"... In PAGE 22: ...1, p = 0.5) = 0.0308. There are two ways STRONG can outperform NEW: sell more items at or above the starting price, and secure a higher price for items sold. Table5 shows the results on frequency of sale. A chi-square test concludes that the probability of sale was not independent between the two sellers (p lt;.... ..."

"... In PAGE 6: ... An attractive feature of DGG-Discover is that it is possible to obtain many di erent summaries for an attribute by simply chang- ing the structure of the associated CH, and these new summaries can be obtained without modifying the underlying data. For example, consider the database shown in Table1 . Using the CH from Figure 1 to guide the generalization, one of the many possible summaries which can be generated is shown in Table 2, where the O ce, Quantity, and Amount attributes have been selected for generalization, and the actual values for the O ce attribute in each tuple have been generalized to the level of West and East.... In PAGE 11: ....3.2. Detailed Walkthrough We now present a detailed walkthrough of the serial algorithm. Consider again the sales database shown in Table1 , and the associated DGGs for the Shape, Size, and Colour attributes shown in Figure 5. The All Gen procedure is initially called with parameters relation = the contents of the sales database from Table 1, i = 1, m = 3, D = the DGGs from Figure 5, and Dnodes = fhShapei; hSizei; hColourig.... In PAGE 11: ... Consider again the sales database shown in Table 1, and the associated DGGs for the Shape, Size, and Colour attributes shown in Figure 5. The All Gen procedure is initially called with parameters relation = the contents of the sales database from Table1 , i = 1, m = 3, D = the DGGs from Figure 5, and Dnodes = fhShapei; hSizei; hColourig. In this walkthrough, we assume that D11 = hShapei, D12 = hANYi, D21 = hSizei, D22 = hPackagei, D23 = hWeighti, D24 = hANYi, D31 = hColouri, and D32 = hANYi.... In PAGE 12: ... We set work relation to the result re- turned from a call to Generalize (line 9) with parameters relation, i = 3, and D32. The value of work relation, shown in Table 6, is the value of Table1 , having selected only the Shape, Size, and Colour attributes, with the Colour attribute generalized to the level of node D32. We call Interest (line 10) with parame- ter work relation.... In PAGE 12: ... We set work relation to the result returned from a call to Generalize (line 9) with parameters relation, i = 2, and D22. The value of work relation, shown in Table 7, is the value of Table1 , hav- ing selected only the Shape, Size, and Colour attributes, with the Size attribute generalized to the level of node D22. We call Interest (line 10) with parame- ter work relation.... In PAGE 13: ... We set work relation to the result returned from a call to Generalize (line 9) with parameters relation, i = 3, and D32. The value of work relation, shown in Table 8, is the value of Table1 , having selected only the Shape, Size, and Colour attributes, with the Colour attribute generalized to the level of node D32. We call Interest (line 10) with parame- ter work relation.... In PAGE 15: ....4.2. Detailed Walkthrough We now present a detailed walkthrough of the par- allel algorithm. Consider again the sales database shown in Table1 , and the as- sociated DGGs for the Shape, Size, and Colour attributes shown in Figure 5. The Par All Gen procedure is initially called with parameters relation = the contents of the sales database from Table 1, i = 1, m = 3, D = the DGGs from Figure 5, Dpaths = ;, and Dnodes = fhShapei; hSizei; hColourig.... In PAGE 15: ... Consider again the sales database shown in Table 1, and the as- sociated DGGs for the Shape, Size, and Colour attributes shown in Figure 5. The Par All Gen procedure is initially called with parameters relation = the contents of the sales database from Table1 , i = 1, m = 3, D = the DGGs from Figure 5, Dpaths = ;, and Dnodes = fhShapei; hSizei; hColourig. We assume that the Dij have the same values as described in Section 4.... In PAGE 22: ...The subset of transactions in the pay-per-view database were selected to facilitate human veri cation of the rankings for the summaries generated, and to validate the serial and parallel algorithms. Tuples were selected with the planned biases shown in Table1 0, where Bias ID is the unique identi er assigned to each planned bias, Attribute is the attribute upon which the planned bias is based, Domain is the pool of possible values for each attribute, No. of Rentals is the number of tuples containing the corresponding domain value, and Analysis describes the planned bias.... In PAGE 23: ... of Attributes No. of Summaries 1 14 2 67 3 126 4 72 Total 279 The 14 unique single-attribute summaries are shown in Table1 2, where the Rank column describes the relative degree of interest of the corresponding summary, the Attribute column describes the name of the remaining non-ANY attribute, the Generality column describes the level of generalization, the Interest column describes the calculated interest based upon the variance measure, and the Bias ID column describes the planned bias, if applicable, of the corresponding summary. Table 12.... In PAGE 23: ...00040 2 B5 All planned biases were identi ed by the discovery task and are shown in Tables 13 through 18. The domains of planned biases 3 and 7 are not generalized, so are not identi ed in Table1 2, but they do correspond to the ungeneralized summaries at nodes Cd and Ad, respectively. In these summaries, the rst column describes the level of generalization, the Count column describes the number of tuples which have been aggregated from the unconditioned data in the original input relation, and the... In PAGE 24: ...e., the summary with rank 12), shown in Table1 5, corresponds to planned bias 5, where the Date, Day, and Time attributes are generalized to ANY, and the Movie attribute is generalized to node A1. This summary, by the classes adult and general, is the least interesting because it is a near uniform distribution of the tuples, with 52% general viewing and 48% adult viewing.... In PAGE 24: ... Table 15. Rentals by classi cation (planned bias 5) A1 Count Count(%) general 26 52 adult 24 48 Total 50 100 The most interesting single-attribute summary, shown in Table1 6, corresponds to planned bias 6, where the Movie, Date, and Day attributes are generalized to ANY (i.e.... ..."