In our ongoing effort to simplify our business, we’ve made several changes to our premium subscription packages in advance of the 2014-15 football season. In short, we’ve:
- Eliminated pricing tiers (e.g. “Lite” and “Pro”) from our office pool picks packages
- Combined our NFL Pick’em and College Football Pick’em products
- Lowered prices on longer-term premium subscriptions (Quarterly and Yearly)
- Moved to consistent $49 pricing for all office pool picks packages
Below is the reasoning behind each decision in case you are curious.
No More “Lite” And “Pro” Packages
We now have a single premium package associated with each type of office pool we cover. In addition, we still offer an awesome bundle deal on our yearly Pool Picks Subscription, which includes access to all four of our pool picks products. You can see all our current office pool packages on our sales page.
Researching and improving our prediction models is an ongoing process, and one to which we devote significant time. The past year has been particularly active, though, as we’ve been rolling out significant pick logic updates to all of our sports. These updates are primarily designed to eliminate a long-standing source of user confusion, conflicting picks, but they should also help to improve our long term pick accuracy.
MLB is the latest (and final) sport to join the club. So, what’s new?
This week we launched a site update to enhance our NCAA Tournament 2013 section. A few weeks ago we introduced improved 2013 bracketology pages, and we’re now rolling out additional analytics to explore how we predict teams to do once the NCAA tournament begins. Remember that all of this new bracket predictions info updates daily, so check back frequently to see what’s changed.
(For more background on our approach to projecting the 2013 NCAA bracket, and why our approach is different and better than that of human bracketologists, you can see our post from November introducing our new NCAA bracket predictions.)
On Friday we launched a site update that has significantly improved our bracketology 2013 pages. For more background on our approach to projecting the 2013 NCAA bracket, and why our approach is different and better than that of human bracketologists, you can see our post from November introducing our new NCAA bracket predictions.
With a lot of back-end predictive analytics for projecting the NCAA bracket now built, we’re starting to focus more on presenting our algorithmic 2013 bracketology information in new and useful ways.
A few days ago when we released our 2012 preseason college basketball top 25, we hinted that we had an exciting new feature that we were just itching to let loose into the wild. Well, here’s what we’re so amped about:
We are now simulating the entire college basketball season every single day, all the way from November through April. Our projections now include:
- Conference tournaments
- NCAA selection and seeding
- The NCAA tournament itself
This means that every single day, you can count on TeamRankings to deliver intelligent, up-to-date, algorithmically derived odds for thousands of future college basketball outcomes, from Kentucky’s chance to land a 1 seed in the NCAA tournament to Grambling State’s odds to make the 2013 March Madness bracket.
Last week we added a feature that’s been on our to-do list for a long time: box scores for all the sports we cover. It’s nothing groundbreaking, but it’s a small step toward exposing as much information as possible.
For the most part, these box scores should look similar to those you see on other major sport sites or in your daily newspaper. (Those still exist, right?) All the basics are there. However, we have added a few pieces of valuable info to each sport’s box score that you won’t be able to get from your typical ESPN report.
In our Monday update post about new college basketball stats and other assorted goodies, we mentioned that we’d be adding player stats for the past few seasons.
Well, our Compy 386 is finished running the numbers, and the stats from the last five years are now available for your perusal.
Our NCAA basketball player stat rankings now go back to the 2006-2007 season. Check out blocks per foul, for example. Make sure you check out the various splits that are available, like performance against Top 50 teams, or in away/neutral games. (Though unfortunately the ‘Last 2 Weeks’ and ‘Last 4 Weeks’ filters will only work once the season starts.)
Also, player pages now have career stats which should go back far enough to cover the full careers of all current players. Here’s preseason All-American Jordan Taylor modeling our new page. [Side Note: That is some seriously impressive improvement in all his shooting percentages from freshman to junior years. Dude must have lived at the practice court.]
So, dive in, unearth some nuggets of wisdom, and pass them on.
The college basketball season starts in only three and a half weeks, which means it’s time to start preparing that section of the site for the upcoming year. We buy it some new back-to-school clothes, make sure it knows where all its classes are, and help it stock up on laundry detergent and quarters.
The first step this year was to add a few new team and player stats, and to make sure the improvements we’ve made to other sports are migrated over to the college basketball side. Here are some features we rolled out today.
Over 20 New College Basketball Team Stats
Some of these are probably familiar to you, but a few won’t be. The more unique stats are defined below:
Assists per Possession [Assists / Possessions]
Effective Possession Ratio [(Possessions + Offensive Rebounds - Turnovers) / Possessions] — This measures how good a team is at actually getting scoring chances out of their possessions. Turning it over costs them a chance, while grabbing an offensive board gains them an extra one. Higher is better.
Extra Scoring Chances per Game [Offensive Rebounds + Opponent Turnovers - Opponent Offensive Rebounds - Turnovers] — Teams always get roughly the same number of possessions. But through rebounding, ball handling, and pressure defense, one team can gain more true scoring chances than the other.
