Recessions and Health: The Impact of Economic Trends on Air Pollution in California

Abstract

Understanding the dynamics of human exposure to pollution is fundamental to the study of environmental epidemiology. Recent evidence suggests that economic activity may play an important exposure-modifying role in the relationship between air pollution and human health.1,2 Community-level air pollution is driven in large part by the economic engine within which these exposures are generated, ranging from nearby polluting factories to truck and vehicle traffic on roadways. When unemployment levels are low and the economic engine is working at full speed, economic output and air pollution–generating activities will likely increase, leading to higher ambient air pollution levels. The opposite scenario will play out during a recession, and regulatory policies aimed at improving air quality may act to speed up or slow down the underlying trend or pace set by the economy. For this reason, the effectiveness of regulatory policies is likely to be overstated when they occur during recessionary periods, and understated when they occur during periods of economic growth. By jointly considering the impact of both the economy and environmental regulations within a framework that models exposure to air pollution, it is possible to disentangle these important effects from one another. I identify the impact of economic activity and regulatory policy on the coefficient of haze (COH), carbon monoxide (CO), and nitrogen dioxide (NO₂) in a set of California counties in the years 1980 through 2000.

METHODS

I obtained historical air pollution data from the California Air Resources Board.3 I used these data to generate monthly median air concentration values, aggregated to the county level, for COH, CO, and NO₂ for the period 1980 through 2000. I chose these 3 pollutants because they share a similar source profile that derives in large part from transportation-related emission sources, which allows for a similar identification strategy and statistical modeling approach. COH in particular has been used to assess historical trends in diesel vehicle emissions in the San Francisco Bay Area4 and in New Jersey.2 I chose the 1980–2000 period because it represents the most complete panel of data for all 3 pollutants (COH data become particularly sparse after 2000). Counties were included in the study if the available data set provided at least 75% coverage over the relevant time period for all 3 pollutants. Table 1 provides a list of these counties by the California regulatory management districts known as air basins.

Modeling Approach

I employed a mixed modeling approach to estimate county-level air pollution concentration changes over time in California using the XTREGAR command in Intercooled Stata version 10.0 (StataCorp LP, College Station, TX). The estimated model included a first-order autoregressive error term [AR(1)] to control for temporal correlation across the monthly air pollution observations. I added a random county effect to control for unobserved differences across the monitoring locations not otherwise captured by the set of included model covariates. I used this approach to estimate the following equation:

where e_it = ρe_it-1+u_it, a_i is random county effect, N_it is weather variables, E_it is economic variables, R_it is regulatory periods, F_it is fire event, P_it is population density, M_t is month, i is county, and t is month plus year.

The air pollution concentration data are approximately log-normally distributed and have been log transformed in the model. a_i represents a random effect that controls for differences across the 19 different counties in the data set, and the error term e_it is a function of both the previous period e_it-1 (where ρ is the correlation across time periods) and a random component (u_it). N includes temperature, precipitation, and wind speed, whereas E incorporates both county-level unemployment and trucking employment statistics. The logged values of the employment statistics are estimated to facilitate direct interpretation of their coefficients as the percentage impact on air concentrations from a 1% increase in the economic variable (known as “elasticity” in economics).10 I have included the regulatory periods as dummy variables in R to control for their impact over time. For estimation purposes, I have excluded the initial regulatory period, and all later periods are judged against the first period. In addition to the baseline comparison provided by the regression estimates, I have calculated the “marginal effect” of each successive regulatory policy to identify the incremental change of each regulation against its predecessor. F represents a dummy variable denoting significant fire events, and P represents population density. Finally, I estimated monthly dummy variables (leaving 1 month out) to control for seasonal variations in air pollution concentrations within a given year.

RESULTS

Table 1 provides the air pollutant summary statistics by county within each air basin for COH, CO, and NO₂. All aggregate differences across air basins were statistically significant (P < .05). Of all the air basins, the San Joaquin Valley had the highest levels of COH and NO₂ and the San Francisco Bay Area had the highest levels of CO. Levels of all pollutants were lowest in the South Central Coast air basin. Figure 1 provides a graphical display of the monthly and annual trends across all California counties included in the study. There is evidence of a downward trend over the 1980 to 2000 period, with wide seasonal fluctuations apparent within any given year. From the beginning to the end of the study period, median annual levels of CO, NO₂, and COH were down approximately 43%, 37%, and 54%, respectively.

Open in a separate window

FIGURE 1—

Monthly and annual trends in air concentrations of (a) haze, (b) nitrogen dioxide, and (c) carbon monoxide: counties included in the study, California, 1980–2000.

Note. CO = carbon monoxide; COH = haze; NO₂ = nitrogen dioxide.

Figure 2 provides a graphical representation of the annual trend in overall unemployment matched with scaled trucking-sector employment (expressed as a percentage of the total county employment). There are broad shifts in unemployment over this period, with highs in the early 1980s and 1990s and lows in the late 1980s and 1990s; the trends in trucking-sector employment follow a similar inverse trend, with high unemployment corresponding to smaller workforce representation in the trucking sector and vice versa.

Open in a separate window

FIGURE 2—

Unemployment and trucking-sector employment trends: California, 1980–2000.

