The Global Environment Monitoring System (GEMS) has been recording SO2 concentrations in major urban areas in developed and developing countries since the early 1970s. Our data set consists of 2621 observations from 293 observation sites located in 109 cities representing 44 countries spanning the years 1971-1996. The GEMS network was set up to monitor the concentrations of several pollutants in a cross section of countries using comparable measuring devices. The panel of countries includes primarily developed countries in the early years, but from 1976 to the early 1990s the United Nations Environment Programme provided funds to expand and maintain the network. The coverage of developing economies grew over time until the late 1980s.
In the 1990s coverage has fallen with data only from the US for 1996. WHO (1984) reports that until the late 1970s data comparability may be limited as monitoring capabilities were being assessed, many new countries were added, and procedures were being developed to ensure validated samples. Accordingly, we investigate the sensitivity of our findings to the time period, but leave the reporting of these (largely confirming results) to Appendix C.
The GEMS data is comprised of summary statistics for several percentiles of the yearly distribution for concentrations at each site together with highest recorded values and both geometric and arithmetic means. In this study we use the log of median SO2 concentrations at a given site, for each year, as our dependent variable. We use a log transform because the distribution of yearly summary statistics for SO2 appears to be log normal (WHO (1984)).
Previous work in this area by the WHO and others has argued that a log normal distribution is appropriate because temperature inversions or other special pollution episodes often lead to large values for some observations. In contrast, even weather very helpful to dissipation cannot drive the natural level of the pollutant below zero.