Decomposing the Impact of Agricultural Mechanization on Agricultural Output Growth: A Case Study Based on China’s Winter Wheat

Abstract

Agricultural mechanization plays a critical role in addressing labor shortages and promoting sustainable agricultural production in developing countries. However, few studies have explored the mechanisms by which mechanization affects agricultural output through various channels, especially identifying the key drivers of this association. Taking winter wheat as an example, this study investigates how and to what extent agricultural mechanization affects agricultural output using county-level panel data from China during 1998–2016. The results show that mechanization has a significant positive impact on winter wheat output. Heterogeneity analysis suggests that this positive impact is more pronounced in the plains with better transport conditions. Further decomposition of winter wheat output growth shows that mechanization drives winter wheat output growth through both expansion in sown area and increase in yield, with the former being the dominant pathway. These findings enrich the understanding of how mechanization increases agricultural output and provide policy guidance for leveraging its potential to promote agricultural development in developing countries.

1. Introduction

Food insecurity remains a global issue, particularly prevalent in developing countries [1]. Improved agricultural output capacity is integral to eradicating hunger and enhancing food security [2]. However, the declining availability of rural labor due to economic transformation does not bode well for this endeavor [3]. Agricultural labor shortages not only contribute to abandonment of farmland but also lead to reduced crop yields. For example, Xu et al. [4] found that a 10% increase in non-farm employment corresponds to a 4% higher probability of farmland abandonment, along with a 3% expansion in the average area of abandoned farmland. Similarly, Paudel et al. [5] demonstrated that labor shortages in Nepal’s mid-hills regions extended land preparation periods, delayed rice transplanting, and ultimately reduced rice yields. These findings underscore the urgent need for developing countries to adopt labor-saving technologies to ensure sustainable agricultural production.

Mounting evidence suggests that agricultural mechanization contributes to enhancing agricultural output in farming systems where labor, rather than land constrains production [6,7]. Mechanization can substitute for labor in agricultural production, significantly reducing labor requirements for farming operations [8,9,10]. For example, wheat farmers in southeast Ethiopia who use combine-harvesters require only 55% of the labor applied by those relying on manual harvesting [11]. This labor-saving advantage of mechanization helps mitigate the risk of cropland abandonment and promote the expansion in cultivated areas [12,13,14]. Moreover, mechanization enhances crop productivity by enabling timely field operations and facilitating production intensification [15,16]. Zhou et al. [17] found that smallholder farmers in China who adopted machinery experienced increased corn yields and reduced yield variability. While the positive impact of mechanization in promoting agricultural production has been well documented, most empirical studies focus on the perspective of either area expansion [8,18] or yield improvements [5,17]. The mechanism of how agricultural mechanization affects agricultural output through its impact on sown area and yield remains understudied in the literature. Understanding these mechanisms helps us to better analyze the underlying drivers of agricultural output growth and inform policies aimed at enhancing food security.

This study aimed to investigate the mechanisms by which agricultural mechanization affects crop output using winter wheat in China as an example. China, the world’s largest wheat producer, contributes approximately 17% of global wheat output (The data are available at https://www.fao.org/faostat/en/#data/QCL, accessed on 3 December 2024 (Food and Agriculture Organization of the United Nations)), with winter wheat accounting for 96% of the nation’s total wheat output (The data are available at https://data.stats.gov.cn/easyquery.htm?cn=C01, accessed on 3 December 2024 (National Bureau of Statistics of China)). These statistics highlight the critical role that winter wheat production in China plays in ensuring both domestic and global food security [19]. However, labor scarcity and rising rural wages due to the migration of agricultural workers to non-agricultural sectors pose challenges to winter wheat production in China [20]. Paradoxically, winter wheat production in China has grown steadily over the past three decades, with total output increasing from 85.2 million t in 1990 to 132.4 million t in 2022 (The data are available at https://data.stats.gov.cn/easyquery.htm?cn=C01, accessed on 3 December 2024 (National Bureau of Statistics of China)).

The development of agricultural mechanization may provide an explanation for the sustained growth of winter wheat output in China [21]. According to data from the National Bureau of Statistics of China, total agricultural machinery power, a key indicator of mechanization level, grew from 287.1 million kilowatts in 1990 to 1.1 billion kilowatts in 2022, a 3.9-fold increase. Also, the overall mechanization rate for ploughing, sowing, and harvesting increased from 19.7% to 73.1% between 1978 and 2022. In particular, the level of mechanization of wheat production has improved dramatically in recent years, with its overall mechanization rate reaching 97.6% in 2022, surpassing that of rice (86.9%) and corn (90.6%) (The data are available from the Ministry of Agriculture and Rural Affairs of China, for further details, refer to http://www.njhs.moa.gov.cn/nyjxhqk/202406/t20240618_6457395.htm, accessed on 3 December 2024).

This study aimed to quantitatively assess how and to what extent mechanization affects agricultural output, and makes two key contributions to the existing literature. First, in contrast to earlier research that examined the overall impact of mechanization on agricultural production, this study identifies the underlying mechanisms by which agricultural mechanization affects winter wheat output through its impact on sown area and yield. Second, this study provides evidence that the primary pathway through which mechanization drives growth in winter wheat output is through the expansion in sown area, rather than improvements in yield. These findings enhance our understanding of the relationship between mechanization and agricultural output, offering valuable insights for policymakers in developing countries to promote sustainable agricultural production and ensure a stable food supply through mechanization.

2. Conceptual Framework

This section introduces a conceptual framework outlining how mechanization affects winter wheat output. Given that both sown area and yield are key determinants of winter wheat output, agricultural mechanization may influence output through two primary pathways.

First, mechanization can expand winter wheat sown area by mitigating labor constraints, which have traditionally limited the amount of land a household can manage when relying on manual tools [14,22]. For example, a study in Bangladesh demonstrated that using a self-propelled reaper for rice harvesting reduced labor demand by approximately 72% and lowered costs by 68% compared to manual harvesting [23]. Such advantages of mechanization not only reduce the likelihood of cropland abandonment but can also incentivize farmers to cultivate more land, primarily fallow plots or land previously rented out [8,12]. However, it is worth noting that mechanization tends to promote a shift toward grain crops that are easier to mechanize, such as wheat, corn, and rice, rather than labor-intensive non-grain crops such as cotton and vegetables [18,24,25]. The disparity in labor demands across different types of grain crops leads to varying labor substitution levels by machinery, which in turn influences the sown area allocated to each type of grain crop. Mechanization is particularly conducive to expanding the winter wheat sown area, largely due to the relatively low labor requirements for its cultivation (analysis of labor use data for grain crops in China from 1998 to 2016 shows that the average labor use per hectare for wheat was consistently lower than that for rice and corn. For example, in 1998, labor use per hectare was 247 days for rice, 213 days for corn, and merely 151 days for wheat. By 2016, labor requirements had declined across all three crops, with rice and corn requiring 87 days and 84 days per hectare, respectively, while wheat required only 68 days per hectare. All data are available from the National Agricultural Product Cost–Benefit Data Compilation).

