Back in 2000, the year of my Pro Tour debut, there were only about 25 Grand Prix per year and GPs with approximately 500 players were the norm. Although average attendance gradually increased over the years, it wasn’t until 2012 that we saw a radical change.
In 2012, the year of the first Modern Grand Prix and of Return to Ravnica’s release, the number of GPs dramatically increased to more than 40 per year. What’s more, attendance boomed.
In celebration of this new era of GPs, I wanted to run the numbers and see which formats and locations drew the biggest crowds. I went over all GPs from January 1, 2012 through December 31, 2017 and obtained statistics on locations, formats, and attendance numbers.
Legacy GPs Are the Largest on Average—Standard the Smallest
Out of all individual Constructed formats, Legacy drew the biggest crowds on average. Standard GPs had the largest overall sum of attendees out of all Constructed formats, but drew the smallest average number of players per GP out of all formats.
|Format||Number of GPs from 2012 till 2017||Average GP attendance|
|All regular GPs||281||1,383|
Note 1: Last year’s Legacy-Limited-Modern event in Las Vegas was excluded from these format statistics as it could unfairly skew the numbers. “Teams” encompass both Team Limited and Team Constructed. “All regular GPs” include the two Block Constructed GPs but exclude all Modern Masters GPs and the aforementioned Vegas triple event.
Note 2: Attendance numbers for 2015-’16-’17 GPs were taken from the main page blurbs on their event coverage pages if available and from the number of players in the final standings of the main event otherwise. Attendance numbers for 2012-’13-’14 GPs were taken from the Wikipedia entry that I discovered along the way. My full data set is available here.
For Legacy GPs, I should point out that its average numbers received a huge boost from the humongous 4,003-player GP New Jersey in 2014. But even if I was to exclude that event, Legacy would still have an average turnout of 1,558—the largest among individual Constructed formats.
The number of Standard GPs has climbed steadily over the years: In 2012-’13-’14, the percentage of Standard GPs was 30%. In 2015-’16-’17, it had grown to 37%. In 2018, this trend will reverse, as only 25% of the GPs will be individual Standard. Instead, we’ll get substantially more Modern GPs (from 15% in 2017 to 20% in 2018) and Legacy GPs (from 4% in 2017 to 7% in 2018). We’ll also get more team events, including Team Trios Constructed, because of the upcoming Team Pro Tour. So fans of the non-rotating formats will have way more GPs to choose from in 2018.
GP Attendance Climbed From 2012 to 2016, but Was Down in 2017
Let’s take a look at overall year-by-year statistics.
|Year||Number of GPs||Average GP attendance|
Note: These numbers include every single event, including such outliers as the Modern Masters GPs. The Modern Masters GPs in Las Vegas 2015 were counted as two separate events.
As the table reveals, average GP attendance saw a steady climb from 2012 to 2016. But from 2016 to 2017, attendance sharply dropped.
Standard, Limited, and Team turnouts in 2017 were down in particular compared to the two years prior. A potential reason could be an oversaturation of events for these formats, perhaps in combination with a rocky Standard and a relatively unpopular Ixalan Limited. But I can only speculate. I’m curious to hear your thoughts on these trends in the comment section!
Japanese GPs Are the Largest on Average—Latin American Ones the Smallest
Japanese GPs have grown massively over the years, both in size and in number. In 2012-’13-’14, their average attendance was 1,708 players. In 2015-’16-’17, average attendance for regular GPs grew to 2,467 players, and that’s without even counting the huge Modern Masters GP in Japan.
|Location||Number of GPs from 2012 till 2017||Average GP attendance|
|Asia Pacific (without Japan)||39||855|
‘Note: Extra-large outlier format GPs (all Modern Masters GPs, and last year’s Vegas event) were excluded from this location analysis. This means that none of the Las Vegas events, with an average attendance of 3,464, are included in the set of GPs in West USA. For the USA, I took the region definitions from the U.S. Census bureau. Southern Europe encompasses Spain and Italy. Northwestern Europe included France, Belgium, Netherlands, Germany (including Bielefeld), Denmark, Czech Republic, and United Kingdom. Eastern Europe comprises Sweden, Russia, Poland, and Austria. This is a weird name for a loose collection of countries, but I lumped them together because they are relatively hard to reach and have smaller turnouts compared to the rest of Europe.
Asia Pacific (without Japan) had relatively small events, including the tiny ones in New Zealand. GP Auckland 2012, at 264 players, was the smallest GP in the data set. In Latin America, the other region with relatively low GP turnouts, it was noticeable that Sao Paulo consistently drew more players than other locations.
I do wish to point out that even if these regions have lower average attendance numbers than the rest of the world, it is my view that they deserve GP support. Having traveled to Latin America as an event coverage reporter multiple times, I experienced first-hand how important these tournaments are for the vibrant Magic community over there. For the Latin American players, a “local” GP is one of the biggest celebrations of the year, and it brings players from multiple countries together.
Predicting via Multiple Regression
Let’s turn from historical data to the future. Suppose that I would like to predict the attendance of GP Houston in January 2018. How could I do that?
