Super Bowl LI is just two days away. It’s time to make a prediction. This is the third article in the series. There is a prediction in this article and it’s close to the bottom if you want to rush down there to see what it is. In my first article “Predicting Super Bowl LI: 4 Quick Lessons in Data Visualization (Part 1)“, I identified four lessons in data visualization while setting up a database, model and the first visualization – a look at the results of all 50 Super Bowls:
In my second article “Predicting Super Bowl LI: 5 Rules for Dashboard Visualizations (Part 2)“, we used this chart as a starting point, creating a dashboard that shows the details of each Super Bowl, the outcome and indicators for each of ten prediction methods. When building this dashboard, I followed my own five rules for dashboard visualizations. Here’s a sample of that dashboard, showing Super Bowl 50:
While this dashboard indicates how each of the predictors worked for one particular Super Bowl, it still doesn’t tell us how well any of the predictors works over time. What we need are more visualizations. In this article, we use 7 different visualizations to showcase the power of visualization while vetting each of the 10 predictors.
Line Visualizations to vet Season Record as a predictor
A simple and sensible predictor; the team with the best regular season record should win the Super Bowl. The team record has been converted to a win/loss percentage with 1 representing a perfect season. The winning team record is a green line. The losing team record, a yellow line.
If this predictor was perfect, the green line would always be above the yellow line. Our graph shows something quite different. 18% of the time both teams have the same record. Of what ‘s left, the team with the better season wins 61% of the time. Even more interesting, if we look at the last ten years alone:
88% of the time the team with the worse season wins the game. That’s a solid number we can work with.
Bubble Visualizations to vet Super Bowl Appearances and Record as predictors
Once again, we’ve used the X axis to display games. The bubble chart allows us to see two measures at once. We’re using Super Bowl record, with 1 being a perfect record, for the Y axis. The size of the bubble indicates how many past appearances the team has had in the Super Bowl. Naturally, bigger bubbles are going to be on the right side of the chart. The top chart shows winning teams, the lower chart shows the losing teams.
What‘s interesting is that there is once again a striking difference between the patterns of earlier and more recent games. Until the latter half of the 90s, teams with perfect Super Bowl records won 100% of the time. Since 2000, a perfect Super Bowl record has become the kiss of death. The only exception, Super Bowl XLVII, was the only time the Super Bowl featured two teams, both coached by a Harbaugh, with perfect Super Bowl records.
Packed Bubble Visualizations to vet Last Super Bowl Appearance as a predictors
Six categories were created:
- Winning team also won the last Super Bowl they played in
- Winning team lost the last Super Bowl they played in
- Winning team had not previous played in a Super Bowl
- Losing team won the last Super Bowl they played in
- Losing team also lost the last Super Bowl they played in
- Losing team had not previous played in a Super Bowl
This Packed Bubble was then created using the total number in each category for bubble size:
Based on what we’ve seen above, I ran the same visualization twice. On the left, all 50 Super Bowls are included. There are no correlations better than a coin flip for that chart. On the left is just the last ten games. Keeping in mind, the sample sizes are small this chart gives us three predictors. Teams at their first Super Bowl won 67% of the time. Teams coming off a loss won 71% of the time. Those coming off a win lost 70% of the time.
- Clustered Columns to vet the video game Madden Football, Chicken Wing sales and the Jobless Rate
A basic clustered bar was built with clusters for each of the three predictors. The bars show both correct and incorrect predictions. Super Bowls without predictions were dropped from these charts.
All three of these predictors have a rate better than 2 out of 3. Jobless rate is the best with an impressive 77% followed by Madden Football with almost 70%.
Simple Line and Columns visualization to vet the Vegas line on the game
The column indicates the point difference between Vegas’ pick and the opposing team. When the column is negative, the pick was wrong. The line shows the Vegas point spread. Bars extending above the line made the spread.
Vegas made the correct pick 64% of the time. The winning team has beaten the point spread 46% of the time. Over the last ten years, Vegas picked the right horse only 4 times. The winning teams have only beaten the spread 2 of those 4 times:
Now that we’ve vetted all the prediction methods, let’s see what those methods predict for Super Bowl LI.
Method 1: Season record
The New England Patriots have a 14-2 record that is considerably better than the Falcons’ 11-5 season. Past history indicates a Falcons win.
Method 2: Super Bowl record
This the second trip to the big game for the Falcons. They lost to the Broncos in Super Bowl XXXIII. The Patriots are one of four teams who have played in eight previous Super Bowls, the first to see nine games. Their record to-date is 4-4. Nobody has a perfect record – history slightly favors the Patriots.
Method 3: Super Bowl appearances
As noted above, the Patriots have much more experience in the big game. That gives them a slight edge.
Method 4: Last Super Bowl win
As noted above, this is a reverse indicator. It does not apply to either team since both recorded a loss in their last Super Bowl.
Method 5: Last Super Bowl loss
This indicator applies to both teams.
Method 6: Madden Football
When the two teams were played in a simulation using Madden NFL 17, the Falcons beat the Patriots badly, 41-11.
Method 7: Jobless Rate
RiseSmart, makes an annual prediction based on jobless rate. This year New England’s jobless rate is 2.4% which beats the rate of 4.8% in Atlanta. Jobless rate picks the Patriots.
Method 8: Chicken Wings
According to the National Chicken Council, the New England region eats 12% more chicken wings than the U.S. average. The South region, including Atlanta, eats 13% more than average. This gives the Falcons a slight edge.
Method 9: Vegas
The current Vegas pick is the Patriots with a spread of -3
Method 10: Vegas with spread
Too close to call.
Summary prediction: Atlanta Falcons
Season record indicates the Falcons. Super Bowl record favors the Patriots, but only slightly. Last Super Bowl appearance is inconclusive. Jobless rate indicates the Patriots with Chicken Wings taking the opposite stance. I’m disregarding both. Madden performs a fairly realistic simulation of the game and pics the Falcons. Vegas went with the Patriots by a thin margin. It’s close, but I’m going with the Falcons.
Bonus Over/Under Analysis
History favors neither over nor under. But looking at the last ten years:
Under was the right call 6 out of ten games. The current Over/under in Vegas is 58. The average total over ten years is 47.1. The only score of 58 happened four years ago when the total was 65. I’d definitely go Under.
Enjoy the big game. Check back with me early next week for part 4. I will review my prediction and showcase all the Super Bowl visualizations that were created, but not used in the preceding articles.
If you haven’t read Part 1 & 2 of this 4 part series, I included links below.
If you have questions please feel free to leave a Comment and I will do what I do best, provide answers.
Most importantly, look for the final blog post in this 4 part series early next week, where I will review my prediction and showcase all the Super Bowl visualizations that were created, but not used in the preceding articles.