The second pathway through which agricultural mechanization influences winter wheat output is through yield improvements. On the one hand, mechanized land preparation improves agronomic practices, including better tillage, weed control, and water management, all of which have been shown to increase productivity [26,27]. On the other hand, growing evidence suggests that mechanization can change the input structure of production factors, optimize resource allocation, and thereby enhance production efficiency [28,29,30,31]. This increased production efficiency directly contributes to higher land productivity [32]. Furthermore, the adoption of agricultural machinery enables timely operations such as field preparation and planting, which contributes to improving crop yields [5,22]. In particular, timely mechanized operations help mitigate weather-related risks, thereby reducing potential crop yield losses [33].

Despite the potential of mechanization to increase winter wheat output through both sown area expansion and yield improvement, as illustrated in Figure 1, concerns remain regarding its environmental and socioeconomic challenges. One environmental concern is that land expansion driven by mechanization may come at the cost of forests and savannahs, contributing to climate change and biodiversity loss [27,34]. Additionally, using heavy mechanization can lead to soil erosion and compaction, causing soil degradation and ultimately reducing yields, thereby threatening the long-term stability of food security [22,35]. Socioeconomically, wealthier and larger farms are more likely to afford mechanization, enabling them to expand at the expense of smallholders [36], potentially exacerbating inequality in land and wealth distribution [37]. The relative importance of these positive and negative effects remains an unresolved empirical question, warranting further investigation.

This study used county-level agricultural production data to examine the impact of agricultural mechanization on winter wheat output, focusing on the relative contributions of sown area expansion and yield improvement (the hierarchy of administrative divisions in China includes provincial, prefectural, county, and township, with the “county” being the basic administrative unit [38]). By identifying the dominant mechanism driving the mechanization–output relationship, it provides deeper insights into agricultural output growth patterns and informs policies aimed at promoting sustainable mechanized winter wheat production.

3. Materials and Methods

3.1. Model Construction of the Impact of Agricultural Mechanization on Winter Wheat Output

We begin by proposing the following regression specification to estimate the causal effect of agricultural mechanization on winter wheat output:

$$Ln Q_{gt}={\beta }_0+{\beta }_{1}Ln M_{gt}+{\beta }_2{\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}+{\beta }_3{\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}+h\left(t\right)+c_g+u_{gt}$$

(1)

where $Q_{gt}$ is the winter wheat output (in 10,000 t) for county $g$ in year $t$. $M_{gt}$ denotes the level of agricultural mechanization, measured by the total agricultural machinery power used in winter wheat production. The parameter of interest is ${\beta }_1$, which is expected to be positive. ${\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}$ is an array of weather variables, including average temperature, cumulative precipitation, and average solar duration during the winter wheat growing season. The winter wheat growing season, defined as from September to the following June, represents the typical nature of winter wheat in most production areas, i.e., planting in September–October and harvesting in May–June [39].

${\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}$ is a set of other productive inputs that may influence winter wheat output, including total fertilizer usage, total labor force, and total cropland area, all in logarithmic form. Furthermore, the ratio of irrigated area, defined as the ratio of irrigated areas to total winter wheat sown area, is included in ${\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}$ to account for farmers’ adjustment to weather shocks [40]. $h\left(t\right)$ represents provincial-level time trends that capture the technological advances associated with winter wheat output, such as the adoption of new crop varieties and conservation tillage techniques. $c_g$ denotes county-level fixed effects, and $u_{gt}$ is an error term with zero mean.

In our baseline specification, standard errors ($u_{gt}$) are clustered at both the county level and province-by-year level. This two-way clustering strategy allows for the consideration of spatial correlations across counties within each province–year cell, as well as serial correlations over years within each county [41].

3.2. Decomposing the Impact of Agricultural Mechanization on Winter Wheat Output

By decomposing the impact of mechanization on winter wheat output, it is feasible to delve into the primary pathway driving this relationship. Winter wheat output can be expressed as the product of the sown area and the corresponding yield per unit area, represented mathematically as:

$$Q=A\times Y$$

(2)

where $Q$, $A$, and $Y$ denote winter wheat output, sown area, and yield, respectively. Previous studies have consistently demonstrated that the sown area of grain crops is primarily driven by the total agricultural machinery power, as it determines the capacity for large-scale land preparation and planting, while the intensity of agricultural machinery usage, reflecting the efficiency and frequency of mechanized operations, plays a critical role in enhancing crop yields [17,18]. Based on these findings, sown area ($A$) and yield ($Y$) can be modeled as functions of mechanization:

$$A=f\left(M\right)$$

(3)

$$Y=f\left(m\right)$$

(4)

where $M$ represents the total agricultural machinery power and $m$ represents the intensity of agricultural machinery use. The relationship between $M$ and $m$ can be written as:

$$M=A\times m$$

(5)

Taking the derivative of $Q$ with respect to the intensity of $m$ yields:

$$\begin{gathered} {\displaystyle \frac{\partial Q}{\partial m}}=\underbrace{{\displaystyle \frac{\partial Q}{\partial A}}\times {\displaystyle \frac{\partial A}{\partial M}}\times {\displaystyle \frac{\partial M}{\partial m}}}+\underbrace{{\displaystyle \frac{\partial Q}{\partial Y}}\times {\displaystyle \frac{\partial Y}{\partial m}}} \\ \hspace{1em}\hspace{1em}\hspace{1em} Area effect \hspace{1em}\hspace{1em}\hspace{1em}Yield effect \end{gathered}$$

(6)

Equation (6) intuitively decomposes the overall effects of agricultural mechanization on winter wheat output into two components: the area effect and the yield effect. The area effect captures changes in winter wheat output attributable to variations in the sown area resulting from agricultural mechanization, while the yield effect captures output changes driven by mechanization-induced yield variations. Quantifying these contributions requires estimating the marginal effects of agricultural mechanization on winter wheat sown area as well as its yield. To achieve this, we establish the following empirical specifications to separately estimate the impact of agricultural mechanization on the sown area and yield of winter wheat:

$$Ln A_{gt}={\gamma }_0+{\gamma }_{1}Ln M_{gt}+{\gamma }_2{\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{A}}+{\gamma }_3{\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{A}}+h\left(t\right)+c_g+{\epsilon }_{gt}$$

