FAQ: Team Rankings Predictions & Picks

posted in > Site Help, College Football, MLB, NBA, NCAA Basketball, NCAA Tournament, NFL

How do your models work?

For descriptions of all the models we use to make game winner, spread, totals, and money line predictions, see our Models, Models, Everywhere blog post.

Why are some of your model predictions listed in green, and others in red?

These colors indicate whether the model’s prediction agrees with our Official TR pick. Predictions shown in green are in concordance with the Official TR pick, and those in red are in disagreement. We feel that color coding the outputs helps users see with a quick glance whether there is consensus among our models, or if there is a split decision.

How do you determine your “Official TR pick” if the models don’t all agree?

For each sport and pick type, we use a technique called logistic regression to combine output of all our models. This technique determines how much weight each of our models should get, in order to maximize our prediction accuracy. In some cases, the Decision Tree rules all, but in other cases, giving partial weight to other models helps. For more information, see About Our Picks.

What do the Confidence star ratings mean?

Our Confidence star ratings tell you how likely we think it is that our prediction will be correct. They are based on precise numerical values (see “What do the Odds column represent?” below), not some wishy-washy gut feelings, and they mean different things for Game Winner picks and for betting picks. For details, please see About Our Picks.

How do you determine the Confidence star rating of your Official TR pick?

The Confidence star rating of an Official TR pick is determined by how accurate we’ve been in the past when predicting games similar to the one in question. We define games as similar based on our projections for them. Those similar games are some of the inputs to the regression model, meaning they will influence the model to make its future estimates closer to the actual success rate that we had on the old picks. For more details on our Official TR picks, see About Our Picks.

What do the Odds column represent? (or Value column for money line picks)

The Odds column is the numerical value upon which the Confidence stars are based. This is the output of the logistic regression that combines info from all our models. It’s our best guess for a team’s win odds or chance of covering, an over/under’s likelihood, or how far above the money line break even point a team’s estimated chance of winning is.

When do you lock your picks?

That’s actually kind of a complicated question. We stop making new predictions 2 hours before a game starts. However, those predictions are initially only in our database. Because our site update scripts (which make sure the new predictions are displayed on the site) run in cycles, it could be anywhere from 5 minutes to nearly two hours before a new prediction is displayed. This means that there is no set lock time for the predictions you see, other than “before the game starts.” Actually, in our ideal world, our models would continue to update right up until tipoff or kickoff, so they could react to last minute line changes, but that’s not technically feasible at this time. Finally, please note that each game locks on an individual basis, according to its own start time — not en masse prior to the start of the day’s first game.

Why is the Game Winner pick sometimes different than the Money Line pick?

A Game Winner pick is simply the team we think has a greater than 50% chance to win the game.

A Money Line pick tells you which team we project to be a better value bet, using the money line odds.

For example, if you bet $200 on the Yankees as -200 money line favorites over the Royals, you would win $100 if the Yankees win, and lose $200 if the Yankees lose. That means you would need the Yankees to win 2 out of every 3 times in order to break even on your bet over the long run. So, if we project the win odds for the Yankees to be greater than 67%, they would be our money line pick. On the other hand, if we project the Yankees’ win odds to be less that 67%, our money line pick would either be the Royals or LAY OFF (depending on what the Royals’ odds are). Let’s say we project the Yankees to have a 60% chance to win the game. The Yankees would be our Game Winner pick, but the Royals would likely be our Money Line pick.

Why do your game winner and ATS picks sometimes conflict?

Basically, because they are the outputs of two different models. We have two separate versions of each of our models, one optimized for picking winners, and one optimized for picking against the spread. Usually, they give the same results, but on occasion, they disagree, just as our Decision Tree and Power ratings models sometimes disagree about which team will win a game.

In situations where picks conflict, one thing you might want to take into account is how confident each pick is. Picks with a confidence rating of only a single star are close to toss-ups in the eyes of the model. Also, looking at our season results page can give you an idea of how often each model is right in a given situation. For example, our Decision Tree ATS model has been correct on 51.1% of games this year, while our Decision Tree Game Winner model has chosen the correct team 60.0% percent of the time when the published confidence is between 50% and 64%.

In the end, which model should take priority is a judgment call. We always advise you take context into consideration when interpreting our model outputs. In this case, that context is that there is another model in disagreement, so you may want to temper your confidence in the pick.

How have your game winner Predictions done in past NCAA Tournaments?

We keep detailed prediction accuracy records for the past 14 NCAA tournaments, broken down by confidence rating. If you want one number, we’ve predicted 71.3% of games correctly over the past 13 years, though our performance has been slightly better in recent years: we’ve gotten 73.1% correct over the past four years. For the complete breakdown, see Bracket Pick Performance.

How have your game winner Predictions done this regular season?

As with the tournament game winner accuracy, we keep detailed records of our regular season accuracy. As I write this on February 9th, 2012, we’ve predicted 5,159 games, and gotten 75.1% correct. For a complete breakdown by month and confidence rating, see College Basketball Picks Season Results.

How have your betting predictions done this regular season?

We keep detailed records of the accuracy of our past betting predictions since the 2007-08 season, which can be found at our College Basketball Picks Yearly Results page. Over that time, our Official TR picks have compiled a 52.1% record against the spread over all games, and we have a 53.3% record on our three-star confidence Official TR picks.

How have your betting predictions done in past Tournaments?

Our Decision Tree model, which is the primary model used to make ATS picks, only started generating predictions in 2008, so our data doesn’t go back very far. We’ve gotten a lot of requests for our NCAA tournament betting picks record, so we’ll post it here, but please keep in mind that a sample size of ~200 games is quite small, and pales in comparison to the amount of data we have on our accuracy during the regular season, which can be found at College Basketball Picks Yearly Results.

During the 2011-12 regular season, through February 9th our 2 star or better totals picks were 771-608-22 (56.1%, +102.2 units at -110 vig). 2-star or better ATS picks were 681-625-23 (52.6%, +6.5 units at -110 vig).

Here is our yearly performance against the spread during the NCAA tournament:

ATS by year3 Star2 Star1 StarTotal
Grand Total23040.0%3433250.7%9094148.9%126130349.2%

Our performance on totals picks during the NCAA tournament:

OU by Year3 Star2 Star1 StarTotal
Grand Total105066.7%3730155.2%8982552.0%136117653.8%

And our money line results during the NCAA tournament:

ML by year2 Star1 StarTotal
Grand Total262254.2%+9.6508337.6%-34.27610542.0%-24.6

I have another question not covered here…

For answer to any other questions you may have, please drop us a comment below, or email us via the contact link on the right.

  • Tom

    Could I see a spreadsheet of your picks with dates. I want to cross check MLB ML picks with another model I have. I’ve been tracking the last 5 days and it is 5-0. I’m happy to share what I am looking at if you can provide me this. Thanks.

  • http://www.teamrankings.com TeamRankings.com

    Tom –

    We are working on being able to provide a complete historical record of our model picks in a spreadsheet format, but we don’t have it available quite yet. This is something we are working on this summer however.

  • Landon

    Perhaps I am missing something, but I notice that when you initially post predictions, you use the opening vegas lines, but by the time you post them, they’ve already moved, usually in the direction you picked. So the lines you post are often unbettable.

    Why not post predictions based on current lines?
    Or post predictions when lines open so they are bettable?

    What am I missing?

  • http://www.teamrankings.com/ David Hess

    Landon, we have had a couple bugs in our automated prediction system this past week or so. We found the source of the problems, so hopefully this issue will be alleviated.
    In a more general sense — we make predictions on a regular cycle based on the most up-to-date lines we have received from our data feed. Generally, those will be accurate but on occasion there are delays. It’s unfortunate, but unavoidable.

  • http://profile.yahoo.com/DYV4TDV5CFMU42LQNKWHGLYMJY David

    Why is it that there is no data on the performance of 4 or 5-star picks historically?

  • http://www.teamrankings.com/ David Hess

    There are no 4- or 5-star gambling picks. We have no category higher than 3-stars (55%+ projected correct rate). As for 4/5 star game winner picks, they are linked above: 

  • Darrendv

    You cover why the Game Winner and ATS picks sometimes conflict, but can you explain why you would have opposing picks for Game Winner and Money Line picks?  

  • http://www.teamrankings.com/ David Hess

    Darren —

    Game Winner picks are simply the team we think has a greater than 50% chance to win the game.

    Money Line picks are which team we project to be a better value bet, using the money line odds.

    For example, if Team A is a -200 money line favorite, you would need to win your bet on Team A two out of every three times in order to break even. So if we project Team A’s win odds to be greater than 67%, they would be the money line pick. If we project Team A’s win odds to be less that 67%, the pick would eaither be Team B or LAY OFF (depending on what Team B’s odds are). So, if we project Team A with only a 60% chance to win, they would be our Game Winner pick, but Team B would likely be our Money Line pick.

  • Weswoolf


      I am using some of your statistical catagories for a project and need to define the catagories used.  Can you tell me what you use to develop the Overall Team Def % and Power rankings?  Thanks.

  • http://www.teamrankings.com/ David Hess

    I am guessing you are asking TeamRankings rather than Darren, since he hasn’t posted any stats/ratings here. If you really did mean Darren, obviously you can ignore this …

    Not sure exactly what you mean by “Overall Team Def %” but my best guess is Defensive Efficiency: http://www.teamrankings.com/ncaa-basketball/stat/defensive-efficiency That is simply points allowed divided by possessions.

    As for the power ratings, we have quite a few varieties. This series of blog posts explains them fairly well: http://www.teamrankings.com/blog/ncaa-basketball/under-the-teamrankings-hood-part-1-power-ratings-basics For a super quick answer, we look at margin of victory, who the opponent is, where the game is played, and the date.

  • Weswoolf


      Thanks for the help.

  • PitchBlack

    Are you guys thinking about adding hockey anytime soon…

  • http://www.teamrankings.com/ David Hess

    PitchBlack — Adding a new sport is a very labor intensive process, so we don’t have any plans to do so at this time. However, if we were to add another team sport, the NHL would be our top candidate.

  • CCPhilly

    Yeah, I would love if you added NHL.

