Assessing the Impact of Farmer Field Schools on Excessive Fertilizer Use in China

Project Overview


Many Chinese farmers use excessively high rates of nitrogen fertilizer, and they often do not know about the adverse effects of excess fertilizer entering soil and water systems. Moreover, the lack of accountability in China's current public agricultural extension system makes that system ineffective at delivering fertilizer training and knowledge to individual farmers.

The Chinese Ministry of Agriculture (MOA) is addressing this problem by instituting farmer field schools (FFS), hoping to avoid the pitfalls of the traditional agricultural extension system by using local farmer-trainers to improve accountability and effectiveness through a participatory training approach. However, rigorous evaluation of the FFS has not been conducted to date.


This project evaluates the environmental and socioeconomic impact of fertilizer-related training provided by the MOA’s FFS program to Chinese farmers. RAND and the Center for Chinese Agricultural Policy (CCAP) designed and implemented a randomized controlled trial (RCT) experiment in 52 villages in two provinces. The results of this impact evaluation will help policy makers design and scale up cost-effective FFS programs throughout China. The findings from this study will also have implications for other developing countries in improving their efficiency in agricultural nitrogen fertilizer use.


We surveyed farmers from Anhui and Hebei provinces, the two provinces in China where our evaluation was implemented. The survey focuses on rice and tomato farming because they are rapidly growing sources of N2O emissions and methane. Moreover, tomatoes are a greenhouse vegetable (GHV) and have significantly different fertilizer needs than rice. By choosing two very different crops in two very different provinces, we sought to identify how the effectiveness of the FFS program varies by crop and location.


This project is sponsored by 3ie.

Preliminary Data

Our baseline survey was done in two stages, the first to collect demographic and other information (Survey A), and the second to collect fertilizer usage (Survey B). We conducted one endline survey (Survey C). Final data and documentation files will be available in 2015. In the meantime, per our contract deliverable, we have posted these preliminary datasets. Note: The main thing that makes these data "preliminary" is the structure of the data, as the files will eventually be combined into more user- and analysis-friendly formats. Some Chinese names (e.g., provinces) will change to English equivalents in the final data.

Dataset Documentation

Study Code: OW3:1216
Country: China
Time range: Dec 1, 2011 – Dec 31, 2013

The zip file contains two folders, one for the rice counties and one for greenhouse vegetable counties. This mirrors the structure of the program and the evaluation. Both of the rice and greenhouse vegetable folders consist of two parts: Baseline and Endline folders. This is consistent with the structure of survey we conducted. The Baseline folders contain all of the pre-intervention data, while the Endline folders contain all of the post-intervention data.

All data are in STATA (13) format.

The name of each individual dataset consists of three parts: 1) the first part is “cln” which means the dataset has been cleaned, rather than original source; 2) the second part is “a/b/c” which represents the survey; 3) the third part is “a/b/c…” which is the name of survey module. Regarding to the second part, since our baseline survey was done in two stages -- the first to collect demographic and other information, and the second to collect fertilizer usage, we denote them as Survey A and B, respectively. We also denote our endline survey as Survey C. The way we name the surveys is consistent with how we name the individual datasets. For instance, “cln_a_c” means this dataset that has been cleaned is from module C of survey A.

In each dataset, each household has a unique household code, called “hcode”, which can be used as match variable to merge different modules. The household code consists of 8 digits. Every two adjacent digits is a group and they represent county, township, village, and household in sequence.

Some datasets are in long form to match the structure of surveys.

The study was conducted in two provinces of China and therefore the data was collected in Chinese. You may need a Chinese version of STATA to check the data.

Corresponding author: Krishna B. Kumar, RAND, 1776 Main Street, Santa Monica, CA 90401,