September 30, 2010

The Impact of Illegal Downloads On Total Sales

One frequent rant of mine is the fact that there's no credible study about the impact of "illegal downloads" (not my wording) on total sales, and that, yet, there are lots of policies being made based on plain wrong "studies".

Lately, many things are happening (including arguments behind the approval of the Gallo Report) around Tera Consultants' "Study on the impacts of digital piracy on the EU's creative industries" (March 2010). If we carefully read the report (yes, reading more than just the abstract), thou, we find lots of interesting things. For instance, the fact that the study didn't research the "impact of illegal downloads on total sales", but, instead, decided to have a non-scientific approach to find some kinds of "average values" from the eight "relevant studies" on the issue. Well, in fact, they used the summary from those eight relevant studies presented on Oberholzer-Gee & Strumpf (2009 - a working draft). And here, I have a couple of things to highlight:

1) Instead of using Oberholzer-Gee & Strumpf latest release paper, they decided to use a working draft;
2) Instead of using Oberholzer-Gee & Strumpf conclusions (that can be summed up as "file sharing can be blamed for part of recorded music revenue declines, [but] that hasn’t eliminated artists’ incentive to create and consumers have benefited from increased choice"), they decided to use only a decontextualized table from that report;
3) Many flaws were pointed to the Oberholzer-Gee & Strumpf research, but let me talk about one: while in their 2005 paper, where they found no negative impact of illegal downloads on total sales, they used their own data, in this new paper - guess what - they used the same as Tera Consultants: a non-scientific approach to find some kinds of "average values" from the eight "relevant studies" on the issue.

So, let's look to those eight "relevant studies":

