Raw counts mislead whenever the groups are different sizes. Here's a small, real example from villa rental data: "which island has the most villas with a private chef?" is easy to answer, but it's the wrong question for most travellers.
The data
These are the published counts from the Gladori catalogue of 782 Caribbean villas as of mid-November 2026: total homes per destination, plus the top three destinations in each collection.
import pandas as pd
homes = pd.Series({
"St. Barts": 313, "Barbados": 160, "St. Martin": 108,
"Turks and Caicos": 78, "Dominican Republic": 66,
"Jamaica": 47, "Anguilla": 10,
})
assert homes.sum() == 782
collections = pd.DataFrame([
("chef", "Barbados", 106), ("chef", "Jamaica", 44), ("chef", "Dominican Republic", 35),
("staffed", "Barbados", 91), ("staffed", "Dominican Republic", 50), ("staffed", "Turks and Caicos", 48),
("beachfront", "Turks and Caicos", 27), ("beachfront", "Barbados", 28), ("beachfront", "St. Martin", 18),
("large_group", "Barbados", 54), ("large_group", "St. Barts", 42), ("large_group", "Turks and Caicos", 35),
], columns=["collection", "island", "n"])
Counts versus shares
collections["share"] = collections["n"] / collections["island"].map(homes)
print(collections.sort_values(["collection", "share"], ascending=[True, False])
.assign(share=lambda d: (d.share * 100).round(1)))
| collection | island | n | share % |
|---|---|---|---|
| beachfront | Turks and Caicos | 27 | 34.6 |
| beachfront | Barbados | 28 | 17.5 |
| beachfront | St. Martin | 18 | 16.7 |
| chef | Jamaica | 44 | 93.6 |
| chef | Barbados | 106 | 66.2 |
| chef | Dominican Republic | 35 | 53.0 |
| large_group | Turks and Caicos | 35 | 44.9 |
| large_group | Barbados | 54 | 33.8 |
| large_group | St. Barts | 42 | 13.4 |
| staffed | Dominican Republic | 50 | 75.8 |
| staffed | Barbados | 91 | 56.9 |
| staffed | Turks and Caicos | 48 | 61.5 |
What flips
- Chef: Barbados has the most chef villas, 106. But in Jamaica, 44 of 47 villas list a chef or cook, 93.6 %. If a cook is non-negotiable, Jamaica is the island where you barely have to filter.
- Beachfront: Barbados has one more beachfront villa than Turks and Caicos, 28 vs 27, yet a beachfront house is twice as likely in Turks and Caicos: 34.6 % against 17.5 %.
- Large groups: St. Barts has 42 villas with six bedrooms or more, but that's only 13.4 % of a very big catalogue. In Turks and Caicos it's 44.9 %.
Caveats
- Only the top three islands per collection are published, so shares for the others are unknown, not zero.
- A share says how likely a random listing on that island is to match. It says nothing about price or quality.
- The denominators move as homes are added or removed, so rerun before quoting.
The pattern is general. Whenever someone ranks by raw count across groups of different sizes, divide first.
Top comments (0)