  • http://goldprices247.com/ Goldie

     IT would be great if you can add hockey.

  • http://silverprices247.com/ Jamescarter

     Totally agree!

  • http://www.teamrankings.com/ David Hess

     Thanks for the feedback, guys. At this point, though, it still doesn’t look like we’ll be able to commit the resources to add another sport any time soon.

  • http://silverprices247.com/ joanna

     I second the hockey!

  • Gianluca

    Would you guys be able to include CFL and NHL into the site? I would really love that!!!!

  • http://www.teamrankings.com/ David Hess

     Gianluca — Adding a new sport is a very labor intensive process, so we don’t have
    any plans to do so at this time. However, if we were to add another team
    sport, the NHL would be one of our top candidates.

  • joe

    great info guys.  regarding the blog post, is there a page with current sweep at home/and away stats?

  • http://www.teamrankings.com/ David Hess

     Joe — Thanks, glad you found it helpful. Due to the very positive response to the MLB sweep data we’ve posted, we’re definitely considering adding a permanent (and auto-updating) sweep data page. If we do add that, it will most likely be an off-season project.

  • Eazy_E

    Why does it show one team on the (ML) MONEY LINE as the dog and the other team on the (RL) RUN LINE -1 1/2 as the favorite? This makes no since to me.

  • EAZY_E

    The game I was talking about earlier was Seattle +102 on the (ML) MONEY LINE and Toronto -120, being -1 1/2 +135 on the (RL) RUN LINE. I don’t understand where you would pick a DOG on the (ML) and then take the FAVORITE on the (RL).
    Would you explain this to me.

  • http://www.teamrankings.com/ David Hess

    Eazy E — See the answer to “Why do your game winner and ATS picks sometimes conflict?” above. This is the same concept, just with money line and run line picks.

  • Stat Geek

     Guys do you have a specific stat on your site that focuses only on the
    middle relief and closer numbers?  Not for the starting pitcher. I want
    to see the rankings for the middle relief and closer rankings of all
    teams. This would be very helpful. If you already have this stat can you
    direct me towards it or inform me if you plan to put it toghether?
    Thanks guys! Keep up the good work.

  • http://www.teamrankings.com/ David Hess
  • Normplante

    iIn the prediction accuracy  breakdown page for nfl ncaa,f what is the all model breakdown based on ?

  • http://www.teamrankings.com/ David Hess

    Norm — If you mean the “All Model Picks” button, that just shows all picks we’ve ever made based on whatever model you have selected at the top. I guess the name is a little confusing — it’s basically “All Picks By This Model” (NOT “All Picks By All Models”).

  • Jak

    What happened with the plan to provide historical picks?

  • oiris

    I noticed your 2* MLB picks for Aug-17 went 2-4, but your prediction accuracy page only changed from 35-30 to 36-31 in August (this is a 1-1), I also checked your 3* for August and you have 6-8, but if you check the last 21 days, it should be 6-9. Are these kind of “innacuracies” about the prediction accuracy page common?

  • PitchBlack

    oiris…sometimes their plays don’t get updated until a couple of days later…I have noticed this…keep watching, probably tomorrow you will see the changes for the 17th…

  • oiris

    So, basically it’s impossible to check the accuracy day to day, since the updating is random?

  • PitchBlack

    Not really random…it’s either on a one day delay or two day delay…but it is definitely accurate…I am not a moderator…just a happy visitor to this site like you…probably they will answer your question tomorrow with more precise info…your looking in the right place though, the Prediction Accuracy charts can show you how their stats work and in what circumstances they work best…

  • http://www.teamrankings.com/ David Hess

    oiris — PitchBlack has it right. We only update the Prediction Accuracy pages once per day, in the early morning, because those pages take a *long* time to compute. Sometimes games from the previous day will make it into the update the next morning, but sometimes it will take an additional day. It just depends on whether we’ve gotten the final box score info from our data provider in time.

  • PitchBlack

    Couple of nights ago, there was a game where San Diego changed their starting pitcher…the best thing to do here would be to lay off because your plays are based on listed starting pitchers…is that what you would reccomend?

  • Sam

    Do you offer the estimates and projections as a (xml) feed, or some other format? It would be a great addition as you could download the data directly to your own prediction tool for further breakdown and analysis.

    Likewise it would be great if you could provide historical data in an easily downloadable format for analysis and process.

    Thanks for a great service!

  • http://www.teamrankings.com/ David Hess

    Yes, there is a good chance that our model would not be able to update to include the new starting pitcher info, so I would recommend laying off in the case of a starting pitcher change.

  • http://www.teamrankings.com/ David Hess

    Sam — No, at this time we only publish our projections and picks on our site. We’re considering a feed or download service in the future.

    Historical data downloads are another service we’d like to offer, but it has kept getting pushed back on the priorities list (plus there have been a few hurdles we still have to clear.

    We’re happy you like the service, and thanks for the feedback. It’s always good to hear directly what you guys are interested in!

  • http://www.teamrankings.com TeamRankings.com

    Jak – sorry for the late reply here. While we had hoped to get it done this summer, this project is currently still on the drawing board. It was a busy last few months and we had to make some tough calls on what to prioritize. Since TR is a full-time business that needs to support itself, that’s the unfortunate reality sometimes.

    With that said, we do realize that several of our most loyal users would like to see complete pick histories (presumably with some ability to sort/filter the data), and that from a transparency perspective having those available makes an important statement. (Although the fact that we currently publish comprehensive prediction accuracy records, updated daily, already puts us at the forefront of transparency.)

    The good news is, we did complete some very important projects, such as making improvements to our prediction logic, and several other things we plan to announce in the weeks and months ahead. I get the sense that a lot of users want to see complete pick history so they can do their own data mining for situational performance trends. Our newest prediction logic updates, which should launch within the next week, essentially do a decent amount of this work for you, and factor the results into our final predictions for games.

    Once we get past the first couple weeks of football season (we’re pretty crushed right now), we’ll revisit this to see if we can’t get something simple up on the site. It’s a fairly popular request and we realize that.

  • PitchBlack

    Do your records take into account when there is a neutral field (or court) game, such is the case tonight with Tennessee and NC State…I nitice you don’t have neutral games listed in your accuracy records…

  • http://www.teamrankings.com/ David Hess

    Hmm, good catch. Our predictive models definitely DO take into account neutral fields, but it looks like our prediction accuracy page does not take that into account when summing up the results. Ill put that on our fix-it list. Thanks for the heads up!

  • bill

    I can’t find the link to creating an mlb model. Please help :)

  • http://www.teamrankings.com/ David Hess

    Bill — We don’t have a section where you can create your own model. I think we did in the past, but that was several years ago (before I came on board).

  • John F.

    My ATS picks pool seem to have different spreads on a lot of the lines, typically .5-1.5 point difference, but I’m assuming that amount may drastically change which side to pick. I realize it locks at different times etc. but is there any way to figure out what your modeling would do with a different spread besides the obvious situations where the spread difference falls in a way that it only makes the case stronger for the same side pick?

  • http://www.teamrankings.com TeamRankings.com

    So we actually worked on this exact project this summer, but will need more time to release it. (That is, you tell us what spread you have on a game, we tell you odds to cover based on our ATS prediction models.) It’s a fairly complicated project as you can imagine.

    In the meantime, for football we typically just use a simple rule of thumb for making adjustments, that a “free point” is worth roughly 2-3% cover odds or so. That value depends on a few things…for example crossing “the 3 or the 7” in football is probably worth a little more. Since most of our point spread winner predictions are in the 55% or lower range, usually getting two “free points” or more is enough to make us switch most picks to the other side in a pick’em contest. For very weak confidence picks, one or 1.5 free points could do it.

    Like I said though, hopefully we’ll have a more sophisticated solution here soon. You are 100% right…factoring in and taking advantage of “free points” in pick’em contests is an absolutely key strategy and a must-do.

  • bh3ka

    how about adding a discussion forum?
    Are the prediction models tweaked/adjusted as the season continues?
    How can I get detailed historical data .. i have a angle which seems good based on last 21 days .. but that’s not really a lot of data.

  • http://www.teamrankings.com TeamRankings.com

    We’ve experimented with a forum in the past, briefly, and the results weren’t great. However, our site is getting much more traffic and interaction now than then, so may be worth another try at some point soon.

    What are you looking for exactly when you say “detailed historical data”?

  • http://www.teamrankings.com TeamRankings.com

    We’ve experimented with a forum in the past, briefly, and the results weren’t great. However, our site is getting much more traffic and interaction now than then, so may be worth another try at some point soon.

    What are you looking for exactly when you say “detailed historical data”?

  • bh3ka

    I am looking for something like last 21 days, but longer. Eg the stats can give 2012/2011 or last 500/1000 etc.

    I realize that it could be intensive on the front end, maybe allow people to submit a request for say MLB, Run Line, 2011. They put their email address and a background job runs, when their request is completed they get the data in their inbox or a link to a url on the site to view it.

    Of course forward testing any angle is 10 times better than back fitting.

  • http://www.teamrankings.com TeamRankings.com

    Not a bad idea, although custom queries like that could be very resource intensive, depending on how many people were doing them, etc. It’s a tricky area, because with a lot of historical picks and infinite ways to slice and dice them, things can get complicated real fast in terms of creating a UI to let a user request EXACTLY what he/she wants to see, and then running a the query. Certainly far from impossible, just saying it would be a major project, which is why we’ve held off so far.

    My guess would be to justify the investment, we’d probably need to make this a subscription service, and the main question is, how many people would pay for it…

  • bh3ka

    Running in the background from a queue with 1or2 jobs active at any one time would limit the resource issues.

    Yeah it is a major project. I would probably think around 0.5% of your repeat visitors might be interested, and even then they may subscribe for one or two months and then stop. For now I will more then to just hear that it’s at right at the bottom of the to do list.

  • bh3ka

    Running in the background from a queue with 1or2 jobs active at any one time would limit the resource issues.

    Yeah it is a major project. I would probably think around 0.5% of your repeat visitors might be interested, and even then they may subscribe for one or two months and then stop. For now I will more then to just hear that it’s at right at the bottom of the to do list.

