Impact Evaluation of Climate Smart Agriculture Program Investments in Food Security Using Machine Learning Estimators

Abstract

Smallholder farmers in rural regions of developing countries are often vulnerable to climate events. Climate Smart Agriculture (CSA) seeks to sustain or improve agricultural yields while mitigating climate change. The CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) has made substantial investments in developing and scaling CSA programs in developing countries. Using a multi-country dataset, this paper runs two complementary analyses. The first uses a double/debiased machine learning approach to estimate the impact of participating in a CCAFS CSA program on household food security. I estimate this impact for the entire sample and within three sub-samples, which categorize households according to their CSA adoption strategy (i.e., non-adoption, specialized adoption, or diversified adoption). Results indicate that the probability of a household being food secure is 6.0 percentage points higher (p<0.05) if it participated in a CCAFS program. The food security benefits of CCAFS program participation are most clearly demonstrated among households that adopted a diverse set of CSA practices, where CCAFS training increased the probability of being food secure by 9.7 percentage points (p<0.05). On the other hand, the food security benefits of CCAFS training were negligible among households that did not adopt CSA practices or adopted a specialized set of practices. The second analysis combines traditional machine learning tools with future climate data to predict and compare the future food security of CCAFS program participants and non-participants. Results show that participating households are more likely to be food secure than non-participating households across all periods. Overall, the food security gap between participating and non-participating households is expected to increase over time.