Read this before quoting any number
This is built from public third-party records (Football Web Pages and Transfermarkt), not from clubs' own audited turnstile data. It is not an official FAW publication. Three limits matter, and they are stated again in full at the end:
- A small number of fixtures have no attendance on record.
- 2020-21 is excluded entirely: played behind closed doors.
- No television effect can be measured. Not one played fixture in the set carries a broadcast flag.
No missing attendance is filled in. Travel and weather are estimates; regression results are model estimates. Records cover 16 August 2019 to 5 September 2026, reviewed on 9 September. Confidence intervals for regression estimates allow for repeated matches at each home club. This is exploratory analysis, with several comparisons and no claim of cause.
What the model suggests
Each estimate compares a category with its reference after adjusting for the listed factors. The full model uses 995 matches. It includes home club, season, kick-off slot, journey band, month, holidays, form, weather and same-day clashes. It does not include ticket prices, away-support counts or broadcast coverage.
Adjusted associations with attendance
Lines show 95% intervals clustered by home club. An interval crossing zero means the direction is uncertain; it does not establish no effect. These are individual intervals, not adjusted for testing multiple hypotheses.
Shorter journeys, stronger crowds
Trips of three hours or more were associated with around 31% lower home attendance than trips under 45 minutes, after adjustment (95% interval −39.5% to −21.6%; n=995). This measures the whole crowd, not away supporters alone. Journeys are estimated between grounds, using one ground per club across seasons.
How far the away team travelled
Bar height is the attendance index; the figure underneath is the real average crowd.
Every fixture, plotted
One dot per match. The line is the average index at each journey length.
Friday football: look beyond the raw gap
The raw attendance-index gap changes when the days are weighted to the same mix of journey bands. That descriptive reweighting is not the regression estimate below: it adjusts travel mix only. Neither comparison proves that moving a match will increase its crowd.
Before and after matching the travel mix
Grey is the raw average. Teal re-weights every day to the same mix of away journeys.
Why they differ: share of fixtures with a 3h+ away trip
Friday against Saturday, like for like
Friday’s adjusted estimate is +9.6% compared with Saturday. The 95% interval is −1.4% to +21.9% on 787 matches. That is a promising point estimate with meaningful uncertainty. The original ordinary-error interval was +3.0% to +16.6%; allowing for repeated observations of the same clubs widens it. The descriptive chart below shows a separate breakdown by journey band.
Kick-off slots
Attendance index by slot, with the real average crowd
Slots resting on fewer than eight matches are flagged and carry no weight.
Day against journey
Compare the attendance indices within each journey band. These are observed averages with unequal sample sizes—not evidence that Friday and Saturday are equivalent, or that one will work best for every club.
North, Mid and South
North, Mid and South are study categories, not official league divisions. The chart keeps all six fixture combinations visible. Geography, journey length and opponent appeal overlap; these averages cannot separate their contributions.
By fixture geography
Club regions follow the supplied club table. Aberystwyth and Newtown are Mid Wales. No claim is made that all within-region trips are short.
North against South draws smaller crowds
North–South fixtures had an average attendance index 22.4% lower than all other fixtures (454 versus 671 matches). The comparison includes Mid–North and Mid–South games. It is descriptive, not a separate causal effect of regional boundaries.
All other fixtures against North–South
Observed attendance indices, not adjusted effects. “All other fixtures” includes games involving Mid Wales.
Friday and Saturday within each comparison group
Observed attendance indices, not adjusted effects. “All other fixtures” includes games involving Mid Wales.
North v South, day by day
Observed attendance indices, not adjusted effects. “All other fixtures” includes games involving Mid Wales.
Derbies
Derby against everything else
“Derby” is a shorthand for an estimated drive under 45 minutes; it does not verify a sporting rivalry.
The strongest repeat fixtures
Minimum four meetings in the dataset.
The shape of a season
Month by month
Holidays and first/last recorded home games
Bank holiday fixtures
First and last recorded home game
“First” and “last” mean the first and last attendance recorded in each club-season in this dataset. Missing matches and the incomplete current season mean these are not necessarily season openers or finales.
Does winning fill the ground?
An extra point per game in recent form was associated with +11.5% attendance (95% interval +5.1% to +18.2%; n=995). Form uses the previous three to five recorded matches. League-position bins are reconstructed without disciplinary deductions or split-phase adjustments; treat them as an approximation.
By league position at kick-off
By form over the last five
Weather at kick-off
Weather estimates from Open-Meteo are matched to ground coordinates and London local time. Conditions use the kick-off hour rounded down. Rainfall totals cover three complete hourly periods ending at that hour; they are not minute-exact measurements or supporter forecasts.