  • Hui and Png (2003) - If you actually read the article you see that the authors acknowledge that:
    • their data is completely dependent of the piracy numbers provided by the IFPI (mind you, the most interested party)
    • they acknowledge that they ignored the need to research the demand-side externalities positive impact of piracy on sales (as proven to be needed in Conner and Rumelt (1991); Nascimento and Vanhonacker (1988); Shy and Thisse (1999); Takeyama (1994))
    • in the same way, they didn't estimate the positive impact of piracy on sales thanks to means of more effective price discrimination (as proven to be needed in Bakos et al. 1999; Besen and Kirby 1989; Jacob and Ben-Shahar 2002; Liebowitz 1985; Takeyama 1997; Varian 2000)
    • Last, and yet the most important (in my opinion), their study only analyses data until 1998 (that's right, 12 years ago!), and they highlight that this is all data "before Napster" (dare I say, now, "before p2p", "before torrents", "before pirate bay", "before web 2.0", "before social music services", ...), and that's a really important behavioural shift, that probably changed all the ecosystem of illegal downloads and their impact on total sales
    In other words, this study should not be considered;
  • Peitz and Waelbroeck (2003, 2004)
    • they use as data source the IFPI World Report of 2003 that provides industry data from 1998 to 2002. As pointed before, IFPI is the most interested party in influencing the results, so the validity can be argued;
    • they assume, because they don't know how to estimate, that the quality of music (which englobes que quality of the music release - compare, for instance, a signed limited edtion 12'' 2LP with the album in DRM-encumbered 128K digital file...) has no influence in buying habits (assuming, thus, that the number of people who use "piracy" to "try before buy" is zero, which is obviously wrong - simply ask your friends);
    • they assume that each "act of piracy" is equivalent to the loss of a CD sale (which is obviously wrong), and then...
    • they assume that you can do a direct correlation between the rise of usage of p2p networks and the fall in music sales, just because they saw a bigger fall in sales at the same time they saw a fall in music sales
    • They state that with broadband, non-piracy times can be spent doing a number of "replacement activities" other then enjoying music, and so people have less time - thus attention - to the music market. Yet, they fail to try to correlate attention-shift with loss of sales
    • In the "Discussion" section, they point out several flaws on their model (like the fact that they don't measure "intensity" in piracy activities, or that they aren't considering the non-renewal factor (the industry failed - at least until 2004, where the study focus - to keep their renewal cycle, when people were buying the same album in a different format, cassette replacing the vinyl record, CD replacing the cassette or vinyl, we just weren't seeing people buying the album in a DRM-encumbered digital format after having it on CD or a "previous" format) as a cause of loss of sales). Yet, they decide to ignore those observations on their conclusion...
    In other words, this study should not be considered;
  • Zentner (2006) - Now, this is actually an interesting study, it just isn't a good study to taking a conclusion on our particular issue. In fact, as the origin dataset did neither contain information on quantities of music purchased nor on intensities of music downloads, he assumed that it was not directly possible to calculate the impact on record sales, and thus, for that specific part, he decided to assume things (and that's stated quite obviously on the article): "if 15 percent of the population downloads music, if downloaders are twice as likely to buy music than nondownloaders, and if – conditional on buying – downloaders and nondownloaders buy the same quantity of units". Notice that he didn't get inspiration for this assumptions in any study at all.
  • Michel (2006)
    • The author states that "a possible problem with this approach is that computer owners who did not engage in file sharing may have decreased their music purchases for reasons unrelated to file sharing (yet inherent to owning a computer)", which is not considered in this study. As stated previously, it is fair to assume that there is an (unaccounted) percentage of non-sales directly related to this fact;
    • An analysis in this study's "Table 1" suggests that trends and influences greatly vary year by year, possibly in a two-years basis. This is actually why the researcher has data from 1995 until 2003, and yet can only assume a relation between file sharing and CD sales in 2002-2003. Also, in "Table 4" analysis, the author concludes that "the relationship between computer ownership and CD expenditures [, the only relationship being studied in this paper,] weakened from 1998 to 2003. In III.B), the author identifies other four reasons why this paper might not be valid, but the really important thing to quote is in the conclusions section, when the author state, above all, that "results should be used carefully when predicting the long-term viability of the music industry in an environment where record labels (or artists) compete directly with free file-sharing services". While it is arguable what the author means with "long-term", with the previous data we can assume that a) we can identify 2 year cycles, b) in 5 years the relationship weakened, so, if that trend is maintained (which is arguable, but wasn't in Mars' report) nowadays, 7 years later, the relationship might be completely dissipated, so...
    ...taking the previous items in consideration, it is fair conclude from the article data itself that this study can't be considered valid nowadays;
  • Montoro-Pons and Cuadrado-Garcia (2006) - This study doesn't appear anymore in the table in the final version of Oberholzer-Gee & Strumpf's paper;
  • Hong (2004, 2008) - This study was centred only and exclusively on the impact of Napster in 2000. As stated before, and shown by the other papers, we can't infer the current correlation between "piracy" and music sales from some event that happened a decade ago;
  • Leibowitz (2008) - the conclusion of this paper, and on Oberholzer-Gee & Strumpf's paper, is quite different from what's presented by Mars' study: in fact it concludes that if we consider all markets (where the universe was 89 different US markets), we cannot relate internet usage with changes in music sales;
  • Leung (2009) - This is a nice study made by using data from college students of seven different undergraduate classes (from the tone of the paper, I'm assuming this is on USA, but I'm not sure). We're talking about heavy music consumers with no money (they buy four or five digital tracks per month, and one CD every two months). The study implicitly also shows, while it's not its focus, that those students would buy more if they had more money (or if the cost of CDs, iTunes tracks and iPods was less than it is). So, I'll argue that this is a nice study but only representative of a small subset of the global "pirate" reality, specially different from the EU reality, where we have differences like the fact that more than 40% of our countries don't have access to the iTunes music store. Actually, I can go ever further, and tell you my belief (no, I don't have any study showing I'm right or wrong, that's the real issue on this blog post...) that, since you can see that this college students consume nowadays much more music than others (for instance, people with 15 more years than them, even if those others have a lot more money) that it was the setting in which they live - including the cultural influences made by the existence of "digital piracy" - that made them such music fans, and when they grow to have more money, they will be better consumers than they are nowadays. In other words, this is a nice study, but one that only proves that there are lots of dimensions to the problem which need to be researched until we can take any scientific conclusion regarding the impact of piracy in music sales.

So, what I have to say about Tera Consultants' study, and everything else that uses it as a base to take a conclusion or decision? Only that I'm really ashamed, as an European citizen, that we pay blind attention to any study, specially studies commissioned by entities like ICC, whose body of members include three of the four major labels.