^aTrucking employment is multiplied by 5 for visual purposes to more closely superimpose the 2 trends.

The results of the random effects models are presented in Table 4. These models explained between 33% and 48% of the variability in air concentrations as estimated by the overall R2 value. All coefficients have the expected sign, with the exception of the fire variable for COH and CO. All monthly dummy variables are significant, which provides further evidence of the strong seasonal component to air pollution concentrations in California. The unemployment rate has the expected negative impact on all pollutants, and the coefficient is statistically significant for CO and NO₂. For every 1% increase in the unemployment rate, the model predicts a decline of 0.11% and 0.08% in CO and NO₂ levels, respectively. When viewed in the context of recent trends in the unemployment rate, these impacts are not trivial. For example, the California statewide annual unemployment rate was 4.9% in 2006 compared with 12.4% in 2010.5 A change of this magnitude would prompt 16.8% and 12.2% declines in ambient concentrations of CO and NO₂, respectively, all else being equal.

The relationship between air pollution levels and trucking-sector employment was stronger than the overall unemployment rate and statistically significant for all 3 air pollutants. For every 1% increase in the share of trucking-sector employment, the model predicts an increase of 0.12%, 0.17%, and 0.11% in COH, CO, and NO₂ levels, respectively. Although the share of trucking-sector employment in a county would not be expected to change as significantly over time as the unemployment rate, a hypothetical 50% increase would be expected to result in 6%, 8.5%, and 5.5% increases in COH, CO, and NO₂ levels, respectively, all else being equal.

The periods of major regulatory actions aimed at reducing mobile-source emissions play a statistically significant role in predicting the declining trend in air concentrations of COH, CO, and NO₂ in California over time. With the exception of the initial regulatory periods for COH and CO, all regulatory periods were associated with a statistically significant decline in air pollutant concentrations against the initial baseline period (1980–1982). The lack of a significant change in the early period is likely due to the fact that the time period is too brief (1 year) or the sample size within that period too small to observe a statistically significant effect of the regulatory intervention. With the exception of the initial diesel engine regulation on COH, the marginal effect of each regulation on air concentrations was negative (marginal effects are noted in parentheses in Table 4), suggesting that the federal and state regulations targeting mobile-source emissions were ultimately successful at lowering ambient concentrations of these air pollutants in California.

DISCUSSION

There is an extensive economics literature exploring the relationship between recessions and human health, with much of the emphasis attempting to explain the counterintuitive effect of improved health outcomes during economic downturns.11–13 In this article, I explore the hypothesis that cyclical economic activity is linked to human health through changes in air pollution exposure, and construct a framework for identifying these effects and disentangling them from environmental regulations.

Relatively few studies are available that have attempted to quantify the exposure-modifying effect of economic trends on air pollution. Using a modeling approach similar to that of the current study, an earlier study showed that changes in the economy had a statistically significant effect on COH in New Jersey in the period 1970 through 2003.2 Another study linked reductions in infant mortality to declines in total suspended particulates resulting from lower industrial output during the 1981–1982 recession, where a 1% reduction in total suspended particulates resulted in a 0.35% decline in the infant mortality rate at the county level.1 Along similar lines, reductions in PM₁₀ (particulate matter < 10 μm in diameter) following the closure of a major industrial polluter in Utah were associated with health improvements in the local population,14 and ozone levels in Europe have been linked to fluctuations in the European Union’s gross national product and industrial production.15 Finally, recent evidence from the US trucking industry further linked occupational exposures to local business activities.16,17

The current study provides evidence of a steady decline in COH, CO, and NO₂ in a subset of California counties over the study period, along with wide seasonal fluctuations in ambient concentrations within any given year. The modeled economic indicators had a statistically significant effect on air pollution concentrations. This evidence supports the overall hypothesis that a strong economy is associated with elevated air pollution levels and, in particular, mobile-source pollutants. Similarly, a weak economy is associated with lower air pollution levels. Although the impact of small changes in employment conditions on air pollution is relatively minor, large economic fluctuations such as those associated with recessionary periods can result in significant declines in air pollution levels. On the basis of these results, recent increases in the unemployment rate in California would be associated with an average 15% decline in the pollutants observed in this study.

Regulatory policies aimed at improving air quality played a statistically significant role in the declining trend in air concentrations of COH, CO, and NO₂ in California over this period. However, given the also significant co-occurring effect of economic conditions, regulatory impact assessment will likely overstate the outcome of a given regulatory policy when it occurs during recessionary periods, and similarly understate the effect when these regulations play out during periods of economic growth. For example, recent efforts to improve air quality in and around the ports of Los Angeles and Long Beach in California have resulted in significant improvements in air quality, with traffic-source pollutants reportedly declining by 45% in the 5 years since targeted reductions began in 2006.18 Although this specific example is outside the scope of the current data set, the evidence presented here suggests that at least part of these reductions are due to the recession that also set in during this time. Overall, the evidence suggests that it is critical for regulatory impact assessments to control for economic trends when evaluating policy effectiveness, especially during periods of strong economic growth or decline.

Acknowledgments

Human Participant Protection

References