(7)

$$Ln Y_{gt}={\alpha }_0+{\alpha }_{1}Ln m_{gt}+{\alpha }_2{\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}+{\alpha }_3{\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{Y}}+h\left(t\right)+c_g+{\vartheta }_{gt}$$

(8)

In Equation (7), $A_{gt}$ denotes the winter wheat sown area in county $g$ at time $t$. The parameters $\mathit{\gamma }$ are to be estimated. ${\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{A}}$ represents a vector of weather variables, including 10-year moving averages of average growing-season temperature, cumulative precipitation, and average solar duration, to capture the impact of past weather experiences on winter wheat sown areas. Experience shapes farmers’ expectations about local climate change, which in turn influence their sowing decisions [42]. ${\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{A}}$ is a vector of socioeconomic variables, including the ratio of lagged wheat price to labor price, the ratio of lagged wheat price to fertilizer prices [43], total cropland area in logarithmic form, and population density in logarithmic form [44]. The remaining variables are defined as in Equation (1).

In Equation (8), $Y_{gt}$ represents the yield of winter wheat (t ha⁻¹) in county $g$ at year $t$. The parameters, denoted $\mathit{\alpha }$, are to be estimated. $m_{gt}$ denotes winter wheat growing-season agricultural machinery usage intensity. The weather covariate ${\mathit{W}}_{\mathit{g}\mathit{t}}^{\mathit{Q}}$ consists of average temperature, cumulative precipitation, and average solar duration during the winter wheat growing season. ${\mathit{Z}}_{\mathit{g}\mathit{t}}^{\mathit{Y}}$ includes other productive inputs, such as fertilizer usage intensity and labor input intensity, all in logarithmic form, as well as the ratio of irrigated area. Descriptive statistics and regression analyses were performed using Stata 16.0.

4. Data

4.1. Agricultural Production Data

We used an imbalanced panel dataset covering 912 counties in China for the period 1998–2016 to examine the impact of agricultural mechanization on winter wheat output. County-level data on agricultural production, including total output and sown area of winter wheat, total agricultural machinery power, total fertilizer usage, and irrigated areas from 1998 to 2016, were provided by the Institute of Agricultural Information at the Chinese Academy of Agricultural Sciences (CAAS). Our analysis covers eight provinces: Hebei, Shandong, Jiangsu, Anhui, Hubei, Henan, Shaanxi, and Shanxi. These provinces are major regions for winter wheat production and mechanization development [19], accounting for over 85% of the national winter wheat output during the study period. Additionally, provincial-level data were incorporated to address the lack of county-specific data. Data such as total labor use for winter wheat production, wheat price, fertilizer price, and labor price were collected from the National Agricultural Product Cost–Benefit Data Compilation. All correlated prices are reported in 1998 prices. Population density data were obtained from the Provincial Statistical Yearbook.

Although the CAAS provides data on total agricultural machinery power for all crops at the county level and provincial-level data on crop-specific machinery usage intensity is available from the National Agricultural Product Cost–Benefit Data Compilation, the lack of county-level crop-specific machinery usage data presents a major limitation for empirical analysis. To address this issue, we developed a maximum-entropy procedure to impute the machinery usage intensity for winter wheat at the county level, leveraging the aforementioned data. This enables the calculation of the total agricultural machinery power specifically related to winter wheat production. The results show that the annual difference between the total agricultural machinery power reported by the CAAS and the aggregated ones imputed through the maximum-entropy procedure is less than 1%. This implies that the imputed machinery usage data can reflect actual use of machinery for winter wheat production in each county. See Supplementary Materials for more details on this procedure.

4.2. Weather Data

Daily weather data were obtained from the China Meteorological Data Sharing Service System, which provided weather information, including maximum, minimum, and average temperatures, precipitation, and solar duration, collected from 820 weather stations across China between 1998 and 2016. Consistent with established practices in prior research examining climate change impacts on agriculture at the county level in China [39,45,46], we used an inverse distance weighting (IDW) method to convert station-level weather data to county-level weather data. This allowed for the calculation of annual weather variables specific to the winter wheat growing season. A summary of the statistics for these variables is provided in

Table 1. Summary statistics of variables used in the analysis for 1998–2016 for all counties.

Variables	Mean	SD	Min	Max
A. Economic variables
Winter wheat output (10,000 t)	11.309	11.885	0.001	42.164
Winter wheat sown area (10,000 ha)	2.231	2.110	0.001	8.435
Winter wheat yield (t ha−1)	4.537	1.822	0.006	12.031
Total agricultural machinery power (10,000 kW)	17.927	17.148	0.001	106.710
Total fertilizer usage (10,000 t)	0.754	0.719	0.001	5.000
Total labor force (10,000 days)	217.024	218.996	0.004	1724.474
Total cropland area (10,000 ha)	4.622	3.141	0.043	15.646
Ratio of irrigated area (%)	41.116	18.690	3.705	98.927
Population density (residents km−2)	448.232	173.172	174.733	798
Machinery usage intensity (kW ha−1)	8.796	3.960	1.260	24.963
Fertilizer usage intensity (t ha−1)	0.360	0.098	0.118	0.660
Labor input intensity (day ha−1)	99.312	35.079	42.750	214.050
Wheat price (CNY kg−1)	1.349	0.218	0.765	1.772
Fertilizer price (CNY kg−1)	4.244	1.258	2.139	6.725
Labor price (CNY day−1)	23.633	16.217	6.998	56.679
B. Weather variables
Average temperatureQ (°C)	10.590	2.107	0.051	15.083
Cumulative precipitationQ (1000 mm)	0.378	0.232	0.023	2.342
Average solar durationQ (hours)	5.620	0.988	1.927	8.171
Average temperatureA (°C)	10.404	2.085	−0.550	14.193
Cumulative precipitationA (1000 mm)	0.370	0.201	0.036	1.637
Average solar durationA (hours)	5.707	0.944	2.195	7.886

The superscript Q denotes the weather variables during the winter wheat growing season of the current year, defined as the period from September to June of the following year. These variables are primarily used to examine the impact of weather variations on winter wheat yield or output. The superscript A represents the 10-year moving average of weather variables during the same winter wheat growing season (September to June of the following year). Variables with the superscript A are primarily used to examine the impact of weather variations on the winter wheat sown area. The number of observations was 15,352. CNY—Chinese yuan.

.

