Studio or Staff? · ORF 387 · Princeton University

The puzzle

Why does success in anime cluster around the same few studios?

A small number of names — Kyoto Animation, ufotable, MAPPA, bones, WHITE FOX — keep showing up at the top of fan rankings. That could mean studios genuinely produce better shows. Or it could mean those studios just keep hiring the same handful of acclaimed directors and designers, and the “studio effect” is really a staff effect in disguise.

To tell these apart we built a panel of 1,875 anime released between 2000 and 2024 (the top 75 each year by AniList popularity) and looked at the same panel through two different network views: who made the show (the studio) and who actually worked on it (the creative staff).

The short answer

Both stories are partly true — with one twist.

Studio Studios genuinely cluster on quality (AniList score, MAL rating, our composite success index). Same-studio anime are about 12% more similar in score than different-studio anime, and studio identity adds 2.7–7.2 percentage points of explanatory power on top of basic title controls.
Staff Studios really do rehire the same key creatives: 47% of all shared-staff connections between titles sit inside the same primary studio. Within a studio, anime that share at least two key staff are 28–42% more similar in outcomes than same-studio anime that don’t.
Twist Being a more “central” person in the staff collaboration network — high degree, high betweenness, high eigenvector centrality — does not predict a staff member’s next-year success once last year’s track record is held constant. Repeated teams matter. Individual network position does not.

In one line

The studio label is doing real work, but a lot of what looks like “studio” is really “the same key team showing up again.” A studio’s edge isn’t any one famous staff member — it’s how stably it keeps the same people together.

Move through the tabs above for the detailed findings, the data and method behind them, the figures from the report, and the limitations.

03 · Data & Method

How the study works, in plain English

A high-level walkthrough of where the data comes from, how we set up the two networks, and how the statistical tests decide whether studios or staff are doing the work. Technical specifics are tucked into more detail expanders so they don’t get in the way.

Where the data comes from

Title metadata, staff credits, and audience signals come from the AniList API. We restrict to the top 75 anime per year by AniList popularity from 2000 to 2024, giving a balanced 25-year panel of well-known titles. We also pair each title with a MyAnimeList rating snapshot as an independent quality outcome. After dropping titles missing the score or studio field, the analysis runs on 1,875 anime.

more detail · what is in the panel

1,840 titles in the similarity tests (after dropping a small number with missing composite-index components) and 1,875 in the regressions. Each title has its primary animation studio, year, source material, format, log episodes, log duration, a franchise/sequel flag, and a multi-studio flag. AniList responses are cached on disk so the panel is reproducible.

The two networks

Same panel, two views. G_studio connects two anime if they share the same primary animation studio. G_staff connects two anime if they share at least two key creative staff. The job is to see whether the similarity-on-success that shows up through the studio lens survives when we look through the staff lens instead.

more detail · what counts as “key staff”

Six creative roles: Director, Series Composition, Character Design, Chief Animation Director, Music, and Sound Director. Voice actors are excluded. We also build a separate person-year staff network, where two people are linked if they were credited on the same title in the same year, used only for the centrality regression.

How we measure success

Four primary outcomes per title: AniList mean score (the audience rating), popularity, favorites, and a composite success index (success_index) that averages standardized score, log popularity, and log favorites. The MAL rating from Kaggle is a robustness outcome we report alongside.

more detail · success_index formula

The report’s success_index is the simple average of three z-scored quantities — score, log(1+popularity), and log(1+favorites) — so a title’s value reflects how it compares to the panel average across all three signals at once.


                success_index_i =
                13
                [ z(S_i) + z(log(1 + P_i)) + z(log(1 + F_i)) ]

S is AniList score, P is popularity, F is favorites, and z(x) = (x - panel mean) / panel standard deviation.

The permutation test

We compare same-studio anime pairs to different-studio pairs on the average absolute difference in outcome — lower means more similar. To check whether the gap is real, we shuffle outcome values across studio labels 251 times and ask how often the shuffled gap is at least as big as the observed one. A small p-value means studios carry real information.

The regressions

On the studio side we fit a title-level OLS with title controls and primary-studio fixed effects, then ask how much the studio block adds to fit. On the staff side we fit a person-year OLS with prior-year success, career age, title count, and year fixed effects, then add centrality features and ask the same question.

more detail · what gets controlled for

Studio model controls: year, source material, format, log episodes, log duration, multi-studio flag, franchise/sequel flag. Studios with fewer than 10 sampled titles are pooled into a single OTHER reference group.

Staff model controls: prior-year success, title count in year t, career age, year fixed effects. Centrality features (degree, betweenness, eigenvector, clustering) get z-scored within year so values are comparable across years.

The within-studio shared-staff test

The cleanest staff-continuity test in the report. Among anime pairs from the same primary studio, we compare pairs that share at least two key staff to pairs that share none. If the shared-staff pairs are systematically more similar, that is direct evidence that repeated teams matter beyond the studio label.