AccessMyLibrary provides FREE access to millions of articles from top publications available through your library.
Designing effective public policy to deal with human health hazards requires understanding how behavior responds to changes in risk. If people adjust behavior to offset changes in risk, then: (1) actual health outcomes of a risk change would be overestimated by technological or close-response models that ignore behavioral adjustments, and (2) changes in time allocations and expenditures on risk-reducing goods may provide useful information about costs and benefits of policy changes (Bartik 1988). Individual averting action also may be a component of socially efficient externality policy (Shibata and Winrich 1983), and some risk communication efforts aim to promote mitigation.
Unfortunately, empirical evidence bearing on connections between risk and behavior is limited and controversial. In consumer product and traffic safety studies such as Viscusi (1984), Evans and Graham (1991), and Keeler (1994), actual behavior is not measured; instead, inferences about behavior are drawn from statistical studies of accident records.(1) Collectively, these studies suggest that a relationship of virtually any strength is possible between risk and behavior, ranging from the technologists' prediction of no effect to Viscusi's (1984) lulling effect in which behavior might more than offset a change in risk. A similar situation prevails among environmental health studies. Although it is sometimes suggested that people take defensive action when pollution increases (Krupnick, Harrington, and Ostro 1990), there have been relatively few attempts to link behavior to measured concentrations of pollution (see Akerman, Johnson, and Bergman 1991; Dickie and Gerking 1991; Doyle et al. 1991; and Smith, Desvousges, and Payne 1995). Related work focusing on actions taken or costs incurred to avoid contaminated water supplies (Harrington, Krupnick, and Spofford 1989; Abdalla 1990; Laughland, Musser, Shortle, and Musser 1996) has provided insights about how behavior responds to environmental hazards, but no clear picture of the connection between behavior and changes in health risk has emerged to assist in formulating public policy.
This paper analyzes unique panel data obtained from a survey of Los Angeles area residents to explain defensive responses to air pollution, especially ozone, using a behavioral model. Panel data are useful here because they allow estimates to control for individual heterogeneity, a potential source of bias in cross-sectional microdata studies. Also, responses to ozone pollution are of interest because: (1) ozone has been linked to acute health impairments in prior epidemiological and medical studies (U.S. EPA 1996); (2) spending less time outdoors effectively reduces exposure (U.S. EPA 1995); and (3) the national ozone standard, currently 12 pphm for maximum one-hour daily concentrations, has been debated intensely since the 1970s and may soon be lowered to 8 pphm (U.S. EPA 1996).
Results indicate that persons who experience smog-related symptoms spend significantly less time outdoors as ozone concentrations exceed the national standard: These individuals are predicted to reduce outdoor time by about 40 minutes on a day when the ozone standard is exceeded, compared to days when the standard is just met. Many people make other behavioral changes to avoid smoggy conditions and the propensity to do so appears to increase with schooling or if health symptoms are experienced. These results support the conclusion that people adjust daily activities to defend against acute health effects of air pollution exposure, but averting decisions appear less closely tied to chronic health impairments.
The remainder of the paper proceeds as follows. Section II outlines an averting behavior model that guides empirical work. Section III describes the data, Section IV presents empirical results, and Section V concludes.
The model follows closely previous work by Gerking and Stanley (1986) and others. Consequently, discussion focuses only on issues relevant to specification and interpretation of equations estimated in Section IV. An individual's utility function is specified as
U = U(X, H, A, [Alpha]), 
where X denotes consumption of a composite good and H represents current health status, [U.sub.X] [greater than] 0, [U.sub.H] [greater than or equal to] 0, and where health, in turn, is produced according to the household production function
H = H(A, [Alpha], K, S). 
In equations  and , [Alpha] denotes the concentration of air pollution, with [H.sub.[Alpha]] [less than] 0, [U.sub.[Alpha]] [less than or equal to] 0, and A represents an activity which may affect both health and utility, such as participation in an outdoor leisure activity. Marginal effects of A on H and U may vary in sign. For example, spending more time outdoors may improve current health status if pollution concentrations are low, but may damage it if concentrations are high. For clarity, A is assumed to reduce H at ambient levels of [Alpha] ([H.sub.A] [less than] 0) so that averting behavior involves decreasing A.
Remaining variables in the health production function denote the stock of preexisting health capital (K) and other human capital (S), where [H.sub.K] [greater than or equal to] 0, [H.sub.S] [greater than or equal to] 0. An individual with a chronic disease such as asthma has a lower stock of preexisting health capital, and all else equal has a lower short-term health status. Likewise, persons with less schooling or other human capital may be less efficient producers of H.
The individual maximizes utility subject to the health production function and full-income budget constraint
I + [Omega]T = [q.sub.X]X + [q.sub.A]A + [q.sub.M]M(H) + [Omega]G(H), 
where I, [Omega], and T, respectively, denote non-labor income, the wage rate, and total time available Also, [q.sub.j] = [p.sub.j] + [Omega][t.sub.j], = X, A, M, denote full, time-inclusive prices of X, A, and medical care M: [p.sub.j] represents the unit money price and [t.sub.j] represents time required to consume one unit of good j. The functions M(H) and G(H) take nonnegative values and, respectively, represent medical care consumption and time lost from market and nonmarket activities as a function of current health status, with [M.sub.H] [less than] 0, [G.sub.H] [less than] 0. Thus, lower values of H lead to greater medical expenses and more time lost from work and leisure activities.
First-order conditions for constrained utility maximization imply
[U.sub.A] + [U.sub.H] [H.sub.A] / [Lambda] = [q.sub.A] + ([q.sub.M][M.sub.H] + [Omega][G.sub.H]) [H.sub.A]. 
As shown, the individual equates the sum of direct and indirect effects of A on monetized utility ([Lambda] denotes the marginal utility of income), to the net marginal cost of A. Alternatively, the term [U.sub.H][H.sub.A]/[Lambda] [less than] 0 could be moved to the right-hand side of equation  and viewed as part of the cost of A. Under standard assumptions, first-order equations can be solved to express optimal choices of X, A, and [Lambda] as functions of all exogenous variables; for example,
[A.sup.*] = A([q.sub.X], [q.sub.A], [q.sub.M], [Omega], T, I, K, S, [Alpha]). 
This equation guides empirical specification in Section IV. Expected signs of partial derivatives of [A.sup.*] with respect to key arguments of equation  are discussed momentarily.
Several variants of this framework have appeared in the literature, including both one-period (e.g., Gerking and Stanley 1986) and multi-period (Cropper 1981) models where medical care is viewed as an input in the health production function, as well as approaches featuring uncertainty about final health outcomes (e.g., Berger et al. 1987). These models often focus on measuring values for pollution changes and thus typically incorporate more restrictive assumptions than those made here. One important condition often assumed is that averting action defends against all adverse consequences of pollution exposure, but provides no further benefit. In some models, then, [Alpha] and A do not enter the utility function directly (Gerking and Stanley 1986) and in others, health affects utility only indirectly through the budget constraint (Cropper 1981). The present model allows for the more likely event that air pollution and actions taken to avoid it have direct impacts on well-being in addition to their effects on health ([Alpha] and A enter the utility function directly). These conditions would seriously complicate application of the model to estimate willingness to pay for air quality improvements.(2)
Additionally, averting decisions may sometimes be discrete choices (Dickie and Gerking 1991). Although data used in Section IV include both discrete and continuous measures of averting behavior, the model presented focuses on the continuous case to simplify discussion of comparative statics and because a discrete choice model leads to an equation like  for the probability of choosing a discrete averting action. An appendix available from the authors …