ANIME STATISTICS, The Anime Laboratory n°3
In this issue, a literature survey of anime studies seeks to highlight a shift brought by artificial intelligence, favoring quantitative approaches. A development follows on relevant statistical indicators and sources for anime production and reception alike. Ensues a discussion on the necessity of articulating inductive case studies with global trends from relevant indicators, by always navigating with nuance every scale of reasoning. Statistical mapping is then employed as part of spatial scientometrics for a case study of a symposium to track its possible evolutions at the national scale.
I. Literature survey
There are various angles one can approach anime as a study object from. Perhaps the most intuitive will be reviews, whether written on websites such as AniList, or video essays on YouTube. Though I mention platforms, word of mouth (Ameri, Honka, Xie, 2019) is a strong factor of socialization between different fandoms, also corresponding to increasingly sophisticated anime-watching decisions. The reason to think so is that anime can be effective at asking thought-provoking questions, making us adopt fundamentally challenging premises.
But there is also a tradition of academic surveys in Japanese animation, though one of its flaws could be that it’s often approaching series only as thematic (or literary, one could say) objects; without, except perhaps sporadically (Lamarre, 2002, Denison, 2023), complexifying one’s approach of the medium from a technological point of view. Another bias that can come up is perhaps using a framework that comes from other disciplines but animation studies, id est film studies or theater studies, to study anime as if it were cinema or incarnated character acting — which it is but in a mediated way (Auger, 2020), explored in the way Lucky Star featured an otaku, fan of a show that the voice actress of the otaku had participated in the voice acting of, that is to say Suzumiya Haruhi no Yuuutsu, meanwhile the otaku, Konata, is cosplaying Haruhi.
In some way, the last article I worked on was an attempt to mix gender studies and animation techniques together, through the quantitative and qualitative study of henshin (Hoarfrost, 2026a). I have also tried at other times to work on real-life places that are featured in anime, in a larger effort to strive for a geography of anime. In this aspect, the various scales of Australian animanga censorship (Remminga, 2026) are contributing to further complexify the idea of anime geography, also with the aspect that gender is a lens to read space with, not just the dessimenation of products but also in cosplay as a production of space (Lefebvre, 1974). Indeed, the Australian scales of censorship of animanga goods is said to “disproportionately restrict content that is queer in nature or intended for a female audience, and reinforce heterosexual, patriarchal norms”, tying in with the development I do on the first issue about the ways far-right-leaning anime fandoms can appreciate the heterosexual, patriarchal stereotypes circulating in media and the policing of girls, women and gender-diverse people in the spirit of Foucault’s biopower (how modern Nation-States control bodies and populations) or natalism politics. (Hoarfrost, 2026b)
Historical studies like in Studio Ghibli: An Industrial History (Denison, 2023) are their own breeds that I would consider akin to documentaries. In fact there is considerable fluidity in the ways one can document animation and filmmaking initatives are good to restitute motion. Erica Friedman’s By Your Side: The First 100 Years of Yuri Anime and Manga (Friedman, 2022), though focusing more on a genre lens and on the animanga industrial complex, also pays attention to extraneous factors impacting transnational receptions of works. For instance, the Lavender Menace of the 1970s, or the Class S (Shamoon, 2021) of the 1920s.
With current technological developments with artificial intelligence applications, specifically with LLMs and transformer/frontier models, albeit far from lacking problematical environmental, sociospatial (one needs but to look at xAI and South Memphis, with a data center that augmented locals’ bills) and political implications (Mistral AI is illustrating itself in France, positioning as a potential European leader by discussing sovereignty), quantitative approaches to Japanese animation or anime have definitely gotten easier. This is because tasks that used to require intense education, like programming, data gathering, sanitization, visualization, now have a lower barrier to entry.
While this can be used for harm, especially considering statistical representations have appeared considerably recently in the evolutionary trajectory of Homo sapiens, and because we are primarily visual species that are easily influenced by manipulatory statistics — though this has to be nuanced with the fact that education is important for anything in the nature versus nurture department (Gigerenzer, Meder, 2014) — , this is also an opportunity for researchers to adopt more easily a statistical, or modelizing approach of phenomena they might have otherwise had to study using more traditional approaches, and maybe notice patterns that can lead to other studies… increasing knowledge in a constructive way.