5. Results and Discussion

5.1. Analysis of the Impact of Agricultural Mechanization on Winter Wheat Output

5.1.1. Main Results

Table 2. Effect of agricultural mechanization on winter wheat output.

Variables	Ln Winter Wheat Output
	(1)	(2)	(3)
Ln total agricultural machinery power	0.8156 ***	0.1540 ***	0.1458 ***
	(0.0256)	(0.0323)	(0.0325)
Ln total fertilizer usage		0.2205 ***	0.2157 ***
		(0.0639)	(0.0631)
Ln total labor force		0.4975 ***	0.5096 ***
		(0.0634)	(0.0628)
Ln cropland area		0.0488 **	0.0466 **
		(0.0192)	(0.0191)
Ratio of irrigated area		0.0005	0.0005
		(0.0005)	(0.0005)
Average temperatureQ			−0.0014
			(0.0102)
Cumulative precipitationQ			−0.1124 *
			(0.0571)
Average solar durationQ			0.0134
			(0.0127)
Constant	−0.1418 **	−1.3348 ***	−1.4032 ***
	(0.0567)	(0.3869)	(0.3888)
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	15,352

This table shows the estimated results of the impact of agricultural mechanization on winter wheat output. The dependent variable in all columns is the logarithm of winter wheat output. Column (1) includes the key explanatory variable while controlling for provincial-level time trends and county fixed effects. Columns (2) and (3) progressively incorporate other productive inputs and weather variables to further refine the model specification. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05, * $p$ < 0.1.

presents the results of estimating the impact of agricultural mechanization on winter wheat output. Column (1) includes provincial-level time trends and county fixed effects, while columns (2) and (3) expand the model by adding variables for productive inputs and weather conditions. The results show that across all model specifications, agricultural mechanization has a significant positive impact on winter wheat output. Specifically, the regression results in column (3), which includes the full set of control variables, reveal that a 10% increase in total agricultural machinery power causes a 1.5% rise in winter wheat output, while holding all other variables constant. Using provincial-level wheat production data from China, Chandio et al. [47] reported similar results, finding that a 10% increase in agricultural machinery power consumption resulted in a 1.1% increase in wheat output.

We used the baseline estimates in column (3) of Table 2 to evaluate the contribution of increased level of agricultural mechanization to the growth of winter wheat output during the period 1998–2016. We began by calculating the changes in winter wheat output for each county, which is achieved by multiplying the estimated coefficient of the total agricultural machinery power by its growth rate and the mean value of winter wheat output. The aggregated increase in national winter wheat output over the sample period is then determined by summing these output changes across all sample counties. The results indicate that increased total agricultural machinery power leads to an increase of 921 million t of winter wheat output, averaging an annual increase of 48 million t, accounting for approximately 46% of China’s annual winter wheat output during the study period.

Regarding the control variables, total fertilizer usage, total labor input, and total cropland area exhibit significant positive effects on winter wheat output, aligning with the findings of Dessale [48]. As per the estimates presented in column (3) of Table 2, a 10% increase in total fertilizer usage, total labor force, and total cropland area corresponds to increases of 2.2%, 5.1%, and 0.5% in winter wheat output, respectively. However, growing-season cumulative precipitation has a significant negative impact on winter wheat output, which is consistent with the findings reported by McCarl et al. [49].

5.1.2. Robustness Checks

To further support the baseline results, we conducted several robustness checks. First, we examined whether the positive impact of agricultural mechanization on winter wheat output is affected by methods used to control for spatial correlation. In the baseline regression specification, the two-way clustering strategy takes into account the spatial correlation between counties within one province in a given year. However, this approach may overlook spatial correlation between observations in adjacent counties across different provinces. As alternative strategies, we controlled for spatial correlation in the regression model using two more conservative methods: (1) a first- or second-order spatial contiguity matrix in the spatial error model (SEM) and the spatial autoregression model (SAR) to better capture spatial correlation between adjacent counties (refer to Yi et al. [50] for the definition of the first-order and second-order contiguity matrices); and (2) Conley standard errors, which account for spatial correlation within a 300 km radius and serial correlation over a 5-year period [51,52]. These results are displayed in the middle section of Figure 2. The estimated coefficients for total agricultural machinery power resemble the baseline results, shown at the top of Figure 2, in both magnitude and statistical significance, implying that our results are robust.

Second, we assessed the sensitivity of the baseline estimates to alternative weather variables. For temperature variables, we first used a piecewise linear approach to construct two growing degree day ($GDD$) variables, i.e., ${GDD}_{l_0:l_1}$ and ${GDD}_{l_1:l_{\mathrm{\infty }}}$, to capture the nonlinear relationship between temperature and winter wheat output [53]. The lower temperature bound ($l_0)$ is fixed at 0 °C, while the upper threshold ($l_1)$ is determined by looping through all possible thresholds between 20 °C and 40 °C, with 32 °C providing the best fit. Therefore, we used ${GDD}_{0–32 °\mathrm{C}}$ and ${GDD}_{32 °\mathrm{C}+}$ as alternative temperature variables to re-estimate Equation (1). Other weather variables, such as cumulative precipitation^Q and its quadratic term and average solar duration^Q and its quadratic term, are also controlled for in the model.

Third, we examined how sensitive the estimates are to the threshold definition in the former approach by directly using ${GDD}_{8–32 °\mathrm{C}}$ and ${GDD}_{32 °\mathrm{C}+}$ as alternative temperature variables [54]. Additionally, we re-estimated Equation (1) using seasonal weather variables, following Tack et al. [55] and Wang et al. [33], as the aggregated weather variables over the growing season of winter wheat may obscure the seasonal variation effects of weather shocks. The seasonal weather variables include average minimum and maximum temperatures, cumulative precipitation, average solar duration in fall, winter, and spring, as well as continuous harvest-season precipitation (fall is a period defined as winter wheat planting to November, winter is a period defined as December to the following February, and spring is a period defined as March to winter wheat maturity). Estimates of these three specifications, presented in the bottom section of Figure 2, show that the key estimates are qualitatively similar to our baseline results.

Fourth, we assessed the robustness of the maximum-entropy procedure used to impute county-level machinery usage intensity for winter wheat. Since the estimated impact of agricultural mechanization on winter wheat may be influenced by the lower and upper bounds of machinery usage intensity derived from the maximum-entropy procedure, we followed agronomists’ recommendations and tested less restrictive bounds. Specifically, we expanded the lower and upper limits by 30% and 50%, respectively, relative to the primary estimates. The results presented in

Table 3. Robustness of causal effects of agricultural mechanization on winter wheat output under various machinery usage intensity bounds.

