IMDb Ratings Are Weighted, Not Averaged, So a Flood of 10s and 1s Moves Them Less Than You Think

"The rating is just the average of the votes." Most readers carry that sentence into a title page, and IMDb's own help pages correct it in plain language: the figure on a title page is a weighted average, not the arithmetic mean.

IMDb accepts every rating from its registered users — one rating per user per title, on a scale of 1 to 10 — and builds a single displayed figure from all of them. The figure is a weighted average: ratings carry different weights, so two votes of 8 don't necessarily move the number by the same amount. The plain arithmetic mean isn't suppressed; IMDb prints it, with the median, on the title's ratings breakdown page, one click below the headline figure. When voting activity turns unusual, IMDb applies a different weighting calculation, and it doesn't disclose the exact method, because a disclosed method could be gamed.

A ballot box of 100 votes shows what a flood does to a plain average

Run the mechanism small, on round numbers. A ballot box holds 100 votes: 50 votes of 8 and 50 votes of 6. The sum runs 50 times 8 — 400 — plus 50 times 6 — 300 — for 700 in total. Divide 700 by the 100 votes and the arithmetic mean is 7.0. The reader's sentence predicts exactly that figure, and on this box the sentence holds.

Now flood the box. A hundred new votes arrive, every one a 10, cast to push the title up. The box now holds 200 ballots and a sum of 1,700. Divided by 200, the mean stands at 8.5. The flood supplies 50 percent of the ballots — 1 in every 2 — and drags the average up a point and a half, and not one of the original hundred voters changed a vote.

Weight the same ballots and the figure moves less. IMDb doesn't publish the weights it assigns; the exact method stays undisclosed by design, so the arithmetic here runs on a stand-in: count each flood vote at one-quarter of a full vote, a weight chosen because the division stays visible. The honest hundred still contribute 700 across 100 counted votes. The flood contributes 100 times 10 times one-quarter — 250 — across 25 counted votes. The weighted sum is 950, the counted votes are 125, and 950 divided by 125 is 7.6. The flood cast 1 in every 2 ballots; under the stand-in weight it supplies 25 of the 125 counted votes — 1 in every 5, or 20 percent. The movement shrinks from a point and a half to six tenths of a point.

The flood running the other way lands softer as well. A hundred votes of 1 against the same honest box pull the raw mean from 7.0 to 4.0 — a sum of 800 divided by 200 ballots, a fall of three full points. Weighted at the quarter, the same flood pulls the figure to 5.8: 725 divided by 125, a fall of 1.2 points. A weighting doesn't refuse the flood. It refuses the flood at full strength.

The documented system runs on larger numbers, and one of its formulas is public

The real electorate is bigger, and IMDb documents its dimensions. A title displays no rating at all until it has collected five ratings. Below a hundred ratings, the FAQ warns, a title is "particularly susceptible to substantial changes in the weighted rating" — at five ratings, one ballot is 1 of every 5 cast, 20 percent of the electorate; at a hundred, the same ballot is 1 of 100. The weighted figure isn't recalculated when a vote lands; IMDb updates it numerous times per day, so a rating cast at noon may not move anything until a later pass. The help pages name the scheme's ingredients without measuring them: a trimmed mean "to reduce extreme influences" and, in IMDb's words, a "complex voter weighting system" that keeps the figure representative of the general voting population.

For its Top 250 charts, IMDb prints the whole formula — a Bayesian estimate, in its own help pages, in full: weighted rating = (v ÷ (v+m)) × R + (m ÷ (v+m)) × C. In the formula, R is the title's own mean, v is its number of ratings, m is the minimum required to be listed — currently 25,000 — and C is the mean rating across the whole report. Run the shares. At exactly 25,000 ratings, v ÷ (v+m) is 25,000 divided by 50,000: one half. Half the chart figure comes from the title's own voters and half from the average title in the report — 1 of every 2 parts is borrowed. At 225,000 ratings the share is 225,000 divided by 250,000, nine tenths: nine of every ten parts are the title's own. At 2,475,000 ratings the share is 99 of every 100 parts. A title holding a mean of 9.0 at the threshold enters the chart arithmetic as 4.5 plus half of C. IMDb doesn't print C's current value beside the formula; at a report mean of 7.0, the chart figure would stand at 8.0 — a full point of the 9.0 held back by the pool until the title's own electorate grows.

IMDb discloses the fact of the weighting and withholds the weights themselves

The withheld half of the system is documented as carefully as the published half. No help page prints a table of weights: which accounts count at full strength, which count at a fraction, and what conduct moves an account from one class to the other all stay inside the company. The trigger for the alternate calculation — the point at which voting activity becomes "unusual" — is never defined. The Top 250 charts count only ratings from users IMDb classifies as regular, and the criteria for that classification are also undisclosed, a second electorate inside the first whose membership rules no reader can inspect. Weights already assigned can change: the FAQ records that a displayed rating can move with no new ratings cast, because the weight on existing ratings shifted, and that the formulas themselves are tweaked regularly. Any account of the internal weights more precise than IMDb's own help pages is an invention, whoever prints it.

Three checks separate the displayed figure from the votes underneath it

The next rating can be tested instead of trusted, and the tests run on numbers IMDb itself prints. First, read the denominator: the vote total sits directly under the displayed figure, and below a hundred ratings the number is a draft, not a settled measurement. Second, open the breakdown page and set the two averages side by side — the arithmetic mean IMDb prints there against the weighted figure on the title page. The distance between them is the weighting, measurable on the page, no disclosure required. Third, read the shape of the distribution the breakdown displays: a ballot box stacked at 10 and at 1 with a thin middle records a polarized electorate, and the weighted figure will sit closer to that electorate's broad middle than the raw mean does. The tests can't grade the film. They identify the instrument — and an average built to absorb a flood is a different instrument from an average that merely counts.