II. Statistical indicators and sources
“Knowledge representation and reasoning have been studied in two research areas: logic and statistics. Broadly speaking, logic describes our knowledge using formal languages and reasons about it using symbolic techniques, while statistics interprets collected data having random variation and infers properties of their underlying probability models. As research advances demonstrate, logical and statistical approaches are respectively successful in many applications, including artificial intelligence, software engineering, and information security.”
(Kawamoto, 2019)
According to this quote, we can oppose logic to statistics, though the study was about integrating both together. Still, we are left to wonder what is meant with “interpret[ing] collected data having random variation and infer[ring] properties of their underlying probability models”. We can break it down in several steps.
1) Interpretation. Data is not neutral in its representation, and what speeches, statements we present with it aren’t either. Therefore it is important to conscientize that any such speech is to be questioned, criticized and recognized in its choice to partition (for partition with symbol Ω is a term used in probabilities) existence according to specific observations.
2) Collected data. If in applied statistics, data visualization is the end product, its manufacture, production, sanitization and manipulation involves a series of prior steps that are critical to get useful data visualizations. In cartography too, this is a problem, and statistical maps create an interesting overlap too.
3) Random variation. There are theoretical models in statistics that try to show perfectly uniform expected values in a distribution. And it is in confronting said perfect models to the disparate natures of some phenomena (ie. trans death in the world according to Transgender Europe [Hoarfrost, 2026c]) that we can see whether a polarization has specific meaning.
4) Inferring properties of underlying probability models. This just means: according to such X distribution of Y phenomenon, derive Z probability. For instance, R², the coefficient of determination, “is the proportion of the variation in the dependent variable that is predictable from the independent variable(s)” (Wikipedia, 2026a). As such it is a unit that tells us something about the behavior of a distribution of two or more variables. The stronger it is, the more interlinked the variables are. It can be used both in a traditional two-dimensional, orthogonal view or in tri-dimensional representations.
This has given us a working definition of statistics, but maybe not of indicator. While it has a more concrete meaning of pointing at, it also has a figurative meaning: that of a statistical index, with a range when it is not an absolute value. And an index permits comparison. But it can have inherent flaws. GDP is an indicator for economics but it fails to account for domestic production, or even the work of some NGOs that can’t be quantified accurately, perhaps due to being tied with environmental preservation, something that is very interdependent with the atmosphere, the lithosphere, the hydrosphere with consequences we can’t reduce to a mathematical representation.
But an index can either concern a snapshot of a given time, or a time range. Taking [Figure 1], that I produced with the help of artificial intelligence, I used a combined index, MyAnimeList’s popularity and score indexes, to represent by clusters a few series of the yuri genre.
One could say the time range of the combined index is 2004–2026, considering Garrett Gyssler created the website then, but the fact is some series were only released later, and I regrouped several entries in franchises. For instance, Madoka Magica has a movie releasing this year.
An indicator, then, can either be made up for a specific representation, or be used from a source as is. Because the former case cannot be predetermined, I am now going to focus on already existing sources or indicators, relevant to anime studies… Or rather, that I deem so as of now. But I can’t imagine what future scholarly generations will create to suit their specific needs as far as data usage goes.
An interesting source of indicators for anime production is Oricon:
“a Japanese corporate group that supplies statistics and information on music and the music industry in Japan and Western music […] which was founded by Sōkō Koike in November 1967 and became known for its music charts. […] The charts are compiled from data drawn from some 39,700 retail outlets (as of April 2011) and provide sales rankings of music CDs, DVDs, electronic games, and other entertainment products based on weekly tabulations. […] The group also lists panel survey-based popularity ratings for television commercials on its official website.”
(Wikipedia, 2026b).
The relationship is a bit complex and I’ll explain. Often, anime is not a profitable industry and studios (40 % according to Anime Senpai, 2022) wind up in the red. They instead work as advertisement for derivated products (dakimakura, figurines, plushies, goodies…) and source materials (manga volumes, VN entries on Steam…). This justifies the complex industry structure of production committees. Tracking sales of source materials thus makes them act as indicators of a franchise’s success, and thus sustainability.
