While most farmers signed land-use contracts with their collectives, some of the signed contracts were lost or damaged over time. Table 1 shows that more than 65% of the respondents had signed land-use contracts.

Another indicator of land tenure security perception is the so-called adjustment. Land distribution requires periodic readjustment as the populations in villages change over time [40]. Generally, land adjustment has been initiated by villagers and implemented by the collective. Because land adjustment affected farmer expectations of land tenure security, the Rural Contracted Land Law in 2002 restricted land adjustment, although small-scale land adjustment was allowed to continue. The current study uses the land adjustment index (Landadjustatt) to measure the farmer’s attitude towards land adjustment. Specifically, the respondents indicated their own attitudes along a five-step scale, where 1 stood for “strongly agreeing” with land adjustment, 5 referred to “strong disagreement”, and 3 indicated a “neutral” view. Generally, the respondents agreed with land adjustment as the average value is 2.29.

The farmers’ privatization assessments of land ownership was another indicator. Landprivate is a categorical variable that ranges from state ownership (coded as 1) to private ownership (coded as 4). The average value was 2.09 (Table 1). More than three out of five (61.28%) respondents indicated state ownership of the land, and 243 (33.02%) respondents believed that the land they cultivated was private. Village community land ownership differs from the ownership of urban land, which is deemed to be state-owned. Regarding the existence of a land rental market in their villages (Landrent), only 18% of respondents answered in the affirmative.

The control variables relevant to individual migration decisions included demographic characteristics, income, migrants’ opinion about integration into urban life, and willingness to abandon rural land-use rights. Among those variables, 82% of respondents were male (Male), the average respondent was 42 years old (Age), and middle school was the average education attainment level (Educ) (Table 1). On average, the annual income earned as a migrant worker exceeded that obtained from farming by CNY 34,970. The annual income differences (Incomediff) were top-coded to eliminate outliers that would exert a disproportionate influence on the estimation results. The lower and upper bounds were set at the 1st percentile and 99th percentile of the original data to reduce the influence of outliers. About 67% of respondents declared that they had integrated into urban life (Integration).

The household-level control variables characterizing the household included the education level of the household head (Hhedu), the number of non-working household members (Hhsizenl), and the annual household income from urban wages and farming (Hhincomediff). On average, household heads had a middle school education and there were three non-working members per household. The average difference in annual household income earned during migration exceeded the income from farming by CNY 54,150. The household annual income difference was top-coded, and the lower and upper bounds were set at the 1st percentile and the 99th percentile of the original data.

4. Methodology

4.1. Classification and Latent Class Model

Latent Class Analysis identifies the underlying data structure to assign each observation to a particular cluster [41,42] and has been widely used to characterize heterogeneity among groups, such as farmers and consumers [43,44,45]. Compared with traditional cluster models, such as those using factor analysis, the structural equation model, or the random effect model, the latent class cluster model is superior in dealing with categorical latent variables. A key advantage of the latent class cluster model lies in its flexibility to accommodate complex distributions and mixed data types (i.e., continuous, categorical, or a combination of the two), enabling the nuanced segmentation of heterogeneous groups. Generally, basic latent class cluster models are structured to estimate the probability of an observation belonging to a specific latent class, conditional on its observed indicators, thereby capturing underlying subgroup structures that may not be apparent through manifest variables alone. The models have the following form:

$$f\left(y_i|\theta \right)=\sum _{k=1}^K{\pi }_k{f}_k\left(y_i|{\theta }_k\right)$$

(12)

where $y_i$ refers to the scores on a set of indicator variables and the θs are the model parameters. The term $f\left(y_i|\theta \right)$ represents the distribution of $y_i$, which is assumed to be a mixture of cluster-specific densities; $f_k\left(y_i|{\theta }_k\right)$ K is the number of clusters; and ${\pi }_k$ denotes the prior probability of belonging to latent cluster k. A maximum-likelihood estimation is used to estimate the parameters.

The criteria applied to select the best-fitting model included the Bayesian Information Criterion (BIC), the number of parameters in the estimation, and a p-value associated with the Chi-squared statistics. The current study used Latent GOLD 4.5 software because it includes model selection criteria and probability-based classification.

