Anime Sales in Japan - 2013 (Released volumes)

[meruruchan]

Actually only include Fall '12 shows because time. Will add more shows/update later maybe. v2+ data no included.
Updated as of 6/3/2013

Key: Show title - Oricon total sales - Stalker final combined points - Difference (units) - Difference (%)
Girls und Panzer (BD only) ________ ___ 26,445 ____18,206 ___+*8,239__ __+*45,25%
Jojo no Kimyou na Bouken _______ ____ 19,370 ____13,709____+*5,661____+*41,29%
Chuunibyou Demo Koi ga Shitai! ___ ____17,520 ____10,185____+*7,335____+*72,02%
Magi___________________________ ___ 13,801 ____10,345____+*3,456____+*33,41%
To LOVE-Ru Darkness __________ _____ 12,674 ____10,011____+*2,663____+*26,60%
Gintama' Enchousen (DVD) _ ___ ______ 11,723 ____11,093_ __ +**,630____+**5,68%
Little Busters! __________________ ____11,609 ____ *7,207____+*4,402____+*61,08%
Initial D 5th Stage __________________ 10,440 ___ _*2,698____+*7,742____+286,53%
Hidamari Sketch x Honeycomb ___ ____ *9,928 ____ *7,201____+*2,727____+*37,87%
[K]__________________________ _____ *9,297 __ _ *2,123____+*7,174____+337,92%
Psycho-Pass________________________ *8,644 ____*4,500____+*4,144____+*92,09%
Kami-sama Hajimemashita____________*6,249 ____*4,439____+*1,810____+*40,77%
Sakurasou no Pet na Kanojo ________ __*3,386 ___ *2,992____+**,394__ _ +*13,17%
Busou Shinki _ ___ _ ___ ____ ___ _ _ _ *2,764 __ _ *2,225____+**,539___ +*24,22%
Oniichan Dakedo Ai [...] (BD only)______*2,497 ____*1,911____+**,586____+*30,66%
Tonari no Kaibutsu-kun____________ __ *2,494 ____*1,831____+**,663____+*36,21%
Jormungand Perfect Order (BD only) _ _ *2,392 __ _ *2,650____−**,258____−**9,74%
Zetsuen no Tempest _________________*1,911 ____*1,850____+**,*61____+**3,30%
Ixion Saga DT ______________________*1,759 ____*1,717____+**,*39____+**2,27%
Hiiro no Kakera Season 2 (DVD only)___ *1,314 ____*1,342__ _−**,*28____−**2,09%
Robotics;Notes (BD only) _____________*1,235____ *1,211____+**,*24____+**1,98%


Average absolute deviation (units): 2789,3
Standard deviation (units): 3982,9
Average absolute deviation (%): 57,34%
Standard deviation (%): 103,63%




Week 1 data only analysis:
Key: Show title - Oricon week 1 sales - Stalker final combined points - Difference (units) - Difference (%)
Jojo no Kimyou na Bouken _______ ____ 19,370 ____13,709____+*5,661____+*41,29%
Chuunibyou Demo Koi ga Shitai! ___ ____14,458 ____10,185____+*4,273____+*41,95%
Magi___________________________ ___ 13,801 ____10,345____+*3,456____+*33,41%
Girls und Panzer (BD only) ________ ___ 13,550 ____18,206 ___−*4,656____−*25,58%___(supply constrained)
To LOVE-Ru Darkness __________ _____ 10,214 ____10,011____+**,203____+**2,02%
Gintama' Enchousen (DVD) _ ___ ______ 10,197 ____11,093_ __ −**,896____−**8,08%
Little Busters! __________________ ____10,115 ____ *7,207____+*2,908____+*40,35%
Hidamari Sketch x Honeycomb ___ ____ *8,916 ____ *7,201____+*1,715____+*23,82%
Initial D 5th Stage __________________ *6,534 ___ _*2,698____+*3,836____+142,18%
Psycho-Pass________________________ *5,659 ____*4,500____+*1,159____+*25,76%
[K]__________________________ _____ *5,197 __ _ *2,123____+*3,074____+144,80%
Kami-sama Hajimemashita____________*4,747 ____*4,439____+**,308____+**6,94%
Sakurasou no Pet na Kanojo ________ __*3,386 ___ *2,992____+**,394__ _ +*13,17%
Tonari no Kaibutsu-kun____________ __ *2,494 ____*1,831____+**,663____+*36,21%
Jormungand Perfect Order (BD only) _ _ *2,392 __ _ *2,650____−**,258____−**9,74%
Oniichan Dakedo Ai [...] (BD only)______*2,231 ____*1,911____+**,320____+*16,75%
Busou Shinki (BD only) ______ ______ _ *2,184 __ _ *1,582__ _+**,602___ +*38,05%
Zetsuen no Tempest _________________*1,911 ____*1,850____+**,*61____+**3,30%
Ixion Saga DT ______________________*1,759 ____*1,717____+**,*39____+**2,27%
Hiiro no Kakera Season 2 (DVD only)___ *1,314 ____*1,342__ _−**,*28____−**2,09%
Robotics;Notes (BD only) _____________*1,235____ *1,211____+**,*24____+**1,98%

