Can less data be better data? The costs, benefits, and issues related to a less expansive data strategy Gary Eilerts, USAID FEWS NET CRED Workshop, June 9, 2011
FEWS NET: Works in conflict environments Provides early warning coverage of 30+ countries Food security focus requires multiple data streams and baselines Favors accuracy over precision Accepts quick and dirty Must inform large resource decisions Is 25 years old Fights every year for funding – needs to show results See
“FEWS NET has a data strategy…?” a skeptic, some years ago
The FEWS NET retort: “Of course we have a strategy!” demographic data nutritional data weather data price data agricultural data livelihoods information
The strategy… “What data do we need?” The automatic answer: “Anything that is related to our primary mission (needs)”
Many partial datasets, many partially answered questions, too much spent on both… and the need for a new strategy… The results?
FROM Everything we need to know to answer any question about ag production in country x’s food security: main season area planted: maize yield: maize area trend over time: maize yield trend over time: maize poor year area planted: maize good year area planted: maize area planted: sorghum yield: sorghum etc. TO: What do we need to be certain enough… if main season agricultural production is vastly above, below, or near average in country X… and what general impact it may have on that country’s food security? Is an alternate strategy feasible (more specific questions, smaller in scope)…??
Real Example: Need: sufficient information about food availability prospects in country X If: Biggest and most significant uncertainty in the production equation for country X is “area planted” If: Wheat production counts for >70% of country’s food availability Then: can we get most of what we need with a good-enough area planted estimate?
ProvinceTotalCropStd. ErrorIrrigatedRainfedCFSAM (km 2 ) Badakhshan Badghis Baghlan Balkh Faryab Hirat Jawzjan Kunduz Samangan Sari Pul Takhar Total Less data, more info: This amount of data is sufficient for a statistically sound country X northern ag area estimate, with known error bars In application, it produces this… compared to the CFSAM
Same amount of weather data for country Y….
… but with a better set of questions, better information about climate change in country Y…
…and countries Z and A… Pattern of rainfall (to <500 mm isohyet) decline Pattern of temperature increase (to >30c average daily temperature)
But, some fundamental issues remain… Is this information good enough?
In light of this information, probably not… Food and agriculture organization B
But where can/do we stop?
We will do what we can to share our strengths and our weaknesses, and you’ll get them all in our data
Perceptions/reality: at little cost and what benefit?
Intermediate conclusions: Less data, used wisely, may be good enough Same data, better questions, may provide better answers A variety of new tools promise both Some fundamental data questions remain to be answered My colleague’s data costs me nothing but a trade; I am happy to trade We all need household, national, regional and global info, and should build/sustain capabilities at all these levels.