Patrick Killeen

Projects Marvel by Phase

DataIn progress2026

Marvel by Phase

The MCU split into its phases and measured: budgets, box office, critic and audience scores, release cadence. Which phase was the golden age, which actors and writers made the studio the most money, and where the critics and the audience stopped agreeing.

Stack
R · ggplot2 · dplyr
A dot plot of the mean rating by MCU phase, across 37 films. Each pale dot is one film and the bar is the phase mean, on a 0-100 scale: Phase 1 scores 74.3, Phase 2 76.1, Phase 3 79.0, Phase 4 72.9, Phase 5 67.7, and Phase 6 73.4 from a single film. Phase 3 is the highest, and the per-film dots keep the small sample sizes visible.

The short version

I pulled 37 Marvel films into one table and measured them by phase: ratings from seven sources, production budgets, worldwide box office, and the gap between each release.

I have watched most of these in the cinema, and the widely held sense that Phase 3 was the good run holds up against the numbers. Phase 3 averages 79.0 out of 100 across its rating sources and $1,227M worldwide per film, and it leads on both. It is also the only full phase where critics scored the films higher than audiences did.

Two stacked dot plots sharing a phase axis, covering 37 MCU films. The top panel is the mean of seven rating sources by phase: Phase 1 scores 74.3, Phase 2 76.1, Phase 3 79.0, Phase 4 72.9, Phase 5 67.7, and Phase 6 73.4 from a single film. The bottom panel is mean worldwide box office by phase: Phase 1 $635M, Phase 2 $878M, Phase 3 $1,227M, Phase 4 $773M, Phase 5 $646M, Phase 6 $522M. Individual films appear as pale dots behind each phase bar, so the small sample sizes are visible. Phase 3 is the highest on both measures.

The honest caveat is that none of the box office is inflation-adjusted. Phase 1 is being compared in 2008 dollars against Phase 5 in 2025 dollars, and I did not apply an index.

The analysis repo is not public yet, so there is no code link on this page. Nothing is deployed either. The charts here are the current output, not a finished site.

How it works

The data began as two CSVs I made years ago and never put in version control: 34 films with 66 columns of ratings, budgets and box office, and a companion file with 332 columns of cast information. I copied both into a new repo and wrote four R scripts that carry them from raw to charts, so the fixes are reviewable rather than buried in a spreadsheet.

Six things were wrong.

Deadpool & Wolverine had been entered on a 0-10 scale for two of its seven rating columns. That dragged its average to 52.73 and made it read as the worst-received film Marvel has made. Corrected, it is 77.71. Guardians of the Galaxy Vol. 3 had a budget of 25,000,000 rather than 250,000,000. The franchise column called Wakanda Forever “Blank Panther”, splitting that franchise in two. The director column was scrambled for the last four films. Six columns of stored differences were empty on every row while a duplicate set held them.

The sixth was the largest. Nineteen of the 34 rows held another film’s cast. Thor’s row held Iron Man 3’s cast, and Doctor Strange’s held Black Panther’s. Four films had lost theirs entirely. The raw cast text on each row had never shifted, so I scored every film-against-block match against it. Four films I entered by hand.

I added the three films released since, and a writers column that did not exist in the original data. Both carry a source URL and a retrieval date of 9 September 2026.

Every rating on this page is a mean of seven sources rather than any one of them. That is not a hedge. Measured against a film’s own average, Rotten Tomatoes audiences sit almost ten points above it and Metacritic critics nine points below, a spread of nineteen points on the same film.

A bar chart of the seven rating sources, each measured as its mean distance from the film’s own seven-source average across 37 films. Bars run right for sources that rate above the average and left for those below. Metacritic critics -9.3, Letterboxd -7.8, Metacritic users -4.4, IMDb -2.0, Rotten Tomatoes critics +5.5, Google +8.8, Rotten Tomatoes audience +9.8. The spread between the highest and lowest source is 19.1 rating points, so the same film can score roughly 19 points higher on one source than another. The chart is the reason every rating elsewhere on the page is a mean of seven sources rather than any single one. Google covers 34 films rather than all 37, because the three 2025 releases have no retrievable Google score.

For the actor and writer rankings I split each film’s worldwide gross across its cast by billing weight, rather than crediting each of them with all of it. Robert Downey Jr leads either way, but the order under him moves: Danai Gurira is seventh unadjusted and nineteenth once an ensemble film counts only for a share.

A slope chart comparing two ways of ranking actors by the box office attributed to them, for everyone in the top 15 on either basis. The left column credits each actor with the full worldwide gross of every film they appear in. The right column splits each film’s gross across its cast by billing weight. Lines connect each actor’s two ranks, and those moving six places or more are drawn in a darker blue. Robert Downey Jr is first on both. The largest falls once the gross is split are Danai Gurira from 7 to 19, Jeremy Renner from 12 to 22, Jon Favreau from 9 to 18. All of them are ensemble-heavy careers, where crediting the whole gross of a team-up film flatters the ranking.

The charts are ggplot2, with every film drawn as its own point behind its phase average, because Phase 5 has six films and Phase 6 has one.

What didn’t work

The old data carried a billed_stars column I had built as a star-power score, correlating with box office at 0.63. It is a function of cast-list length and nothing else: eleven distinct values for eleven distinct cast sizes. It knew how many people were billed and nothing about who. Cast size replaces it as a plain column.

The misalignment sat in the file for two years. The cast was stored as 310 one-hot columns, and in that shape a nineteen-row shift looks like ordinary sparse data. The rebuilt output is one row per actor per film, where it would have been obvious.

What the data says

Phase 3 is the peak on ratings and on gross. The phases either side of it are where the more interesting questions sit.

