*4.2. Benchmark Regression Results*

4.2.1. Analysis of the Impact of Farmland Transfer-Out on Rural Household Consumption

In Table 3, the coefficients of the farmland transfer-out variable are significantly positive, and the coefficient of non-food consumption is 0.118, which is larger than the coefficient of food consumption is 0.016, which verifies hypothesis 1 of this paper, that is, farmland transfer-out can stimulate the consumption of rural households, and rural households are more willing to increase their non-food consumption expenditure, which is beneficial to the optimization of rural consumption structure. The "psychological accounts" theory states that people categorize their income into different accounts according to the way they receive it, which are mutually exclusive and not complementary, and that different income patterns result in different consumption tendencies [1]. The income structure will become richer as rural households generally receive farmland rental income and wage income after their farmland is transferred out, which will continuously strengthen the subjective wealth effect of rural households and induce them to consume. In addition, after rural households transfer out of farmland, they will generally move away from the countryside to engage in non-agricultural production activities in the city. Affected by the new consumption habits of the surrounding people, rural households who transfer-out farmland will gradually change their original consumption habits that prefer to increase food consumption expenditure to those that are more willing to increase non-food consumption expenditure. Therefore, when rural households satisfy the surplus of food consumption expenditure, they are more willing to increase the expenditure on non-food consumption, and their consumption structure will be optimized accordingly.


**Table 3.** Impact of farmland transfer on rural household consumption: Results of benchmark regression.

Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside the regression parentheses are *t* values of coefficients.

#### 4.2.2. Analysis of the Impact of Farmland Transfer-In on Rural Household Consumption

The coefficients of farmland transfer-in variables are significantly positive, and the coefficient of food consumption is 0.028, which is larger than the coefficient of non-food consumption is 0.009. Hypothesis 2 of this paper that farmland transfer can stimulate rural household consumption but rural households who transfer-in farmland are more willing to increase their expenditure on food consumption, which is not beneficial to the optimization of their consumption structure, is confirmed. Schultz's rational theory of small farmers shows that small farmers are poor and efficient, that is, farmers are people with entrepreneurial spirit and can use the right resources [22]. Rural households who transfer-in farmland may engage in moderate scale operation, take advantage of the scale of farmland to reduce the cost of agricultural production, give full play to the rational and effective allocation of resources such as labor and agricultural machinery for agricultural production to bring about an increase in production efficiency, improve the income of agricultural production of rural households and enhance the consumption capacity of rural households. As a matter of fact, agriculture is a very difficult occupation, and rural households cherish the income that is difficult to obtain. In addition, rural households tend to have a high propensity to save preventively for a single productive income from agriculture [47]. After rural households transfer-in farmland, they are still mainly engaged in agricultural production. The consumption habits of the surrounding people and themselves will not change much. Rural households who transfer-in farmland will still maintain their original consumption habits and are more willing to increase food consumption expenditure than

non-food expenditure. Therefore, the increase in income obtained from the transfer-in of farmland to rural households will increase their consumption capacity to a certain extent, but they will save after satisfying the surplus of food consumption expenditure and are generally unwilling to spend too much on non-food consumption, which makes it difficult to optimize their consumption structure.

Furthermore, based on the communication with rural households in the field survey, we know that rural households are mainly engaged in agricultural production before the transfer of farmland. Although the total consumption expenditure of rural households will increase, rural households are more inclined to increase food consumption expenditure, which is not conducive to the optimization of rural households' consumption structure. After the transfer of farmland, the total consumption expenditure of rural households will continue to increase, but rural households who transfer-in farmland are more willing to increase food consumption expenditure, which is not conducive to the optimization of their consumption structure. The rural households that transfer-out farmland are more willing to increase non-food consumption expenditure, which is beneficial to the optimization of their consumption structure. Therefore, the farmland transfer has a heterogeneous impact on the consumption expenditure and consumption structure of rural households of the farmland transfer-out and rural households of the farmland transfer-in.

#### 4.2.3. Analysis of the Impact of Control Variables on Rural Household Consumption

The coefficients of the household head's age variable are all significantly negative, and their coefficients on food consumption are larger than those on non-food consumption, that is, the consumption expenditure of rural households tends to decline as the household head gets older, and their consumption structure also shows a deterioration. The possible explanation is that in rural Chinese society, the head of the household is the mainstay of the family and his income is the most important source of income for the rural household. The coefficients of the family assets per capita variable are all significantly positive, and their coefficients on non-food consumption are larger than those on food consumption, indicating that the more family assets, rural households have the stronger consumption capacity and the more conducive to optimizing their consumption structure. The possible reason for this is that family assets have a certain "wealth effect" and "asset effect", which can bring a stable income stream to rural households, thus enhancing their consumption ability and improving their consumption structure [48].

