The Join Is Dirtier Than the Data

One capital letter manufactures a finding

Villagers Huck, Monty, and Paolo favorite “To The Edge”; the music table spells it “To the Edge”. Joins and GROUP BY match bytes, so the unrepaired join silently drops all three villagers and reports the song as favorited by nobody, which is false. No error is raised anywhere; the output is simply wrong with confidence. After casefolding the key on both sides, the join is total: 0 unmatched rows.

The honest “favorited by nobody” list, repaired and decomposed, is three populations, not one:

The unrepaired list has ten entries; the tenth ("To the Edge", 3 fans) is an artifact of the case mismatch. Favorites are designer assignments, not villager opinions, so "unloved" describes an assignment table, not taste.

Song favorites are mildly concentrated

Forest Life leads with 15 of 391 villagers (3.8%); the top five songs hold 57 of 391 (14.6%), and 89 of the 95 real songs have at least one fan. The denominator is 95, not 98: three rows of music.csv are placeholders.

Personality determines gender, 391 of 391

Every personality is single-gender: Cranky, Jock, Lazy, and Smug villagers are all male (204); Big Sister, Normal, Peppy, and Snooty are all female (187). The chart needs no overlap because none exists. Any gender comparison on this data is a personality comparison wearing a different label, so no finding here is phrased as a gender effect.

Related routing, stated with its confound: 36 of 55 Jocks (65%) have the Fitness hobby; 63 of 104 Peppy-plus-Snooty villagers (61%) have Fashion. Personality or gender, the data cannot say which.

Absence carries the design

391 villagers occupy 361 of 366 possible calendar days; 30 days carry exactly two villagers, none carries three, and five days carry none: 29-Feb, 1-Apr, 15-Apr, 31-Oct, 24-Dec. Four map to game events or rarities (leap day, April Fools, Halloween, Toy Day eve); 15-Apr stays an open question rather than a forced story. Nothing in the documentation says event days get no birthdays; the missing rows say it.

Middle months (1.00–1.13) omitted from the chart for legibility; the full twelve are in the notebook. The spread is a designed near-uniformity: the denominator (days in month) matters more than the count.

Catalog-wide: the types lie before the values do

The same sweeps across all 30 tables: every file opens with a UTF-8 BOM (an export convention, not an accident), and the Buy column mixes integers with the sentinel string “NFS” in 22 of 25 priced tables (6,498 cells), forcing the whole column to load as text. The three tables where Buy loads numeric are clean only because nothing in them happens to be not-for-sale: same pipeline, different rows, different column type.

Study notes: is this data biased?

This case is a clean worked example for the standard bias taxonomy, mostly because of which biases it cannot have. Sampling bias is structurally impossible: the rows are the entire villager roster, a census, and "the sample misrepresents the population" cannot arise when population = dataset. Observer bias is absent: the values were datamined from game files, not measured by people. What remains lives on the analyst's side of the desk.

What the file has instead of bias: a currency limit (a census of mid-2020; later-added villagers are absent) and a faithfully recorded biased generator: the personality-gender partition and the unequal quotas are design decisions the data records accurately. Unbiased file, biased world. The transferable rule: "unbiased" is a property of a question-file pair, not of a file. This data answers "what is the roster's composition?" cleanly and cannot answer "do male villagers prefer fitness?" at all.

Study notes: the data types on display

All 30 tables are structured data; inside them, nearly every textbook type appears, and the two that misbehave here misbehave everywhere.

The two live traps: a sentinel string in a numeric column ("NFS" in Buy) changes the column's type only in the tables where the sentinel occurs, and Booleans-as-words fail equality tests against real booleans. Both are invisible until something downstream computes.

What we cannot conclude

⚠ limits

Villager popularity: this catalog has no popularity signal; rankings elsewhere come from scraped third-party wish-list data. Gender-vs-personality mechanisms: perfectly confounded by construction. Why 15-Apr is birthday-free. And whether the four unloved songs are disliked: favorites are designer assignments, not opinions. The dataset is a census of the roster as of mid-2020; villagers added in later game updates are absent.