Diversity of Agricultural Production and Food Consumption in Rural China: A Dual Analysis of Expenditure and Dietary Structure

Abstract

As rural residents face the dual challenges of transforming dietary structures and addressing nutritional health burdens, establishing a resilient food consumption system for rural households has become an urgent priority. Drawing on micro-level data from the China Land Economic Survey (CLES) for the period 2020–2022, this study employs two-way fixed effects models, an instrumental variable (IV) approach, and seemingly unrelated regression (SUR) techniques to examine the impact of agricultural production diversity on household food expenditure and dietary diversity, as well as the underlying mechanisms. The results reveal that agricultural production diversity yields a significant and robust dual-dividend effect within household food consumption systems: it not only reduces per capital food expenditure but also enhances dietary diversity. Mechanism analysis indicates that diversified production increases food self-sufficiency, thereby reducing cash outflows for essential food items, while simultaneously improving dietary diversity through increased agricultural income and greater agricultural commercialization. Heterogeneity analysis further shows that these effects are more pronounced in villages lacking rural industrial support and among non-ageing households. These findings suggest that, in contexts where market mechanisms remain underdeveloped, the uncritical pursuit of absolute agricultural specialization may not align with the livelihood and nutritional needs of rural residents. From the perspective of fostering a healthy and resilient food system, China should adopt differentiated agricultural support policies, encourage rural households to maintain an appropriate degree of production diversity, and strengthen local agricultural market infrastructure.

1. Introduction

Against the backdrop of the comprehensive drive to revitalise rural areas, the incomes of China’s rural households are gradually rising, and concerns about food security are evolving from merely ensuring basic sustenance to promoting a healthy diet [1]. Since the full attainment of a moderately prosperous society in 2021, people have secured an adequate food supply [2], but they now face the dual challenge of transforming their dietary habits and addressing nutritional health concerns [3]. Meanwhile, the tangible disparities in dietary upgrading between urban and rural areas highlight the distinct challenges faced by residents in different settings during the process of improving dietary structures. Urban residents, supported by well-developed retail and cold-chain systems, as well as stable wage incomes, are able to conveniently access a diverse range of food products from across the country and even internationally through market channels [4,5]. Rural areas, constrained by distribution costs and information access, suffer from dietary imbalances and insufficient intake of high-quality protein, fruit, and vegetables [6], leading to micronutrient deficiencies. Malnutrition, including hidden hunger [7], remains widespread in rural areas. For rural residents, health challenges are particularly pronounced at the household level [8,9,10]. On the one hand, food price volatility and accelerated marketisation have increased the cost of obtaining nutritious food for rural households, thereby squeezing out other opportunities for developmental consumption [11,12]. On the other hand, a monotonous, low-cost diet can no longer meet the growing nutritional demands of these households [13]. In this context, establishing a sustainable consumption system that both alleviates the economic burden on rural households and improves food quality has become an urgent issue that needs to be addressed.

Tracing the root causes of the dietary challenges faced by rural households inevitably leads back to historical transformations in agricultural production patterns. For generations, traditional agriculture operated within a hybrid system combining semi-subsistence production with partial market participation. Households secured their food supply through diversified farming, thereby maintaining an internal cycle of production and consumption [14]. With the shift of labour away from agriculture and the accelerating commercialisation of the sector, a growing divergence has emerged between production and consumption decisions in rural areas. In the pursuit of economies of scale, agricultural production has become increasingly specialised, resulting in a marked decline in diversity and the gradual erosion of the traditional “grow what you eat” model [15]. Alongside this transformation, scholarly debate over the effects of specialisation and diversification in agricultural production on rural diets has intensified. On the one hand, drawing on the theory of comparative advantage, some scholars emphasise the positive effects of specialised production. By increasing agricultural productivity and the degree of commercialisation, specialised production can raise rural incomes, thereby enabling access to a wider variety of food through market purchases [16]. However, the extent of this benefit is heavily contingent on the presence of well-developed local markets and low transaction costs. In rural China, where market infrastructure remains underdeveloped and cold-chain logistics are constrained, excessive specialisation not only limits rural households’ direct access to food but also exposes them to greater market risks and transaction costs [17]. At the same time, a growing body of literature has begun to reassess the supportive role and nutritional value of agricultural production diversity [18,19]. In contrast to specialisation, which is primarily oriented towards profit maximisation, diversification—viewed as a strategy for risk mitigation and more efficient resource allocation—can effectively cushion the effects of market volatility. Moreover, production diversification may help to circumvent the constraints imposed by imperfect rural markets by enabling households to obtain a wider range of foods through self-provisioning, thereby contributing to food security among rural populations [20]. Against this backdrop, it is of considerable practical importance to examine the overall impact of agricultural production diversity on food consumption systems.

There remains no clear consensus within the academic community as to how agricultural production diversity influences food consumption among rural residents [21]. Existing research has extensively examined the welfare effects of agricultural production, broadly following two relatively distinct strands. One adopts an economic perspective, considering whether agricultural production models can generate higher farm income, alleviate poverty, or reduce food expenditure [22,23,24]; the other draws on public health and nutritional perspectives, exploring the potential benefits of production diversity for dietary diversity and nutritional outcomes among rural households [25,26]. Notwithstanding these contributions, the literature is subject to several notable limitations. In particular, most studies treat the economic cost of food and its nutritional quality as separate concerns, and thus fail to examine, within a unified analytical framework, how agricultural production diversity operates across both dimensions. In the context of underdeveloped rural markets in China, production and consumption decisions are closely intertwined. Single-dimensional analyses are therefore insufficient to capture the full decision-making chain from farm to table, and tend to overlook the potential dual benefits of agricultural production in reducing economic expenditure whilst improving dietary quality. Moreover, at the mechanistic level, existing studies have yet to provide a systematic examination and comparison of the underlying transmission pathways [27]. To complement the existing literature, this study seeks to examine in depth the impacts and underlying mechanisms of agricultural production diversity on rural households’ food consumption systems. Drawing on data from the China Land Economy Survey (CLES), an empirical analysis is undertaken to assess its dual effects on household food consumption, specifically its role in enhancing dietary diversity whilst reducing economic expenditure. The study further elucidates the mechanisms through which rural households respond to the combined dietary and economic pressures arising under conditions of imperfect markets. This analysis not only sheds light on the complex decision-making processes within rural households but also provides valuable empirical evidence to inform the formulation of more inclusive agricultural policies and the reconfiguration of modern agri-food systems.

This paper makes three main contributions. First, it is the first study to incorporate both the reduction in food expenditure and the improvement in dietary quality within a single analytical framework. As noted above, previous research has largely focused on either the economic or the nutritional dimension, without offering a comprehensive account of the interaction between the two. This study addresses this gap by systematically evaluating, using micro-level household survey data, the combined effects of agricultural production diversity on reducing the cost of food and enhancing dietary diversity. Second, the paper further disentangles the underlying transmission mechanisms through which agricultural production diversity shapes household food consumption. Household production and consumption are not governed by a single logic, yet existing studies have not adequately captured the complexity of these pathways. This study not only examines the direct effects but also decomposes the transmission process into two channels—self-sufficiency and market-based income generation—and provides a comparative assessment of their respective roles within the food consumption system. Third, the study extends the empirical context by adopting a heterogeneous perspective. By taking into account China’s specific agricultural conditions and socio-economic characteristics, and by incorporating two dimensions—the level of rural industrial development and the degree of household ageing—it demonstrates how the effects of production diversity on food consumption vary across different macro-level contexts and internal household structures. This analysis provides detailed micro-level evidence to support the development of a more resilient and health-oriented food system.

