Internet Service Providers Report

Appendix 1: Cross-Sectional Analysis of Current Copyright Notices

The data for the 51 ISPs reporting subscriber type and copyright infringement notices in 2005 were analyzed using ordinary least squares (OLS), and the results are presented in Table 5. Given the skewed nature of the data on the dependent and independent variables, log transformations were used. A Box-Cox test failed to reject the log-log specification, whereas the linear and log-linear specifications were rejected.

A simple OLS regression of (log) subscribers of each type on (log) infringement notices is Model 1. The number of notices is related to the types of subscribers in the whole sample of firms, with the impact of high-speed subscribers being greater than dial-up subscribers.

The difference in the number of notices received per high-speed subscriber (noted above) prompted exploration of the impact of the size of the ISP on the number of notices. Model 2 includes an indicator variable for large ISPs, which is positive and significant, confirming that large ISPs receive more notices than small ISPs even after controlling for the number and type of subscribers.

Model 3 replaces the ISP size indicator variable with three interaction terms between the size indicator and the number of subscribers of each type. Two of these three indicators are positive and statically significant, demonstrating that large ISPs receive more infringement notices per Cable and DSL subscriber than do small ISPs. There is no difference in the impact of dial-up subscribers across ISPs of different sizes. A model including both the indicator as well as the interactions could not be run due to multicollinearity.²²

Overall, then, analysis of the current relationship between subscriber type and copyright infringement notices received indicate that:

• High-speed subscribers (Cable and DSL) generate more copyright infringement notices than do dial-up subscribers as of January 2005
• Large ISPs receive more notices per high-speed subscriber than do small ISPs as of January 2005.

---

22 The Variance Inflation Factor of such a model was greater than 120, whereas values over 10 are usually considered indicative of multicollinearity.