  • bh3ka

    How about a notification subscription service? I have missed a few picks in MLB that fit my filter due to checking the site frequently enough.

    Say $1 for one sport or $2 for all sports per month. Or $10 or $20 per year.

    The amounts are fairly low so I think the uptake would be decent, and the notifications don’t have to be complex … either the information or simply “MLB Moneyline picks have changed”, “NFL Spread picks have changed”, etc.

  • PitchBlack

    bh3ka…if you check about an hour before each event starts you will get 99.99999% of the updates…I used to miss one here and there so I started checking at 55 minutes before each event starts…when events start close together you can get a bunch at one time…since I started checking in this manner haven’t missed one play that matched my filter…and that’s over a year now…

  • bh3ka

    Agreed .. but looking at today’s MLB.

    1pm , 4pm, 7pm & 8pm is a bit of a mission. Add to that that I am +5 hours it kind of compounds the problem.

    It’s just an idea … and $10/$20 a year is not a lot to pay.

  • http://www.teamrankings.com TeamRankings.com

    I do think the notifications service is an interesting idea. Set your filters, get notified when things match or when things move from a match to a no-match. Business model is a different question, though. That’s pretty specialized functionality — you need to be a big follower of of picks already, and have set up your own systems around them, and decide that you’re willing to pay for a service like that. If it ends up being a few hundred people at $20 a year, it’s just not worth building.

    We’ve run into this challenge when building out some more advanced functionality in the past — we once built a tool that let users create their own predictive models based on stats they chose, and enabled them to backtest their custom models. The hardcore users loved it, but they’re a small minority. So if we can’t get them to see serious value and pony up serious bucks, the overall market to do it is too small.

    Real-time odds services, for example, charge bettors $100-$500 a month for their product — at that level, it changes things. But serving a niche audience with a low-priced product isn’t a recipe for a successful business, no matter how cool we think an idea would be…

  • ludger

    In all your game prediction modules a drop down box with explanations appear at the top a of each column whenever the column is pointed, this drop down box, however the drop down box prevents clicking the” Deciion and Ratings” boxes for sortable data from these columns. Could you have the drop down boxes for Decision Tree and Ratings column drop down a little higher to allow access to click forsortable data ?

  • http://www.teamrankings.com/ David Hess

    Thanks for the heads up. We didn’t realize this was happening in some browsers. We’ll add it to our bugs list, and we’ll get it taken care of as soon as we can.

  • tresmang

    Im trying to figure out why your odds are so different than my books odds…..a 1 point to 2 1/2 pt difference changes a lot in terms of ATS….so your models will more than likely change based on these variables….

  • http://www.teamrankings.com TeamRankings.com

    We use odds from Pinnacle Sports right now. It’s certainly the case those lines could differ from the ones you get, especially if the payout odds are different. We’ve got a few things in the works here though that should help address this…don’t want to promise anything now, but stay tuned.

  • PitchBlack

    What’s the deal with NCAAB Totals…is it just me or are the sportsbooks putting up way less totals so far this season…seems last year Every game had a total…this year hardly any have one…you guys know what’s up with this?

  • http://www.teamrankings.com/ David Hess

    Yeah, we’ve definitely noticed, and we’re not too happy about it, since NCAAB Totals are probably our best pick type. We don’t have any inside info about what’s going on with that, unfortunately.

  • Joe

    What was the strength of schedule at the start of the year?

  • http://www.teamrankings.com/ David Hess

    You’re going to have to be a bit more specific… what sport, for example? If you’re looking for college basketball, here’s a link: http://www.teamrankings.com/ncaa-basketball/ranking/season-sos-by-other?rating_date=2012-01-09

    You can always find ratings as of any date in the past by using the calendar button above any ratings table.

  • PitchBlack

    The last couple of days my sportsbook has had totals on every single NCAAB game…but your predictions only list plays for some of the games, just wondering when your site will be listing predictions for all of the games that have totals available.

  • Batman

    How do you determine your upset odds?

  • http://www.teamrankings.com/ David Hess

    I take it your sportsbook is BetOnline? Sorry, but for now our official line provider is Pinnacle, so until we get a chance to re-fit all our models, or until Pinnacle starts releasing a lot of totals lines again, we’re going to be stuck with whatever they release. … Updating the models is definitely something on our to-do list, though.

  • http://www.teamrankings.com/ David Hess

    Check out the very first answer on this page. :)

  • gumsy

    hi. i notice you mention pinnacle as your “reference” in terms of spreads. when computing your results against the spread in the various sports you cover, which pinnacle number do you use—-the close , the open, 1 hour to game, etc. ? also, do you use their juice or do you assume their line with a -110 pricing structure ?

  • http://www.teamrankings.com/ David Hess

    We use the spread as of when we “lock” the pick, which occurs approximately one hour prior to the game start. And, we assume a -110 pricing structure. Based on a recent evaluation, we think this is not ideal, and one of the projects on our to-do list is to update the prediction accuracy pages to use the true juice for every pick. (We are also considering using additional sources besides Pinnacle, but that’s another topic altogether.)

  • David

    I’m interested in NCAA basketball games. When a player gets injured and is out “today”, but played in the last game, is that information generally included in the Predictive Power Rating?


  • http://www.teamrankings.com TeamRankings.com

    No, it’s not. The predictive ratings are driven by margins of victory and home/away/neutral status of each game. Our New Ratings also factor in a preseason rating for each team whose impact slowly fades away as the season goes on.

    There are no explicit adjustments made for injuries, so a case where, for instance, a star player gets injured the day before a game won’t cause any adjustments. That’s why our game prediction models also incorporate data factors like the Vegas odds — if a player gets injured, our ratings won’t pick up on it, but the betting markets should do a pretty good job of valuing the injured player’s absence in terms of expected points lost.

    It’s also true, as a result, that cases where, say, a team’s star gets injured for 10 games (during which his team plays a lot worse) and then comes back healthy for the rest of the season, can definitely distort the predictive ratings for that team. That’s always something to watch out for.

  • Adrian M

    Hey Guys – first off, thanks for the work you put into this website – it’s amazing 😀

    On to my question: Is there any chance you could add more granularity to the output of your Game Winner model predictions? Currently the site lists performance ranges in clumps of 50-64%, 65-79% and 80%+…What I’d be interested in seeing is performance in smaller increments, like 50-55%, 56-60%, 61-65%, etc.

    I’ve been tracking the output of the models myself and have seen some interesting results…for instance, Decision Tree outputs on NBA Game Winners between 50-55% are correct 59.7% of the time, whereas outputs between 61 and 65% are correct 73% of the time. Unfortunately, I’ve only been able to track this data since Nov. 13/2012 which is a really small sample size – I’d be really REALLY curious to see how the Decision Tree (and the other models) have performed historically in the same detailed performance ranges.

    I realize changing your website to accommodate this is probably too much to ask, but even if you guys could run a one-time calculation in those sorts of detailed ranges (say on NBA game winners) and then post the results on the NBA home page under “analysis and updates” that would be hugely appreciated – it would give some validation/context to the results I’m seeing and allow me to move forward with a little more confidence.

    Thanks again, and keep up the great work!

  • http://www.teamrankings.com/ David Hess


    First, thanks for the kind words; we’re glad you like the site.

    Second, providing a better breakdown of our historical pick data is definitely something on our to-do list. Ideally that would include both more granular summary data like you’ve requested and a record of individual past picks. However, we’ve got our hands full right now with keeping the site running, staying on top of our bowls and NFL playoffs coverage, and preparing our March Madness analytics. So while I’ll add your data-dump idea to our list of possible blog topics, I can’t give you an estimate on when or if we’ll get that done. It could end up being an offseason thing.

  • Adrian M

    Thanks for the reply David. I’ll continue to track the model outputs in that granular fashion on my own time and I’ll keep my fingers crossed that it might be something you guys can get around to. I’m exclusively a money-line bettor and the win/loss % of your output models weighs heavily into my risk analysis and money management strategies (ie. if the model and combined odds indicate a 60% win, I’ll bet up to -150 but not higher (similar concept to the money-line picks you post, though I don’t flip to the dog quite as often as you guys do based on expected value – I just lay-off if the line moves past my acceptable risk limit).

    I’ll keep my eyes on the blogs – thanks again!

  • http://www.teamrankings.com/ David Hess

    Interesting comment about flipping to the dog. I have noticed (over, similarly, a small sample size), that we seem to have a higher ROI on dogs where we predict an outright win, as opposed to dogs where we merely predict a higher-than-Vegas-says chance of winning (even after accounting for our expected edge). This off-season I’ll definitely be checking out a change to our money line logic that puts a much stricter requirement on the win odds being close to or above 50%.

    I am really hoping we’ll be able to get some more granular data up next summer (MLB is our “slow season”). If you continue to track, and see anything startling, feel free to leave a comment or email. It might spur us to rank this higher on the priorities list.

  • Lehigh75

    Just looking at next 7 days NBA Game Winners and noticing that Decision Tree is picking unusually high amount(and %…60.0+) of big underdogs. I know that Decision Tree is quite esoteric but is there any chance there’s an error here or any possible understandable explanation.
    BTW-I’m new here and fascinated!!!

  • http://www.teamrankings.com/ David Hess

    I think the main issue here is that one of the inputs to the Decision Tree model is the Vegas spread. When we don’t have a spread for a game (which we don’t for most of the games on the “Next 7 Days” page), we substitute our power ratings prediction as input instead. Our power ratings alone are historically not quite as accurate as the Vegas spread, so that’s going to hinder the accuracy for those far future games. Once we have a spread in the database, most of those will shift to something more plausible.

    Though now that I think about, how are you defining “big underdogs” in the Game Winners future picks, if there is no line yet? Just by record, or your personal guess, or is there some sportsbook releasing very early lines?

  • Lehigh75

    David-My “big underdogs” comments was just my gut feel and shouldn’t really change much if using power ratings. Look at those 4 star game winner picks: Cha @ S.A., Cha @ Mia, Was @ S.A., Cha @ Hou…….there must be something big in those Decision Tree calculations to warrant picking those away teams to win straight up with 60-70% prob. Even the next six games on the list(all 3 star high quality teams picked to win over low-rated away teams: all Decision Tree picks are the lower-rated away teams with 60-70% prob.
    You have to admit it looks odd(even with power ratings instead of spreads).

