Methodological innovations in estimating the (inverse) relationship between farm productivity and farm size in a developing economy: a case study of Burundi

We use a nonparametric estimation of the production function to investigate the relationship between farm productivity and farming scale in poor smallholder agricultural systems in the north of Burundi. Burundi is one of the poorest countries in the world, with a predominant small scale subsistence farming sector. A Kernel regression is used on data of mixed cropping systems to study the determinants of production including different factors that have been identified in literature as missing variables in the testing of the inverse relationship such as soil quality, location and household heterogeneity. Household data on farm activities and crop production was gathered among 640 households in 2007 in two Northern provinces of Burundi. Four production models were specified each with different control variables. For the relatively small farms, we find clear evidence of an inverse relationship. The relatively large farms show a different pattern. Returns to scale are found to be farm scale dependent. Parametric Cobb-Douglass models tend to over-simplify the debate on returns to scale because of not accounting for the different effects of large farms. Other factors that significantly positively affect production include the soil quality and production orientation towards banana or cash crop production. Production seems to be negatively affected by field fragmentation