This is a critical point, and it would be much appreciated if the authors could address it directly and verify the source for this.

Thanks

This is a critical point, and it would be much appreciated if the authors could address it directly and verify the source for this.

Thanks

First off, thank you very much for your thoughtful reply. I’m glad that you were able to clarify a few points and you answered several questions that I had on the study.

My main problem with the sample being so Google heavy is that, in my personal experience, Google is a target for misguided DMCA notices. In fact, I warn heavily on my site against sending notices to Google for just that reason. In short, if I were going to use the DMCA maliciously to attack a competitor or a detractor, I would do it by filing a notice through Google, not their host and, since Google is so big, I wouldn’t waste time with any other search engines. 512(d) is just too ripe for abuse and Google is the main target for such abuse.

Also, Google’s main hosting service, Blogger, is more prone than most hosts to “hostile” DMCA notices. Blogging is one area where fair use copying, commentary, strong opinions and corporate enemies make it a ripe minefield for DMCA abuses. I would imagine that traditional hosts, like ThePlanet, will probably see fewer such notices.

I do agree that the main skew comes from the voluntarily submitted collection, however, there are elements of Google’s business that make it a more attractive target for malicious and erroneous notices.

I’m going to send you an email in a second regarding the coding. I have a few questions that I want to be off the record as I don’t want to inadvertently mess with your carefully crafted system. I know how hard researchers work to develop coding techniques (having done journalism research projects in the past).

I think it’ll be interesting though to see if Google’s results stack up to other hosts or if there is a skew. Honestly, I’m guessing based upon my own experience dealing with various hosts. One thing research has taught me is that anecdotal evidence can be wrong.

I hope that you have a happy Thanksgiving!

First, I just wanted to note that your presentation of the flaws misses just a bit and to clarify that. Because Google sends everything, the Google notices don’t skew as unreliable within the Google sample. It’s just that Google is only one, albeit important, company, and that it provides primarily search services. (Though it provides more hosting all the time and, in fact, 1/3 of the notices from Google we studied are hosting notices.) Comparison among companies is needed, here, so we weren’t able to say much about notices outside our sample.

The skew you noted actually affects the other –non-Google– notices. This is because the vast majority of these (by the way a much smaller number–Google provides 84% of the sample) are self-reported to Chilling Effects. (The Internet Archive submits notices as well as Google, but not enough to affect the sample.) We note that people who submit notices are more likely to at least think they’re in the right. So, we use these very, very carefully, and draw no general conclusions from them. We do speculate about 512(a), but that’s backed up with confidential information from other ISPs.

On the coding, I will note that we were very careful when we put a notice into the “flawed” category–we did it only in situations where there was a clear defense or other problem. For example, we didn’t code for fair use based only on using just a little bit, though that might be fair use. We stuck to parody, commentary, etc–the favored uses–and to clearcut issue with copyrightability. Additionally, we couldn’t always find out enough information about a notice to decide, so those were not counted as flawed. Sadly, I fear we may have undercounted flaws. So, it’s not that at least 70% are okay, it’s that at least 30% are questionable in our sample.

The main issue with studying 512 notices, of course, is that they are private–it’s been quite difficult to obtain information from OSPs. If we had a randomized sample (and actually 900 or so would probably be fine if randomized, statistically), we’d know more. Still, the fact that we have very good data from at least one company does mean that we were able to make a start.

I want to especially thank you for the thoughtful ideas on reform. We’ve kicked about all but the last one, I think.

First off, thank you very much for your thoughtful reply. I’m glad that you were able to clarify a few points and you answered several questions that I had on the study.

My main problem with the sample being so Google heavy is that, in my personal experience, Google is a target for misguided DMCA notices. In fact, I warn heavily on my site against sending notices to Google for just that reason. In short, if I were going to use the DMCA maliciously to attack a competitor or a detractor, I would do it by filing a notice through Google, not their host and, since Google is so big, I wouldn’t waste time with any other search engines. 512(d) is just too ripe for abuse and Google is the main target for such abuse.

Also, Google’s main hosting service, Blogger, is more prone than most hosts to “hostile” DMCA notices. Blogging is one area where fair use copying, commentary, strong opinions and corporate enemies make it a ripe minefield for DMCA abuses. I would imagine that traditional hosts, like ThePlanet, will probably see fewer such notices.

I do agree that the main skew comes from the voluntarily submitted collection, however, there are elements of Google’s business that make it a more attractive target for malicious and erroneous notices.

I’m going to send you an email in a second regarding the coding. I have a few questions that I want to be off the record as I don’t want to inadvertently mess with your carefully crafted system. I know how hard researchers work to develop coding techniques (having done journalism research projects in the past).

I think it’ll be interesting though to see if Google’s results stack up to other hosts or if there is a skew. Honestly, I’m guessing based upon my own experience dealing with various hosts. One thing research has taught me is that anecdotal evidence can be wrong.

I hope that you have a happy Thanksgiving!

First, I just wanted to note that your presentation of the flaws misses just a bit and to clarify that. Because Google sends everything, the Google notices don’t skew as unreliable within the Google sample. It’s just that Google is only one, albeit important, company, and that it provides primarily search services. (Though it provides more hosting all the time and, in fact, 1/3 of the notices from Google we studied are hosting notices.) Comparison among companies is needed, here, so we weren’t able to say much about notices outside our sample.

The skew you noted actually affects the other –non-Google– notices. This is because the vast majority of these (by the way a much smaller number–Google provides 84% of the sample) are self-reported to Chilling Effects. (The Internet Archive submits notices as well as Google, but not enough to affect the sample.) We note that people who submit notices are more likely to at least think they’re in the right. So, we use these very, very carefully, and draw no general conclusions from them. We do speculate about 512(a), but that’s backed up with confidential information from other ISPs.

On the coding, I will note that we were very careful when we put a notice into the “flawed” category–we did it only in situations where there was a clear defense or other problem. For example, we didn’t code for fair use based only on using just a little bit, though that might be fair use. We stuck to parody, commentary, etc–the favored uses–and to clearcut issue with copyrightability. Additionally, we couldn’t always find out enough information about a notice to decide, so those were not counted as flawed. Sadly, I fear we may have undercounted flaws. So, it’s not that at least 70% are okay, it’s that at least 30% are questionable in our sample.

The main issue with studying 512 notices, of course, is that they are private–it’s been quite difficult to obtain information from OSPs. If we had a randomized sample (and actually 900 or so would probably be fine if randomized, statistically), we’d know more. Still, the fact that we have very good data from at least one company does mean that we were able to make a start.

I want to especially thank you for the thoughtful ideas on reform. We’ve kicked about all but the last one, I think.