Variables	Ln Winter Wheat Output
	Narrow Bound	Wide Bound
	(1)	(2)
Ln total agricultural machinery power	0.1577 ***	0.1368 ***
	(0.0400)	(0.0359)
Other productive inputs	Yes	Yes
Weather variables	Yes	Yes
Provincial time trends	Yes	Yes
County fixed effect	Yes	Yes
Observations	15,352	15,352

This table presents the estimation results of the impact of agricultural mechanization on winter wheat output when using different machinery data imputed using the maximum-entropy procedure. The variation in machinery data is achieved by adjusting the bounds of machinery usage intensity in Equation (S5) of the Supplementary Material. Specifically, the bounds for machinery usage intensity are set as 0.7–1.3 (narrow bound) and 0.5–1.5 (wide bound), respectively. The dependent variables are the logarithm of winter wheat output in all columns. Other productive inputs include total fertilizer usage, total labor force, and total cropland area, all in logarithmic form, as well as the ratio of irrigated area. Weather variables include average temperature^Q, cumulative precipitation^Q, and average solar duration^Q. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01.

show that the direction and magnitude of total agricultural machinery power are similar to those in our preferred estimation specification, shown in column (3) of Table 2. This indicates that our main findings are robust to alternative bounds of machinery usage intensity.

Fifth, an endogeneity issue may arise due to reverse causality, as higher winter wheat output might incentivize farmers to increase their use of agricultural machinery. To address this issue, we used the instrumental variable (IV) method, where the average of the total agricultural machinery power in the previous year in neighboring counties is used as an instrument for the current total agricultural machinery power in each county. As reported in column (1) of

Table A3. Robustness of causal effect of agricultural mechanization on winter wheat output using instrumental variables method.

Variables	IV-2SLS Estimation
	First Stage	Second Stage
	(1)	(2)
Ln lagged average of total agricultural machinery power in neighboring counties	0.1923 ***
	(0.0356)
Ln total agricultural machinery power		0.3521 **
		(0.1354)
Ln total fertilizer usage	0.3980 ***	0.1018
	(0.0733)	(0.0731)
Ln total labor force	0.3707 ***	0.4145 ***
	(0.0676)	(0.0865)
Ln cropland area	0.0199	0.0672 ***
	(0.0149)	(0.0225)
Ratio of irrigated area	0.0001	0.0006
	(0.0003)	(0.0005)
Average temperatureQ	−0.0077	0.0045
	(0.0107)	(0.0094)
Cumulative precipitationQ	−0.0623	−0.0827
	(0.0909)	(0.0579)
Average solar durationQ	0.0265	0.0096
	(0.0199)	(0.0125)
Provincial time trends	Yes	Yes
County fixed effect	Yes	Yes
F-statistic	29.1895
Observations	13,656	13,656

For column (1), the dependent variable is the logarithm of total agricultural machinery power, and for column (2), it is the logarithm of winter wheat output. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05.

in Appendix A, the instrument is a significant predictor of total agricultural machinery power, satisfying the correlation requirement for IVs. Furthermore, the first-stage F statistic is 29, well above the threshold of 10, indicating that the IV is not weak [56]. Column (2) of Table A3 shows the second-stage estimation results, where the coefficient for total agricultural machinery power remains statistically significant, reinforcing the robustness of our findings.

In addition, the IV must satisfy the exclusion restriction assumption, meaning it should only affect winter wheat output through the endogenous variable (i.e., the current total agricultural machinery power). However, in the case of just identification, where the number of instruments equals the number of endogenous variables, the exogeneity of the IV cannot be statistically tested [57]. To address this limitation, we used the local-to-zero (LTZ) approximation method proposed by Conley et al. [58] and its related implementations [59] to test the sensitivity of our estimates when the IV deviates from perfect exogeneity. This method relaxes the exclusive constraint on the IV. Accordingly, when the IV is approximately exogenous, an effective estimator can be obtained by specifying a prior distribution of the correlation coefficients between the disturbance term and the IV. The results in column (1) of

Table A4. Robustness test of causal effect of agricultural mechanization on winter wheat output using local-to-zero (LTZ) method.

Variables	Ln Winter Wheat Output	Ln Winter Wheat Sown Area	Ln Winter Wheat Yield
	(1)	(2)	(3)
Ln total agricultural machinery power	0.4367 ***	1.0380 ***
	(0.0436)	(0.0325)
Ln machinery usage intensity			0.0503 *
			(0.0276)
Other variables	Yes	Yes	Yes
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	13,656	13,656	13,893

Other variables in column (1) include total fertilizer usage, total labor force, and total cropland area, all in logarithmic form, the ratio of irrigated area, average temperature^Q, cumulative precipitation^Q, and average solar duration^Q. Other variables in column (2) include the ratio of wheat price to labor price, the ratio of wheat price to fertilizer price, cropland area in logarithmic form, population density in logarithmic form, average temperature^A, cumulative precipitation^A, and average solar duration^A. Other variables in column (3) include fertilizer usage intensity and labor input intensity, both in logarithmic form, the ratio of irrigated area, average temperature^Q, cumulative precipitation^Q, and average solar duration^Q. “Yes” indicates that other variables, provincial time trends, and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, * $p$ < 0.1.

in Appendix A show that our IV estimates remain robust, even under the assumption that the IV may locally violate the exclusion restriction.

5.1.3. Heterogeneity Analysis

This subsection further investigates the heterogeneity of the relationship between agricultural mechanization and winter wheat output under different environmental conditions, focusing on topographic and transport conditions.

First, topographic conditions play a critical role in shaping the uneven regional development of agricultural mechanization in China. Compared with plains, hilly and mountainous terrains significantly reduce the accessibility and operational efficiency of agricultural machinery in the field [60]. This may suggest that mechanization is more effective in increasing winter wheat output in the plains. Thus, we divide the sample into plains, hills and mountains based on terrain conditions and re-estimate Equation (1) for each group. Columns (1) and (2) of

Table 4. Heterogeneity analysis.

Variables	Topographic Conditions		Transport Conditions
	Plains	Hills and Mountains	High Highway Density	Low Highway Density
	(1)	(2)	(3)	(4)
Ln total agricultural machinery power	0.1734 ***	0.1306 ***	0.1874 ***	0.0521
	(0.0428)	(0.0418)	(0.0375)	(0.0600)
Other productive inputs	Yes	Yes	Yes	Yes
Weather variables	Yes	Yes	Yes	Yes
Provincial time trends	Yes	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes	Yes
Observations	6806	5439	7916	7436

Other productive inputs include total fertilizer usage, total labor force, and cropland area, all in logarithmic form, as well as the ratio of irrigated area. Weather variables include average temperature^Q, cumulative precipitation^Q, and average solar duration^Q. “Yes” indicates that other productive inputs, weather variables, provincial time trends, and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. The number of observations in the topographic conditions groups is reduced to 12,245 due to the exclusion of certain sample counties from the classification provided in the China Statistical Yearbook (Township). *** $p$ < 0.01.

report the results, indicating that the positive relationship between agricultural mechanization and winter wheat output is significant across all terrain types, but this relationship is relatively stronger in the plains.