This study classified migrant workers according to their perceptions of the security of their land tenure. Four land-tenure-related measures were coded, including assessments of land ownership (Landprivate), attitudes towards land adjustment (Landadjustatt), land contract presence (Landcont), and the market for land renting (Landrent). Hypothetically, as the land tenure system evolves, households perceive more individualized land tenure and refuse land adjustment to avoid the risk of the loss of the land allocated to them. Additionally, having a land contract with the collective is a measure of legislative protection, and the existence of a local land rental market implies well-established land tenure.

Table 2. Model summary from four cluster models.

Model	Number of Clusters	Npar	BIC (LL)	L2	p-Value
Model 1	1 Cluster	9	4950.61	132.44	9.6× 10−6
Model 2	2 Clusters	14	4935.45	84.28	0.054
Model 3	3 Clusters	19	4948.17	63.99	0.34
Model 4	4 Clusters	24	4971.27	54.08	0.51

Notes: N = 736. Npar is the number of parameters in the estimation, BIC is the criteria from the Log-likelihood statistics, and L² and p-value are criteria from the Chi-squared statistics.

shows four different cluster models in the latent class analysis, with cluster numbers ranging from 1 to 4. The index Npar indicates the number of parameters in the estimation. The BIC and Chi-squared statistics (L²) were used to identify the best fitting model. The L² statistical significance is indicated by the p-values. The model selection was guided by p-values higher than 0.05, indicating an adequate fit, and the fewest number of parameters. Model 2, with two clusters (cluster 1 and cluster 2), outperformed the other models.

Table 3. Land-tenure-related variables by group, with secured and unsecured land tenure expectations and cluster classification.

Variables	Cluster 1—No of Obs.	Cluster 2—No of Obs.	Cluster 1	Cluster 2	p-Value
Landadjustatt	416	320	3.27 (0.90)	1.00 (0.00)	0.00
Landprivate	416	320	2.09 (1.40)	2.08 (1.41)	0.94
Landcont	416	320	0.67 (0.47)	0.62 (0.49)	0.21
Landrent	416	320	0.26 (0.44)	0.08 (0.26)	0.00
Landadjust	416	320	0.69 (0.46)	0.72 (0.45)	0.46
Landlaw	416	320	0.15 (0.36)	0.09 (0.29)	0.02

Notes: Numbers under cluster 1, col. 4, and cluster 2, col. 5, are indicator means. Standard deviations are in parentheses.

shows the number of observations, means of land-tenure-related variables, and t-tests for each cluster. There were 416 observations (57%) in cluster 1 and 320 observations (43%) in cluster 2. The means of most land-tenure-related variables calculated for cluster 1 were larger than the means of cluster 2, and three of the six were statistically significant. Specifically, the respondents in cluster 1 tended to oppose land adjustment, indicated the existence of a land rental market in their village, and were more likely to have a land-assigning certificate and knowledge of land law than those in cluster 2. Thus, hereafter, cluster 1 is renamed as the secured group and cluster 2 is the unsecured group.

A potential identification issue could arise, because farmers who have secured tenure expectations may have stronger motivations to migrate than those who do not. Should this be true, the causality between the land endowment and household labor allocation will be biased.

Table 4. Results of t-test on differences between means for migrant workers and their land tenure expectations.

Variables	Secured Group	Unsecured Group	p-Value
Migration variables
Share	0.55 (0.24)	0.54 (0.24)	0.65
Plan	0.51 (0.50)	0.41 (0.49)	0.01 **
Control variables
Individual-level
Acreind	2.04 (1.73)	1.84 (1.59)	0.11
Parcel	4.02 (2.91)	3.83 (2.69)	0.37
Male	0.83 (0.38)	0.82 (0.39)	0.69
Educ	2.77 (0.86)	2.66 (0.85)	0.09
Age	42.31 (10.09)	43.11 (9.53)	0.28
Incomediff	34.74 (21.19)	35.27 (20.18)	0.73
Integration	0.67 (0.47)	0.68 (0.47)	0.66
Household-level
Acre	5.53 (4.04)	5.21 (3.92)	0.28
Hhedu	2.68 (0.83)	2.66 (0.82)	0.83
Hhsizenl	3.01 (0.98)	3.08 (0.90)	0.37
Hhincomediff	53.25 (40.81)	55.32 (39.81)	0.49

Note: Standard deviations in parentheses. *** p < 0.01; ** p < 0.05; * p < 0.1.

shows the comparison and t-test between the two groups with different land tenure expectations regarding the dependent and explanatory variables. None of the covariates were statistically significant between the two groups, suggesting that land tenure security was an exogenous factor for both groups in household labor allocation. However, individual migration decisions statistically differed between the two groups. Therefore, the models were specified differently to address the endogeneity in land tenure in their estimations.

