"... In PAGE 31: ... 4.4 Comparison with other studies A comparison of the size of the software considered in other studies and the number of faults classi ed appears in Table3 . Table 4 compares the results of our analysis (column 4) with that of B/P and O/W.... ..."

"... In PAGE 26: ... Table 10 provides the list of consumer attributes. Table 10: Household attributes used in estimation Variable Description Comment Tot Inc total household income Income1 (Tot Inc)*((Tot Inc) lt; 75th percentile) used in spline Income2 (Tot Inc)*((Tot Inc) gt; 75th percentile) used in spline Fam Size family size Adults number of adults 16 Age age of household head Age2 age (squared) of household head Kids number of kids 16 Rural dummy for rural residence Table11 (broken down into 11a and 11b) provides the estimates from our full model (the rst result column), and compares them to those from models that have been used to analyze similar problems in the past. Table 11a presents the estimates of the coe cients of the interactions of the observed household attributes with the vehicle characteristics, the o, while 11b provides the estimates of the interactions with the unobserved household attributes, the u.... In PAGE 26: ... Table 10: Household attributes used in estimation Variable Description Comment Tot Inc total household income Income1 (Tot Inc)*((Tot Inc) lt; 75th percentile) used in spline Income2 (Tot Inc)*((Tot Inc) gt; 75th percentile) used in spline Fam Size family size Adults number of adults 16 Age age of household head Age2 age (squared) of household head Kids number of kids 16 Rural dummy for rural residence Table 11 (broken down into 11a and 11b) provides the estimates from our full model (the rst result column), and compares them to those from models that have been used to analyze similar problems in the past. Table11 a presents the estimates of the coe cients of the interactions of the observed household attributes with the vehicle characteristics, the o, while 11b provides the estimates of the interactions with the unobserved household attributes, the u. There are three comparison models.... In PAGE 26: ... Note that the result is just a logit model for micro data with choice speci c intercepts, though our two stage maximum likelihood procedure does have to account for the fact that the sample is choice based (see above). The column labelled \Logit 1st quot; in Table11 a, provides the estimates obtained when we use only rst choice data to estimate this logit model, while the column labelled \Logit 1st amp; 2nd quot; provides the estimates when we use both rst and second choice data. The implicit estimates of u are all zero in these models, so they do not appear in Table 11b.... In PAGE 27: ...) We can distinguish a signi cant change in curvature for the top quartile of the income distribution, with the marginal disutility of a price increase changing at the 75th percentile. The remaining interactions in Table11 a are also generally of the expected sign, and quite precisely estimated, again in all three speci cations.22 Thus the interactions between Minivans and Kids (+), Age and number of passengers (+), Age and Safety (+), HP and Age (-), and SU and Age (-) were all signi cant in all speci cations.... In PAGE 28: ... Having one or two coe cients of the \wrong quot; sign among twenty coe cients would not disqualify the logits from being used in many practical settings and the increased computational burden of the full model is not obviously justi ed by the pattern of estimated interactions between x and z. Though the demographic interaction terms both seem to make sense and are sharply estimated, Table11 b indicates that they apparently do not explain the full pattern of substitution in the data, for the estimated u coe cients are typically important and very precisely estimated. Nineteen out of twenty two are signi cant at traditional signi cance levels, and over half of these have t- values over twenty.... In PAGE 29: ...Table11 a: Estimates of Interaction Terms, o Vehicle Household Full Logit Logit Characteristic Attribute Model 1st 1st amp; 2nd Price Constant ?0:805 0:092 0:139 (0:047) (0:0001) (0:0003) Price Income1 0:074 0:299 0:344 (0:008) (0:002) (0:001) Price Income2 0:608 0:466 0:603 (0:020) (0:091) (0:007) Price Fam Size ?0:212 ?0:144 ?0:143 (0:010) (0:001) (0:006) Miniv Kids 2:546 0:765 0:771 (0:169) (0:098) (0:323) Pass Adults 0:564 0:018 ?0:067 (0:107) (0:0004) (0:009) Pass Fam Size ?0:104 ?0:055 ?0:006 (0:032) (0:003) (0:0002) Pass Age 0:009 0:002 0:005 (0:002) (0:00001) (0:00001) HP Age ?0:012 ?0:010 ?0:012 (0:001) (0:0004) (0:0001) Acc Age ?0:004 0:001 ?0:002 (0:001) (0:00001) (0:0001) Acc Age2 0:0001 0:000 0:000 (0:00001) (0:00001) (0:00001) PUPayl Age ?0:127 0:512 0:000 (0:010) (0:005) (0:00001) PUPayl Rural 0:843 ?0:043 0:376 (0:121) (0:003) (0:008) Safe Age 0:017 0:403 0:016 (0:0004) (0:007) (0:0004) SU Age ?0:100 ?0:043 ?0:043 (0:005) (0:003) (0:004) SU Rural 0:192 0:403 ?0:016 (0:029) (0:007) (0:002) Allw Rural 0:206 0:142 0:734 (0:176) (0:005) (0:246) OutG Tot Inc ?2:372 0:228 0:305 (0:177) (0:096) (0:063) OutG Fam Size 0:428 ?0:532 0:346 (0:078) (0:057) (0:004) OutG Adults 0:041 ?0:851 ?1:953 (0:134) (0:112) (0:148)... ..."

"... In PAGE 7: ... The real data sets were obtained from several di erent application domains. Table2 shows some characteristics of these data sets. The rst ve data sets in the table, i.... ..."

"... In PAGE 7: ... The real data sets were obtained from several di erent application domains. Table2 shows some characteristics of these data sets. The rst ve data sets in the table, i.... ..."

"... In PAGE 2: ... We discuss the e ect of the PSPC gain drift on the Markarian 766 data in Ap- pendix A. 2 OBSERVATIONS Details of the 9 ROSAT datasets used in this analysis are given in Table1 . We will refer to them as P1, P2, etc.... ..."

"... In PAGE 12: ...emory Cards were described in more detail in section 3.2. Flash Disk Emulators erase a block (usually between 500{1024 bytes) just before writing to it, and so have a lower write bandwidth. Table2 shows the key characteristics of Flash Disks and Flash Cards, as well as the other storage devices we have discussed and is based on data from [12, 4, 7].... ..."