Method
What predicts how critics will review a film? We tested the cast and the director, using only information from before each film.
What we learned
- Who makes a film matters more than who is in it. A director's earlier record tracks a new film's score at 0.56. The cast's earlier record tracks it at 0.06.
- A track record is a lean, not a verdict. Even the best guess is still about 15.8 points off. Streaks do carry over: after a film well above a director's average, the next one averages 9 points above it, and after a weak one, 7.8 below. Individual films still vary widely.
- Sequels tend to score lower. On average 4.9 points lower than the film before, and 60% of them scored lower.
- Critics and audiences mostly agree, and differ in a pattern. Their scores track at 0.72. Critics run kinder on documentary (+14 points) and audiences kinder on horror (-7.1 points).
- Good reviews and big money are mostly separate stories. Among wide releases, critic score and box office barely relate (0.04).
- Several popular hunches show little. Runtime and score: 0.07. Directors improving with experience: first films average 62.9, sixth and later 61.4.
Why it matters
For choosing what to watch, the director is a better clue than the cast. For anyone trying to predict an outcome, the more useful question is who is doing the work, not who is the visible face of it. And it is worth testing a hunch before relying on it, since stars selling reviews, longer films scoring higher and directors improving with experience all came out weak or absent in this data. These are correlations in public data, not proof of cause, and the limits are listed below.
Directors: a clear pattern
For each of 3,273 films (2000 to 2025) whose director had at least three earlier scored films, we took the average critic score of the director's earlier films and compared it with the new film's score. The two track closely (correlation 0.56, range 0.53 to 0.58). Guessing from the director's record was off by 15.8 points on average; guessing the overall average was off by 20.
Five equal-sized groups of films, split by the director's earlier average. Directors with weaker records tend to make films that score lower, and the step up is steady.
Part of this is the type of film, since some genres score higher than others. Comparing each film with its genre's usual score cuts the correlation from 0.56 to 0.47, so genre explains some of it but not most.
Cast: almost nothing
We ran the same kind of test on the cast. For each of 832 films with at least one of 30 well-known actors, we predicted the score from how critics scored those actors' earlier films. The prediction was off by 16.6 points on average. Always guessing the average of earlier films was off by 16.3. The correlation was 0.06.
Because those two tests used different films, we also checked the director measure on only the films that feature our 30 actors (259 films). It still correlates at 0.37 (range 0.26 to 0.47), against 0.06 for the cast.
Each dot is one film. If the cast's history predicted the score, the dots would hug the dotted line.
More questions
We asked a set of other questions of the same data. Each one is reported whether or not it turned up anything: some show a clear pattern and some show very little.
Critics versus audiences
Do critics and audiences agree, and where do they split?
Across 471 films, critic score (Rotten Tomatoes and Metacritic) and the IMDb audience rating (scaled to 100) correlate at 0.72 (0.67 to 0.76). On average critics score 6.4 points higher than audiences. The gap by genre, where positive means critics are kinder than audiences:
Limits: IMDb ratings come from a self-selected group of voters and are on a different scale from critic scores, so the gap has no single 'correct' size. The ranking by genre is more meaningful than the average. Only films with both scores on Wikidata are included.
Sequels versus their predecessors
Do sequels score lower than the film before them?
Among 611 films whose predecessor is also in our data, the sequel scored 4.9 points lower on average (range -6.5 to -3.3). 60% of sequels scored lower than the film before.
Limits: Only sequels that Wikidata links to a predecessor with a score. Franchises, remakes and series entries may be linked differently.
Runtime
Do longer films score higher?
Across 14,758 films, runtime and critic score correlate at 0.07 (0.05 to 0.08). Average score by runtime:
Limits: Runtime may reflect genre and ambition (epics and dramas run long), not length itself. Documentaries and shorts-like features run short and score differently.
Release month
Do films released in some months score better?
Across 9,234 films, Jan releases score highest (66.2) and Oct lowest (61), a spread of 5.2 points. Average score by month:
Limits: Films whose release date on Wikidata is only a year (stored as January 1) are left out, since their month is unknown. The mix of films differs by month (festival and awards releases versus wide commercial releases), so this describes who releases when, not that timing changes a film's quality. Release dates use each film's earliest recorded date, which can be a festival premiere.
Scores over time
Are films getting better reviewed?
The average critic score of films in our data was 55.3 in 2000-2004 and 68.5 in 2021-2025. Average score by release year:
Limits: This does not show films are getting better. The set of films that have a score on Wikidata changes over time, critics' scales may drift, and recent years include films that have fewer reviews so far.
Director streaks
After a director's unusually good or bad film, does the next one follow or snap back?
We looked at directors with at least three earlier films. After a film that scored 15 or more points above their earlier average (676 cases), the next film averaged 9 points above that earlier average. After a film 15 or more points below it (818 cases), the next film averaged 7.8 points below.
Limits: Directors with several scored films are a select group, and the 15-point cutoff is arbitrary.
Production company or director?
Is a production company's record just a proxy for its directors?
On 1,643 films with both records, the company's earlier average correlates with the film's score at 0.46. After removing what the director's record already explains, it still correlates at 0.24 (0.19 to 0.28).
Limits: Companies and directors are linked (studios hire directors they like), so this can't cleanly separate them.
Do directors get better?
Among directors with at least six scored films, how do scores change with experience?
Looking only at directors with six or more films in our data, their average score was 62.9 on a first film and 61.4 from the sixth film on. Average score by film number:
Limits: Directors who got to make six films are a survivor group, so this likely overstates how much anyone improves. Films before 2000 are not in our data, so 'first film' can really mean 'first scored film we have'.
Critic score and box office
Do better-reviewed films earn more?
Across 2,070 films with a recorded worldwide gross, critic score and gross (compared with other films from the same year) correlate at -0.12 (-0.16 to -0.07). Among the 1,635 films that grossed at least $10 million, the correlation is 0.04 (-0.01 to 0.09). Median gross by score group, in millions of dollars (all 2,070 films):
Limits: Gross is not adjusted for inflation or budget, and only films with a recorded gross over $1 million are included. Many small, highly rated films just clear that bar, which pulls the overall correlation down; the wide-release figure above is the fairer comparison. An earlier check on a smaller set of films with well-known actors found a small positive relationship, so the direction depends on which films are included. Treat this as unsettled.
Combining the crew signals
Does combining director, company and cinematographer records beat the director alone?
Trained on films through 2017 (773 films) and tested on later films (463), the director alone missed by 15.4 points, adding the production company brought it to 14.8, and adding the cinematographer changed little (14.8). Guessing the average missed by 18.6.
Limits: Only films with all three records are included, which favors well-documented films.
Data
Film casts, release dates and critic scores (Rotten Tomatoes percentage and Metacritic score, averaged when both exist) come from Wikidata, which is released under CC0. Only films with a score on record are included.
Limits
- This is correlation. A director's record and a film's score may both reflect the kind of film, the studio, or who gets hired for what.
- Only films with a score and a director on Wikidata are included, and directors need three earlier scored films, which favors established ones.
- Budget would be a natural "production quality" measure, but only 85 of our films had one on record, so we can't say anything about it.
- Cast lists on Wikidata don't mark who had the lead, so bit parts and cameos are counted too.
- Wikidata's coverage is uneven. Famous films are more likely to have scores.
- Thirty actors is a small, famous sample. Results could differ for others.
- This is critic scores only. It says nothing about box office or audience scores.
- A null result here doesn't prove casting never matters, only that this simple measure didn't predict reviews.
Try it yourself
The guess-the-score quiz uses the cast-based guess as its opponent.