The “millions” numbers I had were not public; I had them from screenshots from various writers...Ff.net tallies reads but doesn’t—unlike Wattpad or AO3--make them public.
I don’t understand what this means. How can she have screenshots of E.L. James’s FF.net official view statistics, if FF.net does not make the numbers public? Did James hand them out freely to ‘various writers’?
Or, do you/her mean ‘I have FF.net statistics from various random other Twilight fanfic authors’? That’s interesting, I suppose, but I’m not sure what to make of that.
Yes, maybe the way FF.net tracks page views implies that 1 million page views translates to perhaps 4k complete readers of that fic, and so if we assume ‘Master of the Universe’ had 1m page views in total and FF.net was the only source for it, then it had 4k complete readers before it was published as the book 50 Shades of Grey. OK, but why would we assume that? The whole point is that it was an amazingly popular breakout best-selling phenomenon practically without precedent, so why would we think its readership was just like the other relatively popular Twilight fanfics (none of which became NYT bestselling books)?
More likely, given re-readings, users who always go in through chapter 1, users who quit halfway through, browser refreshes, etc., perhaps 4,000 readers finished it.
Interesting approach, but I think this is a severe underestimate because you don’t take into account censoring, the two alternate sources of readers she had before she published the book, and inferring readership from reviews gives a radically different conclusion.
I have a copy of it [‘Master of the Universe’], but the original has been removed—so it is hard to reconstruct how the community responded to the book. It was released serially, and I believe it had over 37K reviews when it was taken down for its ffnet terms of service violations (they don’t allow explicit content, although this rule is widely violated and unenforced). It then moved on to the author’s own website. But reviews represent only a fraction in readership—it varies, but maybe only 1 in 10 readers review—but that was spread out over 100 chapters, and people tend to review regularly. Authors themselves can see all these stats, but readers can’t.
Master of the Universe also had a post on Twilighted, another big fanfiction site. This site is by application only, and it features lots of interaction with authors. So, the reality is, we’ll likely never know for sure how many readers it had—but certainly tens if not hundreds of thousands. It was a huge story.
So there’s some problems here: the FF.net data is severely censored by the takedown; the FF.net review data seems to be inconsistent with your estimated total readership (every single reader would have to leave 2 reviews to make 16k readers match 37k reviews!); and it was 1 of 3 sources of traffic.
This story has over 40,000 reviews!...Happy reading, it’s on chapter 70 and worth every minute of it.
So, if the story had >100 chapters and it had >40k reviews by chapter 70, then on chapter 100 its review count would look more like >57k than >37k. At 120 chapters, it’d be >69k. (EDIT: apparently it was actually 110 chapters? Close enough, I’m not going to revise all the numbers.)
(It’s hard to be more precise as it seems all the relevant sites have been censored from the Internet Archive, and the people with copies apparently didn’t bother to dump the raw FF.net pages such as all the reviews.)
How many readers does >69k reviews translate into?
Now, as it happens, I earlier did some FF.net review analysis for MoR; the median user who ever left a review on MoR did so only once (which is consistent with your estimated attrition of ~50% completers) and the mean of reviews per user is 2.7. So that lets me estimate how many reviewers there we, given the total # of reviews, which gives me an estimated number of reviewers being >25.3k (divide by 3).
The rule of thumb seems to be that <10% of readers will ever leave a review (you could probably get a better number with Fimfiction.net analytics but I don’t know how well that would compare to Twilight fics on FF.net, especially if, as you say, the more awful the story the more people complete it), so to get total readers I multiply by 10 to get >253k.
Then for complete readers, following your estimated 50% attrition, that’s >127k readers—on FF.net.
If the reads on her site and then ‘Twilighted’ were comparable (and remember, the FF.net figures are heavily right-censored due to the takedown, and it didn’t have as much chance to benefit from developing buzz & word of mouth), then multiply by 3 to get >381k complete readers.
(A good try overall, but you probably should’ve read your sources more carefully and done a little more homework before trying to use it as a case study of such claims.)
