Blog · Category
Real datasets analyzed end to end — from messy import to a defensible conclusion.
344 real penguins, a bill correlation that runs backwards until you split by species, then a decision tree, random forest and PCA that classify three species at ~98% — no code.
Dec 1, 2026 · 11 min read
Turn the most famous scatter in statistics into 142 connected trajectories and watch the development gap close — a full Gapminder analysis, no code, no animation.
Jan 4, 2027 · 14 min read
The 2024 General Social Survey opened straight from its .sav and weighted with NORC's survey weight: chi-square, a mosaic, Welch ANOVA and logistic regression show more income doesn't raise the odds of being very happy once money feelings, health and marriage are in. No code.
Feb 1, 2027 · 10 min read
5,479 real days of US births: Friday the 13th looks like it has 5% more babies, until ANOVA, Welch and Games–Howell expose the weekday and holiday effects. Fridays vs Fridays: 5.3% fewer. No code.
Mar 1, 2027 · 12 min read
440 real wholesale customers: raw k-means finds only the whales, a log scale finds two real segments that k-means and Ward agree on, and a label the algorithm never saw validates them at V ≈ 0.72. No code.
Apr 1, 2027 · 11 min read
4,090,836 real NYC taxi rides, opened straight from Parquet: three recording traps that fake a 38% no-tip rate, a 20% default half the city taps, and a logistic regression and random forest that agree on the one thing that moves it — no code.
May 3, 2027 · 11 min read
Card & Krueger's 410 restaurants from a Stata .dta, reshaped to one row per store, then a difference-in-differences with standard errors clustered by store, checked against within-store change scores and the published tables. No code.
Jun 1, 2027 · 10 min read
1,186 people, 14 characters, one chart. Correspondence Analysis maps a whole survey — and Jar Jar sits alone in the corner. No code.
Nov 2, 2026 · 10 min read
53,940 real diamonds, three quality grades that all run backwards, and the charts, ANOVA, regression, PCA and decision tree that crack it — no code.
Jul 31, 2026 · 13 min read
2,139 patients in the ACTG 175 HIV trial: the overall average said the drug worked, one arm said it harmed, and the honest answer lived in the subgroups — paired t-test, effect size, responder rate, ANOVA, MANOVA and ANCOVA. No code.
Sep 1, 2026 · 13 min read
A capability report says Ppk 1.36 — a pass. On 200 real piston-ring measurements, SPC control charts catch the drift that single number hides. The Six-Sigma rule: control before capability. No code.
Oct 1, 2026 · 6 min read
Stratum