The impact of agricultural tourism integration on farmers’ income growth: empirical evidence from China

1 Introduction

Rural transformation has become one of the central development challenges for many developing countries. Despite rapid urbanization and industrial growth, rural areas often remain trapped in low productivity, weak infrastructure, and limited income diversification (Dong and Chunlai, 2023). Agriculture, the backbone of most developing economies, continues to face structural constraints such as small-scale production, low value added, and poor market connectivity (Li et al., 2026). Within this broader development context, the integration of agriculture and tourism is now widely seen as a practical measure to invigorate rural economies, diversify earnings for farming families, and underpin sustainable rural progress (Wang J. et al., 2024; Liu et al., 2026; Liu et al., 2025a; Liu et al., 2025b; Chen et al., 2025; Wang L. et al., 2024; Liu et al., 2025c).

Through the integration of agriculture and tourism, a linkage is built between farming practices and tourism consumption, thus reshaping traditional farmland into multifunctional spatial carriers that incorporate productive, ecological and cultural functions (Chen and Kong, 2021). This integration enables rural areas to capture new sources of value from agricultural resources and to participate in the expanding service economy (De Lapparent et al., 2025; Ma et al., 2024; Wang et al., 2025; Liu et al., 2025a; Liu et al., 2025b; Liu et al., 2025c). By attracting urban visitors, rural regions can generate non-farm employment opportunities (Kolomyts et al., 2020; Veeck et al., 2016) and stimulate local entrepreneurship, thereby enhancing both income stability and community resilience (Vargas Lopez et al., 2026; Bhatta et al., 2026; Fu et al., 2026). For developing countries, agritourism provides a feasible pathway to address the dual challenge of rural poverty and structural transformation (Susila et al., 2024; Camacho and Ramos-Herrera, 2025).

China serves as a highly illustrative empirical context for investigating the developmental effects of agriculture-tourism industrial convergence. As the world’s largest developing economy with a dominant agricultural sector, China is home to extensive rural territories experiencing far-reaching socioeconomic restructuring (Wang et al., 2023; Liu et al., 2023; Liu et al., 2025a; Liu et al., 2025b; Liu et al., 2025c). Since the early 2010s, the Chinese central government has positioned the integrated development of agriculture and tourism as a core policy instrument for advancing rural revitalization and realizing balanced urban–rural development. In 2010, The National Demonstration Counties for Leisure Agriculture and Rural Tourism (LART) program was jointly introduced by the Ministry of Agriculture and the then National Tourism Administration (Huang et al., 2023; Luo et al., 2023). This policy scheme was designed to explore diversified pathways for leveraging rural ecological, cultural and agricultural resource endowments through the integration of productive and service functions. Accompanied by preferential policy arrangements covering fiscal subsidies, infrastructure construction investment and land use governance, eight successive batches of demonstration counties were designated from 2010 to 2017, with geographic coverage extending to nearly all provincial-level administrative regions across the country.

Existing studies provide important evidence on the economic consequences of rural tourism. Huang et al. (2023) examine the urban–rural income-gap implications of the same demonstration-county program, while Luo et al. (2023) study farmers’ income growth in Southwestern China. These studies establish a useful foundation, but two questions remain open. First, does the staggered national rollout of the LART program generate a detectable income response across a broad county sample and a long observation period? Second, do the estimated responses differ with regional development, city size, and transport accessibility, and are they accompanied by changes in investment intensity and environmental conditions? The present study addresses these questions using a balanced panel of 1,532 county-level units from 2000 to 2023. The contribution is therefore incremental: it extends the geographic and temporal scope of existing evidence, examines policy-effect heterogeneity, and reports suggestive channel evidence without claiming to establish a new theory or a formal mediation structure.