  • http://www.teamrankings.com/ David Hess

    Yes, well, that’s why our official TR Pick for those future games ignores the Decision Tree model. :) When we don’t have a spread, we base the official prediction on our power ratings.

    I may be wrong about how the DT model treats games with no spread. It may be using a fake spread of “0”, which would explain the strange odds. Generally, if that game *really* had a spread of 0, it would be because something terribly weird was going on (Charlotte trades for LeBron James or something like that), so the optimistic values could be justified.

    Anyway, as always, our TR Pick is what we believe to be the best prediction we can make at the time, with the given info. SOrry about the confusion!

  • Lehigh75

    David-No reason to be sorry. My interest is purely academic. I thought my question about the Decision Tree results for these games would be answered by someone who knew the model and could drill down to determine the reason for the unusual(IMO) choices. Even a “fake” spread of 0 would just have a neutral impact and wouldn’t, by itself, produce all those “lower-ranked choices”. I just thought someone from your team who was initimate with the decision tree would also be curious to see what factors were the primary cause for those choices. I would love to see the details.

  • http://www.teamrankings.com/ David Hess

    Oh, I am definitely curious, and have access to the model. The problem is that there is no quick/easy way to drill down and determine the reason. The “Decision Tree” label is a bit of a misnomer — the model is actually a random forest (http://en.wikipedia.org/wiki/Random_forest) which combines results from hundreds of individual trees. So it’s tough to put into words why a specific change in the data would have a specific effect.

    It would be possible to look at every individual tree and determine how the prediction changes when a line is changed from something reasonable to 0, and what other tree nodes become more or less important. But it would be difficult and time consuming, as the tree logic is stored in a format that is optimized for machine processing, not for human understanding. I would love to take a day or two to investigate, but when the only real positive to come out of it is to satisfy my/your curiosity, it’s tough to justify the time.

  • twjd

    I can’t believe y’all are charging for your picks, the last time I checked most of your picks were around 50% at best.

  • http://www.teamrankings.com TeamRankings.com

    twjd, I have no idea what exactly you’re checking, but obviously we have winning and losing streaks. All we care about is long term track record. We’ve made over 10,000 playable betting picks in the last 4+ years across all sports, and been profitable against the Vegas lines. Of course, in some sports, we’ve done better than others in recent years, which isn’t really a surprise to us. Users can decide for themselves how to apply our data, and it’s been fun to see all the different ways they use it and find value in it.

    If you’re looking for a guaranteed profit for every sport, every season, and every betting pick type, there are plenty of snake oil salesmen out there willing to sell that illusion to you.

  • PitchBlack

    Been a long time user here, and have been very complimentary towards everyone who runs this site…however have noticed a difference lately in the way the site is being run, couple of flaws here and there…updates not being handled as they once were…pick table many times saying no choices available…and then the answer to the last individuals question (twjd) was very sarcastic and in poor taste (the snake oil statement)…something is different around here and it kind of stinks…and now we have to pay for the info and put up with the new attitude…

  • PitchBlack

    Plus if you played every pick (10,000 across all sports)…you would not have made a profit…52.38% is the number you have to beat…

  • stockjockey

    Actually TR is profitable on playable picks across all sports last 4+ years. It’s there for everyone to see. They have their stronger sports and if you filter through the data, you can find solid situations to increase profit. It’s laughable to hear some people complain about paying a fee for their models which is very reasonable I might add. I’m just surprised it’s taken them this long to implement a subscription service. Keep up the good work TR.

  • PitchBlack

    stockjockey…I guess it depends on what you consider playable…I took it to mean all there plays that they have selected, which if you add it up, does not show a profit…also If you read my comment you would see that I was not complaining about paying a fee, but rather having to put up with the change in attitude, and the errors that have been occurring lately…many times I go to make my selections and it says no picks available…also NBA columns in the NBA section are only accounting for the last twp seasons and this season when it clearly states since 08-09 in the prediction accuracy section…plus if you go to NCAAF in the prediction accuracy section and try to see the plays since 2009, there is nothing there…

  • http://twitter.com/bugaha11 Dan Hoffmann

    I’m entered in a last man standing contest against the spread for the NCAA Tournament. What would be the best way to use your site to do this? I have 5 entries.

  • http://www.teamrankings.com/ David Hess


    What are the rules? Do you have to pick a game on Tuesday? If so, I’d just go with whatever our top ATS pick is the first couple of days:


    I’m guessing, though, that you don’t start til Thursday. Do you make a pick every DAY or every ROUND?

    At any rate, the general strategy is:

    1) Try to avoid picking teams that are popular with your opponents. We don’t have public data on who is popular against the spread, but you can find some at a site like Covers: http://contests.covers.com/Handicapping/consensusPick/daily-consensus-picks.aspx?sport=7 … Or you could assume that the same teams might be overvalued n both bracket picking and spread picking, which would mean our public data is helpful: http://www.teamrankings.com/ncaa-tournament/upset-picks/round-of-64/ (see the Difference column)

    2) Try to avoid picking teams that have high odds to advance deep into the tournament. This can probably help you with that aspect: http://www.teamrankings.com/ncaa-tournament/bracket-odds/

    3) Obviously you’d like to pick teams with high cover odds, but really, even a very good pick is close enough to a coin flip that the rules above are more important. However, her is our ATS picks page: http://www.teamrankings.com/ncaa-tournament/ats-picks/

    4) Diversify your picks. With 5 teams, spreading your picks out will decrease the chance that ALL your entries get eliminated. It’ll decrease the odds of ALL surviving, but that’s a trade off you’ll probably want to take in the early rounds.

  • baldydog

    we do a seed times the round to determine points. is there an algorithem for that type pool.

  • http://www.teamrankings.com TeamRankings.com
  • Darren Wise

    What stats do you consider for your models?

  • http://www.teamrankings.com/ David Hess

    Basically for every team stat you can find on our stat pages (examples: http://www.teamrankings.com/ncb/stats/ ), we use full season stats, rolling averages over a the last X games, last season stats,and other splits. The same goes for our power ratings. Plus we use contextual information like distance traveled to the game, and the spread.

  • Drock

    I understand that your models are running new projections throughout the day and can change up until game time but what causes the projections to change. Assuming no roster changes I was just curious what you look for or inputs can change? Thanks I love all the numbers you guys provide.

  • http://www.teamrankings.com/ David Hess

    Most prediction changes are due to line changes. But they can also change due to power ratings changes (since every team is interconnected, results early in the day could theoretically affect games later in the day) or stats changes (our data provider occasionally updates a box score either to fix a small error, or because the leagues sometimes change the official stats the next day). Finally, if a starting pitcher for an MLB game is scratched, that can result in a change.

  • PitchBlack

    Wow…your numbers have gotten ugly since you went pay…NCAAB and baseball WAYYYYYYYYYYYYY down! What’s up with that!

  • http://www.teamrankings.com TeamRankings.com

    Variance? We had our first losing season in six years with our playable NCAAB ATS and totals picks. Playable NCAAB totals picks still up an average of more than +30 units a year.

    As for MLB, we’re one month into the season. Guess I’m not sure what your point is here.

  • PitchBlack

    Now you are 4 months into the season and over 50 units down on your playable ATS, Total, and ML plays for bases…looks like the books have caught up to your plays, that would be my point here!

  • http://www.teamrankings.com TeamRankings.com

    Our predictions aren’t based on some static formula that the books can “catch up” to and then we say, “Oh no, they figured out our magic formula!” Beating Vegas is a never ending quest — the markets change, the dynamics of the games being bet on change, we adjust our approaches, repeat. Sometimes we win, sometimes we lose, and the goal is to win more than we lose in the long term.

    As for your point about performance, I think we’ve already discussed that. Based on how we happen to publish our pick accuracy (all picks we make, measured by the result of the final prediction we make for a game, usually 1-2 hours before game time), you think our recent performance indicates that our betting picks are bad now — we’ve lost our touch, Vegas has caught up to us, etc.

    We disagree and think you’re making a judgment based on limited data and a short term view, without a great understanding of the potential impact that variance can have on prediction results.

    Great, let’s agree to disagree.

  • Keurig Jones

    Is everything on this site now pay?

  • http://www.teamrankings.com TeamRankings.com

    Keurig, our general team, player, and matchup stats, rankings, data, and blog articles are free. That’s 95%+ of the pages on our site.

    Two types of content require a paid account:

    1) Our office pool picks and analysis, for NFL pick’ems, college football pick’ems, college bowl pick’ems, NFL survivor pools and March Madness bracket contests

    2) Our betting related predictions and analysis for the sports we cover (NFL, NCAAF, NBA, NCAAB, MLB)

    This is the first season our office pool picks and analysis for football will be a pay-based service, though; I assume that’s what prompted your question.

  • Brian

    I thought a few years back there was the ability on your site to build your own predictor model using certain stats, etc. What happened to this?

  • http://www.teamrankings.com TeamRankings.com

    Good memory, Brian. Yes, we did, it was a customizable interface where users could build their own prediction model (based on our Similar Games model) by picking stats they thought were important, and assigning importance weights to each stat. At one point you could even back test your custom model from a game winner, point spread, and totals perspective for the last 100 games! It was cool stuff.

    We eventually just sort of let that product fade away when we made some technology upgrades to the site. It was the type of thing where a portion of our users absolutely loved it, but at the same time it was pretty complex functionality that the tool didn’t really have a super broad appeal. That was also back at a time when we were really trying to build our traffic, so our conclusion back then was that the product was too complex and too niche to be exciting to a lot of people.

    We’re in a different place now business-wise, though, and I’ve definitely had thoughts of resurrecting/updating this tool in the future. The code is old now and we’d need to do some major surgery even just to get it to work exactly as it used to, but maybe we could try to get some sort of test version up and see how users react now. We’ve also learned a lot more about building products and interfaces, so I think we’d do a better job with the product design side of things now too.

    I take it you liked the tool?

  • RCArtsy

    Is there a secret to printing one of your pages? I cannot get it to print.