To properly understand industry data visualizations, like in [Figure 2], one needs to pay attention to units and labels, that may correspond to specific locations. There are three graphs that have different scales. Global manga circulation is a cumulative graph, we see that in 15 years, manga sales for the One Piece franchise have almost tripled from 2010 to 2025. But that annual Japanese sales peaked in 2011, slowly receding to almost 5 million copies. The same story is told for the sales of the top-selling One Piece volume with a peak in Japan in 2011, and a slow decline since. Micro-economics shouldn’t always be put in relationship to a wider macro-economics context, but if we were to do so, we could highlight the stagnation and inflation that has characterized Japan for the decades studied.
[Figure 3] uses MyAnimeList score distributions to characterize some receptions. We see that Berserk (2016) follows a slightly off-centered Gaussian curve, but that the extremes of the discrete scoring range (1 to 10) see a little rise by virtue of being clamped extremes. It is probably to signify that whatever audience MyAnimeList usage represents, this audience wanted to signify extreme reactions of appreciation or rejection. Using Frieren today is not neutral, as it is the first rated show on the website and thus, is a good example of an extreme for a typology. The last one is Ex-Arm, which had a very polarized and negative reaction despite possessing genuine strengths as far as anime direction goes. The problem is that it makes exploratory use of motion capture, which is perhaps at odds with what audiences expected — if one thinks of anime as often defined by limited animation (Lamarre, 2002), and it became too popular, therefore its popularity increased negative ratings in a feedback loop.
As such MyAnimeList, AniList, Kitsu and similar websites are interesting to explore global, overseas receptions. Domestic demand is an important part of Japanese animation reception still, and so we should also look for specifically Japanese reception indicators, which is partly what we actually did in [Figure 2], but derivated. Maybe animation demand can be inferred from manga sales, but that’s kind of a stretch.
[Figure 4] uses data from Niconico anime surveys for the Madoka Magica original series. It is log-scaled therefore percentage of respondents requires some careful reading. Still we see that the reception has global trends and a bit of volatility. But we see that the Japanese reception was excellent and that the series has become a household, beloved classic. Just like MyAnimeList, it has discrete ratings but we could make a curve to smooth it.
But the indicators you use should vary depending on what you want to accentuate. In (Hoarfrost, 2026a) I used Sakugabooru data because it is a website dedicated to curate qualitative animation, which itself introduces different kinds of biases, not to mention that there is a glocalization effect of the medium in online platforms that, albeit seem ubiquitously accessible online, wind up producing very distinct communities based on platform designs.
III. Articulating deductive and inductive
Thus far, remark that I have only been engaging in inductive figures. Case studies, incarnated in individual franchises or series. Madoka Magica series, One Piece manga, a few yuri franchises to make a typology of the genre. And yet, I think any anime studies scholar, or people trying to interact with the production of knowledge around this study object (I’m more of the second type), should also be aware of more global trends. This is because it can be easy to be fixated in a single franchise, or show, and mistake local patterns for something qualitatively unique about a show, when a greater trend is going on in the background. In (Hoarfrost, 2026a), I was trying instead to engage early on in the creation of a model, going from inductive reasoning to generalization, to then use this model for a comparative analysis.
The sources, or indicators, for more global statistics around Japanese animation, can come in many shapes. But they should be relevant for a national scale. If one wants to think about the entirety of Japanese animation, it can be useful to discuss nationwide political indicators, economics, demographics… but also highly polarized geographic places, that also concentrate relevant things for a national imaginary. Hence the place of Akihabara in any otaku-centric anime production. But statistics about places where animators are formed, like vocational schools, can also be useful too.
Because it’s me, and you’re reading The Anime Laboratory, which thinks of itself as an epistemology journal on Japanese animation and anime studies, there’s one more thing you could think about. The geographic repartition of the scholars, scholarly institutions or even just languages dominating the sources of a given bibliography. Should the amount of Japanese researchers matter in a bibliography about anime studies? Is it one of the metrics that matter to avoid any potential of othering the essence of Japan, to reify it as part of Orientalism (Said, 1978)?
