Access, adoption, and diffusion: understanding the long-term impacts of improved vegetable and fish technologies in Bangladesh

Abstract

This paper assesses long-term impacts of early adoption of vegetable and polyculture fish production technologies on household and individual well-being in Bangladesh. In 1996–1997 and 2006–2007, a panel of households were surveyed in three sites where non-governmental organisations and extension programmes disseminated agricultural technologies. Using nearest-neighbour matching to construct a statistical comparison group, the authors find that long-term impacts differ across agricultural technology interventions and across outcomes. Long-term impacts on household-level consumption expenditures and asset accumulation are, in general, insignificant in the improved vegetables sites, but are positive and significant in the individually operated fish ponds sites. However, the impacts on individual nutrient intake, nutrient adequacy, and nutritional status do not follow the pattern of household-level impacts. Differences in long-term and short-term impacts arise from several causes: differences in dissemination and targeting mechanisms that may affect household-level adoption decisions; initial differences between comparison and treatment groups; divisibility and ease of dissemination of the technology; and intrahousehold allocation processes that determine the allocation of gains from the new technology among household members.

Keywords:

• adoption
• agriculture
• Bangladesh
• impact assessment

Acknowledgements

This research was funded by the UK Economic and Social Research Council and Department for International Development under their Joint Research Scheme (Award Number RES 167-25-0361), building on a longitudinal dataset whose collection was funded by the Chronic Poverty Research Centre, HarvestPlus, and the University of Waikato. The authors thank DATA Ltd for their substantive cooperation throughout this project, Dan Gilligan for invaluable advice on methodology, and two anonymous reviewers for helpful comments. They also acknowledge helpful discussions with and comments from Akhter Ahmed, Bob Baulch, Peter Davis, Wahid Quabili, Mohammed Zahidul Hassan, Valerie Mueller, Selim Raihan, and Mohammed Zobair. All errors and omissions are the authors' own.

Notes

1. The indicator of underweight children below age five combines the effects of both long-term/chronic (expressed as ‘low height for age’ or stunting) and short-term/acute (expressed as 'low weight for height' or wasting) undernutrition.

2. Of 43 studies reviewed by Haddad et al. (1996), pro-male bias in nutrient allocations appears to be most prevalent in South Asia; boys in this region are also more favoured in the distribution of non-food health inputs, such as healthcare. Furthermore, this is the only region of the world where girls have higher child mortality rates than boys.

3. The previous evaluation, conducted a few years after the technologies were disseminated, looked at short-term impacts using single-difference analysis, and relied on with-and-without comparisons arising from the evaluation design without using matching methods. Since the interventions were not randomised, the potential for selection bias exists. Using panel data does not completely resolve this issue, but it allows us to control for unobserved time-invariant variables with time-invariant effects. We also match on observables measured at baseline that are not likely to change over time nor be affected by the intervention.

4. Comparisons among programme members with early access to the technology and those waiting to receive the technology are discussed in Kumar and Quisumbing (2010), but are not presented here owing to space considerations.

5. This discussion adapts the exposition of King and Behrman (2008) to the evaluation of two different agricultural technologies.

6. This description draws from Hallman et al. (2007), and recent field visits by the authors.

7. The individual fish ponds households were sampled from Gaffargaon thana, Mymensingh District, and Pakundia and Kishoreganj Sadar thanas, Kishoreganj District; this site is collectively referred to as Mymensingh.

8. The phases of implementation of the Mymensingh Aquaculture Extension Program are described in Orbicon and Lamans Management Services (2009).

9. The attrition analysis reported in 2010 suggests that variables that are significant predictors of attrition including the age of the household head squared, selected household demographic variables and one upazila (subdistrict) and one adoption status dummy.

10. Nominal values are deflated using the rural Consumer Price Index to reflect them in real terms in 2007 prices.

11. We use nnmatch in Stata10 to estimate our matching estimators (Abadie et al. 2004).

12. We do acknowledge that DID estimates only control for time-invariant effects of time-invariant unobservables. Because we do not have a true baseline, it is possible that our estimates underestimate the true impact of the interventions.

13. These are matched observations of early versus late adopters.

14. The change in landholdings is expressed as the percentage change in land area (measured in decimals) while the change in the fraction of fish income is the percentage change in the fraction of total household income from fish ponds.

15. Food consumption expenditures are actually the value of food consumed, valued at market prices. The food consumption module includes both food purchased from the market and food consumed from own production.

16. Conversion factors to nutrient intakes use the International Minilist from the World Food Dietary Assessment System (available from http://www.fao.org/infoods/software_overview_en.stm). Nutrient data have been compiled into an international food composition table, which contains 195 food items that represent basic foods consumed worldwide. For each food on the International Minilist, there are complete values for 52 constituents: the data are taken from published food composition tables, or imputed if no analytic data are available; there are no missing entries. The source of each entry is documented.

17. For the child outcomes, we use the World Health Organisation growth charts, which are now available for children up to age 18. By construction, children in our sample would be between ages 10 and 18.

18. While it could be argued that early adopters may have time-varying unobserved characteristics that affect both early adoption and outcomes, by definition, early adopters are those in villages that had the technology first. The timing of adoption was a result of programme roll-out.