Second, the relationship between agricultural mechanization and winter wheat output may also vary depending on transport conditions, which is motived by evidence of an association between traffic conditions and mechanization development [61]. Following Wang et al. [33], we used highway density as a proxy variable for transport conditions and divided the sample into two groups, high and low highway density. Equation (1) was then re-estimated separately for these two groups, with the results reported in columns (3) and (4) of Table 4. The positive relationship between agricultural mechanization and winter wheat output is statistically significant in areas with high highway density, while it is no longer significant in areas with low highway density.

5.2. Mechanism Analysis

In this section, we delve into the underlying mechanisms driving this positive impact, focusing on the following two aspects.

5.2.1. Winter Wheat Sown Area

A possible mechanism by which agricultural mechanization positively affects winter wheat output is that it expands the area sown to winter wheat. To empirically test this mechanism, we estimated Equation (7). As can be seen in

Table 5. Effect of agricultural mechanization on winter wheat sown area.

Variables	Ln Winter Wheat Sown Area
	(1)	(2)	(3)
Ln total agricultural machinery power	0.9374 ***	0.9433 ***	0.9442 ***
	(0.0183)	(0.0173)	(0.0170)
Ratio of wheat price to labor price		2.6985 ***	2.6736 ***
		(1.0319)	(0.9823)
Ratio of wheat price to fertilizer price		0.6188 **	0.5982 *
		(0.3108)	(0.3102)
Ln cropland area		0.0112	0.0119
		(0.0169)	(0.0171)
Ln population density		0.9986	1.0797
		(1.1405)	(1.1075)
Average temperatureA			−0.0031
			(0.0185)
Cumulative precipitationA			−0.7112 **
			(0.3125)
Average solar durationA			−0.1217 **
			(0.0558)
Constant	−1.6017 ***	−8.2546	−7.7153
	(0.0412)	(6.8535)	(6.6747)
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	15,352

This table shows the estimated results of the impact of agricultural mechanization on winter wheat sown area. The dependent variables are the logarithm of winter wheat sown area in all columns. Column (1) includes the key explanatory variable while controlling for provincial time trends and county fixed effects. Columns (2) and (3) progressively incorporate socioeconomic variables and weather variables to further refine the model specification. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05, * $p$ < 0.1.

, the coefficient for total agricultural machinery power is positive and statistically significant, providing evidence that mechanization contributes to the enlargement of winter wheat sown areas. Specifically, the results in column (3) show that a 10% increase in the total agricultural machinery power is associated with a 9.4% increase in the winter wheat sown area.

Given the scarcity of land resources, the expansion in winter wheat sown area often necessitates a reduction in the cultivation of other crops. To investigate whether agricultural mechanization drives such changes, we re-estimated Equation (7) using two alternative dependent variables: the ratio of winter wheat sown area to the total sown area of all crops and the ratio of other crops sown area to the total sown area of all crops (given the available data, we divided all crops into nine categories: rice, wheat, corn, soybeans, potatoes, cotton, oil crops, sugar crops, and vegetables and fruits. The “other crops” category includes the remaining eight major crops, excluding wheat). The coefficient of the total agricultural machinery power in column (1) of

Table 6. Effect of agricultural mechanization on the share of winter wheat sown area.

Variables	Share of Winter Wheat Sown Area to Total Sown Area of All Crops	Share of Other Crops Sown Area to Total Sown Area of All Crops	Share of Rapeseed Sown Area to Total Sown Area of All Crops
	(1)	(2)	(3)
Ln total agricultural machinery power	0.0500 ***	−0.0500 ***	−0.0101 ***
	(0.0048)	(0.0048)	(0.0027)
Socioeconomic variables	Yes	Yes	Yes
Weather variables	Yes	Yes	Yes
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	9962

Based on the available data, all crops are grouped into nine categories: rice, wheat, corn, soybeans, potatoes, cotton, oil crops, sugar crops, and vegetables and fruits. The “other crops” category includes the remaining eight major crops, excluding wheat. Socioeconomic variables include ratio of wheat price to labor price, ratio of wheat price to fertilizer price, cropland area in logarithmic form, and population density in logarithmic form. Weather variables include average temperature^A, cumulative precipitation^A, and average solar duration^A. “Yes” indicates that socioeconomic variables, weather variables, provincial time trends, and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. The observations in column (3) decrease to 9962, because not all sample counties cultivate rapeseed. *** $p$ < 0.01.

is positive and statistically significant, while that in column (2) is negative and statistically significant. These results suggest that mechanization expands the winter wheat sown area by reducing the share of arable land devoted to other crops.

To further explore land use competition among crops grown in the same season, we examined the effect of mechanization on the share of area sown to competing crops in winter wheat. Specifically, we re-estimated Equation (7) using the ratio of rapeseed sown area to the total area sown to all crops [62]. As indicated in column (3) of Table 6, mechanized winter wheat production significantly reduces the share of rapeseed sown area. These findings suggest that agricultural mechanization promotes adjustment of cropping structure by reducing the cultivation of labor-intensive and less mechanization-friendly crops, such as rapeseed, while increasing the sown area of winter wheat. A possible explanation for rapeseed’s limited adaptability to mechanization is its higher harvest losses associated with combined harvesting methods, caused by uneven maturity, pod shattering, and high temperatures during the harvest period [63]. This finding aligns with conclusions drawn by Qiao [18], who demonstrated that mechanization increases the area sown with grain crops while decreasing that of non-grain crops. While such shifts in cropping patterns driven by agricultural mechanization may reduce crop diversity and thereby increase vulnerability to environmental and market risks [11,64], further research exploring the linkages between farm mechanization, crop diversity, and consumption diversity is essential to better understand whether these concerns are justified [65].

The finding that mechanization drives the expansion in winter wheat sown areas is robust to several sensitivity tests. Our first robustness check involved adjustments to the $K$-year moving average of weather variables. The estimation results for $K=$3, $K=$5, and $K=$15, as presented in

Table A5. Robustness of causal effect of agricultural mechanization on winter wheat sown area using alternative weather variables.