“The Orient has helped to define Europe (or the West) as its contrasting image, idea, personality, experience. Yet none of this Orient is merely imaginative. The Orient is an integral part of European material civilization and culture. Orientalism expresses and represents that part culturally and even ideologically as a mode of discourse with supporting institutions, vocabulary, scholarship, imagery, doctrines, even colonial bureaucracies and colonial styles. In contrast, the American understanding of the Orient will seem considerably less dense, although our recent Japanese, Korean, and Indochinese adventures ought now to be creating a more sober, more realistic “Oriental” awareness.”
This may be a bias by the fact that I only know the scholarship that I have been exposed to, which itself is limited by my grasp of languages. Hence, I would struggle to read through Japanese research were it not to be translated. And the language of choice I use to write, like now, is English, because it is to me a lingua franca. But French is also a demographically growing language, with scholarship that is developing. Anime studies is a growing subject, helped by the globalization of the medium. Still, I believe that to avoid an effect of hegemony whether by a language with its own paradigm or by “scholarship”, as Said puts it, mixing various geographical origins of scholarly influences, at the risk of lowering accessibility by having sources in foreign languages to that of the reader, is a possible mitigation.
However, caution is to be kept at all times when manipulating data, deontologically because more credibility is given to this form of data presentation — mimicking objectivity — the lesser one’s audience is educated in the ways statistics can be manipulative. Ascombe’s Quartet (Ascombe, 1973) shows that datasets can differ in graphs and yet share averages, means…
And so even if a distribution has globally significant attributes, you should still look at it in different representations before you judge of trends, because there could be a signifcant outlier, like in Dataset IV [Figure 8]. The outlier is discussing the particular, or doing a case study. The global parameters like means and averages will be smoothed out by analyzing different instances or observations. In [Figure 1], if you look at Appendix F as well I smoothed out entries that were from the same franchise, so that different franchises can be compared. But this meant manipulating what is seen. Same goes for the maps in [Figures 5, 6, 7] as well where I used a national scale, when I could have differenciated United-States’ States or even Scotland from England in the United-Kingdom. This leads to a risk in generalization of disparate datasets.
IV. Spatial scientometrics: a symposium case study
The Queer and Feminist Perspectives on Japanese Popular Cultures online symposium was as of 2024 “organized by a team of scholars from Concordia University (Canada), UNSW Sydney (Australia) and Tulane University (USA), and supported by the Media, Gender and Sexualities Study Group (University of Tokyo).” (Anime and Manga Studies, 2024). I had the opportunity to participate as a panelist in this edition. I also watched some of the presentations in 2026.
Get Émilia Lilith-Lolita Hoarfrost’s stories in your inbox
Join Medium for free to get updates from this writer.
I believe it is very spatially significant that it is orchestrated online — and with a regular announcement to invite us to think of ourselves also in relationship to the spaces that produce our identities, and to represent ourselves of Zoom Workplace as a material company with material, as well as diachronic, power dynamics — , and with collaboration from North-American, Australian and Japanese scholarly institutions, giving the symposium institutional reputation. And thus, statistical mapping may be an interesting approach to handle this case study. But we’re gonna need a logical framework to approach this.
“A large number of studies in scientometrics deal with the citation impact of different types of collaboration. In particular, several studies have compared the citation impact of internationally co-authored papers with domestically co-authored papers in the light of investigating the rationale for internationalisation policies. Such an effect is expected since more resources are invested in international collaboration and more diffusion channels are activated once results are published. Among the first to assess this issue has been the study by Narin et al. (1991) on a subset of papers covering biomedical research published in 1977. They found that international co-publications are cited on average more often than domestic co-publications, while no differences were found for international co-publications among EU countries, between EU andnon-EU countries and among non-EU countries.”
(Frenken, Hardeman, Hoekman, 2009)
Note that focusing on this symposium in particular is not totally related to anime studies, since it involves discussions on Japanese contemporary culture which takes forms that are very diverse. However, if one approaches anime studies with holism, even a partial relationship is to be analyzed. Which justifies the choice of this case study. Just this quote from (Frenken, Hardeman, Hoekman, 2009) invites to think of the impact in knowledge production of an international symposium held online, thus meaning accessibility, but also potential American imperialism or hegemony.
