The field: how few shows the sellouts belong to

Sellouts are concentrated in very few shows

Selling out is a property of a small set of shows, not something that visits the programme evenly.

59.5% of shows

1,959 of 3,292 have not sold out a single date.

3.0% of shows

produce 41.3% of all sellouts — the top 100 titles.

6.0% of shows

sold out every date that has resolved. A full-run sellout is rarer than a hit.

37.6% of dates

finished under a quarter full; 15.9% finished completely full. The middle is thin — the distribution is barbelled, not bell-shaped.

Across the window that works out at roughly 1.83 million seats sold against 1.79 million empty ones, and about £26.9m of tickets at the cheapest listed price — a floor on gross, not an estimate of it. Almost exactly half of the seats the festival put on sale were used.

The pooled 41.3% is close enough to comedy’s own figure to read as a festival-wide fact, and it is not one. Rank each genre’s own catalogue and theatre turns out to be far more top-heavy than comedy:

Everything on this page explains a seventh of it

Venue, room, price, slot, date, genre and origin together account for about a seventh of how well a show sells. The rest is the act — and that remainder is measurable.

Fit an expectation from everything this page has spent eight chapters on — the operator, the room and its size, the price, the hour, the day, the calendar date, the genre and sub-genre, the country, the running time, whether there was an offer, which box office sold it — and tell it nothing about who is performing. Scored out of fold, on shows it never saw, that model reaches an R² of 0.149 against how deep a show sold into its allocation, and ranks the binary sellout at AUC 0.701.

That is not a weak model. It is the size of the thing this page can talk about. Structure explains roughly one part in seven of how well a show sells; the other six are the show — the act, the material, the reviews, the word of mouth, the luck. Every finding above is a real effect inside that seventh, which is why the largest of them move a rate by tens of points and none of them tell you what to book.

The six-sevenths is not noise, though, and that is the more useful half. Take what a show sold minus what the model expected of it — the part with no explanation attached — split each show’s own dates at random into two halves, and the two halves agree: r = 0.900 in comedy and 0.922 in theatre, against a ceiling of 0.930 for the raw sold-down figure with no model subtracted at all. Venue adjustment costs at most three hundredths of that. Split a run chronologically instead — first half against second, so festival phase, press and word of mouth all move underneath it — and it still holds at 0.837.

So a show’s edge over its room is a stable property of the show, readable off half a run. Rank shows on half their dates, take the top twenty, and on the dates they were not picked on they run 60 points of allocation ahead of everyone else, with 19 of the 20 landing in the held-out top decile where chance would put 2. The acts that beat their room are findable in advance, not only in hindsight.

In full

The material the scroll has no room for.

The section above describes this distribution without naming anyone in it, which is the right call for a scroll and an odd one for a reference page. Here is its head. Ranked on raw sold-out dates, the top 25 listings account for 1,684 of the festival’s 6,597 sold-out dates — 25.5% of everything, from a handful of listings.

That table is mostly a measurement of format. An installation selling twenty-minute slots banks more sold-out dates in a week than a play can in a month, and nothing about that is a demand signal. Rank the same population with the timed-entry listings dropped and the head of the distribution changes completely — the top 25 now carry 492 sellouts, 7.5% of the festival’s rather than 25.5%.

Method, in full

What this chapter measures, what it excludes, and where it is weakest.

A per-performance rate treats every date as one vote, which is the right unit for "will I get in on Thursday" and the wrong one for "was this show a hit". The concentration section publishes both a per-date and a per-show view for that reason, and the two tables above are the per-date view at its most extreme. Neither is corrected for how long a listing ran.

The exposure these tables quantify is stated as a bound in the analysis appendix rather than fixed, because there is no defensible correction factor: excluding a listing by its shape rather than by its genre is a judgement, and the honest handling is to publish the headline both ways and let the size of the gap be the finding. Nothing elsewhere in the analysis has been restated with these listings removed.

The population, the definition of “sold out” and the two bounds every rate sits between are the same across every page here, and are set out once in how this was measured and what these numbers cannot tell you.