Until recently, my job involved statistical prediction of indoor radon concentrations in U.S. homes. This work was performed with several colleagues at LBNL, including Tony Nero, Ken Revzan, and Mike Apte; with U.C. Berkeley statistics professor Andrew Gelman (now at Columbia U.); and with U.C. Berkeley grad students W. John "Rick" Boscardin Jr., Jane "Yvgenia" Fridlyand, Lan Zhou, and others.
More recently, I've been working on mapping air pollution using optical remote sensing data---measure the path-integrated concentration of a pollutant along several paths, and use the results to infer the 2- or 3-dimensional spatial structure of the pollutant. A little like a CAT scan, although we can't use the same methods. This work follows up on the work of Anushka Drescher, who also works at LBNL.
Andrew Gelman and I have also worked on quantifying statistical artifacts in various mapping methods, and using Bayesian statistical modeling to determine inputs for cost-benefit analyses.
For a list of publications (including conference talks and posters), click here.
Radon is a colorless, odorless noble gas that occurs naturally in rocks and soil. High concentrations can build up indoors. The dose-response relationship isn't known very well at low doses, but studies of miners and (sadly) experiments on dogs indicate that radon definitely increases lung cancer risk at high doses. If one extrapolates from high doses to typical indoor doses, then 5000 to 15000 people die per year of lung cancer that they wouldn't have gotten if they hadn't been exposed to radon. Most of these people are smokers---radon is believed to be much more dangerous to smokers than to non-smokers. The EPA recommends that anyone with living-area concentration over 4 pCi/L (150 Bequerels per cubic meter) should remediate their home.
The linear dose-response may substantially exaggerate the risk (or, then again, it may not---nobody knows), and my personal feeling is that I wouldn't spend $2000 to remediate my home if it were at 4 pCi/L. But actually, about 0.5% of homes nationwide have long-term, living-area concentrations over 20 pCi/L (750 Bequerels per cubic meter)! That concentration exposes residents to radiation levels higher than the occupational limit for uranium miners, and at that level there is definitely a substantial increase in risk.
Those very high homes are more likely to appear in some areas than in others, as you might imagine. (E.g., there are "hot spots" near Spokane, Washington; Reading, Pennsylvania; much of Iowa; the Red River Valley in North Dakota, South Dakota, and Minnesota; etc.)
Our mapping program was only moderately successful, mostly because even within a small area like a county there's lots of unexplained variation in indoor radon concentrations, so we could never really home in on the very high homes. There are some known risk factors, though, including: a basement that is used as living space; house located on alluvial geology; house in the northern half of the U.S.; etc.
We used statistical methods generically known as "Bayesian hierarchical modelling" to relate radon monitoring data with explanatory variables such as geologic and climate information, house type, etc.