#### *4.3. Robustness Test and Endogeneity Discussion*

#### 4.3.1. Robustness Test I: Sub-Sample Test

In the field research, we found that a few rural households have two-way farmland transfer behaviors of both farmland transfer-out and farmland transfer-in. However, mixing rural households' two-way farmland transfer behaviors with one-way farmland transfer behavior for regression estimation may affect the authenticity of the results. For this reason, drawing on the study of Yang et al. [21], the data of a sample of 16 rural households with both farmland transfer-out and farmland transfer-in 2020 are excluded from the subsample test. The results in Table 4 show that the significance level of coefficients and the sign of coefficients of the farmland transfer-out and farmland transfer-in variables and the magnitude of coefficients between them on food consumption and on non-food consumption variables are consistent with the results of the benchmark regression, indicating that the benchmark regression results are robust.


**Table 4.** Robustness test I: Sub-sample test.

Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside of regression parentheses are *t* values of coefficients. Control variables are kept consistent with Table 3.

#### 4.3.2. Robustness Test II: Replacing Core Explanatory Variables and Re-Estimating

To exclude the estimation bias caused by measurement bias, this paper uses the method of Hu and Ding [22] to conduct robustness tests using the average per mu income from farmland transfer-out and the average per mu expenditure from farmland transfer-in as proxies for farmland transfer-out and farmland transfer-in5, respectively. The results in Table 5 show that the significance levels of the coefficients and the sign of the coefficients of the variables of the average per mu income from farmland transfer-out and the average per mu expenditure from farmland transfer-in and the magnitudes of the coefficients between the variables of food consumption and non-food consumption are consistent with the results of the benchmark regression, indicating that the results of the benchmark regression are robust.

**Table 5.** Robustness test II: Replacement of core explanatory variables.


Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside of regression parentheses are *t* values of coefficients. Control variables are kept consistent with Table 3.

#### 4.3.3. Robustness Test III: Re-Estimation Using Propensity Matching Score Method

To eliminate the endogeneity problem caused by the possible selectivity bias of the sample, this paper uses the propensity matching score method (PSM) for robustness testing [50]. Based on Table 3 control variables matching control and experimental groups, rural households of farmland transfer-out and farmland transfer-in are set as the experimental group, and rural households of farmland non-transfer-out and farmland non-transfer-in are set as the control group. The average treatment effects (ATT) of farmland transfer-out and farmland transfer-in are estimated using nearest neighbor matching, radius matching and kernel matching, respectively. The results of the common support condition test of Figure 3 show that most of the observations are within the common range of values when matching using the three matching methods of nearest neighbor matching (k = 4), radius matching (caliper = 4) and kernel matching (bwidth = 0.06), and thus the matching quality is reliable.

nearest neighbor matching radius matching kernel matching

nearest neighbor matching radius matching kernel matching

nearest neighbor matching radius matching kernel matching

**Figure 3.** *Cont*.

(**c**)

(**d**)

(**b**)

**Figure 3.** Propensity score distribution and the common support for propensity score estimation. (**a**) farmland transfer-out and farmland not transfer-out, *CS* = total consumption of rural households. (**b**) farmland transfer-out and farmland not transfer-out, *CS* = food consumption. (**c**) farmland transfer-out and farmland not transfer-out, *CS* = non-food consumption. (**d**) farmland transfer-in and farmland not transfer-in, *CS* = total consumption of rural households. (**e**) farmland transfer-in and farmland not transfer-in, *CS* = food consumption. (**f**) farmland transfer-in and farmland not transfer-in, *CS =* non-food consumption.

The results in Table 6 show that the average treatment effects obtained by the nearest neighbor matching, radius matching and kernel matching methods provide further evidence that either farmland transfer-out or farmland transfer-in significantly enhances the consumption capacity of rural households. In addition, taking the nearest neighbor matching method as an example, after excluding other factors, the per capita non-food consumption expenditure of rural households transferred out farmland will increase by 4.081% (exp (0.040) − 1), which is larger than the per capita food consumption expenditure by 0.602% (exp (0.006) − 1), and the per capita food consumption expenditure of rural households transferred in farmland will increase by 1.207% (exp (0.012) − 1), which is larger than the per capita non-food consumption expenditure by 0.401% (exp (0.004) − 1). Therefore, the re-estimation results based on the propensity matching score method (PSM) show that the benchmark regression results are robust.


