Revisiting farm size-productivity relationship: New empirical evidence from Ethiopia

ABSTRACT

Are small farms more productive? With this question in mind, this study revisits the farm size–productivity relationship and explores potential explanations using a unique plot-level data from predominantly wheat producers in Ethiopia. Overall, we find that small plots are more productive than large plots. We next test the conventional explanations hypothesised in the literature – labour market imperfection related to costly monitoring of hired workers and omitted variable bias related to soil quality – and find that neither of them essentially explains the inverse relationship. More importantly, we account for agricultural intensification and found no relationship between plot size and productivity. This suggests that the inverse relationship posited in the literature could simply arise from neglecting the impact of agricultural intensification.

KEYWORDS:

• farm size
• productivity
• inverse relationship
• agricultural intensification
• Ethiopia

JEL CLASSIFICATION:

• O12
• Q12
• Q15

Acknowledgement

The authors would like to thank Professor Keijiro Otsuka, Kobe University, for his insightful and critical comments in the previous version of the paper. All errors are those of the authors.

Disclosure Statement

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

ORCID

Solomon Bizuayehu Wassie http://orcid.org/0000-0002-5200-8851

Notes

1. Previous studies have used farm size, but this study uses plot size to test consistently major conventional explanations provided for the inverse productivity, using a single data set. Specifically, to account for household fixed effect and omitted variable (land quality) bias, that needs plot level observation.

2. There may be a significant wedge (e.g., transport and searching costs) between the opportunity cost of an hour of family labour and wage of hired labour (Newell et al., 1997). Hence, small farms use a larger proportion of family labour, and put in much more effort, compared with hired workers.

3. Kebele and Woreda are the first and the second lowest administrative unit in Ethiopia, respectively.

4. We adopted a log-log specification, the standard practice in the literature. Furthermore, we have also done the analysis using level-level functional form to examine sensitivity of results to functional forms (Appendix, Table A2).

5. We acknowledge that household fixed effects accounts for other factors besides labour market imperfections. However, to be consistent with previous literature (e.g., Barrett et al., 2010), we will use the household fixed effect to account for labour market imperfection.

6. This means that in Equation (1)(1) $\mathrm{l}\mathrm{o}\mathrm{g}\left(Y_{ihk}\right)={\alpha }_{1k}+{\beta }_1\mathrm{l}\mathrm{o}\mathrm{g}\left(Siz{e}_{ihk}\right)+Z_1^{\mathrm{\prime }}ihk\omega +{\tau }_1{W}_{ihk}+{\epsilon }_{ihk}$(1) , ${\epsilon }_{ihk}=\delta Plo{t}_{ihk}+{\upsilon }_{ihk}$, and as a result, the standard assumption $E\left(Plo{t}^{\mathrm{\prime }}\epsilon \right)=0$ does not hold as the correlation between plot characteristics, including land quality and plot size, is likely negative. In that case, the omission of these relevant variables causes the ordinary least squares (OLS) estimator of the true size–productivity relation to be biased; and is also true for Equation (4)(4) $\mathrm{l}\mathrm{o}\mathrm{g}\left(Y_{ihk}\right)={\alpha }_{4k}+{\beta }_4\mathrm{l}\mathrm{o}\mathrm{g}\left(Siz{e}_{ihk}\right)+Z_4^{\mathrm{\prime }}ihk\omega +\gamma A{I}_{ihk}+{\tau }_4{W}_{ihk}+{\eta }_{ihk}$(4) and Equation (6)(6) $\mathrm{l}\mathrm{o}\mathrm{g}\left(Y_{ihk}\right)={\alpha }_{6k}+{\beta }_6\mathrm{l}\mathrm{o}\mathrm{g}\left(Siz{e}_{ihk}\right)+{\varphi }_{6}H{H}_{hk}+{\delta }_{6}Plo{t}_{ihk}+{\gamma }_{6}A{I}_{ihk}+{\zeta }_{ihk}$(6) .

7. In fact, as shown in Table A3 (see Appendix), farm size has a negative and significant (−0.206, p-value = 0.00) effect on agricultural intensification. We have done one more exercise accounting for both land quality and input use, but the results are not different from those reported (Table A2 in the Appendix).

8. Hired labour used due to lack of reliable data on family labour per plot. However, the variation in the availability of household labour by a “family size” variable is included in the socio-demographic variables.

9. The principal component explains 38 per cent of the variance in the five considered variables. And the measure of sampling adequacy (KMO) value is greater than 0.5, which is considered as minimum acceptable value for PCA (Table A4 in the Appendix).

10. In order to conserve space estimation results not reported, but are available upon request.

11. We acknowledge an anonymous reviewer for the suggestion. Full specification of Table 4 appears in the appendix (Table A5).