Ameet Morjaria NSF-AERC-IGC Workshop Mombasa, 4 th Dec 2010 Comments on: “Adoption and Impact of Conservation Agriculture in Central Ethiopia: Application of IV and Control Function Approaches” by Kassie et al.
Overview Motivation –Natural resource degradation is a serious and worsening issue for rural livelihoods in developing countries. –Amplified when agriculture is operated by smallholder farmers (plough based agriculture) especially in Africa (ILRI 2009). –Conservation agriculture (CA) aims at mitigating and making better use of agricultural resources. [CA≡ interventions to deterioration of soil + water resources] What does the paper do? –Empirical investigation into … 1. Adoption of CA = f( ?) 2. Adoption of CA impact on land + labor productivity
Overview Data –2/9 Ethiopian districts where SG 2000 was promoting CA. –Within district choose kebeles – criteria (CA + farmers cooperative) –Within kebeles choose households (random sampling) –Cross-sectional data collected in 2007/8 of HH level characteristics. Methodology –Adoption: Multivariate probit model to estimate different/all adoption of the components of CA estimation using GHK simulator –Impact: estimating structural models of outcomes (crop yield + labor productivity) taking into account endogeneity and heterogeneity concerns by using control functions.
Overview Findings –Adoption of CA = f( location, family size, access to extension, formal education) –Herbicide application (one component of CA) land productivity –Land productivity influenced by location, gender of hh head, livestock wealth, human labor endowment. –None of the above impact labor productivity.
Concerns + Suggestions –The CA implementation how was it done? Phased in? All districts at once? Non-random program placement –Explain institutional set up of SG 2000 –Explain district selection – why 2/9 districts? Are other districts similar? Kebeles selection? Sample selection concern –Show district characteristics to see if similar –IV use suggested – but what endogenous variables are the concern in the context? Discussion is abstract and would be useful to be context-specific.
Concerns + Suggestions –IV solves the endogeneity problem but concern when unobservable factors interact non-linearly with the exogenous regressors. –But why is this a concern in this context - theory? Examples? –The above concern is mitigated by estimating control functions (CF) which are generalizations of IV estimation. Discussion of CF missing. –Relies on same identification assumptions as IV methods (2SLS/GMM) but is based on conditional mean rather than linear projection. –If assumptions hold likely to more efficient but less robust than IV approach.
Concerns + Suggestions –Is the sample size enough to run multivariate probit? Any benchmarks on what is reasonable to have in each choice? –Run probit unpooled regressions i.e. sub-sample as less restrictive –Compare estimation from general IV and CF – to give indication of the bias. –Account for social networks in adoption? e.g. Bandiera Rasul (2006)…. –No findings of CA on labor productivity. Is that surprising? Labor productivity + learning might take longer and cannot be captured in X-sectional data… –Minor point: Woreda dummy significant soaks up anything that is time invariant at district level including the type of crop (page 9).
Conclusion Contribution -Important question -Add to the limited empirical literature on CA -Go beyond adoption of CA and look at outcomes (productivity) -Data contribution as Africa data scarcity