Regional technical efficiency and technology gaps in rural China: evidence from CHIP surveys

Abstract

This paper uses household-level datasets called Chinese Household Income Project (CHIP) surveys to investigate the regional technical efficiency, technology gap, and their determinants in rural China. A metafrontier production function approach is employed to fit the values of the technical efficiency and the technology gap ratio of rural households in six regions of China. The findings indicate that the technical efficiency improved remarkably from 1988 to 2002, but the technology gap ratio among regions changed slightly in the same period. The findings also show a negative relationship between the technology gap ratio and the regional technical efficiency for all regions except the northwest that has the lowest value of both the technical efficiency and the technology gap ratio. Next, the Tobit model is employed to discern the sources of efficiency and technology gaps. The results show that quality of agricultural labor, agricultural infrastructure, natural conditions, and farmer's political status have significant effects on farms' technical efficiency and technology gap ratio.

Keywords:

• Rural China
• regional technical efficiency
• technology gaps
• CHIP
• JEL classification
• O13
• Q16
• R11

Acknowledgment

The author thanks Shoji Nishijima, Souksavanh Vixathep, Takahiro Sato, Tomokazu Nomura and Hiroyuki Kato for valuable comments and suggestions. The author is also grateful to Yutaka Shinada for help in obtaining data. Nevertheless, all remaining errors are my responsibility.

Notes

1. For example, the natural endowment is obviously different between the southwest and the northwest, but in previous studies, both the southwest and the northwest are treated as a region – the west.

2. The three surveys are taken from Griffin and Zhao (1993a); Riskin, Zhao and Li (2000); and Li (2008), respectively.

3. For more information on the first survey, see Khan et al. (1992); on the second, see Khan and Riskin (1998); and on the third, see Gustafsson, Li and Sicular (2008).

4. In this study, the regional economic development level is considered according to the prevalent Chinese regional economy studies, while geographic conditions are considered mainly according to the classification of the National Commission of Agricultural Division (1991: 50–7) that divided the mainland of China into 10 first-level agricultural regions, on the principle that agricultural production should be consistent with (1) regional natural conditions such as climate, soil conditions, and hydrological conditions and (2) regional economic conditions. We matched our samples (19 provinces) and the map of the National Commission of Agricultural Division (1991: 50–7), then classified China into six rural regions.

5. At first, 258 counties were designated as national poor counties in 1986 by the Leading Group for Economic Development in Poor Areas of China, according to the standard that the rural net income per capita of a county was below 150 RMB Yuan in 1985. Then, the number of national poor counties increased to 592 in 1993. In addition, in 1988, 370 counties were designated as provincial poor counties by different provincial standards (Park, Wang and Wu 2002).

6. For the detailed steps of the test, see Battese, Rao and O'Donnell (2004).

7. Because the P-value of the test for southwest stochastic frontier in 2002 is 0.224, we believe that the inefficiency term was present at this level.

8. In this study, both the northwest and the southwest have the lowest .

9. Tobit model was employed by Chavas, Petrie and Roth (2005) and Chen and Song (2008) for the efficiency determinant function, while maximum likelihood estimation was employed by Yao and Liu (1998), and ordinary least squares (OLS) estimation was employed by Tian and Wan (2000). In this study, Tobit and OLS estimators are identical in the and the models since there is no censored observation. Further, the estimators in the model are also similar between Tobit and OLS estimates since there are only 30 right-censored observations in all 22,853 observations.