Empty Stands, Falling Number: A 240-Match Ledger of IPL Home Advantage

On the night of 21 May 2026, at the Rajiv Gandhi International Stadium in Hyderabad, Rising Pune Supergiant made 128/6 in their twenty overs. Mumbai Indians made 129/8 and won by one run. It remains the lowest total ever defended in an IPL final. I did not trust the scoreboard that night. I logged all 240 deliveries by hand — length, line, batter's hand, and the height at which the ball reached the keeper's gloves. I counted 240 balls before I trusted a single number. What emerged was not "pressure." It was dew in the second innings, and a bowling plan that makes even 128 look large. Krunal Pandya's 47 off 38 was the cleanest figure of the evening; the cleanest fact was that the Hyderabad surface behaved differently in each innings. "Home advantage" is spoken about in Indian cricket the way weather is — felt, not measured. To me it is a claim, and every claim needs an n, a date range and a method. Since 2026 I have split the claim into four variables, because home advantage is not one thing but a sum: crowd pressure, toss predictability, host-friendly pitch preparation, and venue identity — size, breeze, the hour dew arrives. I declare the method first, because a method declared later is an excuse. Baseline: the 240 IPL league matches from 2026 to 2026, where each franchise played at its actual home ground. Treatment window: the full season played in the UAE from 19 September to 10 November 2026 — 60 league matches, zero spectators. It is the rarest natural experiment the sport has offered: same rules, same players, no crowd. The result in two lines. Home teams won 54.8% of matches in 2026–19. In the UAE season that figure fell to 47.1%. When the stadiums emptied, the home advantage left with the crowd — at least 7.7 percentage points of it. Not all of that drop belongs to the crowd, but the crowd is the largest single component, and this is the first time we have been able to separate it. One honest caveat, and it is the weakest joint in the comparison. In 2026, "home team" meant the designated franchise, not the stadium. Three venues — Dubai, Abu Dhabi, Sharjah — no travel load, no local pitch preparation. Which means 47.1% is what survives after you strip out both pitch and travel. What remained was crowd and familiarity. On the toss my ledger is as clear as it is dull. The toss effect is seasonal, not structural — no permanent gain, different at every ground. In the UAE, dew in evening games changed the value of the first six overs, and chasing sides won more. Calling that "second-innings advantage" is easy; it was a weather decision, not a captaincy one. Sharjah's short boundaries inflated scores in the same window, so reading Sharjah numbers alongside Dubai numbers is a mistake. Now the part fewer people discuss. When rules change, the dataset resets, and we keep merging old and new numbers without labels. The Impact Player has applied in the IPL since 2026. Its direct effect landed on the death overs: no sixth or seventh bowler is required, so run rates in the last four overs have risen against the pre-rule baseline. An analyst who puts pre-2026 death-over numbers on the same chart as today's is giving one name to two different games. Every table I publish now carries a regime label: pre-Impact, post-Impact. The same logic applies to the two-new-balls rule in ODIs from October 2026, which reset the entire middle-overs spin dataset. Treat the rule as a variable, not as scenery. Here I have to argue against my own conclusion. The fall from 54.8 to 47.1 can be explained by three things other than the crowd: first, slow UAE surfaces where the toss mattered more; second, a tournament compressed into 60 matches, where teams grew familiar with each other and "home" lost value; third, bio-bubble fatigue, visible in the final three weeks. Correlation is not causation, and I will not sell it as such. I also keep a log of the cases where the eye test beat the model, with dates attached. At the 2026 World Cup semi-final, New Zealand's 239/8 looked under-par to my model, but the slow surface and India's top-order pace problem were not properly captured in it. The eye won; I wrote the loss down. Confidence is a variable I now calibrate, not a constant I assume. The forward signals are clear. First: over the next two seasons, does the home win rate in full stadiums return toward the 54.8% baseline? If it does, the crowd component is proven; if it does not, we must blame pitch preparation and travel scheduling instead. Second: WPL data, where seasons are short and samples thin — there I must write n before I write "trend," and where n is small, "insufficient sample" is a complete sentence. The question I am left with: if a large share of home advantage was an attendance figure, then the "fortresses" we wrote about for twenty years — how much of that was cricket, and how much was turnout?