Test data and uniform standards

Example 61 The raw data, given as %-of-nominal values with one decimal place, are found in Table 4.37 For each group of 10 values the mean and the standard deviation were calculated. Using these, the t-values for the differences L - mean, with L = 75, 85, 115, resp. 125% were determined they are all above 2.9, indicating low risk. The corresponding CP-values were calculated the differences ACP75-85 and ACP 15-125 were added and multiplied by 100 to obtain the approximate risk, in %, of finding a result between the inner and the outer limits. For a content uniformity test with n = 10 tablets, a risk of 0.003872% translates into a deviant result once every 20-25 trials, or, with six CU runs per batch, every third or fourth batch. 

It is important to be aware of the criteria used by the authors of the studies cited in selecting their data, because similar requirements should apply to schemes to which uniform standards will be applied. Test protocols, test strains or cell lines, genetic loci, metabolic-activation requirements, mutant-selection protocols, and many other factors were evaluated before data were acceptable for publication. This kind of evaluation is absolutely necessary in constructing a relative-potency scale like the one we propose. Although not all GENE-TOX evaluations are complete, the Committee views the current GENE-TOX effort as a good source of protocol and data evaluation... 

For the purposes of comparing assay, content uniformity, and dissolution data, simple statistics such as sample mean value (SMV) and relative standard deviation (%RSD) derived from experience of performing the tests over long periods of time can be used as acceptance criteria. Alternatively, more sophisticated statistics such as the z-test, F-test or t-test as shown in Table 16-4 can be applied [17-19,25]. In the case of evaluating CU data, it can be concluded that results from two labs are equivalent based on applying the simple statistics of the difference between the SMV from Lab A and Lab B (Table 16-4) not to be more than 2.0%. In the other examples where the more sophisticated statistics such as z-test, F-test, or t-test are applied (Table 16-4), results from two labs are considered to be equivalent because the calculated statistics in each case (z-calculated values of 0.32/0.64, F-calculated value of 0.14, or T-calculated values of 0.30/0.60) are less than the predicted statistics (z-critical value of 1.64, F-critical values of 3.18, or T-critical value of 1.73) [19,25]. 

Consider the following application where uniformity of the blend is often used to determine blend time, and where standard deviations from two data sets are compared. In this illustrative example, 10 samples were collected from two different time points, one sample set from the blender after 10 minutes blend time, the next set after 14 minutes blend time. The first set yields a mean of 1(X) and a standard deviation of 1.9. Tlie next set of data coincidentally yields a mean of 100 but a standard deviation of 3.7. Assume both are normally distributed. In this example, one might believe that, based on this data, the blend was segregating with additional blend time, and one may decide that it is essential to keep the blend time to 10 minutes. Our eyes and intuition tell us these are very different data sets. Yet an F-test would show that these values are not different with 95% statistical significance. In this case, additional samples from each time point, in addition to samples collected from other time points, would help address whether the blend has or has not in fact hit an optimum at 10 minutes. All too often, it is assumed that the lowest RSD necessarily came from the best process (or formulation, time point, etc.), when in fact there may be too much statistical noise to truly make such a determination. 

Having documentation and uniform protocols is important to obtain the most value from test beds and operational deployments. A continually evolving operational deployment such as the Pentagon could provide valuable real-world experience and data comparable to those obtained in test beds if all tests are well documented and standardized. For operational facilities or test facilities, establishment of uniform testing protocols to test effectiveness and validate protection systems would make their results useful to others. For example, information on degradation, maintenance, and operational and life-cycle costs (real... 

Published data on rat toxicity are highly variable because of the diverse test protocols applied and hence are of limited use for the derivation and validation of sound QSAR models. Validated reference data sets are not available for the laboratory rat (unlike fish) and if data are retrieved from the literature, therefore, care should be taken that they have been conducted according to uniform standard protocols and evaluated by consistent criteria. The important test parameters should be identical or comparable throughout the data set. If, however, they vary they should be critically evaluated for their effect on the test results, as such differences may greatly affect the relevance of the QSAR models derived. 

It is evident that a large number of parameters are involved in the fabrication and testing of bulk adhesive specimens and adhesive joints these must be controlled if meaningful experimental data are to be obtained. Joint tests evaluate not only the mechanical properties of the adhesive, but also the degree of adhesion and the effectiveness of surface treatments. The standard test procedures listed by ASTM, BSI, DIN and other official bodies are essentially for testing adhesives and surface treatments rather than joints (e.g. Table 4.3). Unfortunately, most of these tests consist of joints in which the adhesive stresses are far from uniform. The designer and the researcher therefore have to select appropriate tests, and to know what the results mean in terms of their own particular investigations and applications. 

In order to produce truly comparable data, use of uniform standards, uniform test specimens, standard molds, narrow specimen molding conditions specifically defined for each resin family, and uniform test conditions, that is, identical, reproducible conditions is vital. Simply providing detailed information about the specimen geometry, preparation, conditioning, and test procedure is not sufficient to allow true comparability. 

Some of the discrepancies between the Wiepking-Doyle data and the Soden-McLeish data may have been systematic. The Wiepking-Doyle (linear) results were based on a sample population that uniformly had a 12 percent moisture content. The Soden-McLeish (exponential) results,however, were derived in part from the Doyle et al. work based on samples with 9.5 percent moisture content, and in part on their own experiments for which no moisture contents were reported. It is difficult to tell whether these variations in percent moisture could have biased the results significantly. No moisture-related effects on the properties of balsa wood have been evaluated explicitly in any of the literature references surveyed here. To the extent that moisture contents were reported at all, they ranged generally from 8 to 12 percent. This range appears to be a normal condition for kiln-dried balsa wood exposed to ambient atmosphere in temperate climates. (Note The ASTM Standard Method of Testing Small Clear Specimens of... 

---