2. Theoretical Analysis

2.1. Theoretical Framework and Research Hypotheses

The theoretical analysis in this paper is grounded in the realities of market imperfections in rural China [28]. Existing research suggests that, throughout the transformation of China’s agricultural and food systems, production structures, market conditions, and household food procurement practices have remained closely interconnected [29]; this relationship is particularly pronounced in rural areas where market development is uneven [30]. Across large parts of rural China, constraints related to geographical location, transport infrastructure, cold-chain logistics, and access to information continue to impose high costs on rural households seeking to access and utilise markets. As a result, household production and consumption cannot be fully separated [31]. Accordingly, for rural households, agricultural production is not only a source of operating income but also directly shapes the types of food available to them and the channels through which it is obtained.

Given this context, agricultural production diversity shapes food consumption by influencing the ways in which households obtain food. Compared with specialised production, diversification entails that rural households allocate their resources across a broader range of crop and livestock activities, subject to constraints in land, labour, and capital, thereby reshaping the composition of household food sources [32]. On the one hand, diversification increases the range of foods directly available to households, reducing reliance on market purchases through self-production and self-consumption [33]. On the other hand, it may further broaden household food choices by improving resource utilisation, enhancing production stability, and increasing agricultural income. Consequently, agricultural production diversity may affect not only the level of household food expenditure but also the diversity of dietary patterns. More specifically, it exerts a dual effect on rural household food consumption. At the expenditure level, diversification reduces the cash outlay required to meet food needs by strengthening the household’s internal food supply capacity and substituting for market purchases. At the dietary level, it not only enriches the range of foods available but also expands consumption possibilities through income effects, thereby promoting a shift towards more diverse dietary patterns. On this basis, the following hypotheses are proposed.

H1. 

Agricultural production diversity generates significant dual benefits for household food consumption: it reduces per capital food expenditure while at the same time enhancing dietary diversity.

A closer examination of the mechanism through which agricultural production diversity reduces food expenditure highlights the internal supply substitution effect it generates. For rural households with limited incomes, food expenditure is often the most inelastic and difficult to reduce among daily expenses, and, given the coexistence of market transaction costs and liquidity constraints, the role of household self-sufficiency becomes particularly important. As production structures become more diversified, rural households are not only able to satisfy their basic grain requirements but may also achieve self-sufficiency in vegetables, poultry and eggs, and in some cases, meat [34]. This enhanced level of self-provisioning creates a stable, non-monetised supply channel within the household. As the share of home-grown food consumed rises, items that were previously purchased with cash are now supplied internally, reducing households’ reliance on market purchases and, in turn, lowering cash expenditure on food. Thus, the impact of agricultural production diversity is not merely reflected in a greater variety of produce, but in the restructuring of the household food procurement cost structure through higher rates of food self-sufficiency. On this basis, this study advances its second hypothesis.

H2. 

Agricultural production diversity reduces household cash spending on food by increasing the degree of household food self-sufficiency.

Improvements in dietary structure brought about by agricultural production diversity do not rely solely on the logic of self-sufficiency within rural households. Further upgrading of dietary patterns often requires access to markets beyond the household. In addition to the direct consumption of self-produced food, diversified farming and cultivation can also indirectly broaden households’ food choices by raising income and strengthening their links with markets [35]. Diversified production facilitates a more efficient allocation of production factors across seasons, alleviating the resource misallocation and underutilisation associated with monocultural systems. It also helps to mitigate both natural and market risks to some extent, thereby enhancing the level and stability of agricultural income. The resulting increase in household income strengthens purchasing power, enabling rural households to acquire foodstuffs that they cannot produce themselves or that are unavailable year-round, such as off-season fruits and vegetables, processed foods, or high-protein items [36]. At the same time, diversification of production is often accompanied by a higher degree of market participation. As rural households sell surplus produce, they gradually acquire price information, trading experience, and market networks, thereby effectively reducing the information search costs and transactional uncertainties associated with food purchases. Consequently, agricultural diversification not only enhances household consumption capacity through income effects but also improves access to a wider range of foods by increasing integration into the market. The theoretical mechanism diagram for this paper is shown in Figure 1. On this basis, this paper proposes a third hypothesis.

H3. 

Agricultural production diversity enhances household dietary diversity by raising agricultural income and promoting the commercialization of farm production.

2.2. The Household Production and Consumption Model

To translate the theoretical hypotheses outlined above into a more rigorous microeconometric framework, this paper introduces and extends the classical household production and consumption model [37]. Traditional consumer demand theory typically treats households as purely consuming units that make choices from the goods available in a given market, paying limited attention to internal household production processes—from raw ingredients to the preparation of final meals. In the rural context of incomplete markets considered here, households function as both food producers and consumers; their consumption decisions aim to maximise household food consumption utility through the rational allocation of resources, subject to constraints on household production, time, and income. The basic framework of this model can therefore be defined as the maximisation of the household utility function:

$$U=U(Z_1,Z_2,\dots ,Z_N)$$

(1)

$$\mathrm{s}.\mathrm{t}.Z_i=Z_i(x_i,\mathit{Apd},t_{i1},t_{i2},\dots ,t_{\mathit{im}}), i=1,2,\dots ,n$$

(2)

$$T_k=l_k+\sum _{i=1}^n{t}_{\mathit{ik}},k=1,2,\dots ,m$$

(3)

$$I=I(\mathit{Apd})=\sum _{k=1}^m{w}_k{\left(\mathit{Apd}\right)l}_k+v,{\displaystyle \frac{\mathsf{\partial }I}{\mathsf{\partial }\mathit{Apd}}}\ge 0$$

(4)

$$\sum _{i=1}^n{p}_i{x}_i+C_{\mathit{other}}\le I(\mathit{Apd})$$

(5)

Equation (1) represents the household’s final utility function. $Z_i$ denotes the household’s final consumption of dietary bundle $i$, the richness of which directly corresponds to dietary diversity (DDS).

Equation (2) specifies the household dietary production constraint. Each category of household food $Z_i$ is jointly determined by purchased market food $x_i$, household members’ time allocated to domestic tasks $t_{\mathit{ik}}$, and the household’s own agricultural production diversity. A higher $\mathit{Apd}$ indicates that the household has access to a wider range of food sources and can improve dietary quality through internal food supply.

Equation (3) is the time constraint. $T_k$ represents the total time available to household member $k$, $l_k$ denotes the time allocated to agricultural and non-agricultural labour, and ${\displaystyle \sum _{i=1}^n{t}_{ik}}$ is the time devoted to household food production. This constraint reflects the trade-offs in internal time allocation: as more time is assigned to income-generating activities, the time available for food preparation and cooking is reduced, potentially affecting dietary quality.

Equation (4) defines the income constraint. $I$ denotes total household income, $w_k\left(Apd\right)$ represents the average return to labour for household member $k$ and is influenced by the level of agricultural production diversity, and $v$ denotes non-labour income. This setup corresponds to Hypothesis 3: agricultural production diversity, as a mechanism for optimising resource allocation and mitigating risk, can enhance overall household income by generating economies of scope and improving labour allocation efficiency.

Equation (5) represents the expenditure allocation. Total household income is distributed between market food consumption and other non-food expenditures. $p_i$ denotes the market price of food, ${\displaystyle \sum _{i=1}^n{p}_i{x}_i}$ represents total household expenditure on food, and $C_{\mathit{other}}$ denotes expenditure on other non-food items.

