Each season there are thousands of NCAA® basketball games played between Division I women's teams, culminating in March Madness, the 64-team national championship that starts in the middle of March. We have provided a large amount of historical data about college basketball games and teams, going back many years. Armed with this historical data, you can explore it and develop your own distinctive ways of predicting March Madness® game outcomes. You can even evaluate and compare different approaches by seeing which of them would have done best at predicting tournament games from the past.
If you are unfamiliar with the format and intricacies of the tournament, we encourage reading the wikipedia page before diving into the data. The data description and schema may seem daunting at first, but is not as complicated as it appears.
As a reminder, you are encouraged to incorporate your own sources of data. We have provided extensive historical data to jump-start the modeling process, and this data is self-consistent (for instance, dates and team ID's are always treated the same way). Nevertheless, you may also be able to make productive use of external data. If you head down this path, please be forewarned that many sources have their own distinctive way of identifying the names of teams, and this can make it challenging to link up with our data. The WTeamSpellings file, which is listed in the bottom section below, may help you map external team references into our own Team ID structure. You will probably also need to understand exactly how dates work in our data. Remember as well that you are required to disclose your external sources of data prior to the start of the tournament.
We extend our gratitude to Kenneth Massey for providing much of the historical data.
Special Acknowledgment to Jeff Sonas of Sonas Consulting for his support in assembling the dataset for this competition.
Stage 1 - You should submit predicted probabilities for every possible matchup in the past 5 NCAA® tournaments (seasons 2015-2019).
Stage 2 - You should submit predicted probabilities for every possible matchup before the 2020 tournament begins.
Refer to the Timeline page for specific dates. In both stages, the sample submission will tell you which games to predict.
Below we describe the format and fields of the contest data files. The data will likely be refreshed once in early March while Stage 1 of the competition is running. Many of the files are only complete through the end of last season. At the start of Stage 2, we will provide updates to these files to incorporate data from the current season. Right now the WSeasons and WTeamConferences files are the only two files that reference the 2020 season, since those are the only files where the final 2020 data is already finalized; the other files will be updated later on when we provide preliminary data (in early March) and final Stage 2 data (after Selection Sunday) for the current season.
This section provides everything you need to build a simple prediction model and submit predictions.
Special note about "Season" numbers: the college basketball season lasts from early November until the national championship tournament that starts in the middle of March. For instance, this year the first women’s Division I games were played on November 5th, 2019 and the women’s national championship game will be played on April 5th, 2020. Because a basketball season spans two calendar years like this, it can be confusing to refer to the year of the season. By convention, when we identify a particular season, we will reference the year that the season ends in, not the year that it starts in. So for instance, the current season will be identified in our data as the 2020 season, not the 2019 season or the 2019-20 season or the 2019-2020 season, though you may see any of these in everyday use outside of our data.
Data Section 1 file: WTeams.csv
This file identifies the different college teams present in the dataset. Each school is uniquely identified by a 4 digit id number. You will not see games present for all teams in all seasons, because the games listing is only for matchups where both teams are Division-I teams. There are 351 teams currently in Division-I, and an overall total of 365 teams in our team listing. Each year, some teams might start being Division-I programs, and others might stop being Division-I programs, and in fact this year is no exception. Savannah State (TeamID=3366) is no longer part of Division I and so their data in our dataset stops after the 2019 season, and they are no longer part of the Mid-Eastern Athletic Conference. To balance this out, Merrimack (TeamID=3467) has just moved to Division I this year, so they show up as a new team in the data and they are now part of the America East Conference. Other than the teams file and team conferences file, you won't see any data for Merrimack until preliminary 2020 season data starts coming in, as we approach Stage 2 of the contest. Also please note that we recently determined that VMI and Citadel only have men's basketball teams, not women's teams, and so we shouldn't have those schools in the women's data at all. This means all references to TeamID 3154 and TeamID 3440 have been removed from the women's contest data this year, and there are two fewer Division 1 teams this year because of this.
Data Section 1 file: WSeasons.csv
This file identifies the different seasons included in the historical data, along with certain season-level properties.
The game dates in this dataset are expressed in relative terms, as the number of days since the start of the regular season, and aligned for each season so that day number #133 is the Monday right before the tournament, when team selections are made. During any given season, day number zero is defined to be exactly 19 weeks earlier than Selection Monday, so Day #0 is a Monday in late October or early November such that Day #132 is Selection Sunday (for the men's tournament) and Day #133 is Selection Monday (for the women's tournament).
This doesn't necessarily mean that the regular season will always start exactly on day #0 or day #1; in fact, during the past decade, regular season games typically start being played on a Friday that is either Day #4 or Day #11, but further back there was more variety.
Data Section 1 file: WNCAATourneySeeds.csv
This file identifies the seeds for all teams in each NCAA® tournament, for all seasons of historical data. Thus, there are exactly 64 rows for each year, since there are no play-in teams in the women's tournament. We will not know the seeds of the respective tournament teams, or even exactly which 64 teams it will be, until Selection Monday on March 16, 2020 (DayNum=133).
