Overview
The filtered dataset at a glance.
Budget vs gross
Each dot is a film. Films above the dashed line earned more than they cost.
How to read this
Both axes are on a log scale, so equal distances mean equal ratios: each gridline is ten times the one before. The dashed line is break-even (gross equals budget). Gross is US box office only, so many films below the line still made money worldwide.
Rating distribution
Bins of 0.2 rating points.
How to read this
Ratings are recorded to one decimal place, so each bar holds exactly two possible values. The dashed line marks the mean.
Movies per genre
Movies per year
Mean rating by year
Line: mean rating. Band: confidence interval. Years with fewer than 10 films are left out.
How to read this
Older years in this dataset contain only a handful of well-known films, which is why they are excluded: their averages would reflect survivorship rather than the era.
By genre
Mean, spread and confidence intervals of IMDb ratings by genre.
Settings
Mean rating with confidence interval
Dots are genre means; whiskers are t-based CIs; the dashed line is the mean across all films shown.
How to read this
A CI that does not cross the dashed line suggests the genre differs from the average film. Overlapping whiskers between two genres do not prove there is no difference: test the difference directly on the Hypothesis tests tab. Genres with few films have wide intervals.
Rating distributions
Box: median and quartiles; dashed line in the box: mean; dots: every film. Ordered by median.
Average rating by genre and decade
Blank cells have fewer than 10 films.
Summary statistics
Explorer
Pick two genres, directors or decades and compare them.
Selections
Rating distributions
Share of each selection's films in each 0.2-point rating bin, so groups of different size compare fairly.
Box plots
Median, quartiles and every film; dashed line: mean.
Budget vs gross
Log scales; dashed line is break-even.
Mean rating by decade
Error bars are confidence intervals; decades with fewer than 5 films are left out.
Movies in selection
Directors
Directors ranked by mean IMDb rating, with confidence intervals.
Settings
Leaderboard
Mean rating per director with confidence interval.
How to read this
With five or so films per director, the intervals are wide: many apparent differences in the ranking are within sampling noise. Raise the minimum to see the intervals narrow.
Director statistics
Quiz on CI
Ten practice questions on standard errors, confidence intervals and margins of error. Each one uses a fresh random sample, of films or of synthetic business data.
Practice set
Each set has 10 questions. The sample size n is picked at random from 36, 49, 64, 81 or 100, and the page shows n, the sample mean, the sample SD, the degrees of freedom and the t critical value.
Score
Show the sample
Coverage simulation
Draw 100 samples from one genre and build a confidence interval from each. About the chosen percentage should contain the true mean.
How to read this
Each horizontal line is one sample's interval. Red lines miss the population mean. Run it a few times: the miss rate bounces around 100 minus the confidence level. Switching to z makes the intervals slightly too narrow, so coverage drops a little; with n of 30 or more the effect is small.
Hypothesis tests
Is a difference in mean rating larger than sampling noise?
Two genres: Welch t-test
H0: the two genres have the same mean rating.
How the numbers are worked out
How to read this
The curve is the t distribution the test statistic would follow if H0 were true. The red areas are the rejection region at the chosen confidence level; the blue area beyond the observed t is the p-value. Welch's version does not assume equal variances.
All genres: one-way ANOVA and Tukey HSD
H0: every genre has the same mean rating.
How to read this
ANOVA tests whether any genre differs. Tukey's HSD then compares every pair while keeping the overall false-positive rate at 5%, so its intervals are wider than separate t-tests. Eta squared is the share of rating variance explained by genre.
Regression
Fit a line, read the coefficients, check the residuals.
Model
Scatter and fitted line
Uses the first predictor; the line is the simple regression fit.
Residuals vs fitted
A shapeless cloud around zero is what you want to see.
Coefficients
How to read this
Each coefficient is the expected change in y for a one-unit change in that x, holding the other predictors fixed. With log10 variables, one unit means ten times as much. Genre effects are measured relative to the first genre alphabetically. Correlation is not causation: votes, for example, are partly a result of a film being good.
Data
The filtered dataset. Sort, filter by column, or download.