EDIT: see also the comments by a Twilight fanficer excerpted in http://lesswrong.com/lw/kl3/fifty_shades_of_selffulfilling_prophecy/b61d ; particularly note the higher estimate of how many millions of page views top-tier Twilight fics got, the calculated marketing by the author and reuse of the original fans to boost the book, and what the auction revenues imply about number of fans.
I don’t understand what this means. How can she have screenshots of E.L. James’s FF.net official view statistics, if FF.net does not make the numbers public? Did James hand them out freely to ‘various writers’?
She had screenshots that various writers had sent her. I infer that James was one of those writers, but she didn’t say that outright.
I infer that James was one of those writers, but she didn’t say that outright.
That’s the basis for your 2 million number, the number which largely determines the result, some guesswork about something your source never says and probably would have said if it was actually the case? Then the entire analysis is bunk—garbage in, garbage out. And you should especially not infer that low number of total pageviews when everything else disagrees with it.
You can send her an email and ask her yourself. She came up with the 2 million number for this particular story; she said she came up with numbers for stories based on screenshots. It isn’t too hard to connect the dots.
Give me a break—why should I have to do that when a plain reading suggests otherwise and you’re the one trying to make these sweeping generalizations based on your own guesswork?
Fanfiction writers are writers. That’s who the “various writers” are. EL James is a writer.
If the average reader leaves 2.7 comments, and the story had 37,000 comments at the time when people started writing articles about it, I certainly accept 10% of readers leaving comments—the % is lower on most stories—which would indicate 137K readers by that time.
But all that is guess work, whereas it is a fact that that a 70-chapter story with 2 million hits can only have been read in its entirety by at most 28,571 people. That’s the theoretical upper limit. I suppose it’s barely possible that as many as 15,000 people read it on ff.net, but no more than that. And most readers, according to Jamison, read it on ff.net.
Being an active member of the community does not grant knowledge of statistical regularities like you need it to for the argument to work. There is no way she can know ‘most’ readers read it there, because most readers will never say anything and there will be differences in who does say things—the readers that an author hears from are not random readers, to say the least.
The number of reviews that you’re trying to infer more readers from is also from FF.net. The 2 million hits are also from ff.net. There are no numbers from anywhere else. I’ve already demonstrated that it’s theoretically impossible for there to have been even as many as 29,000 readers at that time on FF.net, and you’re apparently still claiming there were 127k on ff.net. It’s your analysis, not mine, that’s been debunked here.
So there’s some problems here: the FF.net data is severely censored by the takedown; the FF.net review data seems to be inconsistent with your estimated total readership (every single reader would have to leave 2 reviews to make 16k readers match 37k reviews!); and it was 1 of 3 sources of traffic.
A “review” on fanfiction.net is a comment. Many readers who leave comments, leave one on many chapters or even every chapter. A reader saying “MOAR PLZ” on chapter 1 and “OMG LOL” on chapter 2 counts as 2 reviews.
Oh, I see you know that. I’m interested in your data on average # of reviews per user. 2.7 seems low. I checked this number by downloading all the comments on HPMOR today for chapters 1-4 by hand, extracting the usernames into one username per line, and comparing line count of that file to line count of sort -u . The result is that, in chapters 1-4, the average user left 1.48 comments. It therefore seems unlikely to me that it could be 2.7 over the entire story.
But these distributions are non-intuitive. Extending this to chapters 1-8 I get no increase in average comments per user; it changes to an average of 1.45.
A command to do this with downloaded chapter comments pages from fanfiction.net is
for i in `ls hpmor*`; do
echo $i;
perl -pi -e "s#<a href='/u/\d+/.+?'>(.+?)</a>.+#\nQZW USER=\$1#" $i;
grep "^QZW USER" $i >> u;
done;
sort -u u > u.uniq; wc u*
A “review” on fanfiction.net is a comment. Many readers who leave comments, leave one on many chapters or even every chapter. A reader saying “MOAR PLZ” on chapter 1 and “OMG LOL” on chapter 2 counts as 2 reviews.
Um… I am well aware of that, and already dealt with that in my analysis. So, any response to any of my other points?
EDIT: when you edit a comment which has been replied to to substantially address that reply, please don’t do that.
