*3.1. Property Rights Exclusion Effect, Land Production Function, and Land "Stickiness"*

The Land Certificated Program has a property rights exclusion effect. The property rights exclusion effect is the exclusivity of contractual rights in farmland. The LCP measured each plot with detailed information, including location, size, and boundaries. The disputes with unclear land information on plot size and boundaries were addressed. Consequently, the LCP was able to guarantee the exclusivity of farmland property rights by clarifying the four areas of farmland and specifying property rights ownership [34]. Now, it continues to promote the improvement of farmland production functions and the redistribution of agricultural labor.

In the past, when farmland property rights were found to be unclear, they were frequently adjusted. On the one hand, farmers could not obtain exclusionary rights through legal empowerment under this system. On the other hand, the exclusionary rights granted by village rules and regulations were not mandatory; therefore, the exclusivity of farmland property rights was weakened [35]. Therefore, in order to avoid the possible risk of land loss, farmers had to bear certain exclusion costs to maintain their contractual rights while going out to work [36]. These farmers were often willing to operate farmland at low cost, or even to transfer farmland to acquaintances for free. As a result, the production function of land and the land "stickiness" of rural labor were weakened.

Therefore, the purpose of implementing the LCP was to permanently determine contractual rights to farmland by measuring each plot. It strengthens the exclusivity of contractual rights to farmland, reduces the risk of land loss, and strengthens the production function of land. This in turn affects the labor allocation decisions of rural households. Laborers are now more inclined to engage in agricultural production, increase the degree of agricultural labor participation, and enhance land "stickiness". At the same time, due to historical legacy problems, such as the poorly defined property rights of agricultural land and the operational flexibility of local governments in the process of implementing the LCP, farmers inevitably pay more attention to the productive function of the land. This enables them to preserve their contractual rights by increasing their labor input, increasing their labor participation, and strengthening the "stickiness" of the land. In this way, farmers can obtain the most favorable "standard of certificated land" and maximize their interests.

#### *3.2. Property Rights Incentive Effect, Land Production Function, and Land "Stickiness"*

The Land Certificated Program also has a property rights incentive effect. The property rights incentive effect is the incentive of farmland returns. The LCP formalized and legalized contract rights though the issuance of land certificates. At the same time, it strengthened farmers' residual claims to farmland returns by stabilizing long-term land contract relationships [37]. Hence the farmers can use the tenure security to mortgage land, stimulate the improvement of farmland production functions and reallocate agricultural labor.

Prior to the program, with the instability of farmland property rights, farmers' residual claims to farmland returns were not stable. This meant that once the farmland was reallocated to adjust to population needs, both the medium- and long-term investments

in agriculture attached to the farmland were lost, and household income from agriculture was reduced. This in turn led to the inefficient use of farmland and agricultural production. Therefore, in order to avoid the possible risk of farmland adjustment, farmers' demand for land production function decreases, thus reducing agricultural labor input and weakening land "stickiness". Driven by comparative interests, farmers then choose to engage in non-agricultural production in urban areas in order to relieve the pressure of survival and improve quality of life, further weakening the land production function and land "stickiness".

The LCP provides an institutional guarantee for agricultural operators to stabilize agricultural production and enhance the land production function by ensuring the stability of farmland property rights and guaranteeing farmers' residual claims to farmland returns. It stimulates and enhances farmers' enthusiasm and stability in agricultural production [29], and the land production function is strengthened. This stabilization of residual claims to farmland and higher expected returns from agricultural operations can, to a certain extent, encourage farmers to increase their participation in agricultural labor, thereby increasing the supply of agricultural labor and land "stickiness" of rural labor.

In summary, the LCP has a property rights exclusion effect and an incentive effect. On the one hand, the LCP confirms contractual rights and ensures farmland exclusivity through the land survey. On the other hand, the LCP guarantees the stability of land property rights and ensures farmers' residual claims to farmland returns through the issuance of certificates. This improves the land production function, thus promoting the reallocation of agricultural labor, increasing the degree of agricultural labor participation, and increasing the land "stickiness" of rural labor. The specific influence path is shown in Figure 2.