2 Policy background and research hypotheses

The national Guidelines for Promoting the Integrated Development of Rural Industries, issued in 2016, further clarified the direction of rural industrial upgrading by emphasizing the conversion of agricultural resources into tourism products, the extension of agricultural value chains, and coordination between production and services. The Rural Revitalization Strategic Plan (2018–2022) subsequently identified leisure agriculture and rural tourism as instruments for increasing farmers’ income, improving living environments, and supporting ecological sustainability. The LART designations were introduced in successive batches rather than simultaneously. This staggered timing creates variation across counties and years that can be examined with a multi-period DID framework. The design is informative under the maintained assumption that, absent designation, treated and comparison counties would have followed parallel income trends after conditioning on county and year fixed effects and the included covariates.

From a theoretical standpoint, the industrial convergence of agriculture and tourism exerts influences on rural residents’ income through a set of interconnected transmission pathways. Among all possible channels, the most straightforward one works via the diversification of household income structures (Incera and Fernandez, 2015). Through participation in tourism-aligned business activities including homestay operation, on-site catering and characteristic handicraft retail, farming households are able to claim a larger proportion of value added within the rural industrial value chain (Fang et al., 2021; Wu et al., 2024). The transformation of native agricultural resources into market-oriented tourism commodities allows rural households to obtain dual income flows from both agricultural production practices and tourism service delivery (Arru et al., 2021). Optimized rural infrastructure and better regional connectivity further cut down transaction costs and broaden market access for local specialties, thus stabilizing the overall earnings of rural households (Bano et al., 2021; Wen-Ta and Tsai, 2025). From the lens of endogenous growth theory, such industrial structural diversification reinforces the local multiplier effect, spurs local consumption demand and facilitates the founding of small-sized enterprises, which jointly support the sustainable growth of rural income (Bai et al., 2024; Zhang et al., 2025). Drawing on the above logical reasoning, this study puts forward the following research hypothesis:

H1: The implementation of agricultural tourism integration policies significantly raises the income level of rural residents in demonstration counties relative to non-demonstration counterparts.

A second mechanism operates through capital attraction. Integrating agriculture and tourism enhances the investment appeal of rural areas by improving expected returns and reducing information barriers (Young and Markham, 2020; Kumail et al., 2023). Public investment in transportation, utilities, and tourism facilities strengthens infrastructure and creates favorable conditions for private participation. The inflow of external capital, including investment from enterprises, local entrepreneurs, and financial institutions, not only expands employment opportunities but also enhances the technological and managerial capacity of rural enterprises. As knowledge and innovation diffuse through these investment networks, local production efficiency improves, leading to industrial upgrading and sustainable income growth (Choo and Park, 2022). On the basis of this analysis, the following hypotheses are proposed.

H2: LART designation is expected to be associated with greater capital attraction and investment intensity, which may support rural income growth.

The third transmission pathway operates through ecological environment enhancement. Given that natural scenic landscapes and regional environmental quality constitute core competitive assets for rural tourism development, the industrial integration of agriculture and tourism creates intrinsic incentives for both local governing bodies and rural residents to engage in ecosystem conservation and restoration efforts (Yasin and Bacsi, 2025; Pérez-Olmos and Aguilar-Rivera, 2021; Sun et al., 2023). Such ecological improvement measures, ranging from soil and water conservation initiatives to biodiversity conservation programs, contribute to higher agricultural production efficiency and reinforce the long-term sustainability of rural socioeconomic development (Liu et al., 2025a; Liu et al., 2025b; Zhao et al., 2025; Liu et al., 2025c). Pleasant and well-preserved rural environments further attract larger tourist volumes, boost overall tourism revenue, and form a positive feedback loop between ecological protection and economic benefit generation (Zhao et al., 2026). Grounded in the theoretical reasoning presented above, the following research hypothesis is formulated:

H3: LART designation is expected to be associated with improvements in the measured ecological conditions, which may support the longer-term development of rural tourism and rural income.

This study makes three incremental contributions. First, it extends existing work by evaluating the LART program with a national county-level panel covering 2000–2023, complementing studies centered on the urban–rural income gap or particular regions (Huang et al., 2023; Luo et al., 2023). Second, it documents how the estimated policy association varies across regions, city-size groups, and transport-accessibility conditions. Third, it examines whether LART designation is accompanied by changes in capital-attraction and ecological-quality proxies. The last set of results is interpreted as suggestive channel evidence rather than formal identification of mediation effects.