By solving the household food utility maximisation problem outlined above, subject to the combined constraints of production, time, and income, dietary diversity and food expenditure of rural households can be expressed in simplified reduced-form equations as functions of agricultural production diversity and other household characteristics.

$$\mathit{DDS}= f\left(\mathit{Apd},I,T,\mathit{ConVar}\right)$$

(6)

$$\mathit{Expenditure}=g\left(\mathit{Apd},I,T,\mathit{ConVar}\right)$$

(7)

Here, $T$ represents the household’s labour and time allocation, while $\mathit{ConVar}$ denotes a set of control variables. Equations (6) and (7) explicitly and rigorously illustrate the dual mechanisms through which agricultural production diversity affects the household food consumption system. On the one hand, diversified production expands the supply of internally produced food via the self-sufficiency substitution channel in Equation (2), thereby enhancing dietary diversity and reducing food expenditure within a given budget. On the other hand, diversified production increases total household income through the income-enhancement channel in Equation (4), relaxing budget constraints and providing households with greater purchasing power to acquire high-value, nutritious foods. Consequently, agricultural production diversity can reduce expenditure via the substitution effect, while simultaneously raising the consumption of higher-quality foods through the income effect. This theoretical model thus provides a robust foundation for the empirical analyses that follow.

3. Data and Model Configuration

3.1. Data Sources

The data employed in this empirical study are drawn from the China Land Economic Survey (CLES). This database not only provides rigorous data cleaning and validation reports to ensure data quality, but its questionnaires also capture household crop and livestock production structures down to the plot level, along with detailed records of household food expenditure and income. These features align closely with the study’s requirements for observing core variables such as agricultural production diversity, food expenditure, and dietary diversity.

The research period is defined as 2020–2022. This choice is partly dictated by the availability of micro-level longitudinal data: the CLES baseline survey was initiated in 2020, and the most recently published official data currently cover up to 2022. Additionally, 2020 marked the culmination of China’s efforts to comprehensively eradicate absolute poverty and to establish a moderately prosperous society in all respects. This period represents a critical juncture in rural China, characterised by the onset of comprehensive rural revitalisation and an upgrading of consumption structures, further complicated by logistics disruptions and market frictions triggered by the sudden public health emergency. The exceptionally high external transaction costs and price volatility observed during this period correspond closely to the real-world constraints outlined in the theoretical framework of this study, providing a clear temporal window for testing the proposed hypotheses.

Regarding survey methodology, the data were collected using the Proportional to Population Size (PPS) sampling method, covering 13 prefecture-level cities and 52 administrative villages within Jiangsu Province, thereby representing the socio-economic characteristics of the region’s rural areas. During data pre-processing, household- and village-level survey data from 2020 to 2022 were merged. After rigorously excluding samples with missing key information, logical inconsistencies, or extreme outliers, the dataset was consolidated into an unbalanced panel comprising 4439 valid observations for subsequent empirical analysis.

3.2. Model Configuration

• Benchmark Regression. This paper utilizes CLES data to examine the impact of agricultural production diversity on rural household food consumption systems. Food expenditure and dietary diversity are selected as dependent variables, representing economic cost and dietary quality dimensions, respectively. With respect to the empirical approach, this paper employs the least-squares dummy variable (LSDV) estimator, incorporating two-way fixed effects for both region and time, as the benchmark specification. This choice is motivated by two main considerations. First, from a methodological standpoint, the core explanatory variable—agricultural production diversity—has a continuous distribution rather than being driven by exogenous binary policy shocks; consequently, quasi-natural experimental designs, such as difference-in-differences, are not suitable. Second, given the heterogeneity in natural endowments and socio-economic conditions across regions, alongside temporal fluctuations in the macroeconomic environment, the use of unbalanced panel data within a two-way fixed-effects framework is more effective than cross-sectional regression for controlling both unobservable regional heterogeneity and time-varying effects. Overall, the two-way fixed-effects specification substantially mitigates omitted variable bias, thereby enhancing the robustness and explanatory power of the core parameter estimates. The specific formula is as follows:

  $$Y_{\mathit{it}}={\alpha }_0+{\alpha }_1{\mathit{Apd}}_{\mathit{it}}+\beta {\mathit{ConVar}}_{\mathit{it}}+{\eta }_c+r_t+{\epsilon }_{\mathit{it}}$$

  (8)

  where $Y_{\mathit{it}}$ is the dependent variable, specifically represented by two indicators in the text: ① $\mathit{ln}\left({\mathit{Expenditure}}_{\mathit{it}}\right)$ denotes the natural logarithm of household i’s food expenditure in year t, used to measure household food expenditure; ② ${\mathit{DietDiversity}}_{\mathit{it}}$ denotes the dietary diversity index (DDS) for household i in year t, used to measure household dietary quality.
  ${\mathit{Apd}}_{\mathit{it}}$ is the core explanatory variable, representing the household’s agricultural production diversity. ${\mathit{ConVar}}_{\mathit{it}}$ represents a series of control variables. ${\alpha }_1$ is the regression coefficient of the core independent variable. ${\alpha }_0$ is the intercept term, $\beta$ is the vector of regression coefficients for the control variables. ${\eta }_c$ is the region fixed effect. In the empirical analysis, regional dummy variables are introduced for control, aiming to absorb regional heterogeneities such as geographical environment and dietary culture that do not change over time. $r_t$ is the time fixed effect, controlled by introducing year dummy variables, aiming to remove the impact of macro time trends such as price fluctuations and policy adjustments. ${\epsilon }_{\mathit{it}}$ represents the random disturbance term.

2.

Addressing Endogeneity Issues. The aforementioned benchmark model may suffer from endogeneity issues stemming from reverse causality. Therefore, the instrumental variables approach is employed to resolve bidirectional causality concerns, utilizing two-stage least squares (2SLS) regression. The first-stage regression model is as follows:

$${\mathit{Apd}}_{\mathit{it}}=\mu +\delta {\mathit{ApdIV}}_{\mathit{it}}+\beta {\mathit{ConVar}}_{\mathit{it}}+{\eta }_c+r_t+{\epsilon }_{\mathit{it}}$$

(9)

In Equation (9), ${\mathit{ApdIV}}_{\mathit{it}}$ is the instrumental variable for agricultural production diversity, and $\delta$ is the regression coefficient of the instrumental variable.

The second stage regression model is as follows:

$$Y_{\mathit{it}}=\mu +{\alpha }_1\widehat{{\mathit{Apd}}_{\mathit{it}}}+\beta {\mathit{ConVar}}_{\mathit{it}}+{\eta }_c+r_t+{\epsilon }_{\mathit{it}}$$

(10)

In Equation (10), $\widehat{{\mathit{Apd}}_{\mathit{it}}}$ is the fitted value obtained from the first-stage estimation, and ${\alpha }_1$ is the estimated coefficient of the fitted value.

3.

Mediation effect model. To gain a deeper understanding of the mechanisms through which agricultural production diversity affects food systems, it should be noted that recent microeconometric research has identified the traditional ‘three-step’ mediation analysis—the classic causal steps approach proposed by Baron and Kenny (1986) [38], which sequentially tests the relationships among the independent variable, mediator, and dependent variable—as being prone to bias arising from the endogeneity of the mediating variable, potentially leading to misleading conclusions [38,39]. To ensure the scientific rigour and robustness of the mechanism analysis, this paper adopts the framework proposed by Jiang Ting (2022) in modern microeconomics [40]. It departs from the conventional testing steps, which are vulnerable to endogeneity concerns, and instead implements a more robust ‘two-step method’ for mechanism testing: namely, after establishing the main effect of the core explanatory variable, the analysis proceeds to examine its causal impact on the mediating variable. Building on this approach, the present paper extends and formulates the following mechanism testing model within the framework of the baseline specification. Building on Equation (8), we extend and specify the following model:

$$M_{\mathit{it}}={\phi }_0+{\phi }_1{\mathit{Apd}}_{\mathit{it}}+\beta {\mathit{ConVar}}_{\mathit{it}}+{\eta }_c+r_t+{\epsilon }_{\mathit{it}}$$

(11)

In Equation (11), $M_{\mathit{it}}$ is the mechanism variable, ${\phi }_0$ is the intercept term, and ${\phi }_1$ is the regression coefficient, reflecting the degree of impact of the core explanatory variable, agricultural production diversity, on the mechanism variable.