**Figure 2.** The impact path of the LCP on land "stickiness".

Based on these observations and previous studies, we hypothesize that the LCP has a significantly positive effect on the land "stickiness" of the rural labor, and land production function plays an important role in it. In other words, the LCP exerts a significantly positive effect on land "stickiness" via its influence on land production function.

#### **4. Data, Variables, and Models**

#### *4.1. Data*

The data used in this paper come from the 2016 and 2018 China Labor-force Dynamics Survey (CLDS) data provided by the Social Science Survey Center of Sun Yat-sen University. This survey is conducted once every two years, with the labor force aged 15–64 as the target population. It utilizes a multi-stage, multi-level sampling method proportional to the size of the labor force. Since the CLDS2018 household samples are all new samples, we were unable to compose panel data. Thus, this paper uses two rounds of surveys consisting of mixed cross-sectional data from 2016 and 2018. This paper focuses on studying the impact of the Land Certificated Program (LCP) on rural households' agricultural labor participation. In order to do so, rural households were used as the research subjects, from which mainly household and community-level data were used. The survey data were then processed and screened to obtain a valid sample of 3936 rural households. The sample farm households covered 247 villages in 26 provinces across China, and the regional

distribution of the sample farm households showed that there were 1399, 1217, and 1320 farm households in the eastern, central and western regions, respectively.

The process of sample selection was as follows: (1) construct the variables needed for this study based on the variables already available in the database; e.g., the variable of the number of rural household members was obtained by using the coding count of each member of the farm household; (2) delete the samples of rural households that were non-agricultural, and keep only those samples whose household type was agricultural; (3) delete the samples of rural households that did not possess farmland; (4) delete the missing values of key variables, e.g., delete the missing value of question F6.2a "Has the household received the Certificate of Rural Land Contract Management Rights" and delete the missing value of question F6.4a "The number of people engaged in agricultural production in your household last year".

## *4.2. Variables*

1. Dependent variable: land "stickiness" of rural labor. According to the above analysis, agricultural labor participation is the external expression of land "stickiness". Thus, this paper adopts the agricultural labor participation rates to represent the land "stickiness" of rural labor. This was determined by calculating the ratio of labor participated in farming activities [38]. On average, the ratio of agricultural participation was 44.3% in this sample.

2. Independent variable: Land Certificated Program (LCP). The CLDS2016 and CLDS2018 surveys provide data on the agricultural production of rural households in 2015 and 2017, respectively. In this period, the LCP was being promoted in the form of a province-wide pilot or county-wide pilot, and had not yet even been extended to the whole country in 2015. Although the LCP had been extended across the entire country by 2017, not all rural households had carried out certificated land work simultaneously for the same pilot unit, which provides a good opportunity for this paper to study the policy effects of the LCP.

This paper draws on the studies of Zhou et al. [39] to measure "whether the farm household has received the Certificate of Rural Land Contract Management Rights" in order to determine whether the farmland was certificated. If the answer to this question was "yes", then the value of LCP was "1"; if not, the value was "0". Using this processing method, the number of observations in the sample of households that had received the Certificate of Rural Land Contract Management Rights was 2335, while the number of observations in the sample of households that had not yet received the Certificate of Rural Land Contract Management Rights was 1601. Thus, the overall certification rate of the sample was 59.3%, which is similar to the results of previous studies [17].

3. Control variables. We also controlled for the characteristics of farmers individuals, households, and villages in our analysis.