**Table 6.** Robustness test III: Re-estimation using PSM.

Note: \*\*\* refers to the statistics being significant at the 1% level. Control variables are kept consistent with Table 3.

#### 4.3.4. Robustness Test IV: Re-Estimation Using Instrumental Variable Method

When examining the impact of farmland transfer on rural household consumption, there may be endogeneity problems caused by reverse causality and omitted variables, and then the direct use of the OLS estimation method is likely to cause bias in model estimation. For this reason, this paper attempts to construct an instrumental variable model to eliminate the endogeneity problem caused by reverse causality and omitted variables. Drawing on the research results of Yang et al. [21] and Hu and Ding [22], the two-stage least squares (2SLS) estimation is conducted using "village farmland transfer-out rate" and "village farmland transfer-in rate" as the instrumental variables for farmland transfer-out and farmland transfer-in. As we all know, a qualified instrumental variable must satisfy two conditions, namely, the instrumental variable is highly correlated with the endogenous explanatory variables (correlation) and the instrumental variable is uncorrelated with the disturbance term (exogeneity). In this paper, the "village farmland transfer-out rate" and "village farmland transfer-in rate" are calculated based on the level of farmland transfer-out and the level of farmland transfer-in in the surveyed villages, which satisfy the requirements of correlation and exogeneity of the instrumental variable. The results in Table 7 show that the one-stage F values are all much greater than 10, indicating that the model does not have the problem of weak instrumental variables. The DWH values reject the original hypothesis that farmland transfer-out and farmland transfer-in are exogenous variables at the 1% level, indicating that the model has endogeneity problems. However, after correcting for the endogeneity problem induced by reverse causality and omitted variables, the significance level of coefficients and the sign of coefficients of farmland transfer-out and farmland transfer-in variables and the magnitude of coefficients between them on food consumption and on non-food consumption variables are consistent with the results of the benchmark regression, which verifies the credibility of the benchmark regression results.

**Table 7.** Robustness test IV: Re-estimation using instrumental variables method.


Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside the regression parentheses are *t* values of coefficients. Control variables are kept consistent with Table 3.

#### *4.4. Endogeneity Discussion*

The endogeneity problem is mainly caused by measurement error, selectivity bias, omitted variables and reverse causality [51]. For the measurement error problem, this paper solves it by replacing the core explanatory variables. For the selectivity bias problem, this paper mitigates it by using the propensity matching score method (PSM), which enables the observations to effectively avoid the estimation bias caused by sample self-selection through matching and resampling [52], thus improving the accuracy of the estimation results. For the omitted variables and reverse causality problems, this paper eliminates them by using the instrumental variables method, while adding as many control variables as possible to exclude the influence of omitted observables on the estimation results of this paper. In addition, the sub-sample of 16 households with both farmland transfer-out and farmland transfer-in is excluded for re-estimation to exclude the influence of different samples with different sensitivity to the obtained results. In summary, strictly speaking, there is no particularly serious endogeneity problem in this paper.

