FIGURE 2

(a) Average deviance of benzene concentration in blood (g/l)² computed by performing jack-knife cross-validation on regression trees of varying size comprised of all available measurements of benzene in environmental and dietary media. The number of terminal nodes (size) of the optimal regression tree model is identified to be 2, based on the minimum value of the average deviance. (b) Dendogram representing the optimal regression tree describing benzene concentrations in blood (g/l) as a function of benzene measurements in environmental and dietary media. The only predictor variable is MEAS180 (Indoor Air Concentration). The distribution of benzene blood concentrations in each terminal node is shown graphically under the node as fractions of the quartiles of the overall distribution. The numbers in parentheses represent the number of observations at each split/node. (c) Average deviance of benzene concentration in blood (g/l)² computed by performing jack-knife cross-validation on regression trees of varying size. The number of terminal nodes (size) of the optimal tree model is identified to be 4, based on the minimum value of the average deviance. (d) Dendogram of the optimal regression tree describing benzene concentrations in blood, as a function of environmental concentrations, baseline and descriptive questionnaires, and time–activity diaries. The values of the predictor variables specifying the observations in the left branch of a split are shown above a split, and the average value of the response variable is shown under the terminal nodes of the tree. The numbers in parentheses represent the number of observations at each split/node. The distribution of benzene concentrations under each terminal node is shown graphically at the bottom of the plot. The predictor variables appearing in the tree are: (a) TA18: the average number of min/day spent indoors with someone who was smoking; (b) TA13: the average number of cigarettes/day smoked by the subject; (c) AB23: a categorical variable describing the year in which the subject moved into the residence.