3 Model construction and variable selection

This study uses a multi-period difference-in-differences (DID) strategy because the LART program was introduced in different counties at different times. The design compares changes in rural residents’ income after designation with contemporaneous changes in counties that had not been designated, while county fixed effects absorb time-invariant local characteristics and year fixed effects absorb shocks common to all counties. The identifying interpretation depends on the parallel-trends assumption. We assess this assumption with an event-study specification and report a placebo exercise based on random policy assignment. The regressions also control for the tertiary-industry share, per capita fiscal expenditure, rural employment, and per capita GDP. Inference must account for serial correlation in panel DID settings (Bertrand et al., 2004). The baseline table therefore reports county-clustered standard errors. In subsequent robustness and subgroup tables, standard errors are clustered at the prefecture level as a deliberately more aggregated inference check that allows disturbances to be correlated across counties within the same prefecture. Because the two clustering choices address different correlation structures, results are interpreted by their direction, magnitude, and precision rather than as mechanically identical estimates.

3.1 Model specification

Following the standard multi-period DID framework, the baseline econometric model is specified as follows:

Y it = α + βPolic y it + γ X it + μ i + λ t + ε it

In this equation, Yit denotes rural residents’ per capita disposable income in county i and year t. Policyit equals 1 from the year in which county i is approved as a national LART demonstration county and 0 otherwise. X it is a vector of county-level socioeconomic controls. County fixed effects and year fixed effects are represented by μ i and λ t, respectively, and εit is the disturbance term. Under the maintained parallel-trends and no-anticipation assumptions, β captures the average post-designation difference in income changes between designated and comparison counties within this specification.

3.2 Variable definition

The outcome is rural residents’ per capita disposable income. The core explanatory variable identifies whether county i has obtained National Demonstration County for Leisure Agriculture and Rural Tourism status by year t; it equals 1 from the official approval year onward and 0 otherwise. To examine potential channels, the study uses two groups of proxy indicators. Capital attraction (Capital) is proxied by the ratio of fixed-asset investment to regional GDP and the tertiary-sector output share. These indicators capture investment intensity and structural change, but they should not be read as direct measures of the capital stock. Ecological quality (Eco) combines standardized measures of green-space coverage, per capita park area, and industrial wastewater-treatment efficiency. These indicators describe observable environmental conditions, although their urban orientation limits their ability to capture all dimensions of rural ecological quality. Accordingly, the channel regressions are treated as suggestive associations rather than formal mediation tests. The controls include per capita fiscal expenditure, rural employment, per capita GDP, and the tertiary-industry share. Continuous variables are transformed using natural logarithms.

3.3 Data sources and sample description

This research compiles a balanced panel dataset that spans 1,532 county-level administrative units across China over the period from 2000 to 2023. Key socioeconomic metrics, such as rural residents’ disposable income, per capita fiscal expenditure, rural employment scale, and per capita GDP, All relevant statistical indicators are retrieved from the China County Statistical Yearbook. The official roster of leisure agriculture and rural tourism demonstration counties is compiled manually based on public policy releases issued by the Ministry of Agriculture and Rural Affairs as well as the former National Tourism Administration. A total of eight groups of national-level demonstration counties were officially authorized and established during the period spanning 2010 to 2017, with geographic coverage extending to nearly all provincial administrative regions nationwide. As illustrated in Figures 1a,b, these designated counties are geographically dispersed across eastern, central and western China, mirroring the stepwise spatial expansion of the policy initiative. These approved units form the treatment group of this study, whereas the remaining non-demonstration counties function as the control group. Following standardized data cleansing procedures and the exclusion of counties with severe missing records, the final balanced analytical sample consists of 1,532 county-level observations. Figure 1c plots the overall upward trajectory of rural per capita disposable income for both groups during 2000–2023, where the treatment group displays a markedly faster growth pace after policy rollout. Figure 1d demonstrates the temporal progression of policy implementation between 2010 and 2017, featuring a steady year-on-year rise in the cumulative number of approved demonstration counties. Figure 1e compares pre-policy socioeconomic attributes across the two groups, indicating their foundational characteristics were largely comparable before policy adoption. This preliminary evidence lends empirical support to the parallel trend assumption that underpins valid DID estimation.

