Abstract
This paper aims to test the effects of social interaction and input spillovers on the agricultural performance of family farms in mixed paddy rice production (organic and conventional farming). Using survey panel data from a Chinese village, we adopt a Spatial Durbin Model (SDM) derived from spatial econometrics to disentangle the social interaction effect from spillover effects and to control for technological heterogeneities. Our analysis reveals a negative social interaction effect indicating that an increase in the yield of neighbouring plots leads to a decrease in the yield of the plot itself. We also find input spillovers. Labour, capital costs, water and organic pesticides have positive spillovers while external nutrients (nitrogen and phosphate) have negative spillover effects. Our study thus calls for a better understanding of social interactions and spillover effects among farmers in the promotion of sustainable family farming.
Acknowledgements
We thank Eric Kéré for their useful comments. We would like to thank the NGO Partnerships for Community Development (PCD) and the Guangxi Maize Research Institute for their valuable technical assistance in the field work. Data and code are available upon request.
Supplementary Material
Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2018.1443206
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. The ‘Number One document’ focuses on what the Central Committee and the State Council view as the most important topics for the year.
2. A ‘village’ in China can either be a natural village (Zirancun), the smallest social entity, or an administrative village (Xingzheng cun), which is a bureaucratic entity. Sancha is a natural village.
3. PCD is based in Hong Kong. More information about this NGO can be found on their website: http://www.pcd.org.hk/eng/index.html.
4. There are two crop seasons of paddy production in one year in Sancha village. The first season is from March to July, and the second season is from August to December.
5. The results of the SLX model are not presented but available upon request.
6. Note that we may introduce group fixed effects to control for unobservable features at the group level but we have to drop plot fixed effects. We thus choose to control for plot fixed effects rather than group fixed effects. All regressions of and Table S1 (in Supplementary Materials) were rerun with group fixed effects. Main results are close to results with plot fixed effects but likelihood ratio tests confirm in all cases that the plot fixed effects model is preferred to the group fixed effects model. Results are available on request.
7. We use the Matlab routine sarregime_panel developed by J. Paul Elhorst.
8. This approach cannot help for investigating whether the social interaction effect has a differential effect according to the nature of the technology of neighbouring plots (cross-technology spillovers).
9. We also employ tests for the presence of spatial interaction effects, which confirm that there are spatial interactions at stake in our case study. More details can be found in Supplementary Materials.
10. We use the Matlab routine panel_effects_sdm written and made available by D. Lacombe and proposed by LeSage and Pace (Citation2009, pp. 114–115).
11. Different self-selection models have been tested (for example, without spatially lagged control variables). The main results remain robust and are available upon request.
12. One important assumption is that group size should be a minimum of two individuals (see Lee, Citation2007, p. 339).
13. Regarding the four subgroups in WF, within one of them, there are both ‘real’ neighbours (plots located in the same place – mountain or plain – and owned by farmers in the same family) and ‘false’ neighbours (plots located in a different place than a given plot but owned by farmers in the same family). Thus, it is possible to find social interactions (from ‘real’ neighbours) but less significant (due to ‘false’ neighbours). To confirm this interpretation, we run an SDM and an SEDM as in Table S3 from a matrix without the mountain reference group (WM). It is worth noting the significance of the negative social interaction effects (SDM) as well as the significance of the negative spatial autocorrelation (SEDM) is greater than those with WF and WG but lower than those with W. The reason is that WM defines neighbours in the same way as W (family and location) but with less information.
14. With WM (without the mountain reference group), we also find non-significance of both the social interaction effect and spatial autocorrelation from a GNS model as in Table S3. Both the reduction of the variation of group size (47, 46, 48 and 42 plots, respectively) and the number of groups (four) compared to the matrix W should explain those results (as with WF and WG). Results are available upon request.