4.2. Empirical Models and Identification

The empirical modeling explored how land endowment affected household labor distribution as well as individual migration decisions, taking the heterogeneity of tenure expectations into account. The household labor distribution model was specified by a fractional probit function:

$$E\left(y_{ik}|\mathit{x}\right)=\mathsf{\Phi }\left(\mathit{x}\mathit{\phi }\right),$$

(13)

where$y_{ik}$ is the share of household labor that navigates to urban jobs divided by the total labor force in household i in group k and characterized by land tenure perception. Given that the sample consisted of migrant workers, the ratio of off-farm laborers in households had the range of ${0<y}_{ik}\le 1$. The vector of the explanatory variables, x, includes land endowment and the number of land parcels in household i and (secured or unsecured group) group k, and characteristics such as household head education, the number of non-working household members, and annual household income from urban jobs and farming. φ is the parameter to be estimated. The reasons for choosing a fractional probit function over the alternatives—such as a two-limited Tobit model, a log-odds transformation, or a Beta distribution—were motivated by two key considerations: the absence of zero observations and our analytical focus on estimating the conditional mean effects. It is well suited to modeling continuous outcome variables bounded between 0 and 1 when zeros are absent and mean-based interpretations are prioritized.

Given that the mechanism of individuals’ time allocation may be different from that of household labor distribution, a separate model examined how land endowment affects an individual’s decision to migrate to work urban job. The model for individual decisions is as follows:

$$E\left(d_{jk}|\mathit{z}\right)={\displaystyle \frac{exp\left(\mathit{z}\mathit{\omega }\right)}{\left[1+exp\left(\mathit{z}\mathit{\omega }\right)\right]}},$$

(14)

where $d_{jk}$ is the decision to migrate for individual j in group k. The explanatory variable (z) in Equation (14) includes a vector of personal characteristics, such as gender, age, education, integration in migrant cities, and household-level variables. The land tenure security perception indicator is an explanatory dummy variable, and ω is the parameter to be estimated.

A potential endogeneity arises in estimating the effect of the income difference between farming and urban jobs on the share of household members employed outside agriculture and individual decisions to migrate. Instead of the income difference changing the share of household members with urban jobs or individual migration decisions, it is possible that greater shares of laborers with urban jobs or positive migration decisions cause income differences to be higher. Ignoring this issue could result in inconsistent estimators of all parameters, φ and ω.

The application of instrumental variables (IVs) is widely recommended to assure consistent estimators. However, obtaining a valid IV is quite difficult and a weak IV will lead to considerable efficiency loss compared to OLS. This study selects monetary gift-giving as an IV, which is highly correlated with income difference but does not affect household labor allocation. China has a long tradition of money gift-giving for interpersonal connectedness, such as at weddings, funerals, and specialized packets for holidays and other big social events or occasions [46,47]. The sample data show that over 97% of respondents gave or received a monetary gift in the previous year. A valid IV must be relevant to the endogenous variable, income difference, at either the household or individual level. The data show that the gross correlation between Moneygift and Hhincomediff is 0.30 (p-value = 0.00) and the correlation between Moneygift and Incomediff is 0.19 (p-value = 0.00). The strict testing for endogeneity is based on the following equation:

$$y_1=f\left(y_2,X,v\right),$$

(15)

where $y_1$ is the dependent variable, i.e., the share of off-farm laborers or individual migration decisions, $y_2$ is the income difference at either the household or individual level, X is a vector of the exogeneous covariates, and v is the fitted residual vector from the first-stage OLS regression of income differences on exogeneous covariates plus the IV, Moneygift. A test of exogeneity would be a test of whether the coefficient of v, defined as $\rho$, is zero. If the null hypothesis of exogeneity is rejected, then the income difference is endogenous.

Table A1. First-stage estimation for IV procedure for household labor allocation.

VARIABLES	HHINCOMEDIFF	SHARE
ACRE	−1.500 ***	0.034 **
	(0.327)	(0.014)
ACRESQ		−0.002 ***
		(0.001)
PARCEL	0.607	−0.011
	(0.619)	(0.008)
HHEDU	2.927 *	−0.046 *
	(1.685)	(0.026)
HHSIZENL	0.218	−0.315 ***
	(1.930)	(0.021)
HHINCOMEDIFF		0.005 ***
		(0.002)
$\widehat{v}$		−0.001
		(0.002)
MONEYGIFT	3.173 ***
	(0.445)
Constant	38.704 ***	0.110
	(5.189)	(0.123)
Observations	736	736
R2/Pseudo R2	0.114	0.045

Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

and

Table A2. Testing for income difference endogeneity in individuals’ migration decisions.

