![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3534244%2Fad6554b6f1900c94092ffa971119a5a5%2Fmmbracket.PNG?generation=1582224736360482&alt=media\")
This data set might be of use to many of you. It contains the tournament matches from 2002 - 2019, complete with Pomeroy, Sagarin and Moore ratings. In addition, it has the associated Ken Pom rating data for both the winning and losing team (at the start day of the tournament each year).
\n\nAs part of this competition, I'm also looking at modelling the outright probabilities of teams winning through a simulation. I'm relatively new to modelling, however, so I was hoping some of you could offer advice. Based on the data above, can you recommend a beginner's approach to finding the probability of a given team X beating another team Y (assuming you have the Ken Pom ratings etc. at that moment in time - I'm scraping them from his webpage)? I'm looking to avoid using any game-specific data (turnovers, assists etc.).
\n\nAlso, could you make the assumption that the pomeroy, sagarin, moore etc. rankings on day 1 of a tournament inherently take regular season matches into account?
\n\nThanks in advance.
", "messages": [ { "id": "752121", "postDate": "02/20/2020 18:51:07", "content": "This data set might be of use to many of you. It contains the tournament matches from 2002 - 2019, complete with Pomeroy, Sagarin and Moore ratings. In addition, it has the associated Ken Pom rating data for both the winning and losing team (at the start day of the tournament each year).
\n\nAs part of this competition, I'm also looking at modelling the outright probabilities of teams winning through a simulation. I'm relatively new to modelling, however, so I was hoping some of you could offer advice. Based on the data above, can you recommend a beginner's approach to finding the probability of a given team X beating another team Y (assuming you have the Ken Pom ratings etc. at that moment in time - I'm scraping them from his webpage)? I'm looking to avoid using any game-specific data (turnovers, assists etc.).
\n\nAlso, could you make the assumption that the pomeroy, sagarin, moore etc. rankings on day 1 of a tournament inherently take regular season matches into account?
\n\nThanks in advance.
", "rawMarkdown": "[This data set](https://www.kaggle.com/dugganl1/march-madness-tourney-x-kenpom) might be of use to many of you. It contains the tournament matches from 2002 - 2019, complete with Pomeroy, Sagarin and Moore ratings. In addition, it has the associated Ken Pom rating data for both the winning and losing team (at the start day of the tournament each year). \n\nAs part of this competition, I'm also looking at modelling the **outright probabilities** of teams winning through a simulation. I'm relatively new to modelling, however, so I was hoping some of you could offer advice. Based on the data above, can you recommend a beginner's approach to finding the probability of a given team X beating another team Y (assuming you have the Ken Pom ratings etc. at that moment in time - I'm scraping them from his webpage)? I'm looking to avoid using any game-specific data (turnovers, assists etc.).\n\nAlso, could you make the assumption that the pomeroy, sagarin, moore etc. rankings on day 1 of a tournament inherently take regular season matches into account? \n\nThanks in advance.", "votes": null }, { "id": "752125", "postDate": "02/20/2020 18:53:07", "content": "
This is an example of how I'm displaying the simulation. At the moment, I've just set the probability of Duke beating everybody to 1, and the rest of the match-ups are coin tosses.
", "rawMarkdown": "\n\nThis is an example of how I'm displaying the simulation. At the moment, I've just set the probability of Duke beating everybody to 1, and the rest of the match-ups are coin tosses.", "votes": null }, { "id": "752519", "postDate": "02/21/2020 05:32:13", "content": "I would expect the Pomeroy, et al, ratings to take the regular season into account.
\n\nAs a first pass, I'd suggest the crude seed-based model that the organizers used to post as a benchmark in the competitions of years past.
\nProb[A beats B] = 0.5 + 0.03*(SeedB - SeedA)\nIt's not sophisticated, but it works alright and it takes the bare minimum of effort. It tends to be conservative (for example, according to this model, a 1 seed has only a 95% chance of beating a 16 seed), but it still tends to do better than about half of the entries in this competition.
