Do Shared Hobbies Mean Better Chemistry?
Decades of social psychology research back the similarity-attraction effect: people consistently rate strangers as more likeable when they share similar attitudes, values, and interests. Byrne (1971) formalised this as a near-linear relationship - the more overlap, the more liking. Shared hobbies act as what researchers call activity anchors, recurring reasons to spend time together that sustain friendships over the long run.
Do Umas follow the same patterns, or did decades of social psychology simply fail to account for a crucial data group?
Methodology note: Agnes Digital excluded
Agnes Digital was removed from all analysis.
Her relation types overlap with nearly every Uma in the roster, giving her a universally high affinity score regardless of shared hobbies. In statistics this is called a high-leverage influential observation. A single data point whose extreme position pulls aggregate results away from the underlying trend for everyone else. In psychology the equivalent is acquiescence bias: a systematic tendency to respond positively to everything, inflating scores across the board and compressing the variance that makes comparison meaningful.
Keeping her in would make her dominate the "most connected" chart, widen every group-average by an unearned margin, and generally make the data say more about Agnes Digital than about affinity.
She is not an outlier because she is bad data. She is an outlier because she is Agnes Digital.
Classification note: Digging holes
Agnes Tachyon's hobby "Digging holes" is classified under Physical & Body. One could argue it belongs in Nature & Outdoors. The authors, however, chose to interpret unannounced excavation as a physical discipline. We stand by this.
Each node is an Uma. Edges connect the top affinity pairs — thicker and brighter means higher score. Node size reflects total affinity across the roster. The big blob in the middle is Social & Interpersonal. On the far left, a few Umas barely connected to anyone.
The average affinity between any two Umas from the same hobby group is 17.1. Between Umas from different groups: 18.2. Small gap, wrong direction. Sharing a hobby group is a mild disadvantage for affinity. The heatmap below shows which group pairings do best.
The strongest cross-group chemistry is Competitive & Athletic and Intellectual & Study (avg 22.0), followed by Competitive & Athletic and Creative & Artistic (avg 21.7). The weakest pairing is Creative & Artistic and Nature & Outdoors at 16.2.
Top 5 most connected: Gentildonna, Vodka, Narita Brian, T.M. Opera O, Mejiro McQueen. Four of the five are Social & Interpersonal. Biwa Hayahide (Domestic & Culinary) is the exception at sixth. The most connected Umas are not necessarily the ones with the highest individual pair scores; they are the ones with consistently high affinity across the whole roster.
The top pair is Gold Ship and Orfevre at 45 pts, same hobby group (Social & Interpersonal). But 9 of the top 15 pairs are different groups. Smart Falcon and Copano Rickey (Social and Domestic, 43 pts) and Hokko Tarumae and Wonder Acute (Nature and Social, 42 pts) both outperform most same-group pairs. The affinity system does not care about hobby alignment.
So What Does Drive Affinity?
Note: Affinity values are hard-coded numbers in the game data. We don't have insight into exactly how Cygames calculated them. The patterns below are our best interpretation based on what the data shows. There may be additional factors we can't see, such as Umastagram follow relationships between characters.
Gold Ship and Orfevre are half-siblings, both sired by Stay Gold, and they raced each other at the 2013 Arima Kinen. The broader Gold Clan (Stay Gold, Orfevre, Gold Ship, Dream Journey, Nakayama Festa, Fenomeno) all share the same sire line and cluster visibly in the network.
Biwa Hayahide and Narita Brian are half-siblings through their dam. They appear frequently in each other's top pairs and in the most-connected list. Symboli Rudolf and Tokai Teio share a parent-child bond. T.M. Opera O and Biwa Hayahide, and T.M. Opera O and Symboli Rudolf, both crack the top 15. Opera O came up in the era shaped by Rudolf's legacy.
Hokko Tarumae, Wonder Acute, Smart Falcon, and Copano Rickey were all dirt specialists who competed at the same time, repeatedly racing against each other across multiple seasons.