Precipitation of at least 4mm over that three-hour window was associated with −23.9% attendance compared with less than 1mm (95% interval −29.9% to −17.5%; full model n=995). This combines the wet and additional-heavy coefficients. Cold also has a negative estimate in the corrected model. Weather is an association, not a guarantee of turnout.
Conditions at kick-off
Raw averages.
What survives the controls
Same club, same month, same opponent's journey. Faded bars are not statistically significant.
Weather estimates depend on the window, thresholds and model. The rain-by-distance chart is descriptive; it does not establish a separate interaction effect.
Does rain punish the long trips hardest?
The intuitive answer is yes. The data leans that way but cannot prove it.
Weather estimates depend on the window, thresholds and model. The rain-by-distance chart is descriptive; it does not establish a separate interaction effect.
Clashing with Cardiff, Swansea, Newport and Wrexham
Same-day clashes were associated with −7.1% attendance (95% interval −13.2% to −0.5%; model n=995). South and Mid Wales clubs are flagged when any of Cardiff, Swansea or Newport plays at home; North clubs are flagged for Wrexham. This is not a nearest-club calculation and does not measure overlapping kick-off times.
Raw comparison
Descriptive comparison. Fixture dates, geography and opponent mix differ between groups.
Saturdays only
Descriptive comparison. Fixture dates, geography and opponent mix differ between groups.
By selected club or combination
Descriptive comparison. Fixture dates, geography and opponent mix differ between groups.
School holidays
The approximate Christmas school-holiday window has an adjusted estimate of +43.1% against term time (95% interval +21.6% to +68.3%; n=995). Holiday dates, festive fixtures and derbies overlap, so this is a planning lead rather than a promised increase. Other holiday estimates remain uncertain.
Attendance by school-holiday window
Raw averages against term-time fixtures.
Welsh school holiday dates are approximated to an all-Wales calendar; individual local authorities vary by a few days, and Easter moves each year. The Christmas figure overlaps the Boxing Day and New Year fixtures, so the two should not be added together, they are largely the same effect.
Is the league actually growing?
Early 2026/27 crowds are encouraging, but the season is incomplete: 56 recorded attendances through 5 September. The charts separate raw averages from a model comparison against 2024/25. New clubs, fixture mix and the short observation window prevent a claim that expansion caused growth.
Average crowd by season
2020-21 omitted: behind closed doors. 2026-27 is a part-season.
Raw change against model-adjusted change
Compared with 2024/25, the model’s chosen reference season. This is not a last-season growth rate. Regression intervals are clustered by home club.
Early evidence after the expansion
Compare clubs appearing in both seasons over 25 July–9 September. This aligns calendar windows, not opponents, match counts or every other factor. It is too early to attribute a change to the expansion.
Change in home average, 2025-26 to 2026-27
The real crowd figures are shown on each bar.
How the fixture list itself changed
Observed share of fixtures by day. A changing schedule can accompany changes in crowd size; this chart does not assign a causal contribution.
Club by club
Average home crowd by season
Darker means a bigger crowd. Blank cells are seasons out of the division. Hover any cell.
Every club in the dataset
Who brings a crowd
These are total home crowds when each visiting club appears. They do not count the visitors’ travelling support separately. Geography, host selection and rivalry can all affect the ranking.
Largest crowds in this dataset
Ideas to test, questions to keep open
Ideas worth testing
What remains uncertain
Data, gaps and method
What is missing, and how it could mislead
Coverage by season
Method and reproducibility
Attendances and fixtures come from supplied Football Web Pages and Transfermarkt extracts. They are published third-party figures, not independently audited attendance records. Missing values remain missing. The dataset has 1,361 played rows: 1,125 with included attendances, 44 otherwise eligible rows without attendance and 192 excluded rows from 2020/21.
The attendance index is the crowd divided by that home club’s recorded home average in that season. Models use log attendance with home-club fixed effects and the predictors listed in the downloadable model audit. Coefficients are converted as 100 × (exp(beta) − 1). They describe conditional log-scale associations; they are not guaranteed changes in an arithmetic crowd forecast.
Regression confidence intervals use home-club clustered covariance, a finite-sample correction and a t reference with the number of clubs minus one degrees of freedom. Ordinary intervals are retained in the audit for comparison. These are exploratory models with multiple comparisons, relatively few clusters and no causal identification. Alternate specifications may change estimates. Raw-chart intervals are ordinary descriptive mean intervals and do not have the same adjustment.
Road times use OSRM routes between the supplied grounds, not actual supporter origins or historic venue changes. Weather uses Open-Meteo reanalysis estimates; three hourly precipitation totals end at the rounded-down kick-off hour. Holiday dates are approximate; clash flags use regional groups. Ticket prices, separate away attendance, reliable television flags and VAR are not measured.
Match dataset · Clubs · Travel matrix · All results and model audit · Reproduction package