4 Empirical results

4.1 Baseline regression results

Table 1 reports estimates of the association between LART designation and rural residents’ per capita disposable income. Column (1) presents the specification without time-varying controls, while columns (2)–(5) sequentially add the tertiary-industry share, fiscal expenditure, rural employment, and per capita GDP.

Table 1. Benchmark regression results.

Standard errors clustered at the county level are reported in parentheses. ^***p < 0.01, ^**p < 0.05, ^*p < 0.1.

Across the five specifications, the coefficient on LART remains positive and statistically significant at the 1% level. In the preferred specification in column (5), the estimate is 0.099 with a standard error of 0.024. Given the log-level specification, this corresponds to an income difference of approximately 10.4%. The estimate declines as controls are added, from 0.164 in column (1) to 0.099 in column (5), indicating that observed county characteristics account for part, but not all, of the baseline association. Fiscal expenditure and per capita GDP are positively associated with rural income. The adjusted R-squared rises from 0.889 to 0.913 as the controls are added. These results support a positive policy interpretation conditional on the DID assumptions; they do not by themselves eliminate all possible sources of endogeneity.

4.2 Parallel trend test and placebo test

Figure 2a presents event-study estimates used to assess the parallel-trends assumption. The pre-treatment coefficients fluctuate around zero and their confidence intervals include zero, providing no clear evidence of differential pre-trends in the plotted periods. After designation, the coefficients generally rise, a pattern consistent with policy effects that develop over time as infrastructure, market connections, and household participation adjust. This visual evidence supports, but cannot by itself prove, the identifying assumption. Figure 2b reports a placebo exercise based on 500 random assignments of policy status. The placebo estimates are concentrated around zero, whereas the baseline estimate lies toward the right tail of the distribution. This pattern suggests that an estimate of the observed magnitude is uncommon under random assignment, providing a supplementary check on the baseline result.

4.3 Supplementary instrumental-variable analysis and limitations

Policy designation may remain correlated with unobserved local characteristics. As a supplementary exercise, the number of traditional villages (NTV) is used as an instrument for LART designation. Counties with more traditional villages may be more likely to receive demonstration status, which motivates the relevance condition. However, relevance and exclusion are distinct requirements. Traditional villages are themselves cultural and tourism resources and may influence rural income through tourism demand or related investment outside the LART designation channel. The exclusion restriction therefore cannot be established solely from a statistically insignificant coefficient in an auxiliary income regression. The IV estimates reported below should consequently be interpreted as sensitivity evidence that relies on a strong and contestable exclusion assumption, rather than as definitive resolution of endogeneity.

Table 2 reports the existing IV estimates. In the first stage, NTV is positively associated with LART designation, and the reported first-stage F-statistic of 870.95 indicates strong statistical relevance. The second-stage coefficient on LART is positive and statistically significant. These results show that the instrument is strongly related to treatment status, but the F-statistic does not validate the exclusion restriction. Because traditional villages may have a direct relationship with tourism activity and rural income, the IV estimates are treated as supplementary and are not used as the sole basis for the paper’s causal interpretation.

Table 2. Supplementary instrumental-variable estimates.

Standard errors clustered at the county level are reported in parentheses. ^***p < 0.01, ^**p < 0.05, ^*p < 0.1.

4.4 Robustness checks

Table 3 reports three robustness checks using the estimates already presented in the manuscript: PSM-DID, exclusion of county-level units administered by municipalities, and restriction of the sample period to 2010–2023. These checks examine sensitivity to observable sample composition, special administrative settings, and the estimation window. No causal-forest analysis is reported; the earlier reference to a fourth test was an editorial inconsistency and has been removed.

Table 3. Robustness checks.