Well, it’s Limited, so I could look at the average attendance of Limited GPs: 1,325. But it’s also in South USA, which had an average attendance of 1,562. I could take the average of the two numbers as my prediction, but that also feels a bit simplistic. It also wouldn’t account for the overall attendance growth, as these averages are influenced by the smaller events in 2012 and 2013.
Instead, I’ll employ a different mathematical tool: multiple linear regression. It can be used when you have a lot of data and want to estimate a formula for the value of a variable (in this case, Grand Prix attendance) based on the value of two or more other variables (in this case, format, location, and date) under the assumption that the relationship is linear (and various other assumptions that I’ll sweep away for now).
For the analysis, I excluded the two Block Constructed GPs and all extra-large outlier format GPs (all Modern Masters GPs, and last year’s Vegas event) because they are too different from other events in terms of scope or format. This means that the predictive value of the model will be limited to “regular” events.
Since regression analysis requires numerical variables as input, I had to transform terms like “Standard” or “West USA” into numerical variables. The standard way to do so is by introducing 0-1 dummy variables. Format, for instance, was coded via four dummy variables: Modern, Legacy, Standard, and Teams. A single GP would then have a 1 for at most one of these dummy variables and 0 for the rest. If all four would be 0, then that would mean the format was Limited. Region was coded in a similar way, with South USA being the base category. I used the method of least squares to estimate the regression coefficients.
The Resulting Model
The regression model or formula that I found (for “regular” events) can be described as follows.
Start with 1,379 players. This is the base level for a Limited Grand Prix in the South USA region.
Then add 5.33 players for every month since the start of 2012. So add 5.33 players if you’re in January 2012, add 10.66 players if you’re in February 2012, and so on. This means that for January 2018 you add 325 players, rounded to complete humans. This element captures the overall growth of GP attendance over the years. It does so in a simplistic, linear way that is not fully accurate but that facilitates the analysis.
Then, check the format:
- If it’s Limited, do nothing.
- Subtract 147 players if the format is individual Standard
- Add 110 players if the format is individual Legacy
- Add 132 players in the format is individual Modern
- Add 150 players if the format is Teams
Though Legacy had the largest average attendance, it doesn’t get the largest boost in the multiple regression model. The intuitive reason is that there were never any Legacy GPs in the smaller regions like Latin America or Asia Pacific.
Afterwards, check the location:
- If it’s South USA, do nothing.
- Subtract 796 players for a GP in Latin America
- Subtract 706 players for a GP in Asia Pacific
- Subtract 421 players for a GP in Eastern EuropeSubtract 299 players for a GP in Canada
- Subtract 227 players for a GP in Midwest USA
- Subtract 213 players for a GP in Southern Europe
- Subtract 80 players for a GP in West USA
- Add 21 players for a GP in Northwestern Europe
- Add 254 players for a GP in Northeast USA
- Add 532 players for a GP in Japan
Most of these location-based adjustments are larger than the format-based ones, indicating that location has a bigger effect on attendance than format.
Prediction is Poor
A regression model is not always a perfect fit—a lot of variation in the observed data can remain unexplained. That’s also the case for this setting. Even though the regression equation is significant with p < .00001, the R-squared value (which indicates the percentage of the variation explained by the model) is only 48%. This sends up warning flags about imprecise predictions.
On average for GPs in my data set, the attendance numbers predicted by the regression formula were off by 309 players. To illustrate what that means: at GP Albuquerque 2016, the regression model predicts an attendance of 1,576 players, but only 1,266 showed up. Meanwhile, at GP Lyon 2015, where the regression model predicts an attendance of only 1,645 players, 1,952 came to play. These are illustrative “average” cases. There were plenty of larger deviations.
So the size of an event remains a bit of a gamble, even with this simple regression model. The region code is arguably too broad, there is inherent uncertainty in player numbers, linear player growth is a big simplification, it doesn’t take into account attendance caps or TOs, and it is hard to fit aspects like region-specific format preferences or the popularity of one Limited format compared to another in a model like this. But hey, we can still make a prediction.
Keeping these weaknesses in mind, let’s try to see what the regression model would suggest for the first three months of 2018. Team Trios Constructed will be treated in the same way as other Team events.
|Grand Prix||Region||Format||Attendance prediction|
|Santa Clara (Jan 6)||West USA||Team Trios Const.||1,774|
|Indianapolis (Jan 20)||Midwest USA||Team Limited||1,627|
|Houston (Jan 27)||South USA||Limited||1,704|
|London (Jan 27)||NW Europe||Limited||1,725|
|Toronto (Feb 10)||Canada||Modern||1,542|
|Lyon (Feb 17)||NW Europe||Modern||1,862|
|Memphis (Feb 24)||South USA||Standard||1,562|
|Santiago (Mar 10)||Latin America||Team Limited||1,069|
|Madrid (Mar 10)||South Europe||Team Trios Const.||1,652|
|Phoenix (Mar 17)||West USA||Modern||1,767|
|Kyoto (Mar 24)||Japan||Team Trios Const.||2,397|
|Amsterdam (Mar 31)||NW Europe||Team Limited||1,886|
I expect to be off by more than 300 players on average. So don’t put too much stock in these numbers. But I do expect that the majority of players attending these events, whether they are GP first-timers to GP veterans, will have an excellent time and return home with awesome Magic memories. In the end, that’s what GPs are all about.