How Our Scores Work
Every film on MovieMagicScore carries a confidence badge. This page explains what those badges mean and why they matter.
The problem with simple averages
Imagine two films. Film A has 50,000 ratings and averages 7.8. Film B has 3 ratings and averages 9.5. Which is the better film?
A simple average says Film B. Your gut says Film A. The problem: small samples are noisy. Three friends could rate anything a 10. With 50,000 people, you get a real picture of how a film lands.
Most platforms just show the average. We think that is not enough. You deserve to know how much to trust the number.
What Bayesian shrinkage does
We use an approach called Bayesian shrinkage (the same method IMDb uses for its Top 250). The idea: pull every film toward the global average, proportional to how few ratings it has.
score = (n / (n + C)) × R + (C / (n + C)) × m
n = number of ratings, R = raw average, C = prior weight (500), m = global mean (~6.0)
A film with 50,000 ratings? Its Bayesian score is almost identical to the raw average. A film with 5 ratings? Its score gets pulled heavily toward the global mean. This prevents unknown films with a handful of perfect ratings from outranking well-tested favorites.
Why we report confidence, not just a number
A score is only useful if you know how stable it is. We compute a 95% confidence interval for every film: the range within which the true score is likely to fall.
A blockbuster with 30,000+ ratings might have a confidence interval of ±0.02. That number is rock-solid. An indie film with 200 ratings might have ±0.15. Still useful, but expect some movement as new ratings arrive.
The three tiers
High confidence
5,000+ ratings and a confidence interval of ±0.10 or less. These scores are extremely stable.
Moderate confidence
200+ ratings and a confidence interval of ±0.30 or less. The score is reasonably stable but may shift with new ratings.
Low confidence
Fewer than 200 ratings or high uncertainty. Treat this score as a preliminary signal, not a verdict.
Honest disclaimers
No scoring system is perfect. Ours assumes that rating populations are roughly representative, which is not always true. Films that attract passionate niche audiences may have inflated scores. Vote brigading, while dampened by Bayesian shrinkage, is not eliminated.
Scores update as new ratings come in. A film tagged "moderate confidence" today may reach "high" in a few months. We recompute the full catalog nightly.
Every film page has a confidence badge. Click it to see the full statistical breakdown for that specific film.