3.3. Variable Specifications

• Dependent variables. To comprehensively examine the multidimensional impact of agricultural production diversity on household food consumption, in terms of both economic cost and dietary structure, this study takes food expenditure and dietary diversity as the dependent variables. On the one hand, food expenditure is used to measure the economic burden of household food consumption. These data are based on respondents’ recall of their households’ purchases of various food items, consumption of self-produced food and eating-out expenditure over the past week. The reported figures are then aggregated and annualised. Since expenditure data typically exhibit a right-skewed distribution, and to mitigate heteroskedasticity, per capita food expenditure is log-transformed. On the other hand, dietary diversity is used to capture the nutritional quality and consumption structure of rural households’ diets. This paper employs the Dietary Diversity Score (DDS) methodology recommended by the Food and Agriculture Organization of the United Nations (FAO) for measurement [41]. First, in terms of the observation period, household food consumption displays a certain degree of randomness and seasonality. To smooth out occasional fluctuations in consumption on any given day, this study considers respondents’ food consumption over the previous seven days, so as to better reflect households’ usual dietary patterns [42]. Second, in constructing the indicator, we referred to the 2022 edition of the Chinese Dietary Guidelines and classified foods into eight groups according to their nutritional attributes: grains and root vegetables, vegetables, fruits, meat, dairy products, eggs, aquatic products, and legumes and nuts. A binary counting method was used for the calculation [43]. If the household consumed any food from a given category within the past seven days, the score for that group is set to 1; otherwise, it is 0. Summing the scores across all eight groups yields the household’s overall dietary diversity score. A higher score indicates a greater variety in the household’s food intake and a more balanced nutritional profile.
• Core explanatory variable. The core explanatory variable in this study is agricultural production diversity, which is intended to capture the richness of rural households’ agricultural production structures. Following the approach adopted in the existing literature [44,45], it is measured by counting the number of crop species cultivated and livestock species reared by each household over the preceding year, with the totals combined into a single count variable. Simple data processing is applied to ensure an accurate assessment, and no complex transformations are performed. Higher values of this indicator correspond to greater diversity in households’ grain and vegetable baskets.
• Mechanism variables. To examine more deeply the mechanisms through which agricultural production diversity affects food consumption systems, and to empirically test the two potential transmission channels—subsistence substitution and income growth—discussed above, two sets of mechanism variables are introduced. First, for the self-sufficiency substitution effect aimed at easing the economic burden, the grain self-sufficiency rate is used as a proxy variable. It is defined as the proportion of self-produced grain in a household’s daily consumption of that food category. This indicator directly reflects a household’s food security capacity: higher values indicate lower dependence on external markets and, consequently, a greater ability to reduce monetary food expenditure through non-market channels. Second, with regard to the income growth effect on dietary quality, agricultural income and the degree of agricultural commercialization are selected as mechanism variables. Log-transformed agricultural income is used to measure net operating income, while the level of agricultural commercialization is captured by the ratio of agricultural product sales to total output value [46]. These variables are intended to reflect rural households’ economic access and market integration, and are used to test whether agricultural production diversification eases liquidity constraints and strengthens households’ purchasing power for diverse foods by increasing agricultural income or raising the level of commercialization.
• Control variables. To minimise endogeneity bias arising from omitted variables, the selection of control variables strictly follows the principle of exogeneity. These variables are drawn from three dimensions: characteristics of household heads and decision-makers, household resource endowments, and the village’s external environment. First, with respect to the characteristics of household heads and decision-makers, we include agricultural training, women’s empowerment and educational attainment. Agricultural training reflects the accumulation of human capital by the household head. Households whose heads have received modern agricultural training are more likely to adopt scientifically informed crop and livestock management practices, which may also foster greater awareness of nutritional health, thereby directly shaping the household’s dietary composition. Women’s empowerment captures the extent to which women participate in household decision-making. Existing research suggests that women are more inclined to prioritise nutritional quality and health when allocating resources, thereby exerting a substantial influence on household food consumption decisions [47]. which has a certain influence on food consumption decisions. Educational attainment is measured by the highest number of years of schooling among household members, in order to control for the effect of knowledge endowment on consumption attitudes. Second, in terms of household resource endowments and facilities, we include labour structure, productive assets, refrigerator ownership and the installation of water purifiers. Labour structure reflects the intensity of household involvement in agricultural production. Higher levels of agricultural labour input generally indicate stronger household management capacity, increasing the likelihood that rural households engage in diversified crop and livestock activities, which in turn influence household food consumption patterns via the self-production mechanism. Productive assets serve as a proxy for household physical capital, capturing the quantity of assets deployed in agricultural production. Existing research suggests that both labour input and physical capital affect the efficiency of factor allocation and labour productivity in agriculture [48], thereby boosting household income and purchasing power, which in turn enhances the household’s economic capacity to improve dietary quality and diversify food consumption. Refrigerators, as important durable consumer goods, can alter households’ food storage and procurement patterns. The use of water purifiers reflects rural households’ concern for living standards and health, which can, to some extent, influence food consumption patterns. Finally, with respect to the external village environment, food accessibility is included as a control variable, proxied by the presence of e-commerce delivery stations in the village. Improved ecommerce infrastructure effectively broadens rural households’ food procurement channels, reduces search costs, and constitutes an important external factor shaping their food consumption behaviour [49]. Descriptions and descriptive statistics for each variable are presented in

  Table 1. Descriptive Statistics of Key Variables.

  Variable Name	Variable Declaration	Mean	SD
  Food expenditure	Per capita food expenditure of rural households (Yuan, logarithmic)	8.105	0.831
  Dietary diversity	Number of food types consumed by rural households in a week	6.645	1.314
  Agricultural production diversity	Total number of different agricultural product types produced by households	1.711	1.415
  Food self-sufficiency Rate	Proportion of self-produced grain in the daily diet of rural households	0.305	0.409
  Agricultural income	Income obtained from agricultural operations by rural households (Yuan, logarithmic)	3.158	4.292
  Agricultural commercialization	Proportion of agricultural products sold through the market by rural households	0.379	0.434
  Agricultural training	Whether the household head has received agricultural technical education or training: 1 = Yes; 0 = No	0.302	0.459
  Women’s empowerment	Whether there are females among household management decision-makers: 1 = Yes; 0 = No	0.274	0.446
  Labor structure	Proportion of household members participating in agricultural labor	0.499	0.413
  Education level	Maximum years of education in the rural household	11.212	4.072
  Refrigerator	Whether the rural household has a refrigerator: 1 = Yes; 0 = No	0.978	0.148
  Water purifier	Whether the rural household has installed a water purifier: 1 = Yes; 0 = No	0.224	0.417
  Productive assets	Total number of productive assets owned by the rural household (Count)	0.530	2.254
  Food accessibility	Whether the village where the household resides has an e-commerce delivery station: 1 = Yes; 0 = No	0.597	0.491

  .