The Mejiro family (McQueen, Ryan, Dober, Palmer, Ardan, Bright, Ramonu) all trace back to the same stable and bloodline. Ardan and Ramonu are sisters. The scores show it.
The Umas with the highest mutual affinity are the ones who share family, a sire, a dam, or a rivalry.
Who is the Most Liked Mejiro?
The Mejiro family has seven members in the dataset. Ranked by total affinity across the full roster:
| Uma | Total Affinity | Top Partner |
|---|---|---|
| Mejiro McQueen | 2437 | Biwa Hayahide (38) |
| Mejiro Dober | 2232 | Still in Love (36) |
| Mejiro Palmer | 2219 | Mayano Top Gun (37) |
| Mejiro Ramonu | 2164 | Gentildonna (35) |
| Mejiro Ryan | 1992 | Ines Fujin (29) |
| Mejiro Ardan | 1959 | Mejiro Ramonu (31) |
| Mejiro Bright | 1848 | Special Week (31) |
McQueen is not just the most connected Mejiro. She sits in the top 5 of the entire roster. Her closest partner outside the family is Biwa Hayahide at 38 pts, higher than any intra-Mejiro pair. Ardan and Ramonu are sisters and their top score is with each other (31), which tracks. Mejiro Bright has the lowest total of the family and her best partner is Special Week, someone with no Mejiro lineage at all.
Does Tracen Academy Do a Good Job Assigning Roommates?
Tracen pairs two Umas per room, and the assignment loosely follows training history. Ritto dorm for horses trained at Ritto, Miho dorm for Miho. But does the academy actually put compatible Umas together?
Out of 40 roommate pairs with affinity data, the average score is 26.0. The overall roster average is 17.6. Roommates score about 48% higher than two random Umas pulled from the dataset.
65% of roommate pairs also share the same hobby group compared to roughly 34% you'd expect by chance given the group distribution. Tracen is putting like with like.
The standouts are genuinely well-matched. TM Opera O and Biwa Hayahide lead at 40 pts. Daiwa Scarlet and Vodka score 39, Grass Wonder and El Condor Pasa score 38.
Not every assignment lands perfectly. Daitaku Helios and Tap Dance City score just 13, and North Flight and Aston Machan also sit at 13, the lowest confirmed pairs with data. Admire Vega and Curren Chan score 14 despite being in the same hobby group. The algorithm, whatever it is, has its off days.
References
Similarity-attraction effect:
- Byrne, D. — The Attraction Paradigm. Academic Press, 1971.
- Sprecher, S. & Regan, P.C. — Liking Some Things (in Some People) More than Others: Partner Preferences in Romantic Relationships and Friendships. Journal of Social and Personal Relationships, 2002.
Hobby classification based on a synthesis of:
- Stebbins, R.A. — Serious Leisure Perspective. seriousleisure.net
- Mokhtarian, P.L. et al. — A Taxonomy of Leisure Activities: The Role of ICT. UC Davis Institute of Transportation Studies, 2006.
- Tinsley, H.E.A. & Eldredge, B.D. — Psychological Benefits of Leisure Participation: A Taxonomy of Leisure Activities Based on Their Need-Gratifying Properties. Journal of Counseling Psychology, 1995.
- An Inventory Study on the Categorization and Types of Recreational Activities. ResearchGate, 2017.
- leisurestacking.com — Hobbies and Leisure: 300+ Hobbies in 5 Core Categories.
Outlier methodology:
- Kruskal, W.H. — Relative Importance by Averaging Over Orderings. The American Statistician, 1987. (influential observation, leverage)
- Paulhus, D.L. — Measurement and Control of Response Bias. In Robinson et al. (eds.), Measures of Personality and Social Psychological Attitudes, 1991. (acquiescence bias)
- Penn State STAT 501 — Distinction Between Outliers & High Leverage Observations
- Wikipedia — Acquiescence bias