	We are sensitive to the appropriate treatment of data reported as detection limit values in the development of MACT floors and emissions limits. We believe that the statistical probability procedures applied in calculating the floor or an emissions limit inherently and reasonably account for emissions data variability including measurement imprecision when the database represents multiple tests from multiple emissions units for which all of the data are measured significantly above the method detection level. That is less true when the database includes emissions occurring below method detection capabilities regardless of how those data are reported. 
      The EPA's guidance to respondents for reporting pollutant emissions used to support the data collection specified the criteria for determining test-specific method detection levels. Those criteria ensure that there is only about a 1 percent probability of an error in deciding that the pollutant measured at the method detection level is present when in fact it was absent. (Reference: ReMAP: PHASE 1, Precision of Manual Stack Emission Measurements; American Society of Mechanical Engineers, Research Committee on Industrial and Municipal Waste, February 2001.) Such a probability is also called a false positive or the alpha, Type I, error. This means specifically that for a normally distributed set of measurement data, 99 out of 100 single measurements will fall within +-2.54 x standard deviation of the true concentration. The anticipated range for the average of repeated measurements comes progressively closer to the true concentration. More precisely, the anticipated range varies inversely with the square root of the number of measurements. Thus, for a known standard deviation (SD) of anticipated single measurements, the anticipated range for 99 out of 100 future triplicate measurements will fall within +- 2.54 SD/√3 of the true concentration. This relationship translates to an expected measurement imprecision for an emissions value occurring at or near the method detection level of about 40 to 50 percent.
      By assuming a similar distribution of measurements across a range of values and increasing the mean value to a higher representative value (e.g., 3 times method detection level or 3xMDL), we can estimate measurement imprecision at other levels. For a measurement at 3xMDL, the estimated measurement imprecision for a three test run average value would be on the order 10 to 20 percent. This is about the same measurement imprecision as found for Methods 23 and 29 indicated in the ASME ReMAP study for the sample volumes prescribed in the final rule (e.g., 4 to 6 dscm) for multiple tests.
      Instead of the method detection limit, analytical laboratories often report a value above the method detection limit that represents the laboratory's perceived confidence in the quality of the value. This independently adjusted value is expressed differently by various laboratories and is called limit of quantitation (LOQ), practical quantitation limit (PQL), or reporting limit (RL). In many cases, the LOQ, PQL, or RL is simply a multiplication of the method detection limit. Commonly used multipliers range from 3 to 10. Because these values reflect individual laboratories' perceived confidence, and, therefore, could be viewed as arbitrary, we decline to adopt the LOQ, PQL, or RL in the database for calculating floor values because such approaches in our view would inappropriately inflate the MACT floor standards. Our alternative to those inconsistent approaches is discussed below.
      We recognize that an emissions limit based on a small subset of data or data from a single source may be significantly different than the actual method detection levels achieved by the best performing testing companies in practice. This fact, combined with the low levels of emissions measured from many of the best performing units, led us to develop a procedure intended to account for the contribution of measurement imprecision to data variability in establishing effective emissions limits. That procedure is for identifying a representative method detection level (RDL). 
      For the total organic HAP emissions data, we applied the following procedure for determining an RDL for methods used in measuring organic HAPs. We determined method detection capabilities for Method 320 and Method 18 as appropriate for the various compounds (e.g., Method 320 for aldehydes, Method 18 for arenes). We believe that this approach is representative of procedures practiced by the better performing testing companies and laboratories using the reasonably applicable and sensitive analytical procedures. We determined for each of the organic HAPs the expected method detection level for the respective method based on internal experience and method capabilities reported by testing companies. With these reported values, we identified the resulting mean of the method detection levels as the representative detection level (RDL) because it is characteristic of accepted source emissions measurement performance. The following table summarizes those findings.
Pollutant
                                  Method used
                                      RDL
                               ppmv, as measured
                                      RDL
                                ppmv @7% O2 dry
Formaldehyde
                                  Method 320
                                                                           0.15
                                                                           0.23
Acetaldehyde
                                      "
                                                                           1.50
                                                                           2.34
Benzene
                          Method 18 (adsorbent tube)
                                                                           0.04
                                                                           0.08
Toluene
                                      "
                                                                           0.04
                                                                           0.08
Styrene
                                      "
                                                                           0.04
                                                                           0.08
Xylenes (m,p,o)
                                      "
                                                                           0.12
                                                                           0.24
Napthalene
                                      "
                                                                           0.03
                                                                           0.06
      
      The second step in the process is to calculate 3xRDL for the sum or total of the organic HAPs to compare with the emissions limit in the final rule. This step is similar to what we have used for other NESHAP including the proposed Portland Cement rule. As outlined above, we use the multiplication factor of 3 to reduce the imprecision of the analytical method until the imprecision in the field sampling reflects the relative method precision as estimated by the ASME ReMAP study. That study indicates that such relative imprecision remains a constant 10 to 20 percent over the range of the method. For assessing the calculated floor results relative to measurement method capabilities, if 3xRDL were less than the calculated floor or emissions limit (e.g., calculated from the upper predictive limit, UPL), we would conclude that measurement variability was adequately addressed with the initial floor calculation. The calculated floor or emissions limit would need no adjustment. If, on the other hand, the value equal to 3xRDL were greater than the UPL, we would conclude that the calculated floor or emissions limit did not account entirely for measurement variability. Where such was the case, we would substitute the value equal to 3xRDL for the calculated floor or emissions limit. This would result in an emissions limit where the method would produce measurement accuracy on the order of 10 to 20 percent similar to other EPA test methods and the results found in the ASME ReMAP study.
      We summed the RDLs adjusted for dilution and moisture as shown in the table above. We determined 3xRDL for the sum of the adjusted pollutants to be 11.2 ppmv @ 7 percent O2, dry. This value is greater than the final 9 ppmv @ 7 percent O2, dry, in the final rule. This result means that we would propose to adjust the total organic HAP limit to 12 ppmv @ 7 percent O2, dry.  At this level, we believe that currently available emissions testing procedures and technologies provide the measurement certainty sufficient for sources to demonstrate compliance at the levels of the revised emissions limit.

