Using typologies for technology targeting
This document discusses using farm typologies and statistical techniques to target technologies to specific farms. It presents a method for assigning farms to existing typology groups based on a small set of farm characteristics. An analysis of farm data from Tanzania, Malawi and Ghana found a high percentage (over 80%) of farms were correctly assigned to typology groups. The next steps discussed are using these typology databases combined with spatial socioeconomic and biophysical data to generate ranked lists of suitable technologies for individual farms based on their features, socioeconomic environment and biophysical conditions. The goal is to move from recommending technologies for broad farm types to targeting technologies tailored to individual farms.
Using typologies for technology targeting
- 1. Using typologies for technology targeting Jeroen Groot (WUR), Beliyou Haile (IFPRI) and Carlo Azzari (IFPRI) Africa RISING ESA project review and planning meeting 12 September 2017 Zanzibar, Tanzania
- 2. Using typologies for targeting We assume that: technologies are suitable for specific farms or farm types, dependent on, e.g.: Farm size Production orientation Farmer perspective and ambitions To scale out to new farms we have to identify the farm type to propose the most promising interventions
- 3. Assigning farms to an existing typology Steps: Statistical typology: 4 farm types in each country Identify 10-15 most discriminating variables RDA Collect data on farm for the 10-15 variables Calculate probability of belonging to each type NBC Farm is assigned to the type with highest probability We performed an in-silico experiment following these steps RDA = redundancy analysis NBC = naïve Bayesian classification
- 4. RDA result: selected variables RDA = redundancy analysis
- 5. NBC result: allocated farm types Used randomly selected 75% of the dataset to train the NBC Tested allocation of 25% dataset to farm types (115- 190 farms, dependent on dataset size) High percentage of correct allocations to types: Tanzania: 83.2% Malawi: 82.1% Ghana: 88.8% NBC = naïve Bayesian classification
- 6. Next step: allocate technologies to farms Using similar approaches and statistical techniques Combining typology (ARBES) databases with spatially explicit socio- economic and biophysical datasets Using a small set of variables collected on farm by advisors or the farmers Generating ranked lists of most promising set of technologies for a farm, given its: Farm features Socio-economic environment Biophysical conditions Move from “suitable for” farm types, to “suitable for” farms
- 7. Targeting: dealing with differences Taking into account: Biophysical conditions (using spatial data) Socio-economic setting (using multiple datasets, ARBES/typology data) Farm structure and endowment and household structure and relations Handheld device App Website PC Application Farm data (10-15 variables, problems, prefs) Cloud DB Matching algorithm Technology suitability ranking Biophysical DB Socio-econ. DB
- 8. Africa Research in Sustainable Intensification for the Next Generation africa-rising.net This presentation is licensed for use under the Creative Commons Attribution 4.0 International Licence. Thank You