Variables	$\mathit{K}$-Year Moving Average of Weather Variables
	3-Year	5-Year	15-Year
	(1)	(2)	(3)
Ln total agricultural machinery power	0.9426 ***	0.9420 ***	0.9465 ***
	(0.0172)	(0.0173)	(0.0169)
Socioeconomic variables	Yes	Yes	Yes
Weather variables	Yes	Yes	Yes
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	15,352

Socioeconomic variables include the ratio of wheat price to labor price, the ratio of wheat price to fertilizer price, cropland area in logarithmic form, and population density in logarithmic form. Weather variables include average temperature, cumulative precipitation, and average solar duration during winter wheat growing season. “Yes” indicates that socioeconomic variables, weather variables, provincial time trends, and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01.

in Appendix A, provide strong evidence for the positive impact of agricultural mechanization on winter wheat sown area. Second, to address the potential endogeneity in the key variable, we used an instrumental variable method. The instrument for total agricultural machinery power is the average of the total agricultural machinery power of their neighboring counties in the previous year. The first-stage results in

Table A6. Robustness of causal effect of agricultural mechanization on winter wheat sown area using instrumental variable method.

Variables	IV-2SLS Estimation
	First Stage	Second Stage
	(1)	(2)
Ln lagged average of total agricultural machinery power in neighboring counties	0.7463 ***
	(0.0386)
Ln total agricultural machinery power		1.0198 ***
		(0.0172)
Ratio of wheat price to labor price	1.2431	2.9967 ***
	(0.8781)	(0.8602)
Ratio of wheat price to fertilizer price	−0.4187	0.3699
	(0.2346)	(0.2617)
Ln cropland area	0.0979 ***	0.0033
	(0.0270)	(0.0169)
Ln population density	−0.3840	−0.7352
	(0.6319)	(0.9086)
Average temperatureA	0.0064	−0.0046
	(0.0175)	(0.0166)
Cumulative precipitationA	−0.3204	−0.2693
	(0.2819)	(0.2715)
Average solar durationA	0.0683	−0.1262 **
	(0.0429)	(0.0557)
Provincial time trends	Yes	Yes
County fixed effect	Yes	Yes
F-statistic	373.0217
Observations	13,656	13,656

For column (1), the dependent variable is the logarithm of total agricultural machinery power, and for column (2), it is the logarithm of winter wheat sown area. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05.

of Appendix A show that the F-statistic exceeds 10, indicating that there is no weak instrument problem. The coefficient of interest, reported in column (2) of Table A6, remains positive and statistically significant, further confirming the robustness of our results. In addition, Column (2) of Table A4 in Appendix A reports the estimated results using the LTZ method. The results continue to support a positive effect of agricultural mechanization on winter wheat sown area, even under the assumption that IV may violate the exclusion restriction locally.

5.2.2. Winter Wheat Yield

The second way in which agricultural mechanization affects winter wheat output is through its impact on winter wheat yield, as described in Section 2. To further examine this relationship, Equation (8) was estimated. The results presented in

Table 7. Effect of agricultural mechanization on winter wheat yield.

Variables	Ln Winter Wheat Yield
	(1)	(2)	(3)
Ln machinery usage intensity	0.1005 ***	0.0768 **	0.0721 **
	(0.0327)	(0.0315)	(0.0318)
Ln fertilizer usage intensity		0.1851 ***	0.1835 ***
		(0.0619)	(0.0614)
Ln labor input intensity		−0.0205	−0.0067
		(0.0761)	(0.0769)
Ratio of irrigated area		0.0015 ***	0.0014 ***
		(0.0005)	(0.0005)
Average temperatureQ			−0.0034
			(0.0089)
Cumulative precipitationQ			−0.0696
			(0.0505)
Average solar durationQ			0.0139
			(0.0103)
Constant	1.0514 ***	1.3674 ***	1.2864 ***
	(0.0598)	(0.4097)	(0.4422)
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	15,352

This table shows the estimated results of the impact of agricultural mechanization on winter wheat yield. The dependent variables are the logarithm of winter wheat yield in all columns. Column (1) includes the key explanatory variable while controlling for provincial time trends and county fixed effects. Columns (2) and (3) progressively incorporate other productive inputs and weather variables to further refine the model specification. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05.

show a significant positive relationship between mechanization and winter wheat yield. The coefficients estimated in column (3), with the full set of controls in the model specification, indicate that a 10% increase in machinery usage intensity is associated with a 0.7% increase in winter wheat yield. This finding is consistent with evidence from other developing countries that demonstrated the positive impact of agricultural mechanization on enhancing crop productivity [5,17].

Table 8. Robustness of causal effect of agricultural mechanization on winter wheat yield using alternative weather variables.

Variables	Ln Winter Wheat Yield
	$\mathit{G}\mathit{D}\mathit{D}$ Variables		Seasonal Weather Variables
	(1)	(2)	(3)
Ln machinery usage intensity	0.0736 **	0.0729 **	0.0803 **
	(0.0316)	(0.0314)	(0.0341)
Other productive inputs	Yes	Yes	Yes
Weather variables	Yes	Yes	Yes
Provincial time trends	Yes	Yes	Yes
County fixed effect	Yes	Yes	Yes
Observations	15,352	15,352	15,352

Other productive inputs include fertilizer usage intensity in logarithmic form, labor input intensity in logarithmic form, and the ratio of irrigated area. Weather variables in column (1) include ${GDD}_{0–32 °\mathrm{C}}$, ${GDD}_{32 °\mathrm{C}+}$, cumulative precipitation^Q and its quadratic term, and average solar duration^Q and its quadratic term. Weather variables in column (2) include ${GDD}_{8–32 °\mathrm{C}}$, ${GDD}_{32 °\mathrm{C}+}$, cumulative precipitation^Q and its quadratic term, and average solar duration^Q and its quadratic term. Weather variables in column (3) include average minimum and maximum temperature, cumulative precipitation, and average solar radiation in the fall, winter, and spring, as well as continuous winter wheat harvest-season precipitation. “Yes” indicates that other productive inputs, weather variables, provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. ** $p$ < 0.05.

presents the estimation results of the robustness tests using alternative weather variables. Following the approach outlined in Section 5.1.2, column (1) incorporates weather variables such as ${GDD}_{0–32 °\mathrm{C}}$, ${GDD }_{32 °\mathrm{C}+}$, cumulative precipitation^Q and its quadratic term, as well as average solar duration^Q and its quadratic term. In column (2), ${GDD}_{0–32 °\mathrm{C}}$ is substituted with ${GDD}_{8–32 °\mathrm{C}}$, while retaining the other weather variables from column (1). Column (3) includes seasonal weather variables. The results consistently show that the positive impact of agricultural mechanization on winter wheat yield remains robust and is not materially affected.

