Chronic vs. Sporadic
February 11, 2011
Fay and I have been going back and forth over this issue for the past couple of weeks as we try to finish a draft of the GBD paper. Condensing 1+ year of work down into 5000 words is never easy, but we want this story to be as clear and straightforward as possible for our readers, which means not telling them about every single thing we did.
In the end we have decided to classify each of the 21 WHO regions as being either chronically- or sporadically-exposed. This allows us to use two different sets of methods, each appropriate to the exposure type. The next step was to make the cut in a reasonable, defensible way. I am writing down here for myself, because with everything I have tried the path has become quite winding.
1) Take the full set of exposure values = 4208 populated cells * 120 months = 504960 values
2) Examine the distribution of these values, which is log normal and looks like this:
3) Take the 90th percentile value (3 ug/m3) as our definition of “high impact”
4) Count the number of high impact months in each of the 10 years in every cell
5) Define cells with >=3 high impact months in >=5 of 10 years as “chronically impacted”. That looks like this:
6) In regions where >= 40% of the population OR 40% of the land area is covered by chronically-impacted cells, call the region chronically-impacted. I went with 40% rather than 50% here because it was a natural break in both distributions. If we went with 50% in both cases Andean South America and Eastern Sub-Saharan Africa would be excluded. The final result looks like this:
