Small and Productive: Kenyan Women and Crop Choice

Abstract

The question of gender differences in agricultural productivity has received particular attention in the development literature. The stylized fact that women produce less than men, while on average occupying smaller farms, presents a quandary as it is also a stylized fact that smaller farms have higher yields per unit of area. Using data from the 2006 Kenya Integrated Household Budget Survey, this study examines whether there is a gap in output per acre between men and women farmers in Kenya. Using ordinary and two-stage least-squares (OLS and 2SLS) analyses, it shows that when crop choice is taken into account, women are as productive as men. Specifically, the study finds that market-oriented crops are the source of differences. This suggests that further research into what determines crop choice is needed, in addition to policy that ensures that women have the same access as men to support for market-oriented crops.

Keywords:

• Agriculture
• women farmers
• Kenya
• Africa
• inverse relationship
• crop choice

JEL Codes:

• B54
• N57
• O13

ACKNOWLEDGMENTS

We would like to thank three anonymous reviewers, the editors, and Michael Carr whose careful comments and suggestions have all served to improve the paper. All errors of course remain ours.

Notes

¹The Land Matrix project is a partnership between the International Land Coalition and a series of European research centers. Since these land deals lack transparency, the confirmed cases of large-scale land acquisitions are, if anything, a subset of the actual cases. For the project's most recent detailed report, see Ward Anseewu, Mathieu Boche, Thomas Breu, Markus Giger, Jann Lay, Leter Messerli, and Kerstin Nolte (2012).

²These calculations assume a population of 41 million of which 78 percent is rural and the average rural household size is 5.6 people (KNBS 2007; World Bank 2012).

³The land laws were updated with the new constitution and remain gender neutral; however, there is no indication of a change in the actual practice of issuance of title deeds.

⁴Of the women farmers in our sample, 74 percent are in female-headed households.

⁵Carmen Diana Deere and Magdalena Leon (2001) and Thomas Masterson (2007) constitute exceptions to the rule. Their work focuses on Latin America, while our focus is Africa. In the African context, see for example Barrett, Belemare, and Hou (2010), who control for female-headed households.

⁶It is possible to have strong tenure security in the absence of formal titles. For example, this can happen in the presence of strong indigenous land governance institutions.

⁷Addis Tiruneh, Teklu Tesfaye, Wilfred Mwangi, and Hugo Verkuijl (2001); James Okuro Ouma, Hugo de Groote, and George Owuor (2006); Julliet Wanjiku, John Uhuru Manyengo, Willis Oluoch-Kosura, and Joseph T. Karugia (2007).

⁸Barrett, Belemare, and Hou (2010), using soil samples from 300 households in two Madagascar villages to impute soil pH and the plots’ carbon, nitrogen, potassium, clay, silt, and sand content, find no support whatsoever for the hypothesis that the inverse relationship between plot size and productivity is due to soil quality. Also, see Surjit S. Bhalla and Prannoy Roy (1988). While our soil controls are not as precise as those in the Barrett, Belemare, and Hou (2010) study, we use a national dataset that allows us to make inferences over the national farming households and different agro-ecological conditions and agricultural practices.

⁹This is the ratio of utilized land (amount of land per crop including multiple season crops) to total land owned. This ratio can be greater than one due to multiple cropping; for example, the planting of more than one crop on the same field.

¹⁰All of these differences are statistically significant at the 5 percent level.

¹¹As part of our robustness checks, we used alternative specifications for the dependency ratio: first, including household members over 65 years of age in the numerator and second, decreasing the cut-off age for dependent members to 12 years of age. The alternative specification did not yield substantive differences to the results.

¹²In literature on African agriculture, a distinction has often been made between cash crops (such as coffee and tea), which were produced for the market and subsistence, and food crops, which, in the past, were produced primarily for home consumption. As mentioned earlier, the markets for food crops have substantially thickened due to increased urbanization as well as export-oriented horticultural production. Therefore, we eschew the use of the term “cash crops” unless discussing an earlier context where the crops are specifically only industrial crops. Our study examines marketed output instead, which is measured as the proportion of total household production that is marketed.

¹³We estimate the impact of the dummy variable in a semi-logarithmic equation based on Peter E. Kennedy (1981).

¹⁴Cooperatives are organized mostly for traditional cash crops (such as tea and coffee), and government extension has tended to focus mostly on crops that are marketed.

¹⁵The instrument for crop choice is the number of other households in the same cluster that produce marketable crops. Clusters have a size of ten households; however, it is not clear that all households engage in farming. To ensure that the results of the instruments are not driven by clusters with a small number of farmers within the cluster, we perform the following robustness test, reported in Supplemental Table A2 in the online appendix, available under the supplemental content tab on the publisher's website. We set a minimum number of households that need to be in the cluster in order for the household to be included in our sample. The minimum number of households is the same as the regression number: regression 1 means that the minimum number of households in order to include the cluster in our calculation is one; regression 2 indicates that the minimum number of households is two; and so on. Our main results are robust.

¹⁶We can test this second assumption empirically. A commonly used rule of thumb is that the Kleibergen–Paap F-statistic from the first-stage regression should have a value larger than ten. James H. Stock and Motohiro Yogo (2002) give a more precise test, providing critical values for the Kleibergen–Paap F-statistic associated with various levels of bias as induced by the endogenous variables. In order to keep bias of the 2SLS results below 10 percent of the OLS bias (the lowest level Stock-Yogo provide), the Stock–Yogo critical value of the F-statistic is 19.93. The Kleibergen–Paap F-statistic in our model is 58.91.

¹⁷For both alternatives, we present the naïve OLS as well as the 2SLS specification.

¹⁸See columns 5 and 6 in Table 4.

¹⁹We use the data from the question “does the spouse/partner live in the household now?” For married households with men primary farmers, 5 percent have absent spouses, while for married households with women primary farmers, the number with absent spouses is much higher at 48 percent. Descriptive statistics for these categories can be found in Supplemental Tables A3.1 and A3.2.

²⁰This gives us the effect of both being a woman farmer and being in a female-headed household to help tease out the difference between being a woman farmer in female-headed versus male-headed households.