Data Section 1 file: WRegularSeasonCompactResults.csv
This file identifies the game-by-game results for many seasons of historical data, starting with the 1998 season. For each season, the file includes all games played from DayNum 0 through 132. It is important to realize that the "Regular Season" games are simply defined to be all games played on DayNum=132 or earlier (DayNum=133 is Selection Monday). Thus a game played before Selection Monday will show up here whether it was a pre-season tournament, a non-conference game, a regular conference game, a conference tournament game, or whatever.
Data Section 1 file: WNCAATourneyCompactResults.csv
This file identifies the game-by-game NCAA® tournament results for all seasons of historical data. The data is formatted exactly like the WRegularSeasonCompactResults data. Each season you will see 63 games listed, since there are no women's play-in games.
Although the scheduling of the men's tournament rounds has been consistent for many years, there has been more variety in the scheduling of the women's rounds. There have been four different schedules over the course of the past 20+ years for the women's tournament, as follows:
2017 season through 2020 season:
2015 season and 2016 season:
2003 season through 2014 season:
1998 season through 2002 season:
Data Section 1 file: WSampleSubmissionStage1.csv
This file illustrates the submission file format for Stage 1. It is the simplest possible submission: a 50% winning percentage is predicted for each possible matchup.
A submission file lists every possible matchup between tournament teams for one or more years. During Stage 1, you are asked to make predictions for all possible matchups from the past five NCAA® tournaments (seasons 2015, 2016, 2017, 2018, and 2019). In Stage 2, you will be asked to make predictions for all possible matchups from the current NCAA® tournament (season 2020).
Since there are 64 teams in the tournament, there are 64*63/2=2,016 predictions to make for that year, so a Stage 1 submission file will have 2,016*5=10,080 data rows.
Example #1: You want to make a prediction for Duke (TeamID=3181) against Arizona (TeamID=3112) in the 2005 tournament, with Duke given a 53% chance to win and Arizona given a 47% chance to win. In this case, Arizona has the lower numerical ID so they would be listed first, and the winning percentage would be expressed from Arizona's perspective (47%):
2005_3112_3181,0.47
Example #2: You want to make a prediction for Duke (TeamID=3181) against North Carolina (TeamID=3314) in the 2005 tournament, with Duke given a 51.6% chance to win and North Carolina given a 48.4% chance to win. In this case, Duke has the lower numerical ID so they would be listed first, and the winning percentage would be expressed from Duke's perspective (51.6%):
2005_3181_3314,0.516
Also note that a single prediction row serves as a prediction for each of the two teams' winning chances. So for instance, in Example #1, the submission row of "2005_3112_3181,0.47" specifically gives a 47% chance for Arizona to win, and doesn't explicitly mention Duke's 53% chance to win. However, our evaluation utility will automatically infer the winning percentage in the other direction, so a 47% prediction for Arizona to win also means a 53% prediction for Duke to win. And similarly, because the submission row in Example #2 gives Duke a 51.6% chance to beat North Carolina, we will automatically figure out that this also means North Carolina has a 48.4% chance to beat Duke.
This section provides game-by-game stats at a team level (free throws attempted, defensive rebounds, turnovers, etc.) for all regular season, conference tournament, and NCAA® tournament games since the 2009-10 season.
Team Box Scores are provided in "Detailed Results" files rather than "Compact Results" files. However, the two files are strongly related.
In a Detailed Results file, the first eight columns (Season, DayNum, WTeamID, WScore, LTeamID, LScore, WLoc, and NumOT) are exactly the same as a Compact Results file. However, in a Detailed Results file, there are many additional columns. The column names should be self-explanatory to basketball fans (as above, "W" or "L" refers to the winning or losing team):
(and then the same set of stats from the perspective of the losing team: LFGM is the number of field goals made by the losing team, and so on up to LPF).
Note: by convention, "field goals made" (either WFGM or LFGM) refers to the total number of fields goals made by a team, a combination of both two-point field goals and three-point field goals. And "three point field goals made" (either WFGM3 or LFGM3) is just the three-point fields goals made, of course. So if you want to know specifically about two-point field goals, you have to subtract one from the other (e.g., WFGM - WFGM3). And the total number of points scored is most simply expressed as 2*FGM + FGM3 + FTM.
Data Section 2 file: WRegularSeasonDetailedResults.csv
This file provides team-level box scores for many regular seasons of historical data, starting with the 2010 season. All games listed in the WRegularSeasonCompactResults file since the 2010 season should exactly be present in the WRegularSeasonDetailedResults file.
Data Section 2 file: WNCAATourneyDetailedResults.csv
This file provides team-level box scores for many NCAA® tournaments, starting with the 2010 season. All games listed in the WNCAATourneyCompactResults file since the 2010 season should exactly be present in the WNCAATourneyDetailedResults file.
