Credentials, press and reviews, in full

What a credential in the blurb is worth

Specific award names carry real signal. Generic “award-winning” carries almost none. Both survive controlling for venue, price and region — or fail to.

The pattern is that naming a specific award beats claiming one. “Edinburgh Comedy Award nominee” nearly doubles the base rate and keeps most of that after you account for the better venue and higher price such an act commands. “Award-winning”, pooled across 479 shows, barely moves.

Pooled is the wrong way to read that last row, but not because one genre carries it. The bare “award-winning” claim runs at ×1.47 in comedy and ×1.96 in theatre and both intervals clear 1 — 1.11 to 1.95, and 1.30 to 2.96. Theatre’s point estimate is the larger of the two; the ratio between them is 1.33 at p = 0.26, which is not a difference this many shows can establish. So: one effect, visible in both genres, larger-looking in theatre, and weaker in either than a named award. What the pooled “weakly” verdict adds is venue operator, region and price held fixed — controls the genre split below does not apply.

A separate question is whether the craft of the writing — not what it claims, but how it is put together — does anything once the room is taken away. The figures below are not the floor sellout rate used everywhere else on this page: they come from a fitted model of each show’s residual, how much deeper into its central-box-office allocation it sold than comparable shows at its own venue, holding room size, price, slot, day, genre, sub-genre, origin, duration and box-office system fixed, expressed in percentage points of allocation sold. A 4-point writing effect on that scale is not the same kind of number as the 38-point spread between venue operators above or the 28-point spread by price — it is a much smaller quantity, measured after the room is already accounted for, and the two should not be read side by side as if they shared a scale.

Of everything tested against that residual — sentence length, word count, readability, second person, opening with a question, opening with a quote, the rule of three, clichés, a book-now call to action — one thing replicates: sentence length. Blurbs in the longest-sentence fifth sell about 4 points less of their allocation than the shortest fifth, and it holds inside comedy and theatre taken separately.

Everything else tested is flat. Second person address (“you/your”, used in 976 of the 2,229 blurbs), opening with a question, opening with a quote, the rule of three, clichés and a book-now call to action all sit inside noise. Blurb length itself — word count, not sentence length — is flat too: a 15-word listing and a 127-word one land within 2.6 points of each other with no monotone shape, once the venue effect an earlier blurb-language pass picked up is controlled for.

One asymmetry is worth naming: printing a press quote is worth about 6 points of allocation in theatre (733 theatre shows) and nothing measurable in comedy (985 comedy shows). Theatre audiences appear to buy on the review; comedy audiences appear to buy on the act.

Which publication’s review is worth having

Listings advertise their press in a parseable shape — a quote, usually a star rating, then an attribution in brackets. Extracting all of it turns “do reviews help” into a question with different answers per masthead, and per genre.

National press buys breadth; the comedy trade buys the last hour. Three in four shows holding a national quote sell out at least one date and they sell out a third of their run, with nearly seven percent going completely clean — three to seven times any other group. A Chortle-quoted show is a different animal: an £11 ticket in a 56-seat room that goes with a median seven hours left on the clock, against 35 for a national-press show.

A review lands, and the tickets do not move

Everything above is blurb text — what a show advertises, and therefore the standing it already had. Reviews carry a date, so they can be tested the other way round: seats moving in the days after the first review against the days before. Nothing measurable happens, and the star count does not order.

For each show’s first professional review, the seats moving in the four days after are compared with the four days before, and that difference is taken against control shows matched on genre and venue tier over the same calendar days, none of them reviewed inside the window. The control group is what makes the number mean anything: a show sells faster as its dates approach and as the festival goes on, so without it every review scores as a hit.

Zero, with an interval running from −0.51 to +0.43 points a day. Over the four days that bounds any effect at roughly ±2.0 points of a show’s house — under one seat a night in a 50-seat room. It holds inside every cut tested: top-tier venues, mid, everything else, and comedy on its own.

Audience reviews do nothing either, on four times the sample: −0.11 points a day across 1,094 events, interval −0.53 to +0.30, and a placebo of +0.33 — bigger than the estimate itself, though still small next to the professional placebos above, and pointing in neither direction. Audience reviews arrive because people came. They do not bring more.

Method, in full

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

The single largest limitation on this chapter is that press is read from blurb text, not from a press archive. A blurb is fixed text written before the festival opened, so what it measures is the standing an act already had, not a review landing in week one and tickets moving in week two. Nobody prints a bad review in their own listing either, so every outlet’s population is selected upward before any measurement happens. Where a claim on these pages can be tested the other way round — reviews carry a date, blurbs do not — it is, and the answer is in the review section: nothing moves.

Two figures in this chapter come from an earlier snapshot than the rest of the page and say so inline rather than being quietly restated: the comedy/theatre press split and the breadth cut. They were computed against a cut two days earlier and have not been rerun, and the honest handling of a figure that has not been refreshed is to date it, not to present it as current. The tier and outlet figures beside them are current.

Odds ratios in this chapter are stratified, not raw. A claim appearing more often at one operator would otherwise measure that operator, so the comparison group for any claim is shows at the SAME operator without it, and the strata are combined rather than pooled. Where an interval is quoted it resamples shows and not dates, for the same reason as everywhere else: a run of twenty dates is close to one observation repeated, and a date-level interval on it comes out about half the width it should be.

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.