Jousting with the Lancet: Pajamas Media Interviews Professor Gilbert Burnham
Having worked with small samples from very large populations, in the financing of international project, I have experienced and caught my own errors due to the large numbers to which data small samples are extrapolated. For example, a minimum food diet requirement of 1800 calories is statistically estimated in a a sample of of 25 clusters of about 50 adults each in for a country of 60 million. Two surveys reveal the following: first, the national food balance sheet shows 2398 cal are available from local food production and imports; second, family budget enquiries in the 25 cluster show actual consumption of 2,418 calories. In a country of 1 million person, the discrepance is not large , only 20 calories per capita, or 1 million cal. per day, which is equivalent to about 500 more tons of food to be produced or imported each day. Over one year, the total becomes 182,500 tons, which is significant for even 1 million persons. At $1,000 per ton, this becomes real money, or about $182.5 million, for a small country. Now friends, apply this “error” to a country of 20 or 25 million persons and observe the magnitude of the difference! It’s devastating. The lesson is that even small sampling differences become magnified/exaggerated when applied to large population (even when sampling errors are estimated).