The individual characteristics were gender, age, age squared, education, political outlook, and health of the household heads. The social and cultural norms were the leading causes of gender inequality [40]. Females' household responsibilities (such as caretaking children, household chores, and so on) limited both their available labor and time to farm [41]. It was therefore expected that males were more inclined to engage in agricultural production. The age of the household head was used as a proxy for the family's farming experience [42]. Therefore, older farmers were expected to be more productive in agriculture. However, as their age increases, the physical ability of farmers who have consistently engaged in agricultural production will naturally be restricted. Hence, the square of age was added to the model to capture the possible non-linearities in its impact [43,44]. The average age of household head is 53.877 years. Household heads with more education were assumed to have more off-farm skills and therefore be more likely to engage in off-farm employment. Zhang et al. [45] found that rural individuals with more education had increasing access to off-farm jobs. In addition, De Rrauw and Rozelle [46] estimated an average return to education of 6.4% in off-farm wage employment

in rural China. Thus, it was expected that the education would have a negative impact on farm employment. The average number of years of education was generally low, at approximately 7 years. Finally, political outlook and individual health played an important role in improving human capital and enhancing the labor market outcomes. Earlier studies found that having a political cadre in the family increased the family members' off-farm employment [47,48]. Shu Lei [44] found that health was an important factor affecting agricultural production. Thus, political outlook and health were also included in the model. On average, the percentage of household heads who were not party members was 8.3%, and 50% of the household heads reported their health status as "healthy".

The household characteristics were total family income, land per capita, labor ratio and the types of farming households. Total family income was determined with respect to household wealth, which may have a negative impact on farm employment, as participation in off-farm employment may require a minimum level of assets [49]. For outcomes measured in terms of monetary value (for example, productivity, value of credit received, and consumption), studies usually reported treatment effects on the scale of the natural logarithm [20]. In order to induce normality in skewed income distribution, this paper treated total family income logarithmically [49]. Land per capita was the ratio of the farmland size and the number of people living in the same family. Large land per capita ratios were expected to increase households' probability of agricultural production, as more agricultural labor was needed to farm more land [50]. The mean of the land per capita was 2.533 mu. Labor ratio was the share of labor population between the ages of 15 and 64. Household labor availability also has an impact on household labor allocation between farm employment and off-farm employment [51]. It was expected that the labor ratio would have a negative impact on farm employment, since the limited amount of arable land per capita in China forced any surplus labor forces to be more inclined to engage in off-farm employment. As for farming type, households were divided into professional households and non-professional households, which accounted for 89.58% and 10.42% of the samples, respectively. The professional households were expected to have more farming skills and therefore to be more productive in agriculture.

The village characteristics were the proportion of the village population engaged in agriculture, the existence of non-agricultural economy in the village, the index of village support services, the distance of the village from the township government, and the topography of the village. As seen in Table 1, the mean of the proportion of the village population engaged in agriculture was 71.919, indicating that approximately 72% of the farmers in the village were involved in agriculture. Where there was a higher proportion of the village population engaged in agriculture, there was less non-agricultural economy in the village, which indicates that the level of economic development in the village was low. When this is the case, the farmers can only engage in agricultural production to maintain demand. On the contrary, the existence of a non-agricultural economy in a village indicates that the level of economic development in the village is high, and that farmers tend to regularly engage in non-agricultural production. In our sample, 18% of villages had a non-agricultural economy. The index of village support services represented the level of support and security for agricultural production at the village level. Village support services are likely to increase the ratio of agricultural production, as it can reduce the sunk costs and realize increasing returns [52]. In this paper, support services mainly included unified irrigation and drainage services, machine plowing services, unified pest prevention services, unified purchase of production materials services, planting planning services, and organization of farmers for agricultural production. Each service available to the farmers was assigned a value of "1", and each service not available was assigned a value of "0". The values were added together to obtain the village support service index. The distance of the village from the township government was also a key factor influencing farmers' agricultural production, as the rural households located further away from the township government were more likely to participate in agricultural production than households located nearby [53]. The average distance of the village from the township was 6.330 km

in our sample. According to Xie et al. [54], we controlled for whether a laborer was from a plain village. If the village was a plain village, it was given a value of "1"; if not, it was given a value of "0".

