How efficient is maize production among smallholder farmers in Zimbabwe? A comparison of semiparametric and parametric frontier efficiency analyses

ABSTRACT

The controversial Fast Track Land Reform Programme in Zimbabwe that redistributes commercially-owned farmland to smallholder households has caused concerns about the efficiency of agricultural production in the country. In this paper, we estimate the efficiency of resource use among smallholder farmers in Zimbabwe when producing maize, the staple crop in the country. Using both a semiparametric model and a fully parametric stochastic frontier model, we find significant production shortfalls for smallholder maize production. While labor, capital, and land all significantly affect the total output, the estimated mean efficiency score for farms with less than 10 hectares of land (A1) appears to be under 0.75, and for the entire sample (A1 and A2) it ranges between 0.595 and 0.772. There clearly exists a great potential for maize farmers to improve the technical efficiency and increase the total output. Gender and age of the household head, access to extension services, and activities of other crops significantly affect the technical efficiency of smallholder maize production in Zimbabwe. We also find that all farms operate under increasing returns to scale and that the technical efficiency score tends to increase with the level of output.

KEYWORDS:

• Stochastic frontier
• maize production
• Zimbabwe
• Fast Track Land Reform Programme
• contextual variables
• semi-parametric model

JEL CLASSIFICATION:

• Q12
• C14
• N57

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Shaw (2003) reports that 45 percent of agricultural land in Zimbabwe was occupied by less than 1 percent of the population in 1980.

² Currently, Zimbabwe’s land ownership falls into four categories: communal, old resettlement, $A1$, and small-scale commercial and $A2$. USDA (2018) reports that in 2018, A1 and A2 farmers produced about 26% and 31% of the country’s maize, respectively.

³ For instance, the offer letter given to the $A1$ farmers explicitly states that the government may withdraw the offer at any time without compensating the farmers for any improvements they made on the land; this provision could disincentivise $A1$ farmers from making investment on the land (Matondi 2012).

⁴ Data from the World Bank (2018) and other sources suggest similar patterns for Zimbabwe’s maize imports and exports.

⁵ While Ndlovu et al. (2014) and Carberry et al. (2013) also discuss maize production efficiency in Zimbabwe, the former focuses on the productivity and efficiency of maize under conservation agriculture, and the latter analyzed the efficiency of maize farmers using crop simulation models without analyzing the factors contributing to the inefficiency.

⁶ A detailed description of the estimation procedure is available in Ferrara and Vidoli (2017).

⁷ The parametric and semiparametric models described in this section can be estimated using the R Environment (R: A language and environment for statistical computing 2017) by exploiting the following packages: frontier (Coelli and Henningsen 2013), semsfa (Ferrara and Vidoli 2015) and gamlss (Stasinopoulos and Rigby 2007).

⁸ As pointed out by one reviewer, the study focuses on Mazowe district, one of the most productive Maize production regions in Zimbabwe. The analysis presented here is therefore limited in scope as it cannot represent the full picture of Maize production in Zimbabwe given the heterogeneity presented across the country.

⁹ These data for $A2$ farmers were not collected in the survey.

¹⁰ Recall that $A1$ farmers are typically only equipped with less than 10 hectares of land, while $A2$ farmers often own over 10 hectares of total land.

¹¹ In the subsequent two years the average maize yield in Zimbabwe had further dropped to below 650 kg/hectare FAO(2018).

¹² We also collected the education background of household head. However, there is little variability with the education variable since most of the household head only received secondary education.

¹³ There are 113 A1 farmers and only 63 A2 households. We find our models perform poorly when applied to A2 farmers alone, perhaps due to the small sample size.

¹⁴ Since the Translog model nests the Cobb-Douglas specification, we can use the Wald and the likelihood ratio tests to determine whether the former provides a better fit than the latter, more parsimonious model.

¹⁵ In other words, farms can adopt different production technologies and as a result, observation-specific measures of the production technology may be estimated.

Additional information

Funding

This work was supported by the West Virginia Agricultural Experiment Station and the U.S. Department of Agriculture National Institute of Food and Agriculture, Hatch project [WVA00683].