September 19, 2010

Evolution of Operating Systems usage

I needed this statistics and I'm sure they're somewhere, but I don't know where. So, I went to Wikipedia which has the updated numbers but not something showing the progress. Fortunately, wiki's have revision history, so it was fairly easy to compile what I wanted. Here it is, for the sake of others (or me) not needing to go to the same hassle again.

Wikipedia page used as a source (by using several different revisions of this document)

Q4   2004 - Windows 96.34%; Apple  3.25%; Linux 0.29%
Q1   2005 - Windows 96.21%; Apple  3.40%; Linux 0.29%
Q2   2005 - Windows 96.10%; Apple  3.50%; Linux 0.31%
Q3   2005 - Windows 96.03%; Apple  3.55%; Linux 0.32%
Q4   2005 - Windows 95.50%; Apple  4.11%; Linux 0.30%
Q1   2006 - Windows 95.29%; Apple  4.30%; Linux 0.31%
Q2   2006 - Windows 95.08%; Apple  4.40%; Linux 0.41%
Q3   2006 - Windows 95.01%; Apple  4.45%; Linux 0.44%
Q4   2006 - Windows 94.10%; Apple  5.41%; Linux 0.38%
Q1   2007 - Windows 93.17%; Apple  6.22%; Linux 0.45%
Q2   2007 - Windows 92.78%; Apple  6.23%; Linux 0.74%
Sep. 2007 - Windows 89.10%; Apple  4.00%; Linux 0.81%
Oct. 2007 - Windows 91.76%; Apple  4.46%; Linux 0.81%
Nov. 2007 - Windows 91.45%; Apple  4.48%; Linux 0.81%
Dec. 2007 - Windows 92.88%; Apple  4.32%; Linux 0.88%
Jan. 2008 - Windows 91.79%; Apple  4.12%; Linux 0.76%
Feb. 2008 - Windows 92.13%; Apple  4.34%; Linux 0.77%
Apr. 2008 - Windows 93.94%; Apple  4.47%; Linux 0.81%
May  2008 - Windows 93.98%; Apple  4.39%; Linux 0.83%
Jun. 2008 - Windows 94.06%; Apple  4.33%; Linux 0.89%
Jul. 2008 - Windows 93.64%; Apple  4.31%; Linux 0.90%
Aug. 2008 - Windows 91.95%; Apple  4.49%; Linux 0.96%
Sep. 2008 - Windows 93.61%; Apple  4.86%; Linux 1.04%
Oct. 2008 - Windows 90.40%; Apple  5.50%; Linux 1.16%
Nov. 2008 - Windows 90.50%; Apple  5.41%; Linux 1.16%
Dec. 2008 - Windows 90.50%; Apple  5.24%; Linux 1.16%
Jan. 2009 - Windows 91.10%; Apple  4.81%; Linux 1.02%
Feb. 2009 - Windows 91.03%; Apple  4.77%; Linux 1.04%
Mar. 2009 - Windows 90.99%; Apple  4.81%; Linux 1.05%
Apr. 2009 - Windows 91.22%; Apple  4.95%; Linux 1.13%
May  2009 - Windows 91.30%; Apple  5.01%; Linux 1.12%
Jun. 2009 - Windows 91.24%; Apple  5.06%; Linux 1.12%
Jul. 2009 - Windows 93.18%; Apple  4.59%; Linux 1.05%
Aug. 2009 - Windows 93.06%; Apple  4.59%; Linux 0.94%
Sep. 2009 - Windows 92.77%; Apple  4.59%; Linux 0.95%
Oct. 2009 - Windows 92.52%; Apple  5.87%; Linux 0.96%
Nov. 2009 - Windows 90.52%; Apple  6.48%; Linux 1.00%
Dec. 2009 - Windows 90.22%; Apple  6.57%; Linux 1.04%
Jan. 2010 - Windows 92.02%; Apple  5.79%; Linux 1.08%
Feb. 2010 - Windows 91.60%; Apple  6.58%; Linux 1.12%
Mar. 2010 - Windows 91.34%; Apple  6.63%; Linux 1.14%