Average absolute deviation (units): 1644,5
Standard deviation (units): 2419,2
Average absolute deviation (%): 31,42%
Standard deviation (%): 50,11%

[something]

Originally Posted by meruruchan

Actually only include Fall '12 shows because time. Will add more shows/update later maybe. v2+ data non included.

Thanks, saves me the time. Just two points:

- Your key has "Stalker final combined points - Oricon sales" backwards, your list has Oricon first, Stalker second.
- Stalker only predicts first week sales, not total sales. Especially now that it stops adding points on release day, rather than 7 days after. That's why it looks like Stalker has undercounted so many in your list. If you just pull first week sales it'll change the data a good bit (of course it will skew GirlPan but that's just trading one skew for another, predictions are useless when stock is inadequate). Notice how much closer the shows that ranked once are than the shows that ranked multiple times.

Also a small posting request: when you do some kind of useful analysis like this, can you do it as a reply to this post? Currently it's been done as a reply to one of the Vol. 1 ranking posts so when I try to quickly skim for a post to do updates, I now see something like this. With just a few replies it's not a huge problem but if whole conversation threads get inserted around there, it makes finding posts quickly more difficult. (To that end I've replied to this under the misc subthread instead of to your original post.)

[something]

Originally Posted by something

If you just pull first week sales it'll change the data a good bit

So you don't have to go digging, here are the first week numbers (without muich formatting:

* - DVD - BTOOOM!
* - BD - BTOOOM!
* - DVD - Busou Shinki
2184 - BD - Busou Shinki
1885 - DVD - Chuunibyou Demo Koi ga Shitai!
12573 - BD - Chuunibyou Demo Koi ga Shitai!
* - DVD - Code:Breaker
* - BD - Code:Breaker
10197 - DVD - Gintama' Enchousen
* - DVD - Girls und Panzer
13550 - BD - Girls und Panzer
* - DVD - Hayate no Gotoku! Can't Take My Eyes Off You
* - BD - Hayate no Gotoku! Can't Take My Eyes Off You
5131 - BD - Hidamari Sketch x Honeycomb Fandisc
1610 - DVD - Hidamari Sketch x Honeycomb
7306 - BD - Hidamari Sketch x Honeycomb
1314 - DVD - Hiiro no Kakera season 2
* - BD - Hiiro no Kakera season 2
6534 - DVD - Initial D Fifth Stage
754 - DVD - Ixion Saga DT
1005 - BD - Ixion Saga DT
14860 - DVD - Jojo no Kimyou na Bouken
4510 - BD - Jojo no Kimyou na Bouken
* - DVD - Jormungand Perfect Order
2392 - BD - Jormungand Perfect Order
1607 - DVD - K
3590 - BD - K
3155 - DVD - Kami-sama Hajimemashita
1592 - BD - Kami-sama Hajimemashita
28166 - DVD - Kuroko no Basuke Fandisc ~Owaranai Natsu~
1757 - DVD - Little Busters!
8358 - BD - Little Busters!
9003 - DVD - Magi
4798 - BD - Magi
* - DVD - Medaka Box Abnormal
* - BD - Medaka Box Abnormal
* - DVD - Onii-chan Dakedo Ai Sae Areba Kankenai yo ne
2231 - BD - Onii-chan Dakedo Ai Sae Areba Kankenai yo ne
1183 - DVD - Psycho-Pass
4476 - BD - Psycho-Pass
* - DVD - Robotics;Notes
1235 - BD - Robotics;Notes
746 - DVD - Sakura-sou no Pet na Kanojo
2640 - BD - Sakura-sou no Pet na Kanojo
* - DVD - Shinsekai Yori
* - BD - Shinsekai Yori
* - DVD - Sukitte Ii na yo.
* - BD - Sukitte Ii na yo.
1928 - DVD - To Love-Ru Darkness
8286 - BD - To Love-Ru Darkness
1237 - DVD - Tonari no Kaibutsu-kun
1257 - BD - Tonari no Kaibutsu-kun
1027 - DVD - Zetsuen no Tempest
884 - BD - Zetsuen no Tempest

[meruruchan]

Originally Posted by something

Thanks, saves me the time. Just two points:

- Your key has "Stalker final combined points - Oricon sales" backwards, your list has Oricon first, Stalker second.