VARIABLES	INCOMEDIFF	PLAN
ACREIND	−1.096 ***	0.067
	(0.399)	(0.130)
ACREINDSQ		−0.005
		(0.014)
PARCEL	0.016	−0.035
	(0.264)	(0.030)
LANDTENURE	0.007	0.408 ***
	(1.496)	(0.157)
MALE	6.360 ***	−0.567 **
	(1.976)	(0.259)
AGE	−0.186 **	−0.026 ***
	(0.077)	(0.009)
EDUC	0.636	0.306 *
	(1.194)	(0.174)
INTEGRATION	3.083 **	0.748 ***
	(1.544)	(0.179)
HHEDU	0.854	−0.274
	(1.205)	(0.171)
HHSIZENL	2.016 *	−0.088
	(1.098)	(0.104)
INCOMEDIFF		0.029
		(0.020)
$\widehat{v}$		−0.024
		(0.020)
MONEYGIFT	0.975 ***
	(0.281)
Constant	28.230 ***	−0.282
	(5.544)	(0.795)
Observations	736	736
R2/Pseudo R2	0.081	0.063

Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

in Appendix A show that the coefficients of Moneygift in both estimations are statistically significant, which validates the relevance of IV. However, the coefficient of the fitted residual, $\widehat{v}$, is not statistically significant. Thus, in the case of the two used data subsamples, the income differences, Hhincomediff and Incomediff, are exogenous in estimating household labor allocation or individual migration decisions, respectively.

4.3. The Test for a U-Shaped Relationship

This study follows the framework for a test for a U-shaped relationship [16] between labor allocation and land endowment in a Chinese farm household. The most common model specification in testing a U-shaped (or an inverse U-shaped) relationship the inclusion of a quadratic term in the standard regression model:

$$y_i=\mathsf{\alpha }+\mathsf{\beta }{w}_i+\mathsf{\gamma }{w}_i^2+{\delta }^{\prime }{C}_i+{\epsilon }_i,$$

(16)

where $w_i$ is the key explanatory variable, $C_i$ is a vector of the control variables, and ${\epsilon }_i$ is the normally distributed error term. Rather than focusing on the sign and significance of β and γ to determine the U-shaped relationship, the current study applies Lind and Mehlum’s method [18], which provides an enhanced test of the relationship. Specifically, if Equation (16) has only one extreme point (either a peak or bottom point), the requirement for a U shape is a negative slope at the starting point ($w_l$) and a positive slope at the endpoint ($w_h$). The start- and endpoints can equal the minimum and maximum values of the data range. A U shape is implied by the condition that $\beta +2\gamma x_l<0<\beta +2\gamma x_h$. To test the condition, the following composite hypothesis is tested:

$$H_0:\beta +2\gamma w_l\ge 0 and/or \beta +2\gamma w_h\le 0$$

$$H_a:\beta +2\gamma w_l<0 and \beta +2\gamma w_h>0.$$

The null hypothesis is rejected if either one or both of the conditions are rejected. However, given the combination of the null hypotheses, the standard test methods are no longer suitable. Sasabuchi [48] provided a framework to construct the rejection area as follows:

$${\mathrm{R}}_{\mathsf{\alpha }}=\left\{\left(\mathsf{\beta },\mathsf{\gamma }\right):{\displaystyle \frac{\widehat{\beta }+2w_l\widehat{\gamma }}{\sqrt{s_{11}+2w_{l}2s_{12}+2w_l{s}_{22}}}}<-t_{\mathsf{\alpha }},\mathrm{a}\mathrm{n}\mathrm{d}{\displaystyle \frac{\widehat{\beta }+2w_h\widehat{\gamma }}{\sqrt{s_{11}+2w_{h}2s_{12}+2w_h{s}_{22}}}}>t_{\alpha }\right\},$$

(17)

where $\widehat{\beta }$ and $\widehat{\gamma }$ are the estimated parameters. The estimated variance and covariance between the $\widehat{\beta }$ and $\widehat{\gamma }$ parameters are $s_{11}$, $s_{22}$, and $s_{12}$, and α is the confidence level. Meanwhile, the estimated extreme point can be calculated as ${\widehat{w}}_{min}=-{\displaystyle \frac{\widehat{\beta }}{2\widehat{\gamma }}}$ and the confidence interval for the extreme point can be constructed using the Fieller method [49].

