EPA interim final report on WTC background study, April 2003

NYC 9/11 Public Portal Document

The program ProUCL tests the normality/lognormality of each data set using the Shapiro-Wilk W Test for sample sizes of 50 or smaller, and using the Lilliefors Test for sample sizes of 51 to 1000. The data sets for this project included samples sizes that are less than 50 and greater than 50, therefore both tests were used depending on the sample size for each analyte. The majority (92%) of the data sets were neither normally (0.5%) nor lognormally (7%) distributed, therefore a non-parametric statistic was used to estimate most of the UCLs.

Descriptive statistics, including the minimum, maximum, mean, median, standard deviation, coefficient of variation, skewness and variance were calculated. In addition, approximately ten different statistical methods were used to calculate UCLs for each analyte’s data set. The output from these calculations also reported descriptive statistics for the log-transformed data, which included mean, standard deviation, coefficient of variation, skewness, median, 80%, 90%, 95%, and 99% quartiles. The output also included a recommendation regarding which calculated UCL was most representative of the data set. This UCL is reported in Tables 6-1 through 6-6.

The UCLs were calculated with the statistical procedures based on the findings of Singh, Singh, and Engelhardt (1997,1999), as referenced in the ProUCL users guide (Lockheed Martin, 2001). For normally distributed data sets, the Student’s t-statistic, based on an arithmetic mean, was used to calculate the UCL. For lognormally distributed data sets, one of five different computational methods was used, based on the log-transformed mean®, depending on the skewness of the data, which was indicated by the standard deviation of the log-transformed data, and the samples size. For data sets that were neither normally nor lognormally distributed, a UCL based on the non-parametric Chebyshev theorem and arithmetic mean was used (Lockheed Martin, 2001).

As stated above, these statistical methods were completed for each analyte, sampling matrix (i.e., bulk dust, air, microvacuum, and wipe samples), and space tested (i.e., resident, common area, and both combined). There were also some matrices in which multiple surfaces were sampled (e.g., counters, floors, and walls), so additional calculations were conducted for different surfaces. The text that follows explains the statistical evaluation of the data and the results of the statistical analyses, as presented in tabular format (Tables 6-1 through 6-6). The values presented in these tables provide estimates of the background concentration for each compound and media.

An additional assessment was undertaken to provide a single recommended value for each analyte for the air, bulk dust, porous materials, and hard surfaces. These values, which are summarized in Table 8-1, can be used as a representative background concentration for Manhattan given the statistical limitations of the data set.

® The anti-log of the log-transformed mean is equal to the geometric mean. WORLD TRADE CENTER BACKGROUND STUDY 21

NYC-WTC_000145896