Factor Market Activity and the Inverse Farm Size-Productivity Relationship in Tanzania

Abstract

Although the inverse farm size-productivity relationship (IR) is sometimes used to motivate arguments in favour of smallholder-led agricultural development, it remains unclear what drives this relationship. It may be attributed to market imperfections that compel small farms to use land more intensively than large farms. Using a three-wave longitudinal household survey from Tanzania, we examine whether the intensity of the IR is related to local factor market activity for land, labour, credit, and animal and machine traction. The IR holds in Tanzania when family labour is either not counted or valued at its shadow cost, though it disappears when family labour is valued at the prevailing local agricultural wage rate. Moreover, the IR is significantly weakened in regions with relatively active agricultural factor markets, such as for land and mechanization services. This suggests that the IR is at least partly driven by imperfections in rural factor markets. As household participation in agricultural factor markets continues to rise, the IR may be expected to weaken or even reverse.

Acknowledgements

The authors are grateful for the encouragement and support of the Food Security Group in the Department of Agricultural, Food, and Resource Economics of Michigan State University. The Stata code used to produce this analysis, along with instructions on how to access the data, are provided in the Supplementary Materials that accompany this paper.

Data availability statement

The data that support the findings of this study (the National Panel Survey of Tanzania) are publicly available from the World Bank at http://microdata.worldbank.org/index.php/home.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2020.1797686.

Notes

1. This pattern is also seen in Tanzania, although it varies across different African countries. Nevertheless, across all countries considered, the IR is evident when GPS measurements are used (Carletto et al., 2015).

2. Recently, potential measurement error in terms of farmer-estimated crop harvest has received attention (Desiere & Jolliffe, 2018; Gourlay, Kilic, & Lobell, 2019; Lobell et al., 2019). However, estimates of crop yield derived from crop cuts may also exhibit bias, may produce results that are not representative of the entire plot (especially when planting densities are uneven), or may be inconsistent with what a farmer would harvest (Diskin, 1997; Fermont & Benson, 2011; Food and Agriculture Organization of the United Nations [FAO], 2017; Sud et al., 2016). Given that many analyses of agricultural outcomes in developing countries rely on estimated yields and often produce logical results (for example, the returns to labour intensification), while these familiar relationships sometimes disappear when using crop cut-based yield measures as the dependent variable, it is not clear that systematically misreported yields are pervasive. Additionally, remotely sensed measures of crop yield alternately show no relation with area planted (Gourlay et al., 2019) or a statistically significant negative relation (Lobell et al., 2019), again suggesting that the IR is not entirely explained by measurement error.

3. It should be noted that a number of analysts have conducted tests for the IR at plot-level with the inclusion of household-fixed effects and concluded that the IR is not attributed (or only limitedly attributed) to market failures (Assunção & Braido, 2007; Carletto et al., 2015, 2013). This is because the market conditions faced by a single household would be controlled for in this model specification, and yet, the inverse relationship between plot area and crop production persists. However, Barrett et al. (2010) do note that controlling for household fixed effects reduces the magnitude of the size-yield relationship by about one-third, and it, therefore, seems likely that multiple factors (potentially including household-level market conditions) drive the IR.

4. Although both rental and sales markets for land and traction would be of interest in a study of market imperfections, data availability limits our focus to rental markets.

5. An alternative model specification was considered involving the individual farms’ factor market participation. However, this was deemed unsuitable in light of the data limitations. As noted in section 3, for both labour and land markets, the mechanism through which the IR would be attenuated may relate to small farms renting out/hiring out their endowments. Unfortunately, this side of the market is crudely captured in the data set, with evident under-reporting of rented out land (as noted by Deininger, Savastano, and Xia (2017)), and only a binary indicator for agricultural wage work (and only if this was the main employer for a household member). The correlation coefficient between the region-level proportion of households that sell agricultural labour as a member’s main occupation and the proportion that hire in agricultural labour is just 0.28, with far more households recorded as hiring in labour.

6. Unfortunately, the data set does not contain a complete picture of the supply side of these two markets.

7. The number of observations in each model can vary, depending on whether we limit our attention to the main growing season or the entire agricultural year.

8. Because the focus of our paper is on the tables that follow, several additional robustness checks of Table 3 are not reported here. These results are quite robust when the smallest 2 per cent of farms are removed from analysis, when household-fixed effects are removed, and when a quadratic term for area is included (with results revealing a convex area-productivity relationship).

9. As noted by Muyanga and Jayne (2019), these net values of production should ideally account for the fixed costs of production, including the purchase of oxen, machinery, and land. In most cases, we lack the necessary information to impute seasonal or annual values.

10. In a final exercise, households are divided by whether they face above- or below-median levels of factor market activity in 2009, based on the household-specific measure of market activity, and equation (1) is applied to each group separately. The results, made available in Table A7 of the Supplementary Materials, further underscore the manner in which the IR is much weaker in the above-median category for each market.

Additional information

Funding

This work was supported by the USAID/Bureau for Food Security through the Feed the Future Innovation Lab for Food Security Policy Cooperative Agreement with Michigan State University, and by the Bill and Melinda Gates Foundation through the Guiding Investments in Sustainable Agricultural Intensification in Africa grant to Michigan State University.