So far, this amazon tracking project has been nominally about finding a way to benchmark the success of various US releases. In practice, it’s been mostly about being wrong.
But I am learning, bit by bit. Each mistake in prediction teaches me about why my current understanding of the rankings falls short, and putting them all together begins to get you something that looks kinda like a complete picture. I’ll only have more datapoints the longer this keeps up.
What datapoints do I have as of now? Aside from the several releases I tracked that actually charted, I’ve learned a couple things from those that haven’t. The following aren’t rock-solid patterns, but they are suggested by how things have gone. Here’s a quick summary of the most important points:
1. Space Dandy didn’t chart against a 6461 BD threshold despite 2 alternate editions, both of which ranked significantly above what I previously suspected the “no sales” line to be.
-This suggests that the non-main editions, which averaged around the 20,000 and 5000 ranks, may not have sold very many copies. To follow up on this suspicion, I tracked 2 post-release disks (Gingitsune on DVD and Devils and Realist on BD) with under-20 stock totals displayed on an irregular basis, getting a total of 43 datapoints for each release over 6 days. Multiple times in that period, the releases made it to ~20,000 in the rankings while selling two or fewer copies per day, more or less confirming my suspicions.
2. Steins Gate Classic didn’t make the charts against a 4761 BD threshold despite outdoing a confirmed seller at 8109 copies (DBZ s3) basically all of its corresponding 8 preorder days and half of its 6 post-release days.
-Add in data on other confirmed sellers, and it’s impossible to take all 14 datapoints for each and produce a result favoring DBZ s3 even a little. It is possible the time of year may be a factor in determining how many copies sold a rank actually means, Attack on Titan’s 2 parts came out in very different parts of the year and notched very similar ranks and sales numbers, making that explanation less likely to account for the observed discrepancy.
However, the Steins Gate release fell off in rank much faster after its release date than DBZ s3 did. This led me to a thought; we know that US anime releases can have fairly large long tails, much more significant than the factor-of-2 amount their Japanese counterparts seem to get from the same sorts of tails. If the US market relies more on tail sales, then it’s not unreasonable to wonder if preorders are less important, since lower costs mean less incentive to go for the early discount. By removing the 8 preorder days from the dataset, I was able to fit a formula that did a better job putting the sales of both closer to the real results (it’s still not a perfect fit) and nailed Attack on Titan and One Piece Film Z’s numbers to boot. Except…
3. DBZ: Battle of Gods shattered everything else’s numbers.
Battle of Gods sold 80k copies, which is a lot. Far more than any formula accounting for the existing ranks and sales of the other 4 releases with known sales could have predicted. The top 100 this film spent its release week rocking has to be worth significantly more than the rest of the rankings. My best guess here is an x^-0.7 power law from mainstream top-20 BD fits, with small-scale adjustments based on Battle of Gods’ own ranks over its first week.
Incorporating all of these observations plus the rank-sales data I know I have, here’s my current best-guess attempt at determining how much a given amazon rank R is worth in terms of on-chart sales S. This formula fits only the 6 days post-release, disregarding pre-orders.
If R is between 1-99: S=252000*R^-0.7
If R is between 100-999: S=10000*R^-0.4
If R is between 1000-9999: S=630-(R-1000)*(620/9000)
If R is between 10000-40000: S=10-(R-10000)*(9/30000)
If R is greater than 40000: S=40000/R
It’s super-ugly, I know, but making it piecewise is the only way to fit all the existing datapoints. We’ll see how good a job it does predicting things over the next 2-3 months. To kick that testing period off, Freezing: Vibration, April’s most likely high-seller, is predicted to total 10,338 copies across regular and limited editions, so that’s got a decent chance of charting if the thresholds are generous and the formula is less wrong this time.
Anyway, here’s the May 2015 releases I’ll be tracking, data for which was first pulled on April 27.