Arsenic levels can vary with time, and old samples often used cruder analytical techniques that could not detect low arsenic levels (below 10 parts per billion). We found that the so-called reporting limits for arsenic (that is, the lowest level of arsenic in the water that states require to be reported) in many states was 5 to 10 ppb in the 1980's and even in the early 1990's. Figure 3 shows that in some states, such as California, many water systems testing their water for arsenic were allowed to report as nondetected any level of arsenic below the state’s relatively high reporting limits.

In many cases, those reporting limits later were lowered, due to improved analytical methods, and arsenic started to be reported in the water of many more communities, as would be expected. This presented a problem for our analysis: when a water system had for years not reported arsenic, and then reported it when the reporting limit dropped, how should we calculate the arsenic level for that system? Additionally, a relatively small number of water systems had very inconsistent reported levels of arsenic over time, and we had to decide how to report their average levels as well. We decided that when a water system conducted multiple tests of its water, we would use two different averaging techniques to estimate the arsenic exposure for consumers of that water:

• First, we calculated a very conservative low average, which assumes that when arsenic was not reported as detected, there was absolutely no arsenic in the water at that time, even if the limit of detection was high (for example, 10 ppb), and even if other tests showed that arsenic was present in the water at levels somewhat below the previous reporting limit. For example, if a water system did five tests when the reporting limit was 10 ppb from 1985 to 1990 and found no arsenic, and then tested twice in 1993 to 1995 when the reporting limit was 3 ppb, and it found 8 ppb both of those later times, the low average calculated for that system would be 2.3 ppb (that is, [0 ppb + 0 ppb + 0 ppb + 0 ppb + 0 ppb + 8 ppb + 8ppb] ÷ 7 measurements = 2.3 ppb).
  
  
• Second, we based our best estimate on a calculated mid-average, which assumes that if at least some arsenic was detected in a water system at some time, then whenever arsenic was not reported as detected, it was present at a level of one half of the reporting limit. Using the same example, if a water system had five tests when the reporting limit was 10 ppb from 1985 to 1990 and found no arsenic, and then tested twice in 1993 to 1995 when the reporting limit was 3 ppb, and found 8 ppb both of those later times, the mid-average calculated for that system would be 5.8 ppb (that is, [5 ppb + 5 ppb + 5 ppb + 5 ppb + 5 ppb + 8 ppb + 8 ppb] ÷ 7 measurements = 5.8 ppb).