Whoa, Ken, that's pretty interesting.
I really enjoyed your reponse, and you've got some great suggestions here! Unfortunately, I put the analysis together as a fun side project in my home office, and most of the suggestions would require far more time and resources than I have at my disposal
Let me briefly answer your questions...-Did you look at day of week and time of day for the games. I imagine some of that effect can get conflated with the nationally-televised binary variable, but that is one that I would initial think to have some effect.
Yeah, I looked at time of day (I was really curious to see if the 4:15 games drew any differently than the 1:00 games, which didn't appear to be the case), as well as the day of week. And you're absolutely right. When you see correlation between increased attendance and Monday Night Football games, for example, it's damn near impossible without quality polling data to say definitively whether it's because a) the game is on national television, b) there's an evening start time, or c) the game falls on a Monday. With a sample so small some of these variables would be perfectly collinear, so there's almost got to be some best guess judgement calls going on. You could probably break down some dummy variables even further, time permitting, and get an even better idea of what's going on, but unfortunately, the time just wasn't there.-I am always so curious about the national games. It does seem anecdotally that more people are interested in attending them. Do you think this is due to simply the fact that they are televised more widely, or because of why they were scheduled to be nationally-televised in the first place (division rival game, replay of the prior season's super-bowl, two teams that are expected to be good before the season)? I feel like there are so many factors that go into the nationally-televised game. Perhaps its a desire to be part of the event, or to fill the seats to project a better image of the Jags nationally, or the reasons discussed above for which the network wants to televise it nationally.
Same on this one. I'd love to see some polling data on it, because especially in Jacksonville, it's really hard to say. For example, on one hand, I think people want to go to these nationally televised games to show how great Jacksonville's fans are. But then you have that super weird Jacksonville thing too where if the game isn't going the Jags way by halftime, 40,000 people just randomly leave. There have been a handful of Monday Night Jags games in recent years where the stadium is almost completely empty toward the end of the game, which seems counter intuitive to the civic pride explanation. Civic pride certainly plays a part, but my best guess is that those with heavy civic pride are attending most games regardless, and that the attendance boost from nationally televised games probably comes from the fringe, non-hardcore fans who just want to be a part of the big thing going on in the city that day. -For winning percentage, is "seasonal" the end-of-season or the season to that point?
I tested a variety of different measures here. Winning percentage on the season can be a little misleading, simply because a 1-0 Jaguars team that fluked into a win to start the season would register as a more winning team than a 15-1 team who destroyed their competition all year. Likewise, for opponents, an eventual Super Bowl winner who loses their first game (0%) would register lower than a one-win (1-15) team that the Jags play on the last day of the season. Quick, dirty, and ad hoc, but in addition to current-day winning percentage, I settled on testing winning percentage over last 10 games (which, for opponents, was pretty close to winning percentage on the season, actually). Like I mentioned, I also tried to test a few momentum based variables as well (record for last three games, did the Jags win their previous game, did the Jags win their previous home game, etc.), as well as a variable for previous season winning percentage and a dummy variable for a winning/losing previous season. I'm sure there's an even better way to weight this variable, but I had a really rough time finding a correlation between any of these home team winning variables and attendance that would suggest that people come to Jags games when the team is winning who otherwise wouldn't
if the team were losing.-Do you have anything like pre-season rankings in there, or some variable that measures the "hype" (media coverage, big pickups/trades, etc)?
None, unfortunately. The rough model that I came up with only accounted for half of the variation in game-to-game attendance. I would guess that these factors, though difficult and time-consuming to quantify, would definitely factor in. Even harder to factor in, but probably what you're referring to here overall, is overall local attitude toward the team, both on the field, and as a stable fixture in Jacksonville. I've noticed, for example, that the city's fanbase has a tendency to turn on the team, both in spending and in attitude, whenever relocation rumors pick up steam (even if they are false). It's quite sad to see. -I would think it might be more telling to measure the relative popularity by a weighted variable (fan base, population in home market, merchandise sales, local fans even [I bet you could get data on that]) rather than ranking. I would expect that to give this variable more resolution.
Absolutely agree. Using the Harris Poll was another quick and dirty shortcut that could likely be done much better (and most certainly has) using a bigger, more thorough data set than I was able to compile. -I don't know about removing 3/4 of the 2009 season from the equation. Perhaps an additional dummy variable thrown in so that you could differentiate what you believe to be an outlier, but there is still differences between the data for those 6 games that might be of interest. Plus, those games were coming off a very bad season, so your trailing-10 percentage may actually show some movement in addition to the "recession" variable.
Fantastic suggestion. I'm going to look into this.-I wonder, does your seasonal normalization for economic effects also remove any annual effects from changes in the team (hype, pre-season rankings, previous season record, hirings/firings, etc)? If every game is normalized by the average attendance that season, then there would be nothing to show for changes between seasons, right?
Aside from the win/loss variables that I spread between over adjacent seasons, you're correct. I'm fairly confident that changes in mean season attendance are influenced first and foremost by economic factors. For example, if you plot mean season attendance for the last few years versus the unemployment rate in our MSA, it's like looking in a mirror. Unemployment goes up, attendance goes down. Unemployment drops, attendance increases. But there are definitely outside factors (such as the Jags relatively historic offseason, for example) that would influence mean attendance for the 2013 season as well. This is good stuff. You should do it for every season AND every team
As a numbers nerd, such a study would be amazing fun. But also an amazing amount of work. I'm going to try doing a journal search tonight, however. I could almost guarantee that someone has already done something similar, with better resources and more data.
Would be a fun crowd-sourcing project too with a large group of people.
If quality polling could be conducted, we could probably paint an even better picture of what truly influences Jags attendance.