32,540 Horror Movies and the October Effect (Live TMDB Data)
Analyze This! is our series of live QuantumLayers demonstrations. We take a public dataset, load it straight into QuantumLayers, and let the platform chart it and test it. Nothing on this page is drawn by hand or typed in from a spreadsheet. This time the data is a public catalogue of horror films from The Movie Database (TMDB), and it is Halloween, so we could hardly pick anything else.
The dataset is TMDB’s list of every film tagged as horror and released between 1950 and late 2022, prepared by the TidyTuesday community project. We imported the CSV into QuantumLayers by URL and marked the dataset public. That one setting is what makes the charts below possible: each one is fetched live from QuantumLayers when you open this page, and each has a button that opens it as a standalone ql-chart you can explore yourself.
We have organized what we found as five scenes. Lights off.
SCENE 1The October effect
Plot horror releases month by month and a heartbeat appears. Every single October from 2012 to 2021, the genre has its biggest month of the year, and it is not close. October 2020 holds the record with 455 releases, roughly three times what a normal month that year produced. In a typical year the October spike lands at about double the surrounding months.
The orange dots mark each October. Two smaller details are worth a look. January is often the second highest month, which is likely a data artifact: films whose exact release date is unknown tend to get stamped with January 1. And the line falls off a cliff at the end of 2022 because the catalogue was captured around the start of November 2022, so late releases simply had not been logged yet.
Figure 1: Horror films released per month, January 2012 to December 2022, fetched live from QuantumLayers. Orange points are Octobers. Source: TMDB via TidyTuesday.
▶ Open the live ql-chart on QuantumLayers
SCENE 2Seventy-two years of fright, in three acts
Zoom out to one point per year and the history of the genre tells itself. Act one is the slow build: just 3 films in 1950, creeping up through the drive-in era to a little over 170 a year by the early 1970s. Act two is the home video boom, peaking at 322 films in 1988, followed by a long 1990s hangover that bottoms out at 134 in 1994. Act three starts around 2004, when releases climb almost every year as cheaper digital production and online distribution opened the door to independent filmmakers. By 2022 the catalogue counts 1,020 horror films in a single year, more than three new ones every day.
Figure 2: Horror films released per year, 1950 to 2022, fetched live. Source: TMDB via TidyTuesday.
▶ Open the live ql-chart on QuantumLayers
SCENE 3Fear has a first language
Horror is made in 97 languages in this catalogue, but English dominates, accounting for roughly 7 in every 10 films. Japanese and Spanish are next, at about 5% each, followed by Italian (home of the giallo), French, Cantonese and German. Everything else, from Thai ghost stories to Indonesian folk horror, shares the remaining slice. Keep in mind that TMDB is an English-led, community-edited database, so English titles are likely over-represented compared with what is actually produced worldwide.
Figure 3: Share of horror films by original language, top seven plus Other, fetched live. en = English, ja = Japanese, es = Spanish, it = Italian, fr = French, cn = Cantonese, de = German. Source: TMDB via TidyTuesday.
▶ Open the live ql-chart on QuantumLayers
SCENE 4The audience verdict: mostly a shrug
Ratings on TMDB run from 0 to 10, and horror’s reviews cluster in a modest place. Among rated films the most common score is 5 to 5.5, about 43% score below 5, and only around 1 in 10 reaches 7 or higher. The tall bar on the far left is not a pile of terrible movies: it is roughly 15% of films that have no rating at all and show up as a zero.
Notice the odd little bump at 9.5 to 10 on the right. Those are not hidden masterpieces either. QuantumLayers’ summary shows the median film has just 6 votes, and a film with one or two enthusiastic voters (often, one suspects, the people who made it) can land a perfect score. A rating is only as trustworthy as the number of people behind it.
Figure 4: Distribution of average audience ratings (0 to 10), fetched live. A score of 0 means unrated. Source: TMDB via TidyTuesday.
▶ Open the live ql-chart on QuantumLayers
SCENE 5Follow the money (if you can find it)
This is where we handed the dataset to QuantumLayers’ automated statistical analysis, which profiles every numeric column, flags skew and outliers, measures correlations and runs analysis of variance across categories. Here is what came back.
Most horror films have no paper trail. The median budget is $0, and so is the median box office. In fact, more than three quarters of the catalogue reports neither number. That is not a genre of free movies: it is a genre of small productions that never publish their figures. The ones that do range all the way up to a $200 million budget and a $701.8 million worldwide gross, which is why QuantumLayers flagged 18% of budgets as statistical outliers.
Popularity is the most lopsided number we have charted in this series. TMDB’s popularity score has a median of 1.7 and a maximum of 5,088.6, giving it a skewness of 70.8. For comparison, meteorite masses in our NASA post came in at 56.9. A handful of current hits soak up nearly all the attention while tens of thousands of films sit in near silence.
Attention predicts earnings better than spending does. In QuantumLayers’ correlation matrix, box office revenue correlates with budget at r = 0.63, but even more strongly with the number of audience votes, at r = 0.78. Votes come after release, so this is not a recipe for a hit, but it does say that the films people talk about are the ones that earn, more than the ones that spend.
Significant is not the same as important. The analysis of variance found that a film’s original language has a highly significant effect on both its budget and its vote count (p < 0.001). Sounds like a headline, until you read the effect size: language explains less than 1% of the variation in either (η² ≈ 0.006 to 0.007). With 32,000 rows, even tiny differences pass a significance test. This is exactly why QuantumLayers reports effect sizes next to every p-value.
▶ Budget vs. revenue: open the live scatter on QuantumLayers
The fine print
Every dataset has its ghosts. In this one, a zero usually means “unknown” rather than “nothing”: zero budget, zero revenue, zero rating and zero runtime all mostly stand for missing values. About 15% of runtimes fall under half an hour, a mix of genuine short films and blanks, while the bulk of features run between about an hour and an hour and a half (see the runtime chart). The catalogue is also a snapshot from early November 2022, so its newest months are incomplete. None of this makes the data less useful; it just means reading it the way an analyst would, which is what the automated checks are for.
How this was built
- Import. The TidyTuesday CSV was loaded into QuantumLayers straight from its public URL. QuantumLayers detected and typed all 20 columns (dates, integers, floats and text) without any manual setup.
- Analyze. We ran the platform’s statistical summary, correlation matrix, analysis of variance and automated insights over the dataset, and used its recommended charts as a starting point.
- Publish. With the dataset set to public, each chart on this page is rendered with Chart.js from a single unauthenticated
POSTto QuantumLayers’ql_get_chart_dataendpoint. The purple buttons open the same charts on QuantumLayers’ hostedql-chartpage, built from nothing but URL parameters.
Want a different cut? Ratings by decade, runtime against popularity, the biggest franchises? The dataset is public, so open any chart above and start changing the parameters.
Data: The Movie Database (TMDB), compiled by the TidyTuesday project (November 2022). This product uses the TMDB API but is not endorsed or certified by TMDB. Counts reflect the dataset at request time.
This post is part of Analyze This!, a series of live QuantumLayers demonstrations. If horror is not your thing, try 10,877 earthquakes in 30 days or 45,716 meteorite landings. To load a public dataset of your own and get the same charts and findings, start building at quantumlayers.com.