Oh, I see you know that. I’m interested in your data on average # of reviews per user. 2.7 seems low. I checked this number by downloading all the comments on HPMOR today for chapters 1-4 by hand, extracting the usernames into one username per line, and comparing line count of that file to line count of sort -u . The result is that, in chapters 1-4, the average user left 1.48 comments. It therefore seems unlikely to me that it could be 2.7 over the entire story.
It may seem low; nevertheless, when I downloaded all of MoR up to ch82 or whatever it was up to when I did that analysis, that was the average count. You can reuse my provided scripts if you doubt it. I don’t find it too hard to believe: there’s high reviewer mortality, and I noted that people tended to leave reviews on either the first chapter or last chapter and avoid middle chapters, so ch1-4 would not be a representative random sample.
EDITEDIT:
But these distributions are non-intuitive. Extending this to chapters 1-8 I get no increase in average comments per user; it changes to an average of 1.45.
Yep, like I said. Reviews are very non-uniformly distributed; besides the start/end effect for completed fics, there’s also the accumulation of reviews on the latest chapter, where chapters posted right before long hiatuses accumulate more reviews than chapters which are part of regular update periods.
I don’t understand what this means. How can she have screenshots of E.L. James’s FF.net official view statistics, if FF.net does not make the numbers public? Did James hand them out freely to ‘various writers’?
Or, do you/her mean ‘I have FF.net statistics from various random other Twilight fanfic authors’? That’s interesting, I suppose, but I’m not sure what to make of that.
Yes, maybe the way FF.net tracks page views implies that 1 million page views translates to perhaps 4k complete readers of that fic, and so if we assume ‘Master of the Universe’ had 1m page views in total and FF.net was the only source for it, then it had 4k complete readers before it was published as the book 50 Shades of Grey. OK, but why would we assume that? The whole point is that it was an amazingly popular breakout best-selling phenomenon practically without precedent, so why would we think its readership was just like the other relatively popular Twilight fanfics (none of which became NYT bestselling books)?
Interesting approach, but I think this is a severe underestimate because you don’t take into account censoring, the two alternate sources of readers she had before she published the book, and inferring readership from reviews gives a radically different conclusion.
To quote from one of your links:
So there’s some problems here: the FF.net data is severely censored by the takedown; the FF.net review data seems to be inconsistent with your estimated total readership (every single reader would have to leave 2 reviews to make 16k readers match 37k reviews!); and it was 1 of 3 sources of traffic.
Actually, the 37k review figure itself seems to be a severe underestimate; http://twilightcupcake.wordpress.com/2010/02/19/fanfic-friday-master-of-the-universe/ is a post from when ‘Master of the Universe’ was still available on FF.net, and the author comments:
So, if the story had >100 chapters and it had >40k reviews by chapter 70, then on chapter 100 its review count would look more like >57k than >37k. At 120 chapters, it’d be >69k. (EDIT: apparently it was actually 110 chapters? Close enough, I’m not going to revise all the numbers.)
(It’s hard to be more precise as it seems all the relevant sites have been censored from the Internet Archive, and the people with copies apparently didn’t bother to dump the raw FF.net pages such as all the reviews.)
How many readers does >69k reviews translate into?
Now, as it happens, I earlier did some FF.net review analysis for MoR; the median user who ever left a review on MoR did so only once (which is consistent with your estimated attrition of ~50% completers) and the mean of reviews per user is 2.7. So that lets me estimate how many reviewers there we, given the total # of reviews, which gives me an estimated number of reviewers being >25.3k (divide by 3).
The rule of thumb seems to be that <10% of readers will ever leave a review (you could probably get a better number with Fimfiction.net analytics but I don’t know how well that would compare to Twilight fics on FF.net, especially if, as you say, the more awful the story the more people complete it), so to get total readers I multiply by 10 to get >253k.
Then for complete readers, following your estimated 50% attrition, that’s >127k readers—on FF.net.
If the reads on her site and then ‘Twilighted’ were comparable (and remember, the FF.net figures are heavily right-censored due to the takedown, and it didn’t have as much chance to benefit from developing buzz & word of mouth), then multiply by 3 to get >381k complete readers.