One downside of the seed based model is that it doesn't treat different teams with the same seed differently (for example, Duke and Virginia were both 1 seeds last year, so this model would predict 50% chance of either team winning).
", "rawMarkdown": "I would expect the Pomeroy, et al, ratings to take the regular season into account.\n\nAs a first pass, I'd suggest the crude seed-based model that the organizers used to post as a benchmark in the competitions of years past. \nProb[A beats B] = 0.5 + 0.03*(SeedB - SeedA)\nIt's not sophisticated, but it works alright and it takes the bare minimum of effort. It tends to be conservative (for example, according to this model, a 1 seed has only a 95% chance of beating a 16 seed), but it still tends to do better than about half of the entries in this competition.\n\nOne downside of the seed based model is that it doesn't treat different teams with the same seed differently (for example, Duke and Virginia were both 1 seeds last year, so this model would predict 50% chance of either team winning).", "votes": null } ], "comments": [ { "id": 752125, "author_name": "dugganl1", "author_url": "", "post_date": "02/20/2020 18:53:07", "content": "
This is an example of how I'm displaying the simulation. At the moment, I've just set the probability of Duke beating everybody to 1, and the rest of the match-ups are coin tosses.
", "votes": null, "replies": [] }, { "id": 752519, "author_name": "goodspellr", "author_url": "", "post_date": "02/21/2020 05:32:13", "content": "I would expect the Pomeroy, et al, ratings to take the regular season into account.
\n\nAs a first pass, I'd suggest the crude seed-based model that the organizers used to post as a benchmark in the competitions of years past.
\nProb[A beats B] = 0.5 + 0.03*(SeedB - SeedA)\nIt's not sophisticated, but it works alright and it takes the bare minimum of effort. It tends to be conservative (for example, according to this model, a 1 seed has only a 95% chance of beating a 16 seed), but it still tends to do better than about half of the entries in this competition.
One downside of the seed based model is that it doesn't treat different teams with the same seed differently (for example, Duke and Virginia were both 1 seeds last year, so this model would predict 50% chance of either team winning).
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "752121": "[This data set](https://www.kaggle.com/dugganl1/march-madness-tourney-x-kenpom) might be of use to many of you. It contains the tournament matches from 2002 - 2019, complete with Pomeroy, Sagarin and Moore ratings. In addition, it has the associated Ken Pom rating data for both the winning and losing team (at the start day of the tournament each year). \n\nAs part of this competition, I'm also looking at modelling the **outright probabilities** of teams winning through a simulation. I'm relatively new to modelling, however, so I was hoping some of you could offer advice. Based on the data above, can you recommend a beginner's approach to finding the probability of a given team X beating another team Y (assuming you have the Ken Pom ratings etc. at that moment in time - I'm scraping them from his webpage)? I'm looking to avoid using any game-specific data (turnovers, assists etc.).\n\nAlso, could you make the assumption that the pomeroy, sagarin, moore etc. rankings on day 1 of a tournament inherently take regular season matches into account? \n\nThanks in advance.", "752125": "\n\nThis is an example of how I'm displaying the simulation. At the moment, I've just set the probability of Duke beating everybody to 1, and the rest of the match-ups are coin tosses.", "752519": "I would expect the Pomeroy, et al, ratings to take the regular season into account.\n\nAs a first pass, I'd suggest the crude seed-based model that the organizers used to post as a benchmark in the competitions of years past. \nProb[A beats B] = 0.5 + 0.03*(SeedB - SeedA)\nIt's not sophisticated, but it works alright and it takes the bare minimum of effort. It tends to be conservative (for example, according to this model, a 1 seed has only a 95% chance of beating a 16 seed), but it still tends to do better than about half of the entries in this competition.\n\nOne downside of the seed based model is that it doesn't treat different teams with the same seed differently (for example, Duke and Virginia were both 1 seeds last year, so this model would predict 50% chance of either team winning)." }, "source": "meta" }