A caveat is that independent researchers, except myself since I know for sure I’m French, aren’t tied to institutions. It excludes some panelists de facto from being accounted for towards statistics related to national representation.
We see that Japan is justly one of the hotspots for panelists in this symposium, but also that Australia and North-America, with the US moreso than Canada, are quite heavily weighting in. Then, Brazil, the United-Kingdom, Italy and France for 2024. Let’s see with [Figure 6] if in 2026, the data looks similar or if new trends occur.
A significant thing to notice is that [Figures 5, 6, 7] demonstrate the evolution of the symposium to account for narratives concerning contemporary Japanese culture from more diverse sources. This is seen with participations from institutions based in Indonesia, India, China, Hong Kong, and a lowered participation from the United-States or France, for instance. In a way, then, the symposium seems to have had a mitigation to the problem of possible Western hegemony in participants’ locations and institutions of affiliation.
However, there is still a potential issue in the apparent absence of Africa as a continent. Because I’m Franco-Moroccan, maybe there is a world in which I would have included Morocco, but my sociological trajectory makes me exclude it. Surely the African continent, 20 % of emerged lands, has interesting, glocalized narratives of interactions with contemporary Japanese culture, including in its queer and feminist dimensions. Maybe it will be addressed in future occasions. But it can be interpreted as the result of world history playing its course and European colonization causing vast inequalities in socioeconomical development. Other areas of research also show this spatial difference in the production of knowledge.
If anything, the shift to more Asian countries in the 2026 edition of the symposium can be perceived as a cultural decolonization of Western hegemony in the production of knowledge around Japanese contemporary culture.
V. Conclusion
Throughout this article, I have sought to demonstrate that statistics could be helpful for a conception of holistic anime studies, provided a certain methodological apparatus was in use. I have first dedicated a literature survey to distinguish some of the current tracks for anime studies, which may or may not employ statistics. And there is meaning in giving to each work the chance to shine qualitatively too, in a postmodern way to return to humanism instead of the gigantism of quantitative turns. Still, I tried to defend the idea that some statistics were interesting to show various things, for instance the difference between production and consumption and reception, or what geographic scale a specific data covered (see Appendices A and B for different geographic areas, respectively Japan and the entire world). For an anecdote, statistical indicators for otaku culture was my first ever topic for an article online when contributing to Les Confins du Monde, at the start of the decade. But there are some limitations to this: I haven’t mentioned things like studying how many animators worked in the field of Japanese animation at any given time in Japanese animation history (Watzky, 2023), which has been done before by studying credits. Another problem is the methodology employed, which requires clarity, transparency too… In the following part, I wanted to address the problem of having to constantly gauge the general and particular, so as to not ignore trends across the entire medium, or to dilute the essence of the medium in a gross generalization. In the end, I also exposed the idea of scientometrics for anime epistemology, with a series of maps concerning the Queer and Feminist Perspectives on Japanese Popular Cultures, but the fact that it focused on English language panels probably influenced on the most represented countries. Still, it showed an attempt between 2024 and 2026 to platform narratives of interactions with contemporary Japanese culture that were perhaps sometimes obscured by the Western world, from speakers affiliated with scholarly institutions in countries such as China, Indonesia, India, in Hong Kong too (which I didn’t aggregate with China, but could have, it just felt like its historical development was a little unique enough to justify it).
VI. Appendices
Appendix A: MangaCodex data for the One Piece manga for the period 2010–2025.
Appendix B: Ex-Arm, Berserk (2016), Frieren score distributions on MyAnimeList.
Appendix C: Madoka Magica ratings by Niconico survey for period 2011.
Appendix D: Symposium case study, data for year 2024.
Appendix E: Symposium case study, data for year 2026.
Appendix F: Data for yuri anime franchises typology.