This section provides city locations of all regular season, conference tournament, and NCAA® tournament games since the 2009-10 season
Data Section 3 file: Cities.csv
This file provides a master list of cities that have been locations for games played. Please notice that the Cities and Conferences files are the only two that don't start with an W; this is because the data files are identical between men's and women's data, so you don't need to maintain separate listings of cities or conferences across the two contests.
Data Section 3 file: WGameCities.csv
This file identifies all games, starting with the 2010 season, along with the city that the game was played in. Games from the regular season and the NCAA® tourney are all listed together. The CityID is present in more than 98% of games since the 2010 season. Games from the 2009 season and before are not listed in this file.
This section provides play-by-play event logs for more than 99% of each year's regular season and NCAA® tournament, and secondary tournament women's games since the 2014-15 season - including plays by individual players.
This year we are transitioning to a different play-by-play source, which includes data since the 2014-2015 season rather than since the 2009-2010 season (that's what we had previously for men's data). However, we are now able to provide play-by-play for both men's and women's data, and there is locational play-by-play detail starting with games from the 2018-2019 season. This includes an X/Y location (ranging from 0 to 100 in each dimension) on the court for each shot attempt, turnover, and foul for many games, as well as an overall categorization of the area on the court that the shot or turnover or foul occurred in (inside left wing, outside right wing, under the basket, etc.) Some games in these recent seasons still lack the locational detail. The data from last year (2019 season) matches what you can expect for the current year (2020 season) as we approach the postseason. Despite the 99.5% coverage, there are still a few games missing annually, and we will try to bring those in as well, if possible.
Data Section 4 file: WEvents2015.csv, WEvents2016.csv, WEvents2017.csv, WEvents2018.csv, WEvents2019.csv
Each WEvents file lists the play-by-play event logs for more than 99% of games from that season. Each event is assigned to either a team or a single one of the team's players. Thus if a basket is made by one player and an assist is credited to a second player, that would show up as two separate records. The players are listed by PlayerID within the WPlayers.csv file.
Event Types and Subtypes:
This is the diagram provided by the play-by-play source:
Here is a diagram of these regions on a typical court:
And you may also have noticed that the three-point-arc is different this year in college basketball. The width of the court is still 15.2 m (50 ft) and the length of the court is still 28.7 m (94 feet), but the distance to the arc has changed from 6.32 m (20 feet 9 inches) to 6.75 m (22.15 feet).
Here is a diagram of the old court, which would apply for play-by-play in seasons 2019 and before:
.svg)
And here is a diagram of the new court, which only applies for season 2020:
.svg)
Data Section 4 file: WPlayers.csv
Note: there are data collection errors within the events, in that they don't necessarily add up to the final stats for the game. In addition, the player name spellings may vary over the course of a season or career for the same player, which would lead to the same player showing up with different PlayerID values. Nevertheless, this was the highest quality data we could manage for play-by-play with the near-complete set of games.
This section contains additional supporting information, including alternative team name spellings and representations of bracket structure
Data Section 5 file: WTeamSpellings.csv
This file indicates alternative spellings of many team names. It is intended for use in associating external spellings against our own TeamID numbers, thereby helping to relate the external data properly with our datasets. Over the years we have identified various external spellings of different team names (as an example, for Ball State we have seen "ball st", and "ball st.", and "ball state", and "ball-st", and "ball-state"). Other teams have had more significant changes to their names over the years; for example, "Texas Pan-American" and "Texas-Rio Grande Valley" are actually the same school. The current list is obviously not exhaustive, and we encourage participants to identify additional mappings and upload extended versions of this file to the forums.
Data Section 5 file: WNCAATourneySlots
This file identifies the mechanism by which teams are paired against each other, depending upon their seeds, as the tournament proceeds through its rounds. It can be of use in identifying, for a given historical game, what round it occurred in, and what the seeds/slots were for the two teams (the meaning of "slots" is described below). You may need to know these specifics if you are trying to represent/simulate the exact workings of the tournament bracket. Unlike the analogous file on the men's side, it is not necessary to provide a Season within this file, because the women's tournament has never had play-in-games and so the 64-team women's bracket has always had the same structure each season.
Data Section 5 file: Conferences.csv
This file indicates the Division I conferences that have existed over the years since 1985. Each conference is listed with an abbreviation and a longer name. There has been no attempt to link up conferences who merged with other conferences, or whose names changed over time. Thus you will see, for instance, a "Pacific-10" conference up through the 2011 season, and then a "Pacific-12" conference starting in the 2012 season, and these look like different conferences in the data, even though it was still mostly the same teams. Please notice that the Cities and Conferences files are the only two that don't start with an M; this is because the data files are identical between men's and women's data, so you don't need to maintain separate listings of cities or conferences across the two contests. However, the Team Conferences data differs slightly between men's and women's, so those files do have the prefixes.
Data Section 5 file: WTeamConferences.csv
This file indicates the conference affiliations for each team during each season. Some conferences have changed their names from year to year, and/or changed which teams are part of the conference. This file tracks this information historically.