5. Results and Discussion

5.1. Household Labor Allocation

Table 5. Estimation results of land endowment’s influence on migration for two model specifications using overall sample and subsamples for secured and unsecured tenure perceptions.

Variable	Full Sample	Secure Group	Unsecure Group	Full Sample	Secure Group	Unsecure Group
Acre	0.014	0.026	−0.005	0.033 **	0.030	0.038 *
	(0.015)	(0.020)	(0.023)	(0.014)	(0.019)	(0.021)
Acresq	−0.001 *	−0.002	−0.001	−0.002 ***	−0.001	−0.002 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Parcel	-	-	-	−0.011	−0.009	−0.009
				(0.007)	(0.010)	(0.011)
Hhedu	-	-	-	−0.043 *	−0.024	−0.065 *
				(0.025)	(0.033)	(0.038)
Hhsizenl	-	-	-	−0.314 ***	−0.309 ***	−0.313 ***
				(0.021)	(0.029)	(0.032)
Hhincomediff	-	-	-	0.004 ***	0.004 ***	0.006 ***
				(0.001)	(0.001)	(0.001)
Constant	0.090	0.054	0.151 *	0.148	0.141	0.114
	(0.055)	(0.073)	(0.084)	(0.092)	(0.126)	(0.134)
Observations	736	416	320	736	416	320
Pseudo R2	0.001	0.001	0.002	0.045	0.038	0.056

Note: Robust standard errors in parentheses. *** p < 0.01; ** p < 0.05; * p < 0.1.

shows the estimation results of two specifications for household labor distribution with and without control variables, i.e., land and household characteristics. Each specification is estimated using the whole sample and the two subsamples with different expectations of land tenure security. The results based on the whole sample represent homogeneity in terms of expectations of tenure security. The estimations without control variables indicate that the estimated models marginally fit the data, as indicated by the low values of the pseudo-R²s. Including control variables increases the goodness-of-fit markedly. Thus, the following discussion focuses on the model with control variables.

In the model with homogeneous expectations of land tenure security, the relationship between land endowment and the share of migrant labor in a household follows an inverse U-shaped trajectory. Initially, the households allocate additional labor to migration in urban areas as their land endowment increases, driven by the high wages from urban employment. However, when the land endowment exceeds a certain threshold, any additional increase in land endowment decreases the amount of labor allocated to jobs in the city, reducing migration. This phenomenon is explained by the complementarity between land and labor input in farming, especially within the family farm system, where larger land endowments require more labor input.

However, land endowment does not significantly affect the share of migrating household members for households with secure expectations of land tenure. In contrast, land endowment is significantly negatively associated with the share of off-farm labor in households with unsecure expectations of land tenure. The results reveal that land endowment hinders rural labor outmigration if land tenure is poorly protected. These findings are consistent with the reports that unsecured land tenure increases the need for “guard labor” to claim land access [20,21].

Table 6. The test results for a U-shaped relationship.

Item	Overall	Secure Tenure	Unsecure Tenure
ACRE	0.033 ** (0.014)	0.030 (0.019)	0.038 * (0.021)
ACRESQ	−0.002 *** (0.001)	−0.001 (0.001)	−0.002 *** (0.001)
Slope at lower bound	0.025 **	0.024 *	0.030 *
Slope at upper bound	−0.048 ***	−0.035 *	−0.069 ***
Appropriate for an inverse U shape (t-value)	1.91 **	1.35 *	1.53 *
Extremum point (mu)	8.191	9.695	7.332
90% confidence interval, Fieller method	[3.332, 10.547]	[-, -]	[-, -]

Notes: The numbers of coefficients and robust standard errors in parentheses come from the estimation in Table 5. The Lind and Mehlum (2010) test provides the slope, test, extremum point, and confidence interval [16]. *** p < 0.01; ** p < 0.05; * p < 0.1.

shows the results of a formal test of the U-shaped relationship. The results based on the whole sample show a positive slope at the lower bound and a negative slope at the upper bound. The t-value for the test of an inverse U-shaped relationship is 1.91 and suggests the rejection of the null hypothesis of a U shape. The extremum point for the overall sample is 8.191 mu, and the 90% confidence interval is about 3 mu to 10 mu. The group with unsecured land tenure expectations shows a similar pattern of an inverse U-shaped relationship to the one identified for the overall sample (although the relationship at the lower bound is weak). The extreme point for the group with unsecured land tenures is 7.32 mu. The lower and upper bounds are set as the minimum and maximum values in the data range. However, the conditions of a positive slope at the lower bound and a negative slope at the upper bound still hold when the lower and upper bounds change within a narrow range. The inverse U-shaped relationship is therefore empirically confirmed.