VII. References
Francis John Ascombe (1973). “Graphs in Statistical Analysis”, The American Statistician, Vol. 27, №1, pp. 17–21. https://www.scirp.org/reference/referencespapers?referenceid=4251767
Mina Ameri, Elisabeth Honka, Ying Xie (2019). “Word of Mouth, Observed Adoptions, and Anime-Watching Decisions: The Role of the Personal vs. the Community Network”, Marketing Science 38(4):567–583, Institute for Operations Research and the Management Sciences. https://doi.org/10.1287/mksc.2019.1155
Anime and Manga Studies (2024). “Online Symposium — Queer and Feminist Perspectives on Japanese Popular Cultures”, Announcements, Programs. https://www.animemangastudies.com/2024/04/01/online-symposium/
Anime Senpai (2022). “40% of Anime Studios Suffered Losses In 2021”. https://www.animesenpai.net/40-of-anime-studios-suffered-losses-in-2021/
Simon Auger (2020). “Le modèle “ base de données ” d’Hiroki Azuma et sa mise à l’épreuve dans la série d’animation japonaise Lucky Star (2007)”, Sciences de l’Homme et Société. https://dumas.ccsd.cnrs.fr/dumas-02506404v1
Rayna Denison (2023). Studio Ghibli: An Industrial History, Palgrave Animation. ISBN: 3031168437
Koen Frenken, Sjoerd Hardeman, Jarno Hoekman (2009). “Spatial Scientometrics: Towards a Cumulative Research Program”, Journal of Informetrics 3(3):222–232, July. https://www.researchgate.net/publication/220065944_Spatial_Scientometrics_Towards_a_Cumulative_Research_Program
Gerd Gigerenzer, Björn Meder (2014). “Statistical Thinking: No One Left Behind”, Advances in Mathematics Education, pp. 127–148, Softcover. https://link.springer.com/book/10.1007/978-94-007-7155-0
Emilia Lilith-Lolita Hoarfrost (2026a). “Animating Queer Phenomenology: Henshin in Yuri Kuma Arashi and Madoka Magica”, Magical Stage. https://magicalstage.moe/2026/09/23/animating-queer-phenomenology-henshin-in-yuri-kuma-arashi-and-madoka-magica/
Emilia Lilith-Lolita Hoarfrost (2026b). “ANIME’S FAR-RIGHT PROBLEM, The Anime Laboratory n°1”, Medium. https://medium.com/@emiliahoarfrost/animes-far-right-problem-11c808ef2f34
Emilia Lilith-Lolita Hoarfrost (2026c). “Transness, death and data (2008–2025)”, Medium. https://medium.com/@emiliahoarfrost/transness-death-and-data-2008-2025-01c9296cf66a
Emilia Lilith-Lolita Hoarfrost (2026d). “CHARTING YURI ANIME: Patterns of Japanese Popcultural Sapphism/Lesbianism Animated Media”, Medium. https://medium.com/@emiliahoarfrost/charting-yuri-anime-patterns-of-japanese-popcultural-sapphism-lesbianism-animated-media-5dceb74991e3
Yusuke Kawamoto (2019). Statistical Epistemic Logic, National Institute of Advanced Industrial Science and Technology (AIST). https://arxiv.org/html/1907.05995v1
Thomas Lamarre (2002). “From animation to anime: drawing movements and moving drawings”, Japan Forum, 14(2), 329–367, British Association for Japanese Studies. DOI: 10.1080/09555800220136400
Henri Lefebvre (1974). “La production de l’espace”, L’Homme et la société, 31–32, pp. 15–32. https://www.persee.fr/doc/homso_0018-4306_1974_num_31_1_1855
Ashley Remminga (2026). “Obscene content: policing queer media, erotica, and fan creations in Australia”, Porn Studies. https://www.tandfonline.com/doi/pdf/10.1080/23268743.2026.2625085
Edward Said (1978). Orientalism, Pantheon Books, United-States.
Deborah Shamoon (2021). “Class S: appropriation of ‘lesbian’ subculture in modern Japanese literature and New Wave cinema”, Cultural Studies, 35(1), 27–43. https://www.tandfonline.com/doi/abs/10.1080/09502386.2020.1844259
Matteo Watzky (2023). “Anime Numbers — 3 — Working conditions through time”, Full Frontal. https://fullfrontal.moe/anime-numbers-3/
Wikipedia (2006–2026a). “Coefficient of determination”. https://en.wikipedia.org/wiki/Coefficient_of_determination
Wikipedia (2006–2026b). “Oricon”. https://en.wikipedia.org/wiki/Oricon
Follow, clap, highlight, repost for more content. Comment to discuss. Look at the Anime list for similar articles.