Standard errors clustered at the prefecture level are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

4.5 PSM-DID

The PSM-DID exercise addresses observable differences between designated and comparison counties. A Logit model is used to estimate policy-assignment probabilities from pre-policy covariates, followed by nearest-neighbor matching. The reported balance diagnostics indicate that standardized differences fall below 10% after matching. Re-estimation on the matched sample produces a positive coefficient consistent with the baseline direction. This exercise reduces imbalance in observed characteristics under the matching assumptions, but it does not remove bias from unobserved confounders.

4.6 Exclusion of counties from directly governed municipalities

In the baseline analysis, our sample includes counties and districts under the jurisdiction of directly administered municipalities, such as Beijing and Tianjin. The administrative frameworks and governance practices in these areas are relatively distinctive, which could introduce potential biases into our estimates. To address this concern, we exclude these county-level units in a robustness check, with the aim of testing whether our core conclusions are sensitive to the inclusion of such special administrative regions. This procedure allows us to verify that the estimated effects are not predominantly driven by the unique managerial features of municipalities, thereby strengthening the credibility of our empirical findings.

4.7 Shortened sample window

To examine whether the result is sensitive to the long pre-policy period, the sample window is restricted to 2010–2023. This specification reduces the influence of distant historical observations while preserving the policy-rollout years and the post-designation period. The estimate is interpreted as a sample-window sensitivity check rather than as a separate causal design.

5 Heterogeneity analysis

This section reports exploratory subgroup estimates by geographic region, population size, and high-speed-rail (HSR) access. The estimates describe differences in policy responsiveness across observed contexts. Because the models do not report formal cross-group coefficient tests, differences in point estimates are interpreted cautiously rather than as definitive evidence that the effects differ statistically across groups.

5.1 Regional heterogeneity

Table 4 reports estimates for eastern, central, and western counties. The point estimates are 0.082 (p < 0.10), 0.061 (p < 0.05), and 0.129 (p < 0.01), respectively. The larger western estimate is consistent with greater scope for place-based tourism policy where initial infrastructure and alternative income opportunities are more limited. Because the table does not report a formal equality test across groups, the differences in coefficient magnitude are interpreted as descriptive heterogeneity.

Table 4. Regional heterogeneity results.

Standard errors clustered at the prefecture level are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

5.2 Population-size heterogeneity

Cities are divided by their 2012 permanent-resident population, corresponding to the period around the first policy batches. Large and medium-sized cities have populations above 500,000, while small cities have populations below 500,000. This fixed pre-designation classification avoids allowing later population changes to redefine the groups (Table 5).

Table 5. Population-size heterogeneity results.

Standard errors clustered at the prefecture level are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

The LART coefficient is 0.070 (p < 0.10) for large and medium-sized cities and 0.127 (p < 0.01) for small cities. The point estimate is therefore larger in the small-city subsample. This pattern is consistent with agritourism representing a larger share of local opportunities in less diversified economies, but the table does not establish that the two coefficients differ statistically.

5.3 Transport-accessibility heterogeneity

Transport accessibility may condition the local market reached by agritourism activities. Using 2012 data, cities are classified into those with an HSR station and those without one. The grouping is used to describe whether the LART estimate varies with initial HSR access; it is not designed to estimate the separate causal effect of opening an HSR line (Table 6).

Table 6. Heterogeneity by high-speed-rail access.

Standard errors clustered at the prefecture level are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

The LART coefficient is 0.095 (p < 0.05) in cities with HSR access and 0.040 (p < 0.10) in cities without HSR access. The larger point estimate in the HSR group is consistent with transport accessibility supporting tourism demand and market connections. As with the other subgroup results, the difference is interpreted cautiously because no formal cross-group equality test is reported.

6 Potential channel evidence

To examine whether LART designation is accompanied by changes in factors that could support rural income growth, Table 7 reports regressions using the Capital and Eco proxy indices as outcomes. These models estimate policy associations with the proposed channel variables. They do not implement a complete mediation design and therefore do not identify the proportion of the income effect transmitted through either channel.

Table 7. Associations with potential channel proxies.