4. Empirical Findings

4.1. Benchmark Regression

Table 2. The Impact of Agricultural Production Diversification on Food Expenditure.

	(1)	(2)	(3)	(4)	(5)	(6)
	Food Expenditure			Dietary Diversity
Agricultural production diversity	−0.027 ***	−0.046 ***	−0.031 ***	0.032 **	0.042 ***	0.060 ***
	(0.009)	(0.010)	(0.010)	(0.014)	(0.015)	(0.015)
Agricultural training		0.037	0.092 ***		0.210 ***	0.181 ***
		(0.027)	(0.027)		(0.041)	(0.041)
Women’s empowerment		0.043	0.012		0.048	0.022
		(0.028)	(0.027)		(0.042)	(0.042)
Labor structure		0.080 **	0.148 ***		−0.240 ***	−0.232 ***
		(0.033)	(0.032)		(0.051)	(0.051)
Education level		0.028 ***	0.023 ***		0.057 ***	0.056 ***
		(0.003)	(0.003)		(0.005)	(0.005)
Refrigerator		0.153 *	0.137 *		0.950 ***	0.864 ***
		(0.082)	(0.081)		(0.153)	(0.152)
Water purifier		0.120 ***	0.107 ***		0.397 ***	0.376 ***
		(0.030)	(0.029)		(0.042)	(0.042)
Productive assets		0.016 **	0.014 *		0.003	0.004
		(0.008)	(0.008)		(0.008)	(0.007)
Food accessibility		0.071 ***	0.042		0.110 ***	0.175 ***
		(0.025)	(0.027)		(0.039)	(0.044)
Regional fixed effects			YES			YES
Time fixed effects			YES			YES
Constant	8.151 ***	7.576 ***	7.818 ***	6.590 ***	4.891 ***	5.065 ***
	(0.019)	(0.084)	(0.094)	(0.032)	(0.160)	(0.172)
Number	4439	4439	4439	4439	4439	4439
R2	0.002	0.036	0.119	0.001	0.093	0.112

Note: Brackets indicate robust standard errors. *** p < 0.01, ** p < 0.05, * p < 0.1.

reports the overall impact of agricultural production diversity on household food consumption systems, focusing on food expenditure and dietary diversity. To ensure the robustness and precision of the estimation results, a stepwise regression strategy is adopted. Regressions are estimated under three specifications: without control variables, with control variables, and with the additional inclusion of fixed effects. In particular, columns (3) and (6) control for both region and time fixed effects, thereby effectively absorbing time-invariant regional endowment differences and time-varying macroeconomic shocks. These two columns constitute the main basis for interpreting the regression results.

First, agricultural production diversity significantly reduces the economic burden of household food consumption and has a marked effect in lowering food expenditure. As shown in columns (1) to (3) of Table 2, agricultural production diversity has a significantly negative impact on food expenditure among rural households. In the specification that includes control variables and two-way fixed effects (column 3), the coefficient on the core explanatory variable is −0.031 and is statistically significant at the 1% level. Therefore, holding other factors constant, a one-unit increase in a household’s agricultural production diversity is associated with an average reduction of approximately 3.1% in per capita food expenditure. This finding reflects the semi-subsistence, semi-market nature of production and livelihoods among smallholder households in rural China. In practice, because market systems in rural areas are less developed than in urban areas, rural households still face substantial transaction costs when purchasing food, including longer travel distances and information search costs. Agricultural diversification therefore enables rural households to substitute market purchases with produce from their own crops and livestock. These self-produced agricultural products can meet household food consumption needs at very low cost, reduce dependence on external markets and, in turn, lower household food expenditure.

Second, agricultural production diversity significantly enhances household dietary diversity and has a clear positive effect on dietary variety. According to the regression results in columns (4) to (6) of Table 2, agricultural production diversity exerts a significantly positive impact on HDDS. Column (6) reports a regression coefficient of 0.060, which is statistically significant at the 1% level. This implies that, for each one-unit increase in agricultural production diversity, the average household dietary diversity score rises by 0.06 points. Third, the regression results with control variables reveal heterogeneous effects on food consumption. Examining the coefficients of the control variables in columns (3) and (6) shows that different household characteristics exert markedly different impacts on both the quantity and the quality of food consumption. First, labour structure exhibits a dual effect: it significantly increases food expenditure while markedly reducing dietary diversity. This finding is consistent with theories of energy conservation and the opportunity cost of time. Household members engaged in intensive agricultural work expend more energy, prompting families to purchase larger quantities of high-calorie staple foods. Meanwhile, the demanding nature of such work may crowd out time spent on cooking, leaving households with insufficient time to prepare more elaborate meals and thereby reducing dietary variety. Second, refrigerators and water purifiers have a significant positive impact on food consumption. As symbols of household wealth and living standards, refrigerators effectively extend the shelf life of food, enabling families to store meat, dairy products, eggs, and fresh fruit and vegetables for longer periods. This makes them a key factor in improving dietary quality. Finally, the significant positive effect of food accessibility on dietary diversity indicates that well-developed market infrastructure—particularly e-commerce distribution stations—can effectively overcome geographical barriers and provide rural households with broader channels for food choice.

In summary, by comparing the regression results on reduced food expenditure and improved dietary diversity, this study finds that agricultural production diversity generates a significant dual-dividend effect on the food consumption system. It not only lowers economic costs by reducing market-based expenditure, but also enhances dietary diversity in the process. This asymmetric effect clearly demonstrates that, in contemporary rural China, promoting moderate agricultural production diversification functions not only as a risk-mitigation strategy, but also as an effective pathway to achieving low-cost, high-level livelihood security and food safety.

4.2. Robustness Test

To further verify the reliability and consistency of the benchmark regression results presented above, robustness tests are conducted along three dimensions: adjusting the time sample, replacing the core explanatory variables, and changing the estimation method. The corresponding results are reported in

Table 3. Robustness Test Results.

	Adjustment Time Sample		Replace the Core Explanatory Variable		Replacement of the Dependent Variable	SUR Model
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Food Expenditure	Dietary Diversity	Food Expenditure	Dietary Diversity	Protein Intake Diversity	Food Expenditure	Dietary Diversity
Agricultural production diversity	−0.041 ***	0.066 ***			0.053 ***	−0.031 ***	0.060 ***
	(0.011)	(0.017)			(0.013)	(0.009)	(0.015)
Planting diversity			−0.041 ***	0.054 ***
			(0.012)	(0.018)
Control variables	YES	YES	YES	YES	YES	YES	YES
Regional fixed effects	YES	YES	YES	YES	YES	YES	YES
Time fixed effects	YES	YES	YES	YES	YES	YES	YES
Constant	7.823 ***	5.273 ***	7.807 ***	5.089 ***	3.630 ***	7.818 ***	5.065 ***
	(0.101)	(0.183)	(0.094)	(0.171)	(0.145)	(0.096)	(0.152)
Number	3642	3642	4439	4439	4439	4439	4439
R2	0.135	0.110	0.120	0.111	0.102	0.119	0.112

Note: Brackets indicate robust standard errors. *** p < 0.01.

.