Apr. 2010 - Windows 90.95%; Apple  6.61%; Linux 1.15%
May  2010 - Windows 90.88%; Apple  6.67%; Linux 1.22%
Jun. 2010 - Windows 89.53%; Apple  6.93%; Linux 1.31%
Aug. 2010 - Windows 88.92%; Apple  7.72%; Linux 1.33%
Sep. 2010 - Windows 88.66%; Apple  7.82%; Linux 1.34%
Oct. 2010 - Windows 88.40%; Apple  8.22%; Linux 1.50%
Nov. 2010 - Windows 88.01%; Apple  8.48%; Linux 1.53%
Dec. 2010 - Windows 87.82%; Apple  8.87%; Linux 1.58%
Jan. 2011 - Windows 87.54%; Apple  8.97%; Linux 1.64%
Feb. 2011 - Windows 87.36%; Apple  9.27%; Linux 1.66%
Mar. 2011 - Windows 87.15%; Apple  9.27%; Linux 1.82%
Apr. 2011 - Windows 87.00%; Apple 10.44%; Linux 1.88%
May  2011 - Windows 86.58%; Apple 10.34%; Linux 1.95%
Jun. 2011 - Windows 86.00%; Apple 10.05%; Linux 2.15%
Jul. 2011 - Windows 83.30%; Apple 10.26%; Linux 2.07%
Aug. 2011 - Windows 82.58%; Apple 10.91%; Linux 2.25%
Sep. 2011 - Windows 82.41%; Apple 11.10%; Linux 2.41%
Oct. 2011 - Windows 86.15%; Apple 10.37%; Linux 2.53%
Nov. 2011 - Windows 85.8%;  Apple 10.7%;  Linux 3.0% 
Dec. 2011 - Windows 80.83%; Apple 11.20%; Linux 3.00%
Jan. 2012 - Windows 78.66%; Apple 13.28%; Linux 3.90%
Feb. 2012 - Windows 82.11%; Apple 10.61%; Linux 2.82%
Mar. 2012 - Windows 73.38%; Apple 15.87%; Linux 5.05%
Apr. 2012 - Windows 82.21%; Apple  9.89%; Linux 3.05%
May  2012 - Windows 83.07%; Apple 12.28%; Linux 3.60%
Jun. 2012 - Windows 78.89%; Apple 15.47%; Linux 3.95%
Jul. 2012 - Windows 79.88%; Apple 13.99%; Linux 4.05%
Aug. 2012 - Windows 79.21%; Apple 14.55%; Linux 3.90%
Set. 2012 - Windows 75.88%; Apple 14.87%; Linux 3.81%
Oct. 2012 - Windows 82.76%; Apple 12.58%; Linux 4.13%
Nov. 2012 - Windows 81.74%; Apple 12.55%; Linux 3.93%
Dec. 2012 - Windows 81.54%; Apple 12.56%; Linux 3.92%
Jan. 2013 - Windows 80.84%; Apple 13.41%; Linux 3.97%
Feb. 2013 - Windows 75.89%; Apple 15.99%; Linux 5.47%
Mar. 2013 - Windows 78.85%; Apple 14.81%; Linux 4.59%
Apr. 2013 - Windows 79.01%; Apple 14.25%; Linux 4.66%
May  2013 - Windows 80.34%; Apple 12.89%; Linux 5.29%
June 2013 - Windows 73.49%; Apple 17.27%; Linux 6.68%
July 2013 - Windows 75.67%; Apple 16.05%; Linux 5.57%
Aug. 2013 - Windows 74.97%; Apple 16.12%; Linux 6.65%
Sep. 2013 - Windows 72.32%; Apple 15.69%; Linux 6.80%
Oct. 2013 - Windows 72.02%; Apple 15.85%; Linux 7.06%
Nov. 2013 - Windows 77.90%; Apple 12.96%; Linux 5.35%
Feb. 2015 - Windows 60.27%; Apple 17.70%; Linux 15.05%
Jun. 2015 - Windows 41.48%; Apple 35.75%; Linux 16.19%
Dec. 2015 - Windows 48.70%; Apple 16.33%; Linux 28.40%
May  2016 - Windows 45.00%; Apple 16.43%; Linux 33.15%
Dec. 2016 - Windows 38.31%; Apple 17.63%; Linux 39.38% 


(I intend to update this article from time to time - hopefully at least once per month.)