Standard errors clustered at the prefecture level are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

Column (1) reports a positive coefficient of 0.123 for the Capital proxy, statistically significant at the 5% level under the table’s reported standard error. The estimate indicates that LART designation is associated with higher values of the investment-intensity and industrial-structure proxy. This pattern is consistent with capital attraction as a possible channel, but the proxy does not directly measure the capital stock and the regression does not establish mediation from policy to capital and then to income.

Column (2) reports a positive coefficient of 0.056 for the Eco index, statistically significant at the 5% level. The estimate indicates that LART designation is associated with improvements in the included green-space, park-area, and wastewater-treatment measures. Because these indicators are partly urban-oriented and do not capture all aspects of rural ecological quality, the result is interpreted as suggestive evidence of environmental change rather than proof of an ecological mediation mechanism.

Taken together, the two coefficients show that LART designation coincides with changes in the selected capital and ecological proxies. These associations are compatible with the theoretical channels discussed above, but they do not determine whether, or by how much, either proxy transmits the estimated income effect. The results should therefore be read as exploratory channel evidence.

7 Discussion and conclusion

7.1 Discussion

Using the national LART demonstration-county program as a quasi-experimental setting, this study examines the relationship between agriculture-tourism integration and rural residents’ income. The subgroup estimates are largest in western China (0.129, p < 0.01), compared with eastern (0.082, p < 0.10) and central counties (0.061, p < 0.05). This ordering is consistent with the possibility that places with weaker initial infrastructure and fewer alternative income sources have greater scope to benefit from place-based tourism investment. It should not be read as a formal test of diminishing marginal returns, because the subgroup models do not directly estimate that mechanism or test equality across coefficients.

The point estimate is also larger for small cities (0.127, p < 0.01) than for large and medium-sized cities (0.070, p < 0.10). One possible interpretation is that agritourism represents a larger share of local economic opportunities where agriculture remains relatively important. In more diversified urban economies, agritourism may play a smaller complementary role. These explanations are consistent with the observed pattern, but alternative differences between the groups cannot be ruled out. The positive coefficients for the Capital and Eco proxies indicate that LART designation is accompanied by changes in investment intensity, industrial structure, and selected environmental conditions. These results add descriptive evidence on possible channels connecting agritourism policy and income growth. They do not establish a dual mediation mechanism, and the ecological measures do not fully represent rural ecosystem quality. Future work using direct rural environmental measures and a formal mediation design could evaluate these pathways more rigorously. The study offers three incremental contributions to the literature:(1) It extends existing evidence on the LART program by examining rural disposable income in a nationwide county panel covering 2000–2023, complementing work on the urban–rural income gap and evidence from Southwestern China.(2) It documents variation in the estimated policy association across regions, city-size groups, and HSR-access groups, highlighting contexts in which the estimates are larger or smaller without treating those contrasts as formal mechanism tests.(3) It reports exploratory evidence on capital and ecological proxies that can motivate future research, while explicitly distinguishing these associations from formal mediation identification.

7.2 Conclusion

The estimates indicate that LART designation is associated with higher rural residents’ per capita disposable income in the study sample. The preferred estimate is 0.099 log points, and the subgroup estimates are larger in western China and small cities. The channel regressions are consistent with changes in investment-related and environmental proxies, but they do not establish mediation. Several limitations define the scope of the conclusions. The DID interpretation depends on parallel trends and the absence of unobserved shocks that differentially coincide with designation. The IV exercise relies on a contestable exclusion restriction and is therefore supplementary. The ecological proxies are partly urban-oriented, and subgroup coefficient differences are not formal cross-group tests. Finally, the estimates arise from China’s institutional setting and may not transfer directly to countries with different land systems, local-government capacity, or tourism markets. Within these boundaries, the results suggest that agritourism-oriented place-based policies can support rural income growth when policy design reflects local economic and resource conditions.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Tingting Bai, Yangzhou University, China

Reviewed by: Chenglong Yao, Shihezi University, China

Youshuai Feng, Huzhou University, China