Budgets climbed through the run while the return on them did not. A Phase 3 film turned its budget into 5.6 times itself. A Phase 5 film managed 3.1. Films also stopped selling for as long after they opened.

Three stacked dot plots sharing a phase axis, each with per-film dots behind the phase mean. The top panel is mean production budget: $160M, $198M, $212M, $207M, $213M, $200M for Phases 1 to 6. The middle panel is worldwide gross divided by budget: 3.7x, 4.5x, 5.6x, 3.8x, 3.1x, 2.6x, with a dashed break-even line at 1.0. The return peaks in Phase 3 and falls in every phase after it. The bottom panel is worldwide gross divided by opening weekend, a measure of how long a film kept selling: 5.9x, 7.8x, 7.7x, 5.3x, 5.2x, 4.4x. Budgets climbed while both the return on them and the length of a film’s run fell. Money is nominal USD and is not inflation adjusted, which flatters the later phases. Phase 6 is a single film.

Critics and audiences agreed least in Phase 5, where audiences scored films nearly eleven points higher on average. Phase 3 is the only full phase that runs the other way; Phase 6 does too, on the strength of a single film. The widest single gap in the data is Captain Marvel at twenty-three points, which is a review-bombing campaign rather than a split in reception, and it is flagged rather than dropped.

A dot plot of audience average minus critic average for each of 37 MCU films, grouped by phase, with a dashed zero line. Phase means are +3.3 for Phase 1, +4.9 for Phase 2, -2.1 for Phase 3, +5.3 for Phase 4, +10.7 for Phase 5 and -1.6 for Phase 6. Two phase means sit below zero, meaning critics scored those films higher than audiences did: Phase 3, and Phase 6, which is a single film rather than a phase average. Captain Marvel is labelled as the widest single gap at -23.0 rating points, and Black Panther at -15.5.

The release cadence explains some of what the ratings do. Phase 4 put out a film every sixty-three days at its median, the tightest run in the whole history, straight after the longest gap in it.

A timeline of 37 MCU films from 2008 to 2025. Each film is a dot on its release date, and its height is the number of days since the previous MCU release. Phases are shown as labelled shaded bands. Most dots sit between 60 and 300 days. One dot stands far above the rest: the 738-day gap between Spider-Man: Far From Home in July 2019 and Black Widow in July 2021, the COVID shutdown. Phase 4 has the tightest cadence at a median of 63 days between films.

The sequel curve does not decline the way the received account says. Third installments rate highest and fourth installments lowest, and only one of the seven franchises drawn here improves at every step. Installment order is also release order, so none of this separates a sequel effect from the period a film came out in.

Eight small line charts in a grid. Seven of them are one MCU franchise each, ordered by mean rating: Spider-Man, The Avengers, Guardians of the Galaxy, Iron Man, Captain America, Thor, Ant-Man. The eighth holds all 37 films as scattered dots. Each panel plots installment number within the franchise, 1 to 4, against the mean of seven rating sources on a 0 to 100 scale. The same mint reference line appears in every panel: the mean across all films at each installment, 75.4, 73.4, 78.0, 67.1. Third installments rate highest and fourth installments lowest. Of the seven franchises drawn, Spider-Man is the only one that rises at every step, from 80.0 to 80.3 to 86.9. Captain America falls hardest at its fourth film, Brave New World, at 52.3. Installment order is also release order, so the chart cannot separate a sequel effect from a period effect.

Splitting each film’s gross by billing weight also lets a rating be attributed the same way. The result is mostly a caution about reading the table too closely: nine of the thirty-four actors sit within a single point of each other at the top.

A dot plot split into two groups: the eight highest and eight lowest of the 34 actors with three or more MCU films, by the mean rating of their films weighted by their billing in each. A shaded band covers the range holding the 9 actors within one point of the highest, from 81.3 to 82.2, which is the point of the chart: the top of the table is a tie rather than a ranking. The highest eight run from Mark Ruffalo at 82.2 to Vin Diesel at 81.5. The lowest eight run from Elizabeth Olsen at 72.6 down to Brie Larson at 62.4. A dashed line marks the all-film mean of 74.6. The middle 18 actors are not drawn.

Only one film in thirty-seven failed to out-gross its own production budget. That is a lower bar than it sounds, because these budgets exclude marketing, which is commonly another half to a whole budget again.

A scatter plot of 37 MCU films, production budget on the horizontal axis against worldwide box office on the vertical. A dashed diagonal marks break-even against budget and a mint diagonal marks three times budget, the rougher industry rule for breaking even once marketing is counted. Only one film sits below the break-even diagonal: The Marvels, at 0.76 times its budget. The highest return is Spider-Man: No Way Home at 9.4 times. Avengers: Endgame is the largest bet and the largest gross, at a $400M budget and $2.8Bn worldwide. Budgets exclude marketing, so the dashed line is a floor rather than the profit line.

The longer films rate higher, at a correlation of 0.45. I do not think runtime is doing the work. A studio grants the extra running time to the films it already expects to carry the most, and those are the films an audience arrives invested in. Eternals is the third-longest film and one of the lowest rated.

A scatter plot of 37 MCU films, runtime in minutes against the mean of seven rating sources on a 0 to 100 scale, with a fitted straight line rising left to right. The correlation is 0.45: longer films tend to rate higher, though loosely. Avengers: Endgame is the longest at 181 minutes and among the highest rated. Eternals is the third longest at 157 minutes and one of the lowest rated, which is the clearest exception on the chart. Runtime has not crept upward across the years: against release year it correlates just 0.15.

The long version

This section is being written.