#### *4.5. Mechanism of Action: Intermediary Effect Test*

The results of the benchmark regressions and robustness tests indicate that farmland transfer (farmland transfer-out or farmland transfer-in) can stimulate rural household consumption, but there is heterogeneity in its effect on the consumption structure of rural households of farmland transfer-out and rural households of farmland transfer-in. Here, we further use the intermediary effect model to test its internal transmission mechanism. However, whether farmland is transferred out or transferred in actually represents the choice of livelihood modes of different rural households. As a result, the income of rural households who transfer-out farmland mainly includes rental income and wage income, and the income of rural households who transfer-in farmland mainly includes productive income. Therefore, this paper is not to test the intermediary effect of rent income, wage income and productive income on rural household consumption in the transfer of farmland, but to test the intermediary effect of income from the farmland transfer-out of rural households (the sum of rent income and wage income) on rural households' consumption and productive income from the farmland transfer-in of rural households on rural households' consumption. The results in Table 8 show that there is a significant positive effect of farmland transfer on the income of rural households under different livelihoods, indicating that rural households after farmland transfer can bring in stable income based on different livelihood strategies. In addition, the fitted regression results show that rural household income under different livelihoods positively affects total consumption of rural households, food consumption and non-food consumption at the 1% level of significance, which indicates that the intermediary effect of rural household income under different livelihoods exists and is significant. Besides this, the optimized consumption structure of rural households of farmland transfer-out and the deteriorated consumption structure of rural households of farmland transfer-in remain consistent with the benchmark regression results. That is, the impact path of "farmland transfer—rural households' income under different livelihoods—rural household consumption" holds. Through calculation, it is found that the intermediary effects of farmland transfer-out on the total consumption of rural households, food consumption and non-food consumption by affecting the income of rural households of farmland transfer-out are 39.014% (39.014% is obtained by multiplying the coefficient 0.257 of the farmland transfer-out variable to the rural household income variable in Table 8 by the coefficient 0.126 of the direct effect of the rural household income variable to the total consumption of rural households variable, and then dividing it by the coefficient 0.083 of the farmland transfer-out variable to the total consumption of rural households variable in Table 3, and then multiplying it by 100%. The rest of the intermediary effect proportion data can be obtained according to this calculation method), 30.519% and 44.648%, respectively; the intermediary effects of farmland transfer-in on the total consumption of rural households, food consumption and non-food consumption by affecting the income of rural households of farmland transfer-in are 38.912%, 40.250%

and 35.389%, respectively, which show that the intermediary effect accounts for a large proportion in the total utility of affecting the total consumption of rural households, food consumption and non-food consumption. The income of rural households based on different livelihood modes is a transmission mechanism that can not be ignored in the impact of farmland transfer on the rural household consumption, respectively. In addition, under different livelihood strategies, rural households' dependence on agricultural income is different, which may also be an important potential reason for the optimization of rural households who transfer-out farmland consumption structure and the deterioration of rural households who transfer-in farmland consumption structure. The intermediary effect of this paper is a partial intermediary effect.


**Table 8.** Intermediary effect test.

Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside the regression parentheses are *t* values of coefficients. Control variables are kept consistent with Table 3.

At the same time, a non-parametric percentile bootstrap sampling method with bias correction is used to conduct 5000 sampling tests to examine the intermediary effect of rural households' income under different livelihoods. In Table 9, the value of *δ*<sup>1</sup> × *φ*<sup>2</sup> does not contain 0 at a 95% confidence interval, and the coefficients of *δ*1, *φ*<sup>1</sup> and *φ*<sup>2</sup> pass the 5% significance level test, and *δ*<sup>1</sup> × *φ*<sup>2</sup> has the same sign as *φ*1, which indicate that the income of rural households under different livelihoods plays a part in the intermediary effect of farmland transfer on rural households' consumption and consumption structure, thus the results of the intermediary effect model test in this paper are valid and robust.

**Table 9.** Robustness test of intermediary effect.


Note: \*\*\* and \*\* refer to the statistics being significant at the 1% and 5% levels, respectively. Inside the regression parentheses are *t* values of coefficients. Control variables are kept consistent with Table 3.

#### **5. Discussion**

In this section, we discuss the potential contributions and limitations of this research. The first discussion concerns the major contributions to the existing literature. This paper contributes to the current studies in four ways. (1) We use the survey data of 537 rural households in 50 villages in Yunnan Province, which is relatively underdeveloped in Southwest China and is located in the Yunnan-Kweichow Plateau, mainly in plateau and mountain terrain, to study the impact of farmland transfer on rural household consumption, which has unique regional characteristics and greater practical significance. (2) By constructing an analytical framework of "farmland transfer—farmland function—income structure—rural household consumption", we comprehensively analyzed the theoretical mechanism relationship between farmland transfer and rural household consumption. (3) Although there is a small amount of literature on the impact of farmland transfer on rural household consumption, this paper more systematically studies the impact of farmland transfer on rural household consumption through benchmark regression, robustness test and intermediary effect test. At the same time, we have achieved more fruitful study results. (4) Based on the empirical study of 537 rural households in 50 villages in Yunnan Province, we have obtained some new findings. For example, rural household consumption expenditure will show a downward trend with the increase in the age of the head of rural household, and the consumption structure will also show a deterioration. Another example is that the more family assets, rural households have the stronger consumption expenditure capacity, which is conducive to optimizing their consumption structure.

