UFC 104 Buyrate Prediction: 398,000
October 27, 2009
MMAPayout.com promised a more detailed explanation of the regression analysis used to derive the linear equation used to predict 398,000 buys for last weekend’s UFC 104 PPV event. We like to keep our promises!
Regression Analysis Explained
Regression analysis is a statistical technique that allows us to explore the relationships between different sets of numbers (or what the stats community call “variables”). Regression analysis pairs the variables into data sets (x and y values), and then derives a linear relationship between those variables. The linear relationship or y = mx + b equation is then used to predict a change in the y variable for every change in the x variable.
In other words, plug in an x-value and out pops a y-value. Or, plug in this week’s viewership to Countdown to UFC 104 and outcomes a prediction of UFC 104′s buyrate based upon the relationship between both variables that regression analysis was able to uncover.
However, for the prediction to be accurate, the variables must show a fairly strong correlation – a change in one variable is met by a similar change in the other. If the variables are only slightly related – or, much worse, have no relation at all – then no linear equation can be established with reasonable predictive power.
Regression of Spike’s “Countdown to UFC” Data
|Adjusted R Square||0.474776506|
Regression of UFC Live Gate Data
|Adjusted R Square||0.490369077|
The regression analysis of both independent variables produced the following predictions:
- Spike’s Countdown to UFC 104 viewership of 524,000 predicted 437,000 PPV buys.
- UFC 104′s live gate revenue of $1.9 million predicted 366,000 PPV buys.
A weighted average of those predictions using 45% Countdown and 55% live gate then predicted the following:
- 398,000 buys
I suppose this is the part where I include a host of disclaimers:
1.) The PPV data used to form the dependent variable are largely all sourced estimates - they aren’t hard numbers. If the MMA community were privy to the UFC’s numbers there wouldn’t be a point to any of this.
2.) The correlation, standard error, and p-values are good for the considered sample size of anywhere between 25-45 data sets. I understand there have been quite significant differences in the variables in the past – they may happen again – but the charts generally speak for themselves.
Side note: I’ve actually considered throwing in log or squared functions on the advice of some statisticians I’ve consulted, but I’ll probably wait until the end of the year before I really mess with anything.
3.) I’m open to suggestions on other variables to consider – preferably something that’s more timely than the Countdown or live gate figures. As I said, this is about giving the MMA community a statistically-backed prediction tool. Unfortunately, the prediction can only be used for forum fodder at the moment, because the prediction itself comes just days before (or after) the event has taken place.
I couldn’t have put together all of this information without the help of some truly dedicated and passionate people within the sport of mixed martial arts:
- Dave Meltzer of Yahoo! Sports and the Wrestling Observer
- David Schwarz of SpikeTV
- Dan Stupp and the great team at MMAJunkie.com
- And, Adam Swift, of course, right!?
Thank-you all very much!