Floor % — Estimates the percentage of possessions on which a team scores at least one point. This stat is key in late game situations. Formula and concept by Dean Oliver.
Free Throws Made per 100 Possessions [100 * Free Throws Made / Possessions] — Formula and concept by John Ezekowitz.
Percent of Points from 2 Pointers [2 * Two Pointers Made / Total Points]
Percent of Points from 3 Pointers [3 * Three Pointers Made / Total Points]
Percent of Points from Free Throws [Free Throws Made / Total Points]
Personal Fouls per Possession [Fouls / Possessions]
Steals per Possession [Steals / Possessions]
Total Rebounding % (Rebound Rate) [Total Rebounds / (Total Rebounds + Opponent Total Rebounds)]
True Shooting % [Points / (Field Goal Attempts + 0.475 * Free Throws)] — This measures how successful a team is at converting scoring opportunities into points, taking into account both three pointers and free throws.
Turnovers per Possession [Turnovers / Possessions] — Some of you may know this as Turnover%. Read more »
It’s update time again! This week we have one big item and one small one.
The big one is something we’ve been developing over the summer — a new power ratings system. It’s not in a completely finished form yet, but it’s good enough that we thought we should let it see the light of day.
The small item is our NCAA Football Polls Comparison page. Read on for a few more details on both.
New Power Ratings
Our existing Predictive Power Ratings work great, but we’re always looking to improve our analytic tools, so we created what we believe is an even better rankings system that we hope will eventually replace our current one.
There are a few main differences between these and our regular ratings:
First, the New Rankings incorporate our preseason projections. For the first few games, the preseason ratings are a big factor, but as the season wears on, they will drop out entirely. This reduces the crazy ratings that you sometimes see in the first couple of weeks.
Second, these rankings are on a different scale: zero equals average, and the rating indicates how many points above or below average a team is. For example, a team with a Predictive Ranking of 10 is expected to beat an average team by 10 points at a neutral location. We think setting the average to zero makes the ratings easier to interpret.
Third, these new rankings should be more predictive than our current ratings. They performed better in our initial analysis, but before we take the big step of replacing our old ratings, we want to do more testing.
The Predictive Rankings are the centerpiece of this new set. These can be used to forecast the winner and score of future games. The rest of the new rankings are derive from these Predictive Rankings.
Most of them are simple splits. For example, the Away Rankings show performance in road game, and the In-Division Rankings rates a team based on how they’ve played against division foes. One of our favorite new splits is the Vs 1-5 Rankings — this shows how a team has fared when playing against the best teams in the league.
There is also a set of Strength Of Schedule rankings (SOS), which indicate how difficult the opponents of a team have been or will be. Beside the normal SOS rating that shows how difficult a team’s past opponents have been, we also have Future SOS, Full Season SOS, and a few other flavors.
Finally, there are two unique ratings. The Luck Rankings compare the number of actual wins a team has earned to the number of wins expected for a team with the same rating.
The Consistency Rankings are simply the standard deviations of the individual game ratings of each team. There, a lower value means a team has been more consistent.
NCAA Football Polls Comparison Page
Here, you can compare the AP Poll or Coaches Poll to our TR Predictive Rating, letting you see which teams are getting too much or too little respect. You can also see the biggest gainers and losers in the polls, and in our ratings.
It’s a quick one-stop shop to get an overview of the college football landscape.
So, please, take a spin through the new features — especially the new ratings — and let us know what you think. It’s your feedback that helps us improve existing tools and add new ones.
Site Update: New TR Pick Logic, 30+ New Football Stats, New Football “Likely Score”, and Earlier Pick Freezing
It’s been a month since we released a notable update to the site, mostly because we were focused on polishing The Predictor. But over the past few weeks we’ve made quite a few changes that should add a lot of value to the site. Check out this list of great new features.
Improved Logic for Official TR Picks
In the past, the logic behind our official TR Picks essentially just copied the pick of whichever model was most reliable in the situation at hand. Starting this week, for NFL and college football, we have a new, smarter algorithm for Game Winner, ATS, Money Line, and Over-Under picks:
- NFL point spread picks
- NFL over under picks
- NFL money line picks
- NFL game winner picks
- College football point spread picks
- College football over under picks
- College football money line picks
- College football game winner picks
We analyzed our historical pick accuracy data to figure out the optimal combination of the predictions of our various models. Now most picks will be a weighted average of multiple models.
Previously, if the Decision Tree model projected 50.1% odds for a team, but the other models both heavily favored the opponent, there was a good chance our logic would still side with the Decision Tree, despite that model basically labeling the game as a toss up. With the new logic, the more confident models will get a bit more weight, and can “out-vote” the less confident model.
Along the same lines, sometimes one model can be overly confident in a pick. If the other two have more conservative predictions, those will dampen the impact of the one outlier. Finally, if the models agree, there may be a synergistic effect, where our official confidence is higher than the confidence of any single model.
Just like trying to predict an election by asking one person who they’ll vote for is silly, but asking a million people can get you a great prediction, combining our models leads to a smarter prediction that listening to any single one. Read more »