• Adjustment of the time sample. Given that agricultural production is vulnerable to external shocks such as climate change and macroeconomic policy adjustments, certain years may display abnormal fluctuations that could distort the estimation results. To eliminate such interference, the data for 2022 are excluded and the sample period is restricted to 2020–2021. This allows us to test whether the findings are driven by specific years and thereby to assess the robustness of the conclusions. If the regression results were significant only in the full sample but became insignificant after excluding 2022, this would suggest that the estimates were driven by that anomalous year. Conversely, if the results remained significant, it would indicate that the conclusions are robust. Columns (1) and (2) of Table 3 report the regression results after adjusting the time sample. The estimates show that the coefficient of agricultural production diversity on food expenditure is −0.041, while that on dietary diversity is 0.066, both significant at the 1% level. These findings are highly consistent with the sign and significance of the baseline regression using the full sample, indicating that the adjustment to the sample period does not fundamentally alter the main conclusions. The impact of agricultural production diversity on food systems thus appears to be robust.
• Replacement of the core explanatory variable. To validate the appropriateness of the core indicator, the explanatory variable is replaced, switching from agricultural production diversity to crop diversity. Agricultural production diversity captures diversification in both crop cultivation and other agricultural activities, and thus provides a more comprehensive reflection of the overall agricultural production structure of rural households. However, given the multiple forms of agricultural diversification, the impact of different agricultural activities is broadly consistent. Moreover, crop cultivation constitutes the fundamental component of rural households’ agricultural production. Therefore, for robustness testing, we focus exclusively on the effect of crop diversity—a core element of agricultural production—on the level of food expenditure. As shown in columns (3) and (4) of Table 3, when crop diversity is used to replace the core explanatory variable, the regression coefficient for food expenditure remains significantly negative, while that for dietary diversity remains significantly positive. This indicates that diversified production reduces food expenditure and enhances dietary diversity, whether assessed from the perspective of overall agricultural production or solely from the angle of crop cultivation. This finding further confirms the robustness of the conclusion.
• The baseline regressions presented above have demonstrated that agricultural production diversity can enhance overall dietary diversity. To further characterise its impact on specific nutrient intake patterns, this study employs protein intake diversity as a more stringent alternative dependent variable. Against the backdrop of ongoing dietary upgrading among rural households in China, the diversity of high-quality protein sources has emerged as a key indicator of nutritional improvement and dietary advancement. Using the CLES data, protein intake diversity scores were recalculated based on household consumption of meat, eggs, dairy products, and plant-based proteins. As shown in Column (5) of Table 3, agricultural production diversity continues to exert a positive and statistically significant effect on protein intake diversity at the 1% level. These results indicate that diversification in agricultural production not only broadens the range of food sources available to rural households but also substantively enhances the quality of nutrient intake.
• Seemingly Unrelated Regression (SUR). In the benchmark regressions, food expenditure and dietary diversity were estimated separately. However, in actual household decision-making, the allocation of expenditure budgets and the choice of food varieties are often made simultaneously, and the two may be correlated. For example, certain unobserved household preferences may simultaneously affect both expenditure and dietary diversity. Ignoring this correlation may reduce the efficiency of the estimates. Therefore, the seemingly unrelated regression approach is employed to estimate the two equations jointly, thereby improving the validity of the results. Columns (6) and (7) of Table 3 report the estimation results of the SUR model. The findings show that the coefficient of agricultural production diversity for food expenditure is −0.031, while that for dietary diversity is 0.060, with both coefficients statistically significant at the 1% level. The estimates from the seemingly unrelated regression model are consistent with those from the two-way fixed-effects specification in the benchmark regression. This indicates that the core conclusions of this study remain valid even after accounting for correlation in the disturbance terms across equations, thereby demonstrating the robustness of the findings.

4.3. Endogeneity Analysis

Although the baseline regression models account for unobserved regional endowments and temporal trends through two-way fixed effects, the relationship between agricultural production diversity and the food consumption system of rural households may still be subject to endogeneity arising from reverse causality and omitted variable bias. Specifically, household food consumption preferences may feed back into production decisions, while unobserved characteristics, such as risk preferences and management capacity, may simultaneously influence both production and consumption behaviours, potentially biasing the estimates. To mitigate endogeneity and obtain consistent and unbiased estimators, an instrumental variable approach is employed, with strict adherence to the principles of relevance and exogeneity in instrument selection.

This study constructs distinct instrumental variables for these two dependent variables. First, for the effect on food expenditure, household size is chosen as the instrument. In terms of relevance, within the institutional context of smallholder agriculture and given households’ resource endowments, larger households are more likely to possess surplus labour and human capital, enabling them to engage in labour-intensive, specialised, and diversified agricultural production. In terms of exogeneity, although household size may influence overall household food demand, this study uses per capital food expenditure as the dependent variable and controls for household labour structure and other relevant characteristics in the regressions. This approach, to some extent, severs the direct pathway through which household size could affect food expenditure via a scale effect. Consequently, once relevant household characteristics are controlled for, household size does not directly influence per capital food expenditure, thereby satisfying the exogeneity requirement for the instrument.

Second, for the effect on dietary diversity, land transfer is selected as the instrumental variable. In terms of relevance, land transfer represents a crucial institutional mechanism that helps households overcome initial constraints arising from fragmented land endowments and serves as a key driver of the optimisation of agricultural production factor allocation. Compared with households that do not engage in land transfer, operators who actively acquire land generally demonstrate stronger intentions for scale-based management and greater capacity to integrate production factors. They are able to plan production layouts and adjust crop and livestock structures over a larger spatial scale, thereby reducing reliance on monocultural staple crops and facilitating a shift towards diversified agricultural production. In terms of exogeneity, land transfer constitutes a structural reallocation of agricultural production factors among households and occurs at the upstream stage of production decision-making, primarily affecting production scale and the allocation of resources. In the model, key variables that could directly influence dietary choices, such as asset endowments, are strictly controlled, effectively eliminating confounding effects arising from wealth and preference differences. On this basis, land transfer is unlikely to directly affect households’ final dietary composition and can therefore be regarded as satisfying the exogeneity requirement for an instrumental variable.

The first-stage regression results reported in columns (1) and (3) of

Table 4. Endogeneity Test: Instrumental Variable Method.

	(1)	(2)	(3)	(4)
	Phase One	Phase Two	Phase One	Phase Two
Household size	0.199 ***
	(0.013)
Transfer of agricultural land			0.846 ***
			(0.042)
Agricultural production diversity		−0.403 ***		0.186 ***
		(0.048)		(0.050)
Control variables	YES	YES	YES	YES
Regional fixed effects	YES	YES	YES	YES
Time-fixed effect	YES	YES	YES	YES
Underidentification test (Kleibergen–Paap rk LM statistic)	208.27 ***		336.98 ***
Weak instrumental variables test (Kleibergen–Paap rk Wald F statistic)	230.78		397.66
Number	4439		4439
R2	0.275	−0.190	0.304	0.098

Note: Brackets indicate robust standard errors. *** p < 0.01.

show that the instrumental variables—household size and farmland transfer—both exert a significant positive effect on the endogenous explanatory variable, namely agricultural production diversity. This suggests that the chosen instrumental variables are strongly correlated with agricultural production diversity. Further statistical tests confirm that the instruments are valid. First, the LM statistics for the underidentification tests are 208.27 and 336.98, both significant at the 1% level, strongly rejecting the null hypothesis of insufficient instrument identification. Second, the Wald F statistics for the weak instrument tests are 230.78 and 397.66, far exceeding the Stock–Yogo critical value of 16.38 at the 10% significance level, indicating that the instruments employed in this study are highly explanatory and that weak instrument concerns are absent. Columns (2) and (4) of Table 4 present the second-stage estimates, which further corroborate the robustness of the core findings. With respect to food expenditure, column (2) shows that the coefficient on agricultural production diversity is significantly negative at the 1% level. Moreover, compared with the benchmark regression, the absolute value of the second-stage estimated coefficient is substantially larger. This change suggests that the benchmark regression underestimates the effect of agricultural production diversity in reducing household food expenditure. With regard to dietary diversity, column (4) shows that the coefficient on agricultural production diversity is 0.186 and remains significantly positive at the 1% level. These results confirm that the conclusion—that agricultural production diversity significantly enhances households’ dietary diversity—remains robust. After rigorously testing the relevance and strength of the instrumental variables and applying two-stage least squares (2SLS) to address potential endogeneity, the findings demonstrate greater reliability and explanatory power. The empirical evidence consistently shows that agricultural production diversity both reduces households’ food expenditure and improves overall dietary diversity.