I'm stupid. Fixed that.

Originally Posted by something

- I believe Stalker only predicts first week sales, not total sales. That's why it looks like Stalker has undercounted so many in your list. If you just pull first week sales it'll change the data a good bit (of course it will skew GirlPan but that's just trading one skew for another).

Not really sure about what Stalker wants to predict but yes, using wk1 data only reduces undercounting by a good bit.
I went with "total" sales because some shows add A LOT of copies in wk2+, and because the total is what matters in the end (also because it saved me time). Will add wk1 data later.

Originally Posted by something

Also a small posting request: when you do some kind of useful analysis like this, can you do it as a reply to this post? Currently it's been done as a reply to one of the Vol. 1 ranking posts so when I try to quickly skim for a post to do updates, I now see something like this. With just a few replies it's not a huge problem but if whole conversation threads get inserted around there, it makes finding posts quickly more difficult.

Your link it's a 404 for me but OK, I will remember it.
EDIT: OK, I see it now.

Originally Posted by something

So you don't have to go digging, here are the first week numbers (without muich formatting:[...]

I love you.

[something]

Originally Posted by meruruchan

Not really sure about what Stalker wants to predict but yes, using wk1 data only reduces undercounting by a good bit.
I went with "total" sales because some shows add A LOT of copies in wk2+, and because the total is what matters in the end (also because it saved me time). Will add wk1 data later.

Okay, there's use for total too, I'm just saying that if anyone wants to draw conclusions about how accurate or not Stalker is (I'm guessing ultimatemegax wants the data in hopes of proving Stalker is deeply untrustworthy =P) then they do have to use first week data only. It's impossible, by definition, for Stalker to predict sales beyond first week if they stop counting points at release day.

Your link it's a 404 for me but OK, I will remember it.

Ah, I fixed the link after you read the post, should work now.

[meruruchan]

Updated with wk1 data and Busou Shinki that I skipped since something listed it as v2 data here for whatever reason.

[something]

Originally Posted by meruruchan

Updated with wk1 data and Busou Shinki that I skipped since something listed it as v2 data here for whatever reason.

When a title only ranks on one format for Vol. 1 but ranks on both for Vol. 2, I'll generally show the Vol. 2 data instead, provided the format that ranked both times has not changed drastically. Busou Shinki's BDs were consistent from v1→v2 so presumably the DVDs were as well.

And it looks like going with week 1 data cut down the offset by half or more.

This makes me wonder if Stalker made the right choice in stopping point accumulation on release day. Including it through release week could result in overestimating, but I think it's unusual for Stalker to underestimate this many. Jojo in particular would have been much closer if release week points were added. But then a number that were close might have gone over - it's very hard to say. No single approach will work for everything.

The holy grail of predicting would be working B&M retailer rankings into the algorithm somehow but that's just too difficult and most of the data is limited or unavailable. There are probably some (crude) tweaks you could apply to the daily point values based on ranking well at Animate or HMV or whatever but it wouldn't be systematic.

I was adding final Stalker points to my spreadsheet and calculating these percentages until recently, when Stalker took away the "past releases" page. Annoyingly, there's now no easy way to keep track of which titles have finished accumulating points so I have to do it more manually... just a bit behind on that now. Depending on whether you intended to do earlier seasons or not, I can throw something together too.

[ultimatemegax]

Originally Posted by something

Okay, there's use for total too, I'm just saying that if anyone wants to draw conclusions about how accurate or not Stalker is (I'm guessing ultimatemegax wants the data in hopes of proving Stalker is deeply untrustworthy =P) then they do have to use first week data only. It's impossible, by definition, for Stalker to predict sales beyond first week if they stop counting points at release day.

Never said it was deeply untrustworthy, just that I feel it's idiotic to compare Stalker data 2 months out like some are doing. It's seen as this gold standard without any type of actual testing, which I want to start to see if it's a valid method.

Here's what I've got thus far via Autumn wk1s. One of the most used statistical tests for these types of data is an ANOVA, an analysis of variance between the projected and actual numbers. Anything above a probability value of 0.05 is seen as significantly different from the projection. Using the wk1 data for Autumn, the p value comes out to 0.49, way less than significantly different than the projected values. It increases to 0.35 when you remove the Girls und Panzer listing due to being supply restrained, but not significantly different. Anything under 0.05 is not significant, end of story, with this value.

