21.5 Correlation vs causation

उदाहरण A — Basket size vs minutes (fictional Pune).

import pandas as pd

sample = pd.DataFrame({
    "order_id": [f"BLK-PUN-{i:03d}" for i in range(1, 11)],
    "items": [2, 3, 3, 4, 5, 5, 6, 8, 9, 12],
    "delivery_mins": [9, 10, 11, 12, 13, 14, 15, 18, 20, 28],
    "evening": [0, 0, 1, 0, 1, 1, 1, 1, 1, 1],
})
print("corr items vs mins", round(sample["items"].corr(sample["delivery_mins"]), 2))
print("corr evening vs mins", round(sample["evening"].corr(sample["delivery_mins"]), 2))

Both correlations can look “real”. Evening rush may drive minutes more than item count. Ravindra Bagale’s note to the team: “Association seen; cause not claimed.”

उदाहरण B — City averages are not personal blame.

city_avg = pd.DataFrame({
    "city": ["Pune", "Solapur", "Nashik", "Kolhapur"],
    "avg_mins": [14.2, 11.0, 13.5, 10.8],
    "rain_days": [4, 1, 3, 0],
})
print(city_avg["avg_mins"].corr(city_avg["rain_days"]))

A positive corr across four cities is a weak classroom toy — too few points. Never publish “rain causes delays” from four rows.

उदाहरण C — Safe wording bank.