4.4. Heterogeneity Analysis

4.4.1. Heterogeneity in the External Industrial Environment

Given that the level of industrial development across villages may significantly influence rural households’ livelihood patterns and channels for obtaining food, this study divides the sample into two groups according to whether the village has rural industries: villages with rural industries and villages without rural industries. Grouped regression analysis is employed to investigate the moderating effect of the external industrial environment on the impact of agricultural production diversity. Columns (1) and (3) of

Table 5. Heterogeneity of Rural Industries.

	(1)	(2)	(3)	(4)
	Food Expenses		Dietary Diversity
	Industrialized Villages	Non-Industrialized Villages	Industrialized Villages	Non-Industrialized Villages
Agricultural production diversity	−0.016	−0.034 ***	0.023	0.079 ***
	(0.023)	(0.012)	(0.033)	(0.017)
Control variables	YES	YES	YES	YES
Regional fixed effects	YES	YES	YES	YES
Time fixed effects	YES	YES	YES	YES
Constant	7.806 ***	7.875 ***	5.227 ***	5.078 ***
	(0.244)	(0.105)	(0.444)	(0.188)
Number	987	3375	987	3375
R2	0.164	0.111	0.139	0.107

Note: Brackets indicate robust standard errors. *** p < 0.01.

report the regression results for villages with rural industries, while columns (2) and (4) present the results for villages without rural industries. The regression results reveal clear group differences in the effect of agricultural production diversity on food expenditure, with each subsample displaying markedly distinct characteristics. In the group without rural industries, the regression coefficient of agricultural production diversity for food expenditure is −0.034, while that for dietary diversity is 0.079, both significant at the 1% level. These results are consistent with the benchmark regressions and indicate that in traditional villages lacking industrial support, agricultural diversification remains a key strategy for households to secure food. By contrast, in the group with rural industries, although the coefficient of agricultural production diversity has the same sign, it fails to reach statistical significance. This suggests that in villages with more developed industries, the degree of production diversity within households does not directly determine their level of food consumption. For households participating in rural industries, the development of such industries generates additional non-agricultural employment opportunities and wage income. This not only eases household budget constraints but also raises the opportunity cost of agricultural labour. As labour and commodity markets become more mature, households’ production and consumption decisions gradually diverge. Better-off rural households increasingly rely on cash income to purchase a wider variety of foods directly from the market, thereby reducing their dependence on self-provisioning. Consequently, the welfare benefits of agricultural diversification serve as an essential safeguard for households in less developed villages, where industrial foundations are weaker and market dependence is lower.

4.4.2. Heterogeneity in Aging Levels

Agricultural production is typically labour-intensive, and households at different life-cycle stages display fundamental differences in both labour supply and consumption demand. In this study, the sample is divided into an ageing group (columns (1) and (3)) and a non-ageing group (columns (2) and (4)), according to whether individuals aged 60 or above account for more than 50% of household members. Using grouped regression analysis, we then examine how demographic structure moderates the impact of agricultural production diversity on household food systems. The empirical results reported in

Table 6. Heterogeneity of Aging Levels.

	(1)	(2)	(3)	(4)
	Food Expenses		Dietary Diversity
	Ageing Households	Non-Ageing Households	Ageing Households	Non-Ageing Households
Agricultural production diversity	−0.016	−0.045 ***	0.022	0.094 ***
	(0.014)	(0.016)	(0.023)	(0.020)
Control variables	YES	YES	YES	YES
Regional fixed effects	YES	YES	YES	YES
Time-fixed effect	YES	YES	YES	YES
Constant	7.908 ***	7.496 ***	5.087 ***	4.997 ***
	(0.111)	(0.192)	(0.208)	(0.334)
Number	2214	2225	2214	2225
R2	0.139	0.106	0.105	0.094

Note: Brackets indicate robust standard errors. *** p < 0.01.

indicate that the effects of agricultural production diversity differ markedly across households with varying degrees of ageing. The welfare benefits of agricultural production diversity are mainly evident in the non-ageing group, but fail to materialise in the ageing group. For non-elderly households, the regression coefficient of agricultural production diversity for food expenditure is −0.045, while that for dietary diversity is 0.094, both statistically significant at the 1% level. This suggests that for younger households with abundant labour resources, diversified agricultural production can significantly reduce living costs and improve dietary diversity. For ageing households, the absolute value of the regression coefficient for agricultural production diversity declines substantially and fails to reach statistical significance. This indicates that agricultural production diversity does not generate significant food-consumption benefits for ageing households. The underlying reasons may be considered from two perspectives. First, on the production side, agricultural diversification is inherently labour-intensive and requires producers to possess substantial physical strength and initiative. Young workers in non-ageing households are able to allocate resources efficiently so as to maximise output, whereas in ageing households, deteriorating health conditions raise the marginal cost of diversified production, thereby eroding its cost-saving advantages. Second, from the consumption perspective, non-ageing households tend to have more young members, who require a higher level of dietary diversity and nutrition, making it easier for agricultural diversification to be translated into more varied diets. By contrast, ageing households tend to exhibit more stable dietary habits. As household members grow older, their energy expenditure falls and their sensitivity to dietary variety correspondingly declines. As a result, agricultural production diversity has a more limited impact on food consumption in ageing households. Overall, the impact of agricultural production diversity on food systems is strongly conditioned by household demographic structure. This underscores that, in the context of a rapidly ageing rural population, production-focused interventions alone may be insufficient to effectively improve the welfare of older people. It is therefore essential to complement them with age-friendly social services and support measures.

5. Analysis of the Impact Mechanism

Although previous studies have explored the impact of agricultural production diversity on household food expenditure and dietary diversity from multiple perspectives, empirical evidence on its underlying mechanisms remains limited. Existing research predominantly focuses on the direct relationship between production and consumption, with few studies probing the complexity of the decision-making process from field to table. To date, no study has clarified how rural households adjust their agricultural production structures in order to reduce living costs while simultaneously improving dietary composition. To this end, this paper further investigates the specific mechanisms through which agricultural production diversity shapes household food consumption systems, focusing on two dimensions: food self-sufficiency and income effects.

The diversity of agricultural production determines the extent of a household’s capacity for self-provisioning. The level of self-sufficiency directly affects the household’s dependence on external markets and its food expenditure. To investigate whether agricultural production diversity influences household food expenditure by altering food self-sufficiency, this study conducts a mechanism analysis. The estimation results are reported in

Table 7. Mechanism Analysis of the Impact of Agricultural Production Diversity on Food Expenditure.

	(1)	(2)
	Food Expenses	Food Self-Sufficiency Rate
Agricultural production diversity	−0.031 ***	0.103 ***
	(0.010)	(0.004)
Control variables	YES	YES
Regional fixed effects	YES	YES
Time fixed effects	YES	YES
Constant	7.818 ***	−0.059
	(0.094)	(0.042)
Number	4439	4439
R2	0.119	0.341

Note: Brackets indicate robust standard errors. *** p < 0.01.

