Experiment · Apple News · Views
Possessive named entity formula — does it drive more Apple News views than baseline?active
Type: formula comparison ·
Generated: 2026-04-08
Result: Not statistically significant (p=0.072). Observed lift: 1.39× — but could be noise at current sample sizes (n=1,573 before, n=296 after).
Hypothesis: Non-Featured articles with possessive named entity framing (e.g. 'Savannah Guthrie's husband') have higher median views than untagged baseline headlines
The main site reports 1.94× lift for possessive named entity (n=75, not yet significant).
This runs the formal significance test with the current dataset and documents the result.
When the Apple News dataset grows month-over-month, re-run this experiment to track
whether significance is reached at n≥100. The formula classifier matches headlines that
contain a full named person (two capitalized words) plus a possessive construction.
Comparison
Statistics
| Group | n | Median views | Lift | p-value | Significant? |
|---|
| A — Untagged | 1,573 | 1,692 | — | 0.0725 | No |
| B — Possessive Named Entity | 296 | 2,344 | 1.39× |
Test: Mann-Whitney U (non-parametric).
Platform: apple_news.
α = 0.05.