The second discussion is about the limitations of this study. (1) The results of this study are based on the corresponding empirical analysis of 537 rural households survey data in 50 villages in Yunnan Province. There are certain regional limitations, and whether it is applicable to other regions remains to be discussed, but the significance of the results of this study is not to be underestimated. (2) Based on the cross-sectional data of 537 households in 50 villages in Yunnan Province in 2020, the research conclusion is that the static impact of farmland transfer on rural household consumption and consumption structure cannot reflect the trend of time dynamic impact of farmland transfer on rural household consumption and consumption structure. This requires our team to conduct a continuous follow-up survey on these rural households and use panel data to overcome the limitation of this study.

#### **6. Conclusions and Policy Implications**

#### *6.1. Conclusions*

Based on the first-hand survey data of 537 rural households in 50 villages in Yunnan Province, this paper constructs an analytical framework of "farmland transfer—farmland function—income structure—rural household consumption", uses the OLS model to deeply explore the impact of farmland transfer on rural household consumption, and further uses the intermediary effect model to explore its internal transmission mechanism. The following conclusions are drawn: First, farmland transfer (farmland transfer-out or farmland transfer-in) can stimulate rural household consumption. Second, the coefficient of farmland transfer-out to non-food consumption is 0.118, which is larger than the coefficient of farmland transfer-out to food consumption is 0.016; rural households who transfer-out farmland are more willing to increase non-food consumption expenditure, which is beneficial to the optimization of their consumption structure. Third, the coefficient of farmland transfer-in to food consumption is 0.028, which is larger than its coefficient of non-food consumption is 0.009; rural households who transfer-in farmland are more willing to increase food consumption expenditure, which is not conducive to the optimization of their consumption structure. The above research results are still robust after excluding possible endogenous problems through four robustness tests, namely, sub-sample test, replacement core explanatory variables test, propensity matching score (PSM) test and instrumental variable test, which shows that the conclusions obtained from benchmark regression are true and reliable to a large extent. Fourth, rural household consumption expenditure will show a downward

trend with the increase in the age of the head of the rural household, and the consumption structure will also show a deterioration. Fifth, the more family assets, rural households have the stronger consumption expenditure capacity, which is conducive to optimizing their consumption structure. Sixth, the results of the intermediary effect model show that the transfer of farmland affects rural households' consumption and consumption structure by affecting rural households' income under different livelihood modes. At the same time, using the non-parametric percentile bootstrap sampling method of deviation correction, the results of 5000 sampling tests show that the effect of the intermediary effect model is effective and robust.

#### *6.2. Policy Implications*

Improving the consumption capacity and consumption level of rural households is not only a strong response to the major strategic deployment of "accelerating the construction of a new development pattern with domestic big cycle as the main body and domestic and international double cycles promoting each other" put forward in China's 14th Five-Year Plan, but also conducive to the orderly advancement of China National New Urbanization Plan and China Rural Revitalization Strategy. Therefore, in order to further release the consumption capacity of rural households and improve their consumption level, this paper draws the following enlightenment: first, it is necessary to establish the interest coordination mechanism of farmland transfer, constantly reduce the transaction cost of farmland transfer and guide rural households to carry out farmland transfer in an orderly manner, so as to realize the optimal allocation of farmland resources. Second, improve the non-agricultural employment mechanism of rural surplus labor force, reasonably arrange rural households of farmland transfer-out and strengthen their skills training, so as to ensure the stability of their non-agricultural employment and obtain higher income. Third, improve the stability of farmland property rights, promote rural households of farmland transfer-in for moderate scale operation and constantly encourage them to improve the expected return on investment in farmland, so as to ensure the sustained and stable growth of their agricultural production. Fourth, social security shoulders the major responsibility of ensuring people's livelihood, promoting social equity and meeting the needs of the people for a better life. In the new era, rural areas should build a multi-level social security system in an all-around way, so as to lay a foundation for promoting the improvement of rural households' consumption ability and the optimization of consumption structure. Fifth, rural households should save an appropriate amount of their income and appropriately increase their family assets.

**Author Contributions:** Econometric analysis and writing the original draft, L.L.; conceptual formulation, survey design, data collection, modeling, manuscript editing and revision, M.H. and L.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Natural Science Foundation of China (72163003), Three Dimensional Practice Model for Cultivating Innovative and Entrepreneurial Talents in Agriculture and Forestry (2020346) and the Foundation of Postgraduate of Guizhou Province (YJSKYJJ[2021]035).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
