Household Welfare Effects of Agricultural Productivity: A Multidimensional Perspective from Ghana

Abstract

Although poverty reduction policies in developing countries hinge on the link between smallholder productivity and welfare, micro level empirical evidence on this is limited in sub-Saharan Africa and neglects the multidimensionality of welfare. This article contributes through investigating the productivity-poverty relationship using a number of welfare measures with a three-period panel dataset for Ghana. The findings are that welfare is increasing with labour productivity irrespective of the measure used, but that dramatic increases in productivity would be required to achieve meaningful poverty reduction.

Acknowledgements

Financial support from the Institute of Statistical, Social and Economic Research (ISSER) is highly appreciated. The author is grateful to Chris Garforth who arranged for additional funding from the University of Reading to support part of the 2010 data collection. Thanks to two anonymous referees and Oliver Morrissey (the managing editor) who provided very useful comments and suggestions. Not the least, we are grateful to the households and community leaders who have made this research possible. The data and codes used in this study are available upon request.

Notes

1. Nonlabour income is assumed to be exogenously given.

2. Consumption expenditures include both actual and imputed. The imputed expenditures involve consumption of own produced goods. These are aggregated using market prices. Total consumption expenditure was adjusted using the Paasche price indices.

3. Most developing countries use a calorie requirement threshold of 2,000–2,300 kcal per capita per day. The Ghana Statistical Service uses 2,900 kcal per adult equivalent unit per day, which translate to about 2,202 kcal in per capita terms (Ghana Statistical Service, 2007, p. 6).

4. The assets used are telephone, television set, radio, tape recorder, bicycle, sewing machine, cooking stove and battery torch light. These assets are considered as wealth status symbols in the villages. The index was calculated separately for each village to avoid location bias.

5. See Filmer and Pritchett (2001), for example, who applies a similar definition.

6. Greene (2004) has shown that parameter estimates from a fixed effects (FE) probit model are biased in the presence of the incidental parameter and small T. The conditional fixed effects logit is not applied here because it requires that the dependent variable necessarily vary across time for each unit of observation. Given the small time dimension of the panel relative to the number of households and the time-invariance of some responses a large proportion (43%) of the data points drop out, making the conditional FE procedure inefficient in the current application.

7. The fact that linear combinations of normals are normally distributed aids numerical integration.

8. During the farming season in our study, villages’ young adults sometimes absent themselves from school to undertake important farming activaties such as planting and weeding, making this variable likely correlated with farm productivity.

9. There are no markets for organic manure in most of the study villages.

10. Households in the study villages generally cultivate two types of plots: those that are a long distance away from the homestead and those at the homestead.

11. The larger IV estimates in absolute value are common when there may be measurement error, which is corrected by the IV procedure (Griliches & Hausman, 1986).