To address potential endogeneity issues, the relationship between agricultural mechanization and winter wheat yield was further examined using the instrumental variable method. Specifically, the one-year lagged average machinery usage intensity in neighboring counties was used as an instrument for the current machinery usage intensity in each county. The first-stage estimation results, presented in column (1) of

Table A7. Robustness of causal effect of agricultural mechanization on winter wheat yield using instrumental variable method.

Variables	IV-2SLS Estimation
	First Stage	Second Stage
	(1)	(2)
Ln lagged average of machinery usage intensity in neighboring counties	0.5089 ***
	(0.0623)
Ln machinery usage intensity		0.1531 *
		(0.0874)
Ln fertilizer usage intensity	0.2235 ***	0.1735 ***
	(0.0512)	(0.0630)
Ln labor input intensity	−0.2530 *	0.0864
	(0.1201)	(0.0905)
Ratio of irrigated area	0.0005	0.0015 **
	(0.0003)	(0.0006)
Average temperatureQ	−0.0117	0.0014
	(0.0089)	(0.0090)
Cumulative precipitationQ	0.0219	−0.0675
	(0.0675)	(0.0512)
Average solar durationQ	0.0125	0.0153
	(0.0157)	(0.0108)
Provincial time trends	Yes	Yes
County fixed effect	Yes	Yes
F-statistic	66.7197
Observations	13,893	13,893

For column (1), the dependent variable is the logarithm of machinery usage intensity, and for column (2), it is the logarithm of winter wheat yield. “Yes” indicates that provincial time trends and county fixed effects are controlled for in the model. Standard errors, enclosed in parentheses, are two-way clustered at the county level and at the province-by-year level. *** $p$ < 0.01, ** $p$ < 0.05, * $p$ < 0.1.

in Appendix A, show that the F-statistic of 67 exceeds the critical threshold of 10, supporting the strength of the instrument. The estimation results of the second stage in column (2) show that the positive relationship between agricultural mechanization and winter wheat yield holds when using an instrumental variable estimation. In addition, the estimated results of the LTZ method in column (3) of Table A4 in Appendix A show that our IV estimates remain robust under the assumption that IV may violate the exclusion restriction locally.

5.3. Decomposition Results

Building on the previously estimated marginal effects of agricultural mechanization on the sown area and yield of winter wheat, we decomposed the overall effect of agricultural mechanization on winter wheat output into area effect and yield effect components according to Equation (6).

Table 9. Decomposition results of marginal effects of agricultural mechanization on winter wheat output for a 1 kW ha⁻¹ increase in machinery usage.

Various Channels	Winter Wheat Output
	(t)	(%)
Output increase	14,115	100
Area effect	13,286	94
Yield effect	829	6

These marginal effects are calculated using Equation (6) in Section 3.2 based on the average values of winter wheat output, sown area, and yield for the years 1998–2016, as well as the estimated coefficients presented in Section 5.2.

presents the decomposition results of contributions from the area and yield effects to mechanization-driven changes in winter wheat output during the period of 1998–2016. The results indicate that an increase of 1 kW ha⁻¹ in machinery usage for winter wheat production leads to 14,115 t increase in winter wheat output. Approximately 94% of this growth is attributed to the expansion in the sown area, while the remaining 6% stems from improvements in yield. These findings indicate that the expansion in sown area is the primary driver of winter wheat output increases associated with advances in agricultural mechanization over the study period. This result is not surprising since agricultural mechanization primarily functions as a labor-saving technology, with minimal impacts on land productivity unless it is combined with high-yield and efficient agronomic practices designed to optimize yield attributes and increase wheat yield [36,37,66].

6. Conclusions and Implications

Taking winter wheat as an example, this study used county-level panel data from China to empirically examine the impacts and underlying mechanisms of mechanization on agricultural output. The results indicate that mechanization significantly increases winter wheat output, with a 10% increase in total agricultural machinery power leading to a 1.5% rise in winter wheat output. However, the impact of mechanization on winter wheat output is heterogeneous due to different environmental conditions. This positive impact is greater in the plains, where transport conditions are better. By decomposing the overall impact of mechanization on winter wheat output into its area and yield effects, we find that the output growth driven by mechanization is predominantly attributed to the expansion in sown area.

These findings provide valuable insights for policy discussions on enhancing winter wheat output and securing food supplies through mechanization in China as well as other developing countries. The positive impact of agricultural mechanization on winter wheat output underscores the urgency of its promotion and widespread adoption. Particularly in regions experiencing agricultural labor shortages, mechanization can ensure the timeliness of winter wheat production, thereby mitigating labor-related threats to food security. Furthermore, developing specialized machinery for hilly and mountainous areas would improve the accessibility and operational efficiency. The government should collaborate with relevant institutions to support private-sector research and development, as well as fund applied research focused on designing machinery tailored to the specific needs of these regions. Targeted investments in transportation infrastructure could reduce machinery access costs, facilitating broader adoption of mechanization. These measures would further enhance the benefits of mechanization for winter wheat production.

Given that mechanization primarily drives output growth by expanding the sown area, optimizing land-use policies becomes imperative. Key measures include refining land transfer systems to enable efficient land allocation and advancing farmland consolidation to support mechanized operations. In regions with severe land fragmentation, policies should prioritize land consolidation by simplifying land transfer and consolidation processes. Educational campaigns should raise awareness among farmers of the benefits of land consolidation and encourage the voluntary merging of small plots into larger, contiguous units suitable for mechanized operations. Addressing farmland abandonment and encouraging the cultivation of marginal lands with arable potential are also practical options to increase cultivated land area.

Additionally, the relatively limited contribution of mechanization to yield enhancement emphasizes the need to integrate mechanization with complementary agronomic technologies, such as conservation tillage and precision agriculture. Future agricultural policies should prioritize the synergistic development of mechanization and innovative agronomic technologies, including the adoption of suitable varieties, optimized planting schedules, and improved wheat establishment methods, to enhance winter wheat yields and ensure long-term stability in food supplies.

While this study provides valuable insights into the impact of agricultural mechanization on winter wheat output, it is important to recognize some limitations. Due to the differences in planting patterns among China’s three major crops (maize, rice, and wheat), the relative importance of mechanization may vary for other crops [67]. Future research should explore the effects of mechanization on other key crops to gain a more comprehensive understanding of its role in food production.