  • http://www.teamrankings.com TeamRankings.com

    RCA — up, up, down, down, left, right, left, right…sorry, got that confused with my old Nintendo secret codes there…

    Unfortunately, the site could be a mess to print right now depending on what browser you’re using and what page you’re trying to print. It’s just not something we get a lot of requests for these days, so it’s tough to justify spending a lot of time making sure all our pages print very well.

    What page are you trying to print?

  • Chris

    Ur college football stats are way off.. Correct me if I’m wrong.. I’m comparing it to sports reference.com… I would really love to buy the football package.. Only thing holding me back is the football team stats is wrong.. Rushing receiving stats

  • http://www.teamrankings.com/ David Hess

    A lot of the differences you see are probably because we don’t count games against non-FBS teams in our stats. Our models use the stats as inputs, and giving teams credit for destroying FCS opponents can skew the predictions.

    There are bound to be one or two mistakes besides that, simply because at some point along the line, human hands (not ours, but somebody that works for the colleges, or our data provider) have to type in every piece of information. But we have data integrity scripts that alert us to “weird” values (like a player getting 20 sacks in a game), and if we are alerted to any errors, we make sure they get corrected.

  • Chris

    I understand.. Makes sense.. One more thing though.. Predictions grid.. I noticed the 5 star game winner picks is not in order.. It usually shows from 5 star(top of page) to one star(bottom of page).. Can you guys fix that.. It’s easier to look at….thank you!

  • http://www.teamrankings.com/ David Hess

    If this is the page you’re talking about, I’m not seeing any problem:

    Can you email a screen shot to support AT teamrankings DOT com? Also, if you go to this page and press the blue “Send Details” button, that would be helpful: http://bit.ly/1djI353

    Also, on that page, you should be able to click the column headings to sort, so even if it doesn’t load correctly be default, you can sort it by Game Winner:

  • Chris

    Yup that’s the page predictions grid..the game winner stars are not in order..even if you click the column to organize it

  • http://www.teamrankings.com/ David Hess

    Chris, I was able to reproduce the bug, and I think I fixed it. Please give it a try, and let me know if you are still having problems.

  • Chris

    U fixed it.. Thanks

  • nick andriotis

    please explain official tr pick ny giants -1 and then power rating of +8 carolina in red?what does this mean in numbers? thanks Nick A.

  • http://www.teamrankings.com/ David Hess

    The Power Rating column shows the prediction of the Power Ratings model. +8 Carolina would mean that the Power Ratings model favors Carolina to cover the listed spread by 8 points. If it’s listed in red, that’s because the Power Ratings model prediction conflicts with the official TR Pick.

  • Jay Stokes

    How do you manage changes to your methodology? I imagine there could be tweaks and improvements. Do you lock things down and implement changes in batches? Or do you just keep adding changes? Whatever your policy is, how do you communicate it and how does one manage conflicts with historical data.



  • http://www.teamrankings.com/ David Hess

    We generally update the models each offseason by adding new data to the training sets. We occasionally make larger changes to the way we make predictions, which we announce in blog posts. For example: http://www.teamrankings.com/blog/site-updates/updated-mlb-pick-logic

    For historical data, we always show our prediction that we made at the time of the game. We do not go back and update predictions with the new models.

  • Jay Stokes

    Thank you for your reply. I have done a cursory search, but have you made any material changes to your NFL model for this season?

  • http://www.teamrankings.com/ David Hess

    Nope. We’re using the same framework as last season — basically what was outlined in the MLB post I linked.


    college football power ranking? does the power ranking factor in the Schedule of Strength ?

  • http://www.teamrankings.com/ David Hess



    whats the difference yard per point and yards per margin…also thanks for the fast response!

  • http://www.teamrankings.com/ David Hess

    “Yards per Point Margin” is [defensive yards per point] minus [offensive yards per point]

  • Bill

    In your scoring statistics, particularly the first half/second
    half/overtime scoring per game in NFL, NCAAFB, NBA, NCAABB, do your
    scoring figures in the second half include overtime or are these three
    mutually exclusive categories so that overtime scoring is not reflected
    in the second half figures as well as being their own category at the
    same time?

  • Bill

    In your scoring statistics, particularly the first half/second
    half/overtime scoring per game in NFL, NCAAFB, NBA, NCAABB, do your
    scoring figures in the second half include overtime or are these three
    mutually exclusive categories so that overtime scoring is not reflected
    in the second half figures as well as being their own category at the
    same time?

  • http://www.teamrankings.com/ David Hess

    They are mutually exclusive. So any overtime scoring does not count in the second half scoring average.

  • RonG

    I apologize in advance. I know that this has been asked and answered but I’ve looked and can’t find it. What adjustments should I make to what TR shows when there are discrepancies between TR and Vegas odds. Tonight is a great example. TR picks NYG, 3 stars, +9.5 points. Right now Vegas odds are +8. How does that change the TR pick?

  • http://www.teamrankings.com TeamRankings.com

    We use Pinnacle as our book of record on the site, and they stayed at 9.5. Pinnacle’s lines will sometimes be a bit off from Vegas because they tend to move payout odds sometimes before the spread, but in this case, according to SBROdds most offshore books were at 9.5 or 9:


    What we need to do here is just have a tool on the site that translates our projections at a given spread, to odds for whatever spread you’re getting. That’s on our list of future enhancements. In the meantime, rough rules you can go by are listed below.

    For totals, a decent rule of thumb is that each half-point is worth about 1.5%. So if we have the odds of an Over 45 at 55%, and your line is 46, a good estimate for the odds of an Over 46 would be about 52% (so, not a profitable bet at typical -110 lines).

    For spreads, it’s trickier, as there are key numbers that are very valuable:

    — The half-point on either side of 3 is worth about 5%
    — The half-point on either side of 7 is worth about 3%
    — The half-point on either side of 10 is worth about 2.5%

    For most other spreads, you can use a rule of thumb that says a half-point is worth about 1% to 1.5%. … Except around 0. Going from -0.5 to +0.5 is basically worthless, because ties are so rare.

    So, say our spread is -7, and we have the favorite covering 55% of the time. If your spread is -8, you would subtract 3% for the half-point move from -7 to -7.5, then another 1% for the move from -7.5 to -8, which would give you an estimated cover rate of 51% for the favorite (so, not a profitable bet at typical -110 lines).

    Hope that helps.

  • bobby

    in the cfb stats, in the scoring defense area, there is a stat entitled “opponent offensive points per game (ESTIMATED)” —i capitalized estimate. what does this stat refer to and how is it derived ? without an explanation from you, i interpret it to be a (expected) defensive points allowed per game stat, incorporating the strength of offenses already faced, and assuming the defense would face the average college offense this week—maybe the 62nd ranked one—-this is how many points you’d expect them to allow. also, whether i’m way off or not, and you can explain what the stat means—–do your defensive scoring stats account for (not include in defensive points allowed, actual and expected) teams which give up fumble, interception, kick, punt, and other td’s not actually given up by the actual defense ? north texas is allowing about 23 a game on defense , but really only about 18 a game on defense, based on the bevy of return td’s against them. last thing relating to scoring defense ( you think that’s my that’s my number 1 value??!!) , do you tease out overtimes from scoring stats—ytd wise, and on the EXPECTATION model ? thanks

  • http://www.teamrankings.com/ David Hess

    OK, several implied questions here…

    1) Everything in our “Stats” section is simply a record of what happened on the field. There are no opponent adjustments there. The Ratings and New Ratings sections show power ratings, which do take opponent strength into account.

    2) Offensive Points Per Game is simply points per game scored by the offense (so, ignoring defensive and special teams touchdowns). The catch is … we don’t have data on whether each touchdown is followed by an extra point or a two-point conversion (or whether the conversion fails), so we just count every touchdown as 7 points. That’s where the “ESTIMATED” comes in. We’re estimating that a team gets 7 points per touchdown, rather than using the true score.

    3) No, we do not remove overtime stats.

    So, “Opponent Offensive Points per Game (Estimated)” is points allowed per game by a team’s *defense* (throwing out return TD’s, and using a value of 7 points per TD). As you can see, we have North Texas at 18.3, which matches up with what you said above.

  • JEFF

    What would be the formula to compare 2 teams yards per point by the strength of schedule?

  • JEFF

    What would be the formula to adjust yards per point to strength of schedule
    When comparing two teams that’s going to play

  • http://www.teamrankings.com/ David Hess

    We publish the YPP stat as a record of what happened, and we publish the SOS rating as our evaluation of a team’s schedule. We haven’t done any research how those interact, so there’s no set way to compare yards per point *and* account for SOS.

  • Rich

    I signed up primarily to bet your NFL totals predictions (my own model has been unreliable). Here’s the problem: Currently, for instance, 5 of 1pm games are showing as 2 star or above. Conventionally this would say there is value in these so bet your chosen stake size. However, I have noticed these predictions are revised right up until the last hour before kickoff. As per your FAQ you mention this. So what benefit is there in posting lines hours before, when people will follow them and bet them, only to hours later, before kickoff, say these picks are no longer recommended?

  • http://www.teamrankings.com/ David Hess

    “So what benefit is there in posting lines hours before, when people will follow them and bet them, only to hours later, before kickoff, say these picks are no longer recommended?”

    I think this is easier to answer if I first explain WHY the predictions change:

    Most of the time a prediction changes on the day of the game, it’s because the line changed. What line you bet a game at has a massive impact on whether it’s a smart bet or not. In NFL totals, every half point makes roughly a 1% to 1.5% difference in over/under odds. So a single point move could be the difference between, say, a 2-star Over at 52.5% and a 1-star Under at 50.5% (a move of 3%). In other words, these two predictions would not actually be a change in our forecasted total:

    2-star Over 46 (52.5% odds
    1-star Under 47 (50.2% odds)

    There are also cases where the line moves, but then our prediction DOES change in a way that conflicts with the earlier one. For example, if we originally had Over 46 at 52.5% odds, then later we have Over 45 at only 51%. This can happen because the line itself is an input to our models, and it serves as basically a catch-all variable to get a baseline for how the market thinks that things outside the scope of our model are affecting the game. There are cases where our model will predict a game is a good play, because our stats/ratings based forecast is moderately different than the betting market consensus. However, if the consensus moves even further away from our stat/rating based forecast, then that can sometimes be interpreted as a sign that there is some significant info that we’re not taking into account (often an fresh injury, which our stats are blind to), and in that case the prediction might change significantly.