The specific definitions of the above variables and the results of the descriptive statistical analysis are shown in Table 1.


**Table 1.** Definitions and Descriptions of Variables.

#### *4.3. Models*

#### 4.3.1. Propensity Score Matching (PSM) Method

In order to test the impact of the LCP on the land "stickiness" of rural laborers, this paper constructs an empirical model of land "stickiness" in the following form.

$$Y\_i = a\_0 + a\_1 R\_i + a\_2 X\_i + \varepsilon\_i \tag{1}$$

where *Y* denotes the labor land "stickiness" of farming households *i*; *Ri* denotes whether the LCP was implemented; *Xi* is a series of control variables, including individual characteristics, household characteristics, and village characteristics; *a*<sup>1</sup> and *a*<sup>2</sup> are the coefficients to be estimated for LCP and control variables, respectively; *a*<sup>0</sup> is a constant term; and *ε<sup>i</sup>* is the error term.

Notably, as the LCP was gradually implemented using a moderated piloting system, "selective" bias might arise as a result. The LCP was affected by its circumstances, such as the social, historical, or economic conditions [17]. On the one hand, the local governments of the initial pilot regions were required to invest copious human, material, and financial resources into supporting the LCP, which placed stress on these regions in terms of their economic development. On the other hand, after the LCP, the farmland would no longer be adjusted, which ultimately affected the existing interest pattern in rural areas and even hinder the implementation of the policy [55]. Therefore, the promotion of the LCP inevitably put certain demands on the strength of a village's collective organizations. Thus, in order to ensure the effectiveness of the LCP pilots, the government tended to select the areas with relatively high levels of economic development and the villages with stronger rural collective organizations or less traditional farmland adjustment as the pilot areas for the LCP. This ultimately resulted in the problem of sample selection bias faced in this study. At the same time, there was a level of heterogeneity in land "stickiness", individual characteristics, household characteristics, and village characteristics between the groups with and without LCP (see Table 2). Therefore, if a simple regression analysis were used to estimate the policy effect of LCP on the land "stickiness" of rural laborers, the estimation results may be biased.


**Table 2.** Average individual, household, and village characteristics by LCP status.


#### **Table 2.** *Cont.*

Note: Standard deviations are shown in parentheses. \*\*\*, \*\*, and \* denote significance at 1% level, 5% level, and 10% level, respectively. Significance levels are obtained from *t*-tests or chi-square tests, depending on whether the variable is categorical or continuous.

Based on the possible "selective" bias of the sample and the heterogeneity of individual characteristics, household characteristics and village characteristics, this paper adopted the propensity score matching (PSM) method to estimate the effect of the LCP on the land "stickiness" of rural labor. The basic idea of this method was to construct a sample of uncertificated farmland for the sample of certificated farmland by introducing a counterfactual framework, and to ensure that the characteristics of both samples are similar except for the farmland rights [56]. The difference of land "stickiness" between the two samples can be regarded as the result of two different experiments (with and without LCP) on the same individual. Therefore the difference was the policy effect of the LCP on labor land "stickiness".

Specifically, in this paper, the land "stickiness" of farmer i with confirmed farmland rights (*R* = 1) was set as *Y<sup>R</sup> <sup>i</sup>* , which was the treatment group; the land "stickiness" of farmer *i* with unconfirmed farmland rights (*R* = 0) was set as *YNR <sup>i</sup>* , which was the control group. The effect of farmland rights on labor land "stickiness" was thus determined as:

$$T\_i = Y\_i^R - Y\_i^{NR} \tag{2}$$

In Equation (2), since it was not possible to observe the land "stickiness" of farmer *i* both before and after the farmland was certificated, the counterfactual framework was constructed in this paper as:

$$\begin{split} T\_i &= E(Y\_i^R | R=1) - E(Y\_i^{NR} | R=0) \\ &= E(Y\_i^R | R=1) - E(Y\_i^{NR} | R=0) + E(Y\_i^{NR} | R=1) - E(Y\_i^{NR} | R=1) \\ &= E[(Y\_i^R - Y\_i^{NR}) | R=1] + [E(Y\_i^{NR} | R=1) - E(Y\_i^{NR} | R=0)] \end{split} \tag{3}$$