This is what the values look like when charted. The line itself is a perfect 1:1 ratio of Oricon wk1 sales and final Stalker predictions. As you can see, most of the dots are under the line (the ones at the left are difficult to tell, but they are underneath it). This indicates that Stalker is under-predicting the values for this season. If a dot is over the line, Stalker over-predicted its value (Hi Gintama!). The former would be your example of "storefront effect" while the latter would be an example of people buying at Amazon way more than at other stores. A coefficient of determination (usually called r-squared) sees how well the data fits the line from 0-1 with 0 meaning not at all and 1 meaning perfectly. The r-squared of this line is 0.8. Pretty good, but not perfect. Excluding the supply-constrained Girls und Panzer increases r-square to 0.9. Not bad at that point.

Would love to add more values to help strengthen or weaken these numbers with more n, but this is what I've got thus far. Basically supports what I've felt. It's a decent estimator (when total numbers have been estimated), but you can't rely on it to give you exact values.

[something]

Originally Posted by ultimatemegax

Never said it was deeply untrustworthy, just that I feel it's idiotic to compare Stalker data 2 months out like some are doing. It's seen as this gold standard without any type of actual testing, which I want to start to see if it's a valid method.

I would definitely agree (and have always said) that the "最終予測累積ポイント" predictions are useless (moreover, probably not even meant to be used) more than a few weeks out, and that Stalker's real value is in automating what would otherwise be very tedious to do: aggregating all the hour by hour rankings to see trends. So it definitely is useful at this point just to see what is trending up and what is trending down.

So, like, I can't say for certain that Tamako Market will sell badly in the end, but I can say that it's not doing very well, and its trajectory will need to change significantly from its present course to be on par with recent KyoAni titles. Certainly the other option is to just say nothing at all, but if we've got a whole separate thread devoted to speculation, then I'm gonna speculate, darnit. There's not too much to talk about otherwise.

One other thing to consider with Stalker: not all predictions are created equal. It is known ahead of time that some titles will not be accurate at all. After the first season of Dog Days, we knew for a fact Dog Days' would be greatly underestimated. But what's an algorithm to do? You could adjust based on previous seasons, but that's very labor intensive and only helps with sequels. You could also try to adjust for director, studio, source material target audience, say assuming ahead of time that Sexy Boys II is a fujoshi title with a really kickass Animate-exclusive Sexy Boy Bedsheet... but these are very arbitrary things to factor into a calculation.

Sometimes it's better to keep the algorithm simple and let human judgment explain the expected outliers. Of course, a statistical analysis aggregating predictions across multiple series will of necessity penalize it for the same, though it's still interesting to see. I especially like the scatter plot graph. I'll be curious if the systematic underestimation happened in Winter/Spring before Stalker stopped counting earlier.

Would love to add more values to help strengthen or weaken these numbers with more n, but this is what I've got thus far. Basically supports what I've felt. It's a decent estimator (when total numbers have been estimated), but you can't rely on it to give you exact values.

See the link at the end for an extraction of the relevant 2012 Vol. 1 data. Note that Stalker has changed its methodology over time, the biggest change being, as I've mentioned, when it stops calculating points. I don't remember when exactly it made the switch to not count anything after release day, but they did make the change retroactively to all previous releases. However I was recording them as they came out, so I still have the pre-adjustment values.

The Aniplex/Amazon issue also renders all BD data points for the affected titles unusable. I've tried to exclude those when I remember.

There are also occasionally releases that just seem to disappear from the database after release, so there are maybe 2 I don't have points for because of that. But they tend to be low-selling DVDs, I've never seen a BD go missing.

I've included:
Series title
Format
Week 1 sales per format
Week 1 total sales
Season
Stalker Points
% difference, negative means underestimates

Though I imagine you'll use your tools to calculate the last. 190 total releases (counting formats separately), 142 of which ranked and can provide data points. Another reason for separating formats is to see if the trends for DVD and BD are the same, or differ somehow.

Spreadsheet (xlsx)

[meruruchan]

Originally Posted by something

When a title only ranks on one format for Vol. 1 but ranks on both for Vol. 2, I'll generally show the Vol. 2 data instead, provided the format that ranked both times has not changed drastically. Busou Shinki's BDs were consistent from v1→v2 so presumably the DVDs were as well.

Oh, that's OK then.

Originally Posted by something

Depending on whether you intended to do earlier seasons or not, I can throw something together too.

I was thinking to add the rest of 2012's shows, whenever I find some time to spare, but that's going to be a bit long.

EDIT: Just saw something's link. That's A LOT of time saved, definitively helpful.