    OK, so that’s why a prediction changes. Your actual question is, what’s the benefit of posting early predictions? Several reasons:

    1) The early lines should be softer (easier to beat) than the late lines. We are just beginning a research project to look at how our initial prediction do, compared to our final predictions. All we’ve examined so far is college basketball spread picks, but in that case, betting our initial picks (if you can find them at the initial lines) is much more profitable than betting the final picks. That’s mostly because the early lines are simply not as accurate. … This is a perfect example where it would make total sense for an early pick (at a soft line) to be rated above 52.4% odds, but a later pick (at a more accurate line) to be below 52.4%.

    2) Some people want to bet earlier, of any number of reasons. Maybe they are working Sunday, so have to bet Saturday. Who knows. So we want to post predictions as early as we can. Those will, of course, be our best predictions we can make, given the current data. Now, suppose closer to game time, the market consensus changes (maybe due to injury info being released), and our model prediction changes. We have two options: A) stick our heads in the sand and ignore the new data, or B) make new predictions that will, of course, be our best predictions we can make, given the current data. If we chose A (which is what it seems you are suggesting), that would be doing a disservice to anybody betting closer to game time, as we’d be intentionally making a worse prediction, just so that people who bet early don’t see the line change. We instead choose to to do B.

    I like to use a weather forecast analogy. Say you’re going on a picnic Sunday afternoon, and you check the weather Wednesday, and it says “clear skies”. Then on Saturday afternoon, an unexpected storm rolls in. If you check the weather again on Sunday morning, should the weatherman stick with his “clear skies” forecast or switch to “rain”? Obviously he should go with “rain”, regardless of what his Wednesday forecast was.

    I hope this made sense, but if anthing wasn’t clear, or you have any questions, fire away.


    Totally unrelated to your question, I want to comment on this line:

    “Conventionally this would say there is value in these so bet your chosen stake size.”

    We would also caution you to pay attention to the actual odds we publish for each game, not just the number of stars. A 2-star pick could be projected with anywhere from 52.5% odds (only a hair above the break even point for a -110 bet) up to 54.9% (nicely profitable in the long run, if the prediction is accurate). Betting the same amount on both those games would probably not be the best long term strategy, if your main goal is growing your bankroll. If you aren’t familiar with the Kelly Criterion, we recommend you check it out: http://en.wikipedia.org/wiki/Kelly_criterion … however, please note that many people suggest using “half Kelly” or “quarter Kelly” systems, where you bet 1/2 or 1/4 the amount recommended by Kelly (this reduces short term volatility, and helps reduce the damage of any errors in the win odds predictions).

    For example, I just used this calculator, and entered the current odds for our 2-star totals picks (and assumed -110 odds): http://www.sbrforum.com/betting-tools/kelly-calculator/

    The result shows that it might make sense to bet far less than normal on some of the games with values barely over 52.4%:

    Obviously there are other reasons to gamble, and if part of your reason is entertainment … well, it’s not very fun to try to figure out optimal bet size. So a rule of thumb is probably good enough. You’ll have to determine what your own rule of thumb is, but betting a full unit on a game predicted with 52.5% under odds may not be a great rule. :)

    Just some food for thought.

  • Rich

    Thanks for the response. Appreciate it.

    Yeah I just read my original q back and realized it was riddled with ambiguity.

    I’ll try to be a little more lucid this time. What troubled me was that I could bet a line, say, on Tuesday (say with a 5% edge) and then by Friday your prediction had changed so dramatically that I was on the wrong side.

    Surely you can understand the premise. A subscriber doesn’t want to be told ‘this bet is +EV’ then two days later told he’s on the wrong side.

    I understand the whole line movement angle obviously affecting your estimation of the value but I was more concerned that other factors would substantially change your total points forecast.

    When, in your opinion, is the ideal time to get down?

    ‘Now, suppose closer to game time, the market consensus changes (maybe due to injury info being released), and our model prediction changes.’

    This, in particular, worried me. You’re not saying that consensus shapes your predictions, are you? After all, Warren Buffet and George Soros didn’t become billionaires because they followed consensus.

  • http://www.teamrankings.com/ David Hess

    “When, in your opinion, is the ideal time to get down?”

    We’ve recently done some research in college basketball that shows our earlier picks have been more profitable than our later picks, almost entirely because the earlier lines are softer than the later lines. We haven’t yet done the same analysis for football (it’s on our to-do list, but we’ve got some other deadlines to meet right now), but my best guess is that the same would be true for football. So earlier is *probably* better, but I can’t say that with confidence at this point.


    “You’re not saying that consensus shapes your predictions, are you? After all, Warren Buffet and George Soros didn’t become billionaires because they followed consensus.”

    Sorry, I was a bit vague there. Let me clarify.

    By “market consensus” I simply meant that we use the latest spread and/or total line as an input to the models, *not* that we actually look at which side the public is betting for or against.

    We include the line for a couple reasons.

    1) We don’t directly incorporate injury info, weather, “motivation”, and other contextual info into the models. Including the betting line essentially acts as a catch-all for that. If the line begins only moderately off from where our stats/ratings-based prediction would be, but then moves further way, that’s often a sign that there is some important piece of info that’s not in the model (QB injuries are a common example in the NFL).

    2) The line is used to group or bin historical games for analysis by the Similar Games and Decision Tree models (more info on those here: http://www.teamrankings.com/blog/ncaa-basketball/under-the-teamrankings-hood-part-4-models-models-everywhere ) … This allows for findings along the lines of (totally made up stat here, just for an example): Road favorites of -0.5 to -2.5 with an extremely low turnover margin have an average MOV of 6 points. I’m vastly oversimplifying here, but the general idea is that there are market inefficiencies related to certain stats/ratings, and how they affect close games as opposed to blowouts, or road teams as opposed to home teams, and they become more apparent if you can bin similar games together when examining the results. Sometimes, a line change will shift a game into a different bin, and the models will have an non-linear response to that.

  • Rich

    Excellent. Appreciate the correspondence. That help’s to clarify a few things.

  • http://www.teamrankings.com/ David Hess

    I just wanted to add that this rule of thumb only works reasonably well for line moves of a couple points or less. The further away from the original line you get, the lower the value of a point. But there’s no easy way to express that in a rule of thumb, unfortunately.

  • Dave

    just curious as to what your calculation is for calculating possessions per game for college basketball? i spot checked a couple of your figures using the well known “FGA-OR+TO+(.44*FTA)” and came up with different answers (off by 2 or 3) in each case. Thanks!

  • http://www.teamrankings.com/ David Hess

    Dave, I believe the commonly accepted FTA multiplier for college basketball is 0.475, not 0.44. See:



    That said, we’re actually using that 0.475 FTA multiplier as a global default for basketball, which was set to that value when our prediction models were originally trained. So you will probably see our NBA possession totals a bit higher than some other sites’.

  • bt

    do your strength of schedule ratings in college football get “recalibrated” weekly to account for earlier season mis-judgements about teams that, it turns out later in the season, are better or worse than was perceived at the time the initial rating given to the game in the sos rating ? one example, to be clear–does the memphis sos get tougher based on playing duke in week 3, a time when your ratings may not have caught up to them. that’s the concept, even if the example may not reflect your feelings about duke in particular— it could be auburn or some other unexpectedly improved or declining team.

  • http://www.teamrankings.com/ David Hess

    Yes, our SOS ratings are always based on our most current power ratings, not power ratings at the time of the earlier game.

  • Sportsbook

    Just having fun. Do you incorporate referee stats into your NBA predictions.


    I have found that this is a big help on total betting.

  • http://www.teamrankings.com/ David Hess

    We don’t at this time, but it’s definitely something on our radar that we’d like to look into incorporating in the future.

  • Dan

    How do you calculate your SOS? How does it differ from other websites calculations?

  • Dan

    I’m especially concerned with college football and college basketball. Excellent site by the way.

  • http://www.teamrankings.com/ David Hess

    Most of the SOS power ratings we publish are answers to the question:

    “If you replaced all of a team’s past games with games at a neutral location against a single hypothetical opponent, what predictive rating would that opponent need to have in order for the team to have the exact same number number of expected wins as they do against their real past schedule?”

    The except is the “Raw SOS” in the New Rankings section. That’s just the average of the predictive ratings of the past opponents.

  • Daniel

    Awesome. Way better than espn’s and cbs’s over-simplified formulae.

  • Daniel

    I assume you are using current predictive ratings for the “Raw SOS” and not the ratings at the time the games were played.

  • http://www.teamrankings.com/ David Hess

    Yes, all ratings are recalculated using our current knowledge of the teams.

  • http://www.teamrankings.com/ David Hess

    Thanks, glad you like it. It’s definitely not for everyone. It makes comparing schedules of equal strength teams fairer, we think. But it makes it tougher to compare schedules for two teams of very different strengths. Luckily, SOS matters less for teams of widely different quality, we think.

  • Daniel

    I see. So how does that fact that the #1 team is going to have a similar win% against the #250 team and #350 team affect your SOS? I really like using SOS come tourney time, but haven’t done so the past several years because most models don’t take something like that into account and will reward playing the #250 team way more than playing the #350 team. By the way, thanks for answering all of my questions.

  • http://www.teamrankings.com/ David Hess

    Yes, that’s exactly the type of situation that led us to design it this way. For the #1 team, The number of expected wins again the #250 team and the #350 team are both going to be very close to 1. So switching one out for the other wouldn’t change the SOS much (for the # team). On the other hand, for a team ranked #300, switching #250 for #350 would be a big difference. So this SOS isn’t great for comparing the SOS of teams with very different predictive ratings. However, it should be better than a simple average if you’re trying to compare teams of roughly similar strengths.

  • Daniel

    Awesome, I don’t mess with large point spreads so that’s perfect. Thanks.