In Equation (3), *E Y<sup>R</sup> <sup>i</sup>* − *<sup>Y</sup>NR i <sup>R</sup>* <sup>=</sup> <sup>1</sup> is the average treatment effect on the treated (ATT) of LCP on land "stickiness", and *E YNR i <sup>R</sup>* <sup>=</sup> <sup>1</sup> − *E YNR i <sup>R</sup>* <sup>=</sup> <sup>0</sup> is the selection bias.

In the above analysis, the certificated farmland group and the uncertificated farmland group were not randomly assigned, so they had a level of selection bias. The PSM method constructs the control group with similar characteristics to the treatment group as much as possible by matching the scores of the whole sample. In this way, we were able to effectively reduce the sample's selection bias and obtain the effective estimate of the average treatment effect on the treated (ATT). The main estimation steps were as follows.

First, we used a logit model to estimate the conditional probability fitted value of rural households to carry out certificated farmland, i.e., the expression of the propensity score value is

$$PS = \Pr(R = 1 | X\_i) = E(R = 0 | X\_i) \tag{4}$$

In Equation (4), *PS* is the propensity score value; *R* = 1 indicates certificated farmland farmers; *R* = 0 indicates certificated farmland farmers; and *Xi* indicates observable individual, household, and village characteristics.

Next, the treatment group was matched with the control group. To verify the robustness of the matching results, two methods, nearest neighbor matching (NNM) and kernel-based matching (KBM), were selected for matching in this paper.

Then, the common support test and the balance test for propensity score estimation were performed. The common support test was utilized to determine whether the treatment and control groups had a common support region and whether there was any partial overlap in the range of values. The balance test was used to determine the matching quality by comparing whether there was a significant difference between the treatment and control groups in terms of explanatory variables.

Finally, the ATT was calculated to determine the effect of the LCP on the land "stickiness" of rural laborers.

#### 4.3.2. Mediation Effect Model

In order to further study the influence path of the LCP on rural labor land "stickiness", and test whether there was a mediating effect of land production function between the LCP and labor land "stickiness", we constructed the following model.

$$M\_i = b + \beta\_2 R\_i + \lambda\_2 X\_i + \varepsilon\_{2i} \tag{5}$$

$$Y\_i = \mathcal{c} + \mathcal{B}\_3 \mathcal{R}\_i + \mathcal{B}\_4 \mathcal{M}\_i + \lambda\_3 X\_i + \varepsilon\_{3i} \tag{6}$$

where the meanings of *Yi* and *Ri* are kept consistent with Equation (1); *Mi* is the mediating variable, i.e., the land production function; *β*1, *β*2, *β*<sup>3</sup> and *β*<sup>4</sup> are parameters to be estimated; *a*, *b* and *c* are constant terms, and *ε*1*i*, *ε*2*<sup>i</sup>* and *ε*3*<sup>i</sup>* are random error terms.

#### **5. Analysis and Results**

#### *5.1. The Estimation Results of the PSM Model*

#### 5.1.1. Propensity Score Estimation

The most direct way to measure the impact of the LCP on labor land "stickiness" is to compare the land "stickiness" of certificated rural households with that of non-certificated rural households. However, these two categories (certificated and uncertificated) are mutually exclusive, which makes it impossible to observe the effects of the same household's "certificated" and "uncertificated" farmland at the same time. Therefore, this paper constructed a logit model and a marginal effects model with certificated farmland as the dependent variable, and then estimated individual, household, and community characteristics to obtain the propensity scores of each variable. The estimation results are shown in Table 3. The explanatory variables all had variance inflation factors (VIFs) of less than ten, except for age and age squared. Although the VIFs of age and age squared were more than ten, according to Allison's research, multicollinearity will have no adverse consequences when high VIFs are caused by the inclusion of powers or products of other variables [57]. Hence, the estimated model was free of any serious multicollinearity.