Given their inverse U-shaped relationships, Figure 2 shows the plotted predictions for land endowment and the share of migrating household members for both the overall sample and the unsecure tenure group. Both lines display clear inverse U-shaped relationships (Figure 2). That is, households tend to allocate about half of their labor to jobs in urban areas when the land area they operate is relatively small. The motivation to migrate is driven by the income difference between urban and rural areas. With the increase in land endowment, farm households tend to allocate more labor to jobs in urban areas, and when the land endowment is abundant, households are more likely to retain labor for farming. The group with unsecured land tenure expectations has a similar pattern of an inverse U-shaped relationship. However, the group with unsecured land tenures is associated with a curve that has a much steeper slope on the right side (Figure 2), which reinforces the negative relationship between land endowment and the share of household members with off-farm urban jobs.

Additionally, the household income difference between migrant employment in urban areas and farming has a significant positive effect on migration to urban areas in both the full sample and the subgroup with unsecure land tenure expectations. This finding is consistent with the Todaro model, which argues that income disparity between urban and rural areas drives migration [23]. Conversely, the presence of non-working household members—specifically children and elderly dependents—exerts a significant negative influence on migration likelihood. The number of non-working members in a household prevents migration. Children and the elderly require care, which could limit labor mobility.

5.2. Individual Migration Decisions

This section examines how land endowment affects individuals’ decision to migrate to urban areas. The dependent variable is the choice between remaining in an urban area or returning to a farm.

Table 7. Estimation results of land endowment effects on individuals.

Variables	PLAN	PLAN
Acreind	0.016	0.041
	(0.118)	(0.128)
Acreindsq	−0.006	−0.005
	(0.014)	(0.014)
Parcel	-	−0.034
	-	(0.030)
Landtenure	-	0.394 **
		(0.156)
Incomediff	-	0.005
		(0.004)
Male	-	−0.408 *
		(0.223)
Age		−0.030 ***
		(0.009)
Educ	-	0.323 *
		(0.174)
Integration	-	0.823 ***
		(0.170)
Hhedu		−0.246
		(0.170)
Hhsizenl		−0.034
		(0.094)
Constant	−0.121	0.400
	(0.167)	(0.575)
Pseudo R2	0.001	0.062

Notes: n = 736. Robust standard errors in parentheses. *** p < 0.01; ** p < 0.05; * p < 0.1.

shows the results of two relationships, estimated with and without control variables. The discussion focuses on the results of the former specification.

Land endowment and its squared term do not significantly influence individuals’ decisions to migrate under specifications with or without control variables. This finding confirms that distinct mechanisms underpin individual and household migration decisions. Migrants perceiving their land tenure as secure are more likely to stay in urban areas. The income difference between urban jobs and farming positively affects migrant workers’ decisions, but the measure is statistically insignificant. Notably, gender disparities emerge: female migrants are more likely to remain in urban areas as opposed to males (Table 7). Since there are fewer women among the rural migrants employed in urban areas, this effect is differently experienced by those remaining on farms. The differences across gender for migration will benefit from additional future research regarding cultural aspects. Age is strongly negatively associated with individual migration decisions, whereas education helps migrant workers to compete in the urban labor market and increases their likelihood of remaining in urban areas. These findings are further supported by the statistically significant variable indicating integration into city life. The higher the level of integration into urban life, the higher the likelihood that migrant workers choose to stay in urban areas.

5.3. The Land Rental Market

The assumption of the above model is that the land rental market is not active. Although land transfer in rural areas across the country has begun to take shape, land transfer is not strongly market-dominated and, in practice, is still dominated by the government; at the same time, due to the high degree of government financial bondage [50], land transfers from small farmers to other operators tend to be relatively limited, which is at variance with national statistics. We also add that, in China, farmland ownership is retained by rural collectives, and only the management right can be circulated as an institutional arrangement [51,52]; meanwhile, while studies have explored the impact of farmland-use right transfer on collective action on public land from an irrigation perspective [53], the unique perspective of this study better reflects the relevant themes. Moreover, many studies have confirmed the enhancing effect of land tenure security on land transactions [54]. On the contrary, unsecured land tenure reduces the activity of the land rental market. Therefore, the inactive land market and unsafe land rights discussed in this paper are meaningful.