And >381k completers is very different from your estimated >4k; it does not tell the story you tell. Given all this, it seems like the title is unwarranted and this does not demonstrate the irrationality and randomness of media markets the same way as experiments like the 2006 study “Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market” and Salganik & Watts 2009 do.
(A good try overall, but you probably should’ve read your sources more carefully and done a little more homework before trying to use it as a case study of such claims.)
EDIT: see also the comments by a Twilight fanficer excerpted in http://lesswrong.com/lw/kl3/fifty_shades_of_selffulfilling_prophecy/b61d ; particularly note the higher estimate of how many millions of page views top-tier Twilight fics got, the calculated marketing by the author and reuse of the original fans to boost the book, and what the auction revenues imply about number of fans.
She had screenshots that various writers had sent her. I infer that James was one of those writers, but she didn’t say that outright.
That’s the basis for your 2 million number, the number which largely determines the result, some guesswork about something your source never says and probably would have said if it was actually the case? Then the entire analysis is bunk—garbage in, garbage out. And you should especially not infer that low number of total pageviews when everything else disagrees with it.
You can send her an email and ask her yourself. She came up with the 2 million number for this particular story; she said she came up with numbers for stories based on screenshots. It isn’t too hard to connect the dots.
Give me a break—why should I have to do that when a plain reading suggests otherwise and you’re the one trying to make these sweeping generalizations based on your own guesswork?
Fanfiction writers are writers. That’s who the “various writers” are. EL James is a writer.
If the average reader leaves 2.7 comments, and the story had 37,000 comments at the time when people started writing articles about it, I certainly accept 10% of readers leaving comments—the % is lower on most stories—which would indicate 137K readers by that time.
But all that is guess work, whereas it is a fact that that a 70-chapter story with 2 million hits can only have been read in its entirety by at most 28,571 people. That’s the theoretical upper limit. I suppose it’s barely possible that as many as 15,000 people read it on ff.net, but no more than that. And most readers, according to Jamison, read it on ff.net.
And how would she know?
She was there. She was an active member of the community when the story came out.
Being an active member of the community does not grant knowledge of statistical regularities like you need it to for the argument to work. There is no way she can know ‘most’ readers read it there, because most readers will never say anything and there will be differences in who does say things—the readers that an author hears from are not random readers, to say the least.
The number of reviews that you’re trying to infer more readers from is also from FF.net. The 2 million hits are also from ff.net. There are no numbers from anywhere else. I’ve already demonstrated that it’s theoretically impossible for there to have been even as many as 29,000 readers at that time on FF.net, and you’re apparently still claiming there were 127k on ff.net. It’s your analysis, not mine, that’s been debunked here.
A “review” on fanfiction.net is a comment. Many readers who leave comments, leave one on many chapters or even every chapter. A reader saying “MOAR PLZ” on chapter 1 and “OMG LOL” on chapter 2 counts as 2 reviews.
Oh, I see you know that. I’m interested in your data on average # of reviews per user. 2.7 seems low. I checked this number by downloading all the comments on HPMOR today for chapters 1-4 by hand, extracting the usernames into one username per line, and comparing line count of that file to line count of sort -u . The result is that, in chapters 1-4, the average user left 1.48 comments. It therefore seems unlikely to me that it could be 2.7 over the entire story.
But these distributions are non-intuitive. Extending this to chapters 1-8 I get no increase in average comments per user; it changes to an average of 1.45.
A command to do this with downloaded chapter comments pages from fanfiction.net is
Um… I am well aware of that, and already dealt with that in my analysis. So, any response to any of my other points?
EDIT: when you edit a comment which has been replied to to substantially address that reply, please don’t do that.
It may seem low; nevertheless, when I downloaded all of MoR up to ch82 or whatever it was up to when I did that analysis, that was the average count. You can reuse my provided scripts if you doubt it. I don’t find it too hard to believe: there’s high reviewer mortality, and I noted that people tended to leave reviews on either the first chapter or last chapter and avoid middle chapters, so ch1-4 would not be a representative random sample.
EDITEDIT:
Yep, like I said. Reviews are very non-uniformly distributed; besides the start/end effect for completed fics, there’s also the accumulation of reviews on the latest chapter, where chapters posted right before long hiatuses accumulate more reviews than chapters which are part of regular update periods.