From Table 3, we can see that age, political outlook, and health status among individual characteristics had significant effects on the LCP. There was an inverted "U"-type relationship between age and the LCP. This indicates that as the age of the head of the household increased, so did the probability of the household participating in the LCP. However, when the age increased to a certain degree, this same probability decreased in turn. Political outlook and health status had positive effects on the LCP and were significant at the 5% and 1% statistical levels, respectively. In terms of household characteristics, land per capita and household type had positive effects at the 1% statistical level, and the probability of the LCP increased by 0.7% for each unit increase in land per capita. Compared with non-professional households in agricultural production, the probability of the LCP was 11.3% higher for professional households. In terms of village characteristics, the non-agricultural economy had a negative effect on the LCP, and this was significant at the 5% level. This means that villages with a developed non-agricultural economy had less dependence on farmland, and a low probability of participating in the LCP to secure farmland property rights. The support service index and distance to township government had positive effects on LCP, as they were significant at the 1% and 5% levels, respectively. The topography of the village had a negative effect on the LCP at the 10% statistical level, and the probability of the LCP in plain areas was lower than that in non-plain areas. This is because the LCP is easier to promote and implement in the non-plain areas as these farmers are less dependent on land.


**Table 3.** Logit model results of factors determining LCP.

Note: \*\*\*, \*\*, and \* denote significance at 1% level, 5% level, and 10% level, respectively.

#### 5.1.2. Matching Quality Tests: Common Support Test and Balance Test

The common support test aimed to determine whether there was a significant difference between the propensity score values of the treatment group (the certificated farmland group) and the control group (the uncertificated farmland group). If the common range of propensity scores of the two sample groups was large, it indicated a good matching result. Otherwise, it would lead to biased estimation results. The results of the common support test were matched using the nearest neighbor matching and kernel-based matching methods. As shown in Figure 3, most of the observations were within the common support region and only a small number of samples were not in the common support region after matching. Therefore, the common support test was satisfied by the matching estimation using the nearest neighbor matching and kernel-based matching methods in this paper, indicating a good matching effect.

The balance test aimed to examine whether there are significant systematic differences in the variables of individual characteristics, household characteristics and village characteristics between the treatment groups and the control groups. If there was no significant difference after matching, this indicated that individuals with the same characteristics could be found to match between the two groups and the matching effect was good; if not, the matching effect was poor.

**Figure 3.** Propensity score distribution and common support for propensity score estimation: (**a**) with nearest neighbor matching; (**b**) with kernel-based matching.

In Table 4 it is shown that after applying the nearest neighbor matching method, the standardized deviation of each variable after matching was controlled within 10%. Following this, the standardized deviation of most of the control variables decreased compared with that before matching. Except for the variables of total household income, proportion of working population, and village topography, the deviation of all variables after matching decreased substantially. In addition, the variables including gender, education, and political outlook in individual characteristics, and type of farm household in household characteristics, as well as the variables of population engaged in agriculture ratio, non-agricultural economy, support services index, and distance to township government in village characteristics, were significantly different between the treated and control group samples before matching, and there were also no systematic differences after matching. These indicate that the differences in characteristics between the treated and control groups were basically eliminated, and the two groups of samples became more similar and comparable.


**Table 4.** PSM quality indicators before and after matching.


#### **Table 4.** *Cont.*

Note: \*\*\*, \*\*, and \* denote significance at 1% level, 5% level, and 10% level, respectively.

Further, similar results were also obtained using the kernel-based matching method. Compared with before matching, the standardized deviations of each variable were significantly reduced, and all variables were controlled within 10%. Except for income and labor ration, the percentage of deviation reduction in all variables decreased significantly. These indicate that the systematic difference changes between the treatment and control groups before and after matching were consistent with the nearest neighbor matching.

