Femslash week in review: 2019-01-15 - 2019-01-21

High-level stats for week of 2019-01-15 - 2019-01-21

• Total works categorized F/F on AO3: 4095 (+115 from last week)


• Works I classified F/F: 2504 (+4 from last week) (1167 new, 1337 continued)


• 1.04% of all 239900 AO3 works I've classified F/F were updated this week

A few callouts this week:

• This is a "ripping the bandaid off" week for the stats, so I'm going to be paying attention to any positive or negative reactions to this I get from anyone, for future reference the next time I have to do something similar. Short version: I fixed what I consider to be a bug in how fandom ratings are tabulated, which of course means that all past fandom ratings are wrong, and I have to retcon everything. So I did: the reason there are two graphs this week is so you can see the effects of the change for yourself: the first graph shows the past eight weeks under the new methodology I'll be using going forward; the second graph shows the past eight weeks the old way. I feel like making a change like this begs questions such as "Why should we pay any attention to your graphs if you are just making things up?", so I want to take a little time to answer the most frequent questions I have asked myself about this topic. If you're not interested, there are a couple of recs below so you can skip down.
  • What was the bug? For this to make sense, we need a refresher on ratings. It would be very reasonable not to have ratings at all, and just to count the number of works in each fandom and have the top fandom be the one with the most works. But I built the stats for me, and what I like (that I thought I could measure easily, and that I thought might possibly match what anyone else likes) is ensemble fandoms with a lot of different F/F ships and not a lot of ships containing male characters. So I make two adjustments to the raw count of number of works: one is that I weight the rating to discounting fandoms where a high percentage of works that are tagged F/F seem to have only cis male characters. (That's not relevant to today's bug, just full disclosure.) The other adjustment is that I group works by ship, and if there are N works with a given ship, instead of adding N to the rating, I add log(N + 1, 2). This was a simple attempt to reward ensembleness by penalizing fandoms that have exactly one dominant ship. If one fandom has two works this week with different ships, and another fandom has three works all with the same ship, their ratings will be identical, which matches my expectation of my likelihood of finding something I want to read. Your next question is "What do you do if a work has more than one ship?", or at least that should have been my next question when I wrote this algorithm. But it wasn't. In December, I noticed a surprise top 20 fandom whose rating was higher than its number of works, and, indeed, it's because that fandom had a number of works with 3 or 4 ships, and I was multiple-counting those works. This behavior was reporting that a fandom with one work containing two ships, should interest me just as much as a fandom with two works containing one ship each. That's not true; the latter fandom seems clearly more interesting. And the idea of a single work contributing much more than 1 to a rating also seems wrong.
  • What does the algorithm do now? The new behavior is, when I'm adding up works in each ship for the log(N + 1, 2), each of the N works contributes 1/S to that sum, where S is the number of ships in that work. This doesn't get quite exactly the numbers you'd expect, but it's pretty close and is easy to measure. If a fandom contains one work with S ships, its rating is going to be S * log(1 / S + 1, 2), which converges to, uhh, something. (It's approximately 1.44 for implausibly large numbers of ships; I should know what the limit is mathematically, but that's really not relevant to the stats, so I'm not going to figure it out. For our purposes, it's 1.28 for a single work with 4 ships, which is not too bad.)
  • This bug seems obvious; do you not have unit tests at all? That is an extremely personal question which I refuse to answer on the grounds that we don't know each other that well.
  • I've been reading the stats for awhile; is everything I know a lie? That's a bit philosophical, but I'm going to argue not really. First of all, for top 20 purposes, this changes things, but not much, as you can see from the comparison charts. Five weeks ago, I reported that Steven Universe had displaced RWBY as the number two fandom, but SU actually stopped at number three. Probably nobody noticed at the time, and it doesn't really matter anyway; RWBY and Steven Universe are both major fandoms. I fixed the bug because I wanted to smooth out the behavior for smaller fandoms: things lower in the top 20, things not in the top 20 at all. If I'm trying to do an over-all-time historical report on a particular fandom (spoiler: I'm hoping to start doing some of that), I want those numbers to make sense and not have inexplicable jumps caused by a single multi-ship work. Looking at the top 20, and especially at the top 5, fandoms that are in those slots have always been big and they're still big. I don't think anyone needs to panic, unless the idea that this is a filtered rating and not a raw work count makes you nervous. That's always been true, though, and it's very intentional: if you just go to AO3 and filter by category, you will learn that my nemesis Steve/Bucky is the third-biggest F/F ship in MCU. Correcting for how marginal femslash is relative to fanfic in general by providing information about non-male (also non-OC and non-RPF) ships, is one of my most explicit goals in the stats, and I have to put my thumb on the scales to do that. Hopefully wherever my ratings differ from raw work counts, it's an improvement for the utility of the data to fans like me, and where it isn't, I'll try to fix it over time.


• Debbie/Lou Groundhog Day AU


• Molly Weasley/Poppy Pomfrey canon-compatible pre-canon

( Full top-20 table and description of methodology after the jump )