Isolating the calendar shift effect on attendance before you blame marketing

The show closed on the Friday, down 6.4 per cent, and by Monday lunchtime somebody has written the word underperformance next to the audience acquisition line.

The edition before ran Tuesday to Thursday in mid-October. This one ran Wednesday to Friday, a week later, and the Friday landed inside the English school half-term. Nothing about the marketing programme changed. The budget was flat, the channels were the same, the creative was refreshed but not rebuilt.

Before anybody defends anything, the calendar deserves an hour. Isolating the calendar shift effect on attendance is a solved problem in official statistics, and borrowing their framing turns a defensive meeting into an arithmetic one. It is also one of the few pieces of attendee analytics you can do with four numbers and a diary.

Official statistics treats the calendar as a known effect

The US Census Bureau produces, distributes and maintains X-13ARIMA-SEATS, the seasonal adjustment software behind its own economic series. Two of the things it corrects for are directly relevant to a show.

Trading day adjustment removes the effect of weekday composition. A month with five Saturdays behaves differently from a month with four, and the difference is a property of the calendar rather than of the economy. Moving holiday adjustment handles holidays whose position shifts between periods, Easter being the standard example. Both are estimated as regression effects fitted alongside the model of the series, and the Census Bureau ships a separate utility, Win Genhol, purely to generate user-defined moving holiday regressors.

The working assumption underneath all of that is the part worth stealing. Comparing a period against the same period last year means very little until the calendar difference between the two has been estimated and removed. Statistical agencies do not treat that step as optional, and they do it before anybody is allowed to interpret the series.

Nobody runs X-13 on a trade show, and they should not. A show has one observation a year and a handful of days, which is nowhere near enough series to fit anything. What transfers is the discipline: identify the calendar differences between the two editions, size each one, and subtract before you interpret. Doing that by hand on four numbers is entirely feasible.

List the calendar differences before you look at the result

There are four kinds of date effect on a show, and they need separating because they behave differently.

The day pattern is the first: which weekdays the run occupies. A Tuesday to Thursday run and a Wednesday to Friday run are not the same product, because the closing day of a trade show is always the weakest and a Friday closing day is weaker again.

The gap since the last edition is the second. Editions are rarely exactly 365 days apart. A show that ran on 14 October and then on 21 October the following year has a 372 day gap, and any per-day rate of demand accumulation is being measured over a longer window.

Position in the wider calendar is the third: school holidays, public holidays, religious observance, and the end of a quarter or a fiscal year for the buying audience.

Competing events are the fourth, and they belong in this list rather than in a market narrative, because a clash is a date fact you could have looked up.

Write all four down for both editions before opening the attendance file. Doing it afterwards, once you have seen the number, produces a list that mysteriously explains exactly the shortfall you observed.

Should you compare by weekday or by day position?

Here is the data. The earlier edition ran Tuesday to Thursday with daily unique attendance of 18,600, 21,300 and 12,900. The later edition ran Wednesday to Friday with 19,100, 20,400 and 8,700.

Distinct people were 26,400 and 24,700, so the headline is minus 1,700, or minus 6.4 per cent. Person-days were 52,800 and 48,200, so minus 4,600.

The tempting comparison is weekday against weekday. Wednesday against Wednesday gives 21,300 against 19,100, which is minus 10.3 per cent, and looks alarming. It is also the wrong comparison, because the earlier Wednesday was day two of the run and the later one was day one, and position within the run dominates weekday at a trade show. Day one is arrival day, day two is peak, and the closing day empties by mid-afternoon regardless of what it is called.

So compare by position. Day one 18,600 to 19,100, up 2.7 per cent. Day two 21,300 to 20,400, down 4.2 per cent. Day three 12,900 to 8,700, down 32.6 per cent.

Days one and two together went from 39,900 to 39,500, which is minus 1.0 per cent. The entire decline sits on the closing day.

Size the calendar component and report the residual

Now put a number on the calendar effect rather than gesturing at it.

Days one and two say the underlying show was flat, near enough, at minus 1.0 per cent. If day three had behaved the same way, it would have come in at 12,900 times 0.99, which is 12,771.