Meanwhile, according to the distribution of standardized deviations used for both methods (see Figure 4), the distribution of standardized deviations of each variable before matching was relatively discrete. This indicates that the individual characteristics, household characteristics, and village characteristics of the certificated farmland group and the uncertificated farmland group before matching were significantly different. After applying the nearest neighbor matching and kernel-based matching methods, the standardized deviations of each variable were less than 10%, and most of them were concentrated around 0, which shows a significant reduction compared with the pre-matching period. This also indicates that the matched group of the LCP and the group without the LCP were better balanced at the level of control variables, and there was no longer a significant difference.

In summary, the results of the balance test using the two matching methods of nearest neighbor matching and kernel-based matching remained consistent. This indicates that the sample matching passed the balance test.

**Figure 4.** Standardized % bias across covariates: (**a**) with nearest neighbor matching; (**b**) with kernel-based matching.

#### 5.1.3. Analysis of Matching Results

The nearest neighbor matching and kernel-based matching methods were used to assess the effect of LCP on the land "stickiness" of rural labor (see Table 5). In general, the ATT values obtained by both matching methods were relatively close to each other, indicating that the analysis results were robust. In the nearest neighbor matching, the ATT value of LCP was 0.049. Meanwhile, in the kernel-based matching, the ATT value of LCP was 0.048. Both ATT values were statistically significant at the 1% level. In other words, the implementation of the LCP will contribute to an increase in the agricultural labor force participation rate by about 4.8% to 4.9%. This indicates that the LCP has a positive effect on agricultural labor participation and enhances the land "stickiness" of rural labor.


**Table 5.** The ATT of LCP on land "stickiness".

Note: Treatment group (T) and control group (C). \*\*\* denote significance at 1% level.

In addition, the unmatched ATT value of 0.061 was slightly higher than the matched result. This suggests that sample selection bias and variability in sample characteristics can overestimate the policy effect of LCP, and that simple regression model estimates are biased.

#### *5.2. Heterogeneity Analysis*

With the development of urbanization and industrialization, rural society has become divided. The main agricultural operators present a situation of co-existence of professional and part-time households. At the same time, with the improvement of the farmland rented market, farmland resources can be reallocated in the rural market, which in turn provides the possibility for professional agricultural production households to increase their farm sizes. To this end, this section examines whether the policy effects of the LCP on labor land "stickiness" are heterogeneous in terms of the rural household types and farm size. Does the policy effect of the LCP have preferences in terms of household types and farm size?

In this paper, the types of rural households are divided into professional and nonprofessional households in agricultural production. At the same time, drawing on the idea of grouping in Opler's study [58], the samples of farm size less than 25% are defined as small scale, the samples of farm size between 25% and 75% are defined as medium scale, and the samples of farm size more than 75% are defined as large scale.

From Table 6, we can see that in terms of rural household type, the promotional effect of the LCP on agricultural labor participation of non-professional household groups was significant, while the promotional effect on agricultural labor participation of professional household groups was not significant. In terms of farm size, the promotional effect of the LCP had the largest effect on medium-sized farms, with an ATT value of 0.060, and it was significant at the 1% statistical level; the second largest effect was on small-sized farms, with an ATT value of 0.042, and it was significant at the 10% level; and the effect on large-sized farms was not significant.

**Table 6.** Heterogeneity analysis results.


Note: Standard errors are shown in parentheses. Professional households (PH) and non-professional households (NPH). \*\*\*, \*\*, and \* denote significance at 1% level, 5% level, and 10% level, respectively.

The above analysis shows that the policy promotion effect of the LCP on land "stickiness" of rural labor is mainly reflected in the group of non-professional agricultural production households, medium-sized farms, and small-sized farms, while the promotion effect on professional agricultural production households and large-sized farms is not significant.
