Survey on EPO — DG1 search quality

Measuring patent quality: A claim and search report approach

“…The variance associated with the subjectivity of the inventive step has been observed in other studies as well [10]. The variance is much larger for "A" documents because according to the IPI patent search personnel these can be used to: signal potential infringement, i.e.…”

“…That is to say, if we use OLS, we would obtain the correct marginal effect, but be more likely to find a null result. In order to correct for the known source of the heteroskedasticity in the linear model, we present both feasible generalized least squares (FGLS), 9 the fractional logistic model pioneered by Papke & Woolridge, and the most appropriate b-distribution model 10 [12]. Table 6 presents a comparison of these estimators.…”

“…11 The models in Table 6 indicate that granted EPO patents have about 7e12% lower rejected claim score, with the FGLS on the low 9 With u ¼ 1/p(1Àp). 10 Since a b-distribution is only defined within the unit interval, those patents with either no claim rejections or all claim rejections were assigned a value of 0.001 and 0.999 respectively. The same transformed dependent variable was also used for benchmarking the other models.…”