6. Conclusions

Land endowment, labor migration, and their relationship are key issues in many developing countries. Understanding those relationship helps policymakers formulate effective policies to stimulate migration processes and, eventually, the permanent urbanization of rural jobseekers. The previous studies on the effects of land endowment on labor allocation provide checkered support for their U-shaped relationship. This study contributes to the literature by examining the relationship between land endowment, labor migration, and land tenure perceptions among rural–urban migrants in China.

The current paper recognizes the heterogeneity of land tenure security perceptions among rural migrants employed in urban areas as the key motivating factor in decisions to undergo rural–urban migration. By classifying migrant workers into secured land tenure and unsecured land tenure groups, we examine how land endowment affects two groups of migrants with differing perceptions of land tenure security, revealing subtle institutional influences on migration behavior. The chosen empirical approach controls directly for the perceptions of land ownership and administrative land adjustment schemes and identifies how the amount of acreage affects household labor distribution as well as individuals’ decisions to migrate to urban areas in search of jobs. Though rural land is constitutionally defined as being owned by village collectives in China, rural migrants exhibit heterogeneous perceptions of their land tenures, reflecting strong attachments to the land.

The estimation results confirm an inverse U-shaped relationship between land endowment and household labor distribution. Moreover, this study distinguishes between the effects for households who perceive the security of their land tenures differently. Specifically, the operated land area only hinders labor allocation primarily among rural households who perceive their land tenure to be unsecured, where it displays an inverse U-shaped relationship. Conversely, land endowment does not constrain labor allocation if the household perceives their land tenure to be secured. Notably, there are limits to the land endowment that can be allocated to farm households by village collectives in China. Within the collective framework, increases in land endowment to individual farms induce outmigration due to perceptions of land tenure security. The study further distinguishes between household-level and individual migration decisions. The individual decision to migrate to urban areas to seek a job is more likely if land tenure is perceived as secure. Additionally, another finding from this study is that the level of integration into urban life encourages migrant workers to remain in the cities, underscoring the dual role of land tenure and urban adaptation in shaping migration outcomes. Collectively, these methodologies not only serve to ameliorate the limitations inherent in earlier research, but also enhance the empirical rigor of labor allocation dynamics analysis, thereby advancing the field’s capacity for reliable inquiry.

Securing land tenure should be prioritized in rural policy design if more rural migrants are needed to fill jobs in urban areas. Rural labor outmigration, encouraged by secure land tenure perceptions, may stimulate the development of local land renting markets. The land titling program, conducted at the end of 2018 in response to widespread perceptions of unsecured land tenure, may have to be augmented by a program of compensation for titled land. These measures would strengthen tenure security, reducing rural households’ reliance on land as a sole safety net and encouraging labor mobility. Additionally, the expansion of social welfare benefits accessible to rural migrants in urban areas could encourage them to remain in cities permanently. As shown during the recent financial crisis and COVID-19 pandemic, rural migrants are most likely to lose urban jobs, forcing them to return to farms. Such unforeseen events prompt the return of millions of people back to rural areas, potentially slowing agricultural productivity growth and economic development in rural and urban areas. To mitigate these risks, policy framework design must clarify whether land donations serve as a transitional buffer or a permanent fallback. Land endowment is viewed by rural migrants and their households as a safety net unless the urban hukou sustains urban residency. Therefore, the ongoing second-round contracting renewal policy, as well as rural revitalization initiatives, should secure land tenure for migrant workers. In this context, the rural social security system reform and the development of the land leasing market are essential to achieve reasonable labor mobility and sustainable rural development. These recommendations align with current policy trajectories, such as land tenure confirmation and urban–rural integration initiatives, while offering measures to address systemic barriers to equitable migration outcomes.

While this study enriches our understanding of land tenure’s role in migration dynamics, its limitations are its focus on a specific institutional context and reliance on cross-sectional data. Future research could extend this inquiry to diverse institutional settings to test the robustness of the observed relationships. Also, longitudinal research or panel data would offer deeper insights into how evolving tenure security influences migration decisions.