. The estimation results in column (2) show that the coefficient for agricultural production diversity is statistically significant at the 1% level, indicating that, holding other explanatory variables constant, a one-unit increase in agricultural production diversity is associated with a 10.3% increase in household food self-sufficiency. This suggests that agricultural production diversity helps to reduce household food expenditure by enhancing food self-sufficiency. The results indicate that households with higher levels of agricultural production diversity are more likely to attain a greater degree of food self-sufficiency. Combined with the preceding discussion, it can be inferred that one reason why agricultural production diversity lowers food expenditure is that it increases rural households’ reliance on their own agricultural products, thereby reducing the need to purchase food from external markets.

The preceding analysis has shown that agricultural production diversity can enhance dietary diversity. To examine whether diversified production structures can indirectly promote dietary improvement by strengthening household agricultural performance, two mechanism variables—agricultural income and agricultural commercialization—are introduced for testing. The results are reported in

Table 8. Mechanism Analysis of the Impact of Agricultural Production Diversity on Dietary Diversity.

	(1)	(2)	(3)
	Dietary Diversity	Agricultural Income	Agricultural Commercialization
Agricultural production diversity	0.060 ***	1.133 ***	0.094 ***
	(0.015)	(0.050)	(0.005)
Control variables	YES	YES	YES
Regional fixed effects	YES	YES	YES
Time fixed effect	YES	YES	YES
Constant	5.065 ***	0.899 **	−0.091 **
	(0.172)	(0.403)	(0.041)
Number	4439	4439	4439
R2	0.112	0.260	0.468

Note: Brackets indicate robust standard errors. *** p < 0.01, ** p < 0.05.

. First, from the perspective of agricultural income, Column (2) in Table 8 shows that the regression coefficient for agricultural production diversity on agricultural income is 1.133, and it is statistically significant at the 1% level. This indicates that agricultural production diversity can ease the liquidity constraints faced by households by increasing agricultural income, thereby strengthening their purchasing power for food and broadening dietary diversity. Second, in terms of rural households’ market participation, column (3) of Table 8 shows that the regression coefficient of production diversity on agricultural commercialization is 0.094, which is also statistically significant at the 1% level. This implies that each additional unit of production diversity is associated with an average increase of 9.4% in the agricultural product sales rate. On the one hand, when rural households produce a wider range of agricultural products, their trading activities become more frequent and their connections with markets grow stronger. Greater market exposure, in turn, reduces the information and transaction costs they incur when purchasing food. Through this process of market integration, commercialization further diversifies household dietary patterns. On the other hand, the deepening of agricultural commercialization facilitates the monetisation of agricultural products, and improvements in household farm performance lead to higher incomes, thereby promoting dietary diversity.

6. Discussion and Conclusions

Against the broader backdrop of comprehensive rural revitalisation, the food consumption system of rural households in China is undergoing profound transformation. With the continued deepening of marketisation and the evolving dynamics of external transaction costs, households’ allocation choices between self-produced food and market purchases not only reshape the scale and composition of food consumption expenditure but also, by influencing sources of food acquisition, serve as key determinants of both economic welfare and dietary diversity. Building on this context, the present study systematically examines the overall impact of agricultural production diversity on the food consumption system of rural households in China. The analysis draws on micro-panel data from the 2020–2022 China Land Economic Survey (CLES), incorporating both food consumption expenditure and dietary diversity within a unified analytical framework. The main findings and discussion are presented below:

First, agricultural production diversity generates a significant dual dividend effect on the overall food consumption system of rural households. The analysis indicates that greater agricultural production diversity not only substantially reduces household food consumption expenditure but also effectively enhances dietary diversity scores. Second, agricultural production diversity reshapes households’ food consumption patterns through two distinct mechanisms. On the internal supply side, diversified production increases food self-sufficiency, generating a pronounced self-sufficiency substitution effect that directly lowers financial outlays for basic food acquisition. On the external market side, rural households engaged in diversified production can alleviate liquidity constraints by increasing total agricultural income and deepening market participation. Through the combined effects of expenditure substitution and consumption expansion, agricultural production diversity exerts a profound influence on the optimisation and upgrading of household food consumption structures. Third, the welfare effects of agricultural production diversity on household food consumption display significant heterogeneity. For traditional rural households with limited non-farm opportunities and relatively young labour forces, diversified production strengthens self-sufficiency and supplements income, thereby more markedly improving dietary patterns and reducing expenditure. In contrast, for households with abundant non-farm employment opportunities or a higher proportion of elderly members, labour and time constraints heighten reliance on market provision, weakening the link between production and consumption and rendering the effects of diversified production comparatively limited.

This study offers a constructive reflection on the prevailing emphasis on agricultural specialisation. Some research suggests that excessive diversification may give rise to diseconomies of scale and efficiency losses [50]. However, in the context of smallholder farming within a large country, land fragmentation and seasonal labour idleness mean that moderate diversification can offset the inefficiencies associated with specialisation and, indeed, enhance agricultural efficiency [51]. In rural China, production and consumption decisions are closely intertwined, and moderate agricultural production diversity constitutes a rational economic strategy [52]. This approach holds practical significance for securing household food security and improving overall welfare. The study confirms that production diversity promotes the optimisation of dietary patterns, aligning with research emphasising that, in areas with limited market accessibility, diversified cropping structures help safeguard household nutritional intake [3,20]. Furthermore, existing studies on food consumption often focus on a single dimension, such as expenditure variation or nutritional improvement, lacking a systematic examination of their interaction [53]. This study innovatively demonstrates that optimising diversity at the agricultural production stage can simultaneously reduce food consumption expenditure and achieve nutritional upgrading in household dietary structure.

These conclusions offer important insights for enhancing household food security and optimising dietary structures. First, at the micro-production level, agricultural structural adjustments should avoid an excessive focus on large-scale specialisation. For rural households with limited capacity to manage risk, targeted subsidies and technical support should encourage the retention of diversified production systems, such as grain–cash crop rotations and integrated crop–livestock systems, thereby consolidating the foundation of household food security. Second, at the market level, rural cold-chain logistics, primary markets, and e-commerce networks should be continuously strengthened to reduce the costs and frictions associated with the circulation of high-quality foods, fostering a virtuous cycle between production and consumption. Third, at the public policy level, attention should be given to the food access vulnerabilities of disadvantaged groups, including the elderly, through social interventions such as the establishment of community canteens, the promotion of modern dietary and nutritional knowledge, and the provision of targeted nutritional subsidies, thereby facilitating more inclusive and sustainable improvements in rural livelihoods.

Finally, it is important to acknowledge the limitations of this study and to highlight directions for future research. First, constrained by the regional coverage of the microdata, the sample is primarily drawn from Jiangsu Province in eastern China, where relatively high levels of agricultural modernisation and well-developed market systems may limit the generalisability of the findings to central and western regions. Future research should incorporate nationally representative panel data to enable cross-regional comparative analyses, thereby examining the welfare effects of agricultural production diversity on rural households across diverse contexts. Second, although the dietary diversity index employed in this study is widely used, it does not capture finer details such as energy intake or micronutrient consumption. Subsequent studies could integrate 24 h dietary recalls or health monitoring data to more precisely evaluate the long-term impacts of agricultural production practices on individual health. Finally, although this study employs instrumental variable methods to mitigate endogeneity concerns, the non-experimental nature of the data makes it difficult to fully eliminate potential household self-selection bias. Future research could exploit quasi-natural experiments or randomised controlled trial designs to obtain more robust causal identification evidence.