  • JB

    Can you explain your power rating system? Say you were trying to get an advantage in the line would you use the Home pwr. + Home field adv. Pwr. – Away Pwr? Or Subtract Predective pwr Home – Away? I guess what I’m asking is how to use your power ratings. Most sites say add 3 to 4 points for home field but you have your own home field adv rating which is nice because not all home field advantage is the same. But how is that information intended for use? Is that information included in the Home pwr? Is all the information included in the Predective pwr?

  • http://www.teamrankings.com/ David Hess

    We do not suggest using the Home Advantage power rating for predicting games. It’s merely a retrodictive record of what has happened this season. It’s not designed to be at all predictive.

    If you’re interested in what our win% predictions for games are, we suggest you check out our prediction pages. For NCAA basketball, that would be: http://www.teamrankings.com/ncaa-basketball-win-picks/

    If “you were trying to get an advantage in the line” then we’d recommend you check out our ATS predictions. For NCAA basketball that would be: http://www.teamrankings.com/ncaa-basketball-ats-picks/

    To make predictions using the power ratings, we suggest subtracting the away team’s Predictive Power Rating from the home team’s Predictive Power Rating, and adding the published Home Advantage (from the right side of the Predictive Power Ratings page).

    For more info on our ratings, please see http://www.teamrankings.com/blog/ncaa-basketball/under-the-teamrankings-hood-part-1-power-ratings-basics and the rest of the posts in that series.

  • bub

    how does a team have a “true shooting percentage” of over 100% ? you list the Miami Heat as having a “true shooting percentage” of 135% in their win against san Antonio on January 28, 2014.

  • http://www.teamrankings.com/ David Hess

    Yeah, that name is a bit of a misnomer. That stat is more like “True Shooting Percentage” multiplied by 2. So it’s basically points per True Shooting Attempt. See here for more info on the standard definition of True Shooting Percentage: http://www.sportingcharts.com/dictionary/nba/true-shooting-percentage-ts.aspx

    As for *why* it’s still like that … basically just because we created other tools/pages/models that depend on that value, and those were made before we discovered that it was off by a factor of two. So correcting the definition would mean we have to adjust all the downstream code. It’s on our to-do list, but it’s just a time consuming project.

    Sorry for the confusion!

  • bub

    don’t go changing for me (kidding)….as long as it’s explained, it makes sense now. I say just leave it. thanks for the reply, as It was a head scratcher for a stat I happened upon by chance.

  • CalamusAppello

    I LOVE the new rankings list and specifically use Luck and Consistency everyday. Please tell me these individual stats will be available well into the future. They are helping me to make my bankroll explode!!

  • http://www.teamrankings.com/ David Hess

    Obviously it’s tough to predict too far into the future, but at this point we have no plans to get rid of them.

  • Daniel

    Are your defense efficiency ratings in college basketball some combination of points per possession allowed and strength of schedule? Obviously one would want to factor in blowout wins/losses. Anyways, just looking to understand this particular model, mainly whether or not you have factored in SOS and points per possession allowed.

  • http://www.teamrankings.com/ David Hess

    No, the only defensive efficiency we have posted is in the Stats section, not the Ratings section. Everything in the Stats section is unadjusted. It’s not a model, it’s just raw points allowed per possession.

  • Daniel

    Awesome, thanks!

  • IzzoneBill

    When do your stats update? I am a nite owl and use your stats in handicapping games. I like to get my handicapping done as early as possible so I can watch the lines.

  • http://www.teamrankings.com/ David Hess

    We update team stats each morning, but the amount of time it takes can vary based on how many games there were the night before. Generally it’s safe to assume that most team stats from the previous day will be up to date by about 9AM Eastern.

    Player stats also generally update once a day, but the scripts can take a very long time, so it may take a full day or more for player stats from a game to be reflected.

  • IzzoneBill


  • Charles

    What formula do you use for calculating your NBA Team Possessions Per Game?

    The formula that I’ve seen used most often is

    Possessions = FGA – OffReb + TOV + (.4 or .44 x FTA)

    However, when I do the math, I never seem to get the same result as your “Last 1” statistic..

  • http://www.teamrankings.com/ David Hess

    We use 0.475, per the conservation here: http://www.teamrankings.com/blog/ncaa-basketball/betting-predictions-faq#comment-1122420767

    In addition I should point out that we calculate the value for both teams, then average the two values.

  • Derek

    Why is Louisville (as of Feb 13) tops in your NCAA picks (12% likely to win), while Syracuse is so low (3% to win) despite being #1 in most of your power ratings? Just curious; your power ratings are obviously used but are not the top consideration.

  • http://www.teamrankings.com/ David Hess

    The projection pages are based on our “New Rankings” which incorporate information from our preseason ratings. In those, Louisville is #1: http://www.teamrankings.com/ncaa-basketball/ranking/predictive-by-other

  • Justin

    I’ve looked around but haven’t found anything on injuries, trades, etc. Can you briefly explain how these things are generally dealt with? Are the models totally based on a team’s performance as a whole or are they based on the sum of each team’s individual parts for a specific game. Thanks

  • http://www.teamrankings.com/ David Hess

    The MLB models use starting pitcher stats, but aside from that all the models use team stats as a whole. However, besides full season stats, we also use rolling stat averages and power ratings over the Last X games (X varies by sport) as input, plus the spread and total for the game in question (when these differ greatly from what’s expected based on power ratings, that can be a sign of an injury, motivational issues, inclement weather, etc).

  • Clem

    how did the Shockers go from 2.4% to get a 1-seed last week to 50.2% this week? They’ve been projected to go 31-0 the whole time. Was there an adjustment to your seeding program?

  • http://www.teamrankings.com/ David Hess

    Yes, we made a change to the formula.

    We have a variable in the seeding algorithm that punishes teams for being from smaller conferences. In our test data, this worked well, but Gonzaga last year and Wichita State this year were being obviously underrated by our projections. So we manually removed the penalty from Wichita State for this season, and plan to make a more systematic adjustment to the formula in the offseason to handle cases like this (where a team is obviously FAR better than the rest of their conference and/or has a history of success that could lead to their record being viewed as more “legitimate” by the selection committee).

    This is the only manual tweak we’ve made in the 2 years since we created the bracketology projections. Obviously we would have preferred not to manually intervene here, but with a model this complex, there are bound to be novel scenarios that the projections aren’t perfectly calibrated for. We felt it was better to make this change than to continue publishing such an unrealistically low value for Wichita State.

  • VegasCali

    Why does you NCAAB standing not include the AAC

  • http://www.teamrankings.com/ David Hess

    Thanks for the heads up there. Looks like when we updated the site to reflect the new American conference, we somehow missed the standings page. I’ve updated the page to include the AAC — though for now it’s listed last on the page.

  • Clem

    Just curious, do you know where Wichita State would be right now, by the old formula? It seems like everything worked out for them to be as high as possible, at least before the large conferences have had their tournaments.

  • http://www.teamrankings.com/ David Hess

    I don’t have an exact number, but my best guess is in the 10% to 15% range. And yes, they are pretty well positioned. I would be surprised if the committee didn’t give them a 1 seed.

  • Harvey

    My regular brackets pool has disbanded. Is there a pool I can join?

  • http://www.teamrankings.com/ David Hess

    We don’t host any pools ourselves, but maybe some other readers will chime in.

    You might have more luck if you post this question on our March Madness forum: http://www.teamrankings.com/ncaa-bracket-picks/forum/open-discussion/

  • eric

    nhl in the future???

  • http://www.teamrankings.com/ David Hess

    No current plans for it.

  • JTB

    Can you explain the Off/Def Efficiency numbers. Similar at all to EFG%? Appreciate all the stats, Thank You

  • http://www.teamrankings.com/ David Hess

    Those are points (scored or allowed) per possession. Possessions are calculated as laid out here: http://www.teamrankings.com/blog/ncaa-basketball/betting-predictions-faq#comment-1243112802

  • Chris

    Do you have 2012-13 NBA data that can be accessed via this site?

  • Chris

    i.e. last year’s version of, for example, this page: http://www.teamrankings.com/nba/team/indiana-pacers/ats-results

  • http://www.teamrankings.com/ David Hess

    We don’t have past data formatted like that page. We do have old stats, rankings, and trend data. For any of the following types of pages, you can view past seasons (or past specific dates in the case of the rankings and team stats) by using the year links, calendar button, or drop down menu above the main data table:

    All ratings pages — http://www.teamrankings.com/nba/ranking/predictive-power-ranking-by-team

    All single team stat pages — http://www.teamrankings.com/nba/stat/three-point-pct

    All player stat leader pages — http://www.teamrankings.com/nba/player-stat/assists

    All trends record pages — http://www.teamrankings.com/nba/trends/win_trends/

    All team-specific player stat pages — http://www.teamrankings.com/nba/team/atlanta-hawks/roster

    All team-specific trends pages — http://www.teamrankings.com/nba/team/atlanta-hawks/win-trends

  • James

    Is there a page that has definitions for the various statistics. I am looking for return yards in college football. You have a great site!! Thanks!!

  • http://www.teamrankings.com/ David Hess

    This blog post has definitions for some of the newer basketball stats we’ve added, but no, we don’t have a definitions page right now.

    Return yards are simply the total number of yards a player has returned kicks/punts/interceptions for.

  • Paul

    Love your site… Where you do you get your NCAA football player stats? Particularly fumbles. Thanks!

  • Sara

    do you have information on injury, age, experience, salary, division by NFL players? help on any or all the above mentioned would be greatly appreciated!

  • http://www.teamrankings.com/ David Hess

    We get most of our data (including the NCAA player stats) from Sports Direct, Inc: http://www.sportsdirectinc.com/

  • http://www.teamrankings.com/ David Hess

    We do have some of that info, but it’s not all organized in one place.

    Injury — for any *upcoming* game, if you go to the matchup section for that game, there is a gray menu on the left side of the page. Under the “Player Stats & Data” section, there is a link to “Injury Report”. For example, from the Super Bowl: http://www.sportsdirectinc.com/ … however, this is only accurate for *future* games … for past games it will show the *current* Injury Report

    Age — on player pages, we show the age at the top of the screen, if we have the data for that player. For example: http://www.teamrankings.com/nfl/player/marshawn-lynch

    Experience, Salary — I don’t believe we don’t have info on this published anywhere.