The gap between that and the actual 8,700 is 4,071 person-days. Two calendar facts are competing to explain it: the closing day moved from Thursday to Friday, and it landed in half-term. On this show's own history, a Friday closing day has run at about 0.68 of the level a Thursday closing day reached, which would predict 12,771 times 0.68, or 8,684, against an actual 8,700. The calendar model accounts for essentially all of it, with 16 person-days left over.

That is a suspiciously clean fit and it should be treated as one, because the 0.68 factor was estimated from a handful of editions and carries a wide error band. The useful version states the range: at 0.62 the model predicts 7,918 and the show over-performed by 782 person-days; at 0.74 it predicts 9,451 and the show under-performed by 751. The honest sentence is that the calendar explains most of the closing-day loss and the residual is smaller than the uncertainty on the estimate.

Then convert person-days back into people, because the headline is in people. On this show's history about 22 per cent of closing-day attendance is single-day attendance, people whose only visit was that day. That is 2,838 people in the earlier edition. If the closing day fell 32.6 per cent, roughly 925 of them did not come at all, while the rest of the person-day loss is repeat visits by people who attended anyway.

So of the 1,700 person decline, about 925 is calendar. The residual is about 775 people, or 2.9 per cent of the prior edition. That is the number to hand to the audience acquisition team, and it is a completely different conversation from minus 6.4 per cent. Splitting that 775 into retained, lapsed and new attendees is D13's job, and it should be run on the corrected figure rather than the raw one.

How much does the date gap actually matter?

Two of the four calendar factors deserve separate handling because organisers routinely get their sizes backwards.

The 372 day gap sounds like it should matter and mostly does not for attendance. It matters for registration pacing, where an extra week of promotion window shifts the curve, and the year over year registration comparison in A5 covers that properly. By show open the audience has largely accumulated regardless, and a week either way moves the total by less than the noise in it.

A competing event in the same month is the opposite. Calling it a soft market gives it far too much room. It is a nameable, checkable event that removed a specific population from your hall, and it can be sized. Take the badge file from your last edition, match it against the competing show's exhibitor list and published audience profile where you can, and count how many of your lapsed attendees work for companies with a strong reason to be at the other event. That is a partial answer rather than a complete one, and it is still far better than the sentence about market conditions that would otherwise appear.

Write the correction into the report template

The reason this analysis rarely happens is that it has no home. It is nobody's job, it is not a section in the template, and by the time anyone thinks of it the narrative has set.

Give it a home. A four line calendar block at the top of the attendance section of every post-show report, filled in before the attendance numbers are read: run days and weekdays for both editions, gap in days, calendar collisions, competing events. Filling it in takes five minutes and it is completed by whoever books the venue rather than by an analyst. It sits above the registration to attendance reconciliation that D9 sets out, because the calendar block explains movements that the ledger will otherwise absorb silently.

The block does two things. It stops the narrative forming before the correction, and it accumulates. After four editions you have a small history of day-position and weekday patterns on your own show, which is what turns the closing-day factor from a guess into an estimate with a range around it.

The industry-level context is genuinely useful for expectation-setting and useless for this correction. UFI's 36th Global Exhibition Barometer, published in January 2026 from 378 companies across 57 countries and regions, reports 47 per cent of respondents seeing activity up more than 5 per cent in 2025. Knowing the sector grew tells you nothing about whether your Friday was a bad Friday.

Where this stops

The correction rests on a historical factor for how a closing weekday behaves on your show, and most organisers do not have enough editions to estimate one.

With four or five prior editions you have perhaps two observations of a Friday close, which supports a rough factor and no confidence interval worth printing. State the factor, state that it comes from two editions, and let the reader discount it. Do not present a point estimate from two observations as though it were measured.

The second limit is that calendar effects and real effects are entangled and cannot be fully separated by arithmetic. If the show moved into half-term because the venue had no other slot, and the venue had no other slot because a larger competitor took the week, then the calendar effect and the competitive effect are the same event seen twice, and adding them separately double-counts.

The third is that this analysis can become an excuse machine. Every edition has some calendar difference, and a team that corrects for all of them every year will never record a bad year. The control is to fix the correction rules in advance, in the template, and apply them symmetrically. A calendar adjustment that only ever gets calculated when attendance falls is not an adjustment.

Fill in the four line calendar block for your last two editions this week, then recompute the day-position comparison. If days one and two are flat and the whole change is on the closing day, you have a date problem, and the marketing review can be cancelled.