    Division — On the player list page, we have current rosters (as of the end of last season) by team: http://www.teamrankings.com/nfl/players/ … if you know the divisions for teams, you can get lists for each team there. (Our projections show the divisions: http://www.teamrankings.com/nfl/projections/standings/)

  • Nick

    Hey, on the mlb team home rankings page in the v 1-5 column if a team has a record of 3-0, that is 3 wins vs an opponent that was ranked in the top 5 of team away rankings correct?

  • http://www.teamrankings.com TeamRankings.com

    Nick, the “vs” columns show record against teams CURRENTLY ranked in those bands, in the exact same ranking. So it’s admittedly sort of useless for home/away rankings — it’s showing record against teams currently ranked 1-5 in the home rankings.

    It makes sense in most of the rankings, home/away is just an exception, so it’s probably not worth us changing that.

  • Hopalong

    Your current (pre game 3) odds for NBA finals is WAY off of the Las Vegas odds. You have SAS 67.5% (Wow! Huge favorites!) even though MIA now has home court advantage and has been more impressive in playoffs. What is driving this? Possibly season performance (in a season where the Heat laid back to rest DWade)?

  • http://www.teamrankings.com/ David Hess

    Those odds are based entirely on our power ratings, which do take the whole season into account. So you are correct that one reason for the disparity is that the Heat’s full season performance was probably more affected by injuries than the Spurs’ was. In addition, some research has shown that past champions may perform better in the playoffs than you’d expect from their regular season performance — as if they have “another gear” to shift into for the playoffs. This effect won’t be well-accounted for by full season power ratings.

    EDIT: It’s also worth mentioning that the power ratings don’t adjust for shortened lineup rotations that occur in the playoffs, or for specific team/player matchup effects.

  • Ashu

    You prediction accuracy of ATS/money line picks, hovers around ~50% over last few seasons.. however, one needs to be in the range of ~55% to be profitable in the betting business… In that event, what should drive me to pay for monthly subscription to get the picks from you..

  • http://www.teamrankings.com TeamRankings.com

    Ashu, first to clarify, the percentage win rate of money line picks isn’t necessarily related to profit. If you bet all underdogs on the money line, you may only win 40% of the time, and can still make a lot of money. Also, 52.4% would be the break-even win rate for other bets at typical -110 juice; I think most professional bettors would consider 55% long term to be exceptional, not just satisfactory, performance.

    To the general question, my short answer is that our approach from the beginning has been to publish a lot of objective predictions and data, and let customers decide how they want to use it. We’ve built algorithmic models using general frameworks that we can apply to multiple sports and pick types. We publish all the results for everyone to see, and continue to make refinements, but the bottom line is it’s pretty clear that our methods to date have worked better for some sports and pick types than others.

    That’s not a surprise to us, but interestingly, we’ve also found that in sports/pick types where our official “TR Picks” for games haven’t done all that well, we have many users who are nonetheless having success integrating our underlying model predictions (e.g. our Decision Tree, Similar Games, and/or Power Ratings model odds) into their own data-driven betting systems that they’ve created. They are using our raw model projection data, but not our official picks.

    All that is perfectly fine to us — I’m certain there are plenty of ways to get an edge from all our data that we haven’t discovered yet, and there are plenty of smart people in the world.

    So to recap, the customers we talk to who have gotten the most out of our service right now from a betting picks perspective are typically doing one of two things:

    1) Using our TR picks directly, but leveraging our prediction accuracy reports to selectively focus on sports/pick types where there is a strong multi-year performance record.

    2) Using our raw model projection data in “custom” ways.

    For our part, we continue to refine and improve the logic behind our TR picks and confidence odds every year, but it takes years before we can truly read the results of any given refinement.

    For example, we updated our logic for NBA totals picks this season, and our playable picks (2-star or better, as 1-star picks are expected to be unprofitable long term at -110 juice) have done much better this season (http://www.teamrankings.com/nba/betting-models/yearly-results/?type=ou), but we’ll need to wait several seasons to get a big enough sample of results to feel more confident about that.


    college conviction pool picks. I adjusted the opponents % picks ( since I know them) but the picks are the same as the national % picks. That wasn’t the case yesterday. Help is the site not working today?

  • http://www.teamrankings.com TeamRankings.com

    Michael, I reposted your question in the Pick’em Week 1 forum, and answered it there: http://www.teamrankings.com/football-pool-picks/forum/2014-ncf-week-1-discussion/#comment-1565100176


    I still don’t get it. The adjusted opponent % yesterday and today are about the same but the pick results today vs yesterday are not even close. I don’t know which one to use. Picks have to be in by 2:00 today.

  • http://www.teamrankings.com TeamRankings.com

    Michael, did you catch David’s reply in the Q&A Forum of the Football Pick’em Pools product? That’s the place to ask any questions like this, or to follow up: http://www.teamrankings.com/football-pool-picks/forum/2014-ncf-week-1-discussion/#comment-1565100176

  • http://www.teamrankings.com TeamRankings.com

    If your picks are due, btw, it looks like you set up a pool with both weekly and season prizes. If you only have one entry, you need to decide which prize you want to go after — you can use the “Maximize Profit” picks if you go for the Season Prize option, or the “Optimal Picks” for Weekly Prizes if you want to focus on those.

    General point — picks WILL change over the course of the week as data inputs change. Your best bet is to just wait until the last minute before you pool’s picks are due, then go grab picks.


    yes I read it but it doesn’t answer the question.
    when I change to opponents % the picks are the same as the national picks. That was not the case yesterday. I got very different results for the picks and confidence rank;

  • http://www.teamrankings.com TeamRankings.com

    OK — let’s take this discussion over to the Q&A forum thread. I will answer you there.


    I understand the concept of season vs weekly.
    I don’t understand why the national % picks and the adjusted opponents % picks are the same when the difference in the two % is significant.

  • http://www.teamrankings.com TeamRankings.com

    Michael — I’m still not exactly clear what you are saying here. Just sent you an email.

  • c21terry

    I am confused regarding the ATS picks and what the power rating number means. Next week Old Dominion is +16 going to play at NC State. The power rating number is green Old Dominion +25.8. What exactly does that 25.8 mean? Does it factor into the current spread of +16?

  • http://www.teamrankings.com TeamRankings.com

    That means the Power Ratings model projects Old Dominion to cover the spread by 25.8 points. With a line of +16, that would equate to an average margin of victory of 9.8 for Old Dominion.

    Also, if you haven’t done so yet, I strongly encourage you to read our help text about the predictions (linked immediately above the picks table on every picks page): http://www.teamrankings.com/about/about-our-predictions/ … The TR Pick and Odds columns represent our best projection for the game.

  • Kyle Eckstein

    I’m not exactly sure how to phrase this question, so we’ll see if you understand it. OK. What I’m wondering about is using the predictive models for something like bowl games. I like to do different types of contests where you rank the games from confidence of 1-35 and in some instances once the first game is played, every pick is locked. Now I notice that it seems like the Decision Tree model does fairly well in bowl games and was hoping to use the DT to make the picks. What I don’t like is that picks can change all the way up until each individual game time, so while the DT may go 22-13 overall, it may only go 18-17 based on the info prior to the first bowl game. I guess I just don’t know how to use the models how to predict the bowl games all at once, when the records are misleading due to being a fluid prediction.

  • http://www.teamrankings.com/ David Hess


    I’m not exactly sure what you’re asking. For game winner picks (which I assume you are talking about), yes, it would make sense that the final pick has a better record than the initial pick, since more information is available to the model when it makes the final pick (the results of more games, team stat updates, betting market line changes, etc).

    Conversely, for betting picks, the early predictions ought to be better, since earlier lines are generally less efficient and easier to beat than closing lines.

    Of course, everybody else in your pool *also* has less info available to them before game 1 than before game 35, just like our models.

  • JD

    Need help
    Iowa St -10
    power ranking has georgia st+2.3 in RED
    what does this mean?

  • http://www.teamrankings.com/ David Hess

    That would mean that the Power Ratings model projects Georgia St to cover the spread (+10) by 2.3 point.

    It’s read because that model disagrees with the TR Pick.

  • Andy

    what column shows you your prediction against the spread? for example I am looking at the kentucky CBB game against miss. on 1/6/15 and the odds columns shows 98% but they did not cover the spread. the TR pick shows that it was right so i assume the TR pick is a straight up pick on winner not against the spread.

  • http://www.teamrankings.com TeamRankings.com

    You’re just on the wrong page — you’re looking at our game winner picks page, not our spread picks page. The spread picks page is here:


    Like game winners, you do need a subscription to access picks for the current day’s and any future day games.

  • Mike

    What time are the closing lines set at before the game starts? Are the linew based off of the odds on your website or a cretain sports book?

  • http://www.teamrankings.com TeamRankings.com

    Hopefully this clears things up:

    1) There is no “set” time before a game starts where we freeze our final line listed on the site for a pick, and the corresponding pick. Our site is updated using scripts that run continuously throughout the day, and depending on how much info they need to process during a specific update cycle, they can take from a 20-30 minutes to significantly longer to run. The vast majority of final picks on the site are “frozen” within 1-2 hours of the start of the game.

    Because it takes our models time to predict games (since they incorporate a lot of data and processing) AND because it takes time to push new content out to live web pages, it’s simply not possible for us to react to line changes that occur very close to game start time. (Longer term, we do want to get faster at making predictions, though.)

    2) Pinnacle Sports is the sports book of record for our lines and picks in the vast majority of cases. We will occasionally use BetOnline or BookMaker as an odds source if Pinnacle hasn’t released a line for a game yet.

  • Mike

    Is there a contact number I can call someone at for a further minor discussion very difficult to understand by email.

  • http://www.teamrankings.com TeamRankings.com

    Mike, sent you an email about this.

  • http://www.ibworldacademy.com IB Tutor

    Its amazing post. I love basketball. I blog often and I truly appreciate your
    content. The article has truly peaked my interest. I’m going to take a note of
    your website and keep checking for new details about once a week. in IB World Academy