{
  "id": 131397,
  "title": "Questions regarding kenpom statistics methodology",
  "url": "/competitions/google-cloud-ncaa-march-madness-2020-division-1-mens-tournament/discussion/131397",
  "author_name": "",
  "post_date": "2020-02-19T16:57:01.764209600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>For the widely used kenpom statistics, is there any publicly available documentation on how each variable is calculated / the exact definition? Most of the information appears to be behind a paywall. In particular, how is the 'Luck' factor calculated.</p>\n\n<p>Also, do kenpom years correspond to seasons in this dataset? ie: is 2019 in kenpom data actually referring to the 2018-2019 season as in this dataset, or does it just mean the calendar year?</p>\n\n<p>Thanks</p>",
  "messages": [
    {
      "id": "750749",
      "postDate": "02/19/2020 16:57:01",
      "content": "<p>Hello,</p>\n\n<p>For the widely used kenpom statistics, is there any publicly available documentation on how each variable is calculated / the exact definition? Most of the information appears to be behind a paywall. In particular, how is the 'Luck' factor calculated.</p>\n\n<p>Also, do kenpom years correspond to seasons in this dataset? ie: is 2019 in kenpom data actually referring to the 2018-2019 season as in this dataset, or does it just mean the calendar year?</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "Hello,\n\nFor the widely used kenpom statistics, is there any publicly available documentation on how each variable is calculated / the exact definition? Most of the information appears to be behind a paywall. In particular, how is the 'Luck' factor calculated.\n\nAlso, do kenpom years correspond to seasons in this dataset? ie: is 2019 in kenpom data actually referring to the 2018-2019 season as in this dataset, or does it just mean the calendar year?\n\nThanks",
      "votes": null
    },
    {
      "id": "750771",
      "postDate": "02/19/2020 17:16:57",
      "content": "<p>Interesting question about luck. I did a Google search (because I couldn't find anything on the site even though I'm a subscriber) and found the following:</p>\n\n<p>Luck rating (Luck): KenPom defines this as a measure of the deviation between a team's actual winning percentage and what one would expect from its game-by-game efficiencies. Essentially, a team involved in a lot of close games should not win (or lose) all of them. Those that do will be viewed as lucky (or unlucky).</p>\n\n<p>There are a lot of ways to quantify luck:\n-You could look at Pythagorean Expectation (<a href=\"https://en.wikipedia.org/wiki/Pythagorean_expectation\">https://en.wikipedia.org/wiki/Pythagorean_expectation</a>)  vs actual winning pct. \n-I also looked at the pct of games won that were decided by less than x points and tried different x's and then compared to actual win pct. \n-You could even calculate in game win probabilities and define luck as any team that won a game with less than a 5% - 10% win probability.  </p>\n\n<p>I'm not aware of any place where the actual formulas are available. Maybe a lot of Google sleuthing will get you there.</p>\n\n<p>To your second question, I believe the answer is that 2018-19 refers to the 2019 season.</p>",
      "rawMarkdown": "Interesting question about luck. I did a Google search (because I couldn't find anything on the site even though I'm a subscriber) and found the following:\n\nLuck rating (Luck): KenPom defines this as a measure of the deviation between a team's actual winning percentage and what one would expect from its game-by-game efficiencies. Essentially, a team involved in a lot of close games should not win (or lose) all of them. Those that do will be viewed as lucky (or unlucky).\n\nThere are a lot of ways to quantify luck:\n-You could look at Pythagorean Expectation (https://en.wikipedia.org/wiki/Pythagorean_expectation)  vs actual winning pct. \n-I also looked at the pct of games won that were decided by less than x points and tried different x's and then compared to actual win pct. \n-You could even calculate in game win probabilities and define luck as any team that won a game with less than a 5% - 10% win probability.  \n\nI'm not aware of any place where the actual formulas are available. Maybe a lot of Google sleuthing will get you there.\n\nTo your second question, I believe the answer is that 2018-19 refers to the 2019 season.",
      "votes": null
    },
    {
      "id": "750808",
      "postDate": "02/19/2020 17:50:41",
      "content": "<p>Looks like the residuals of a gaussian model fit to some basic stats around points scored/allowed.</p>\n\n<blockquote>\n  <p>...The easiest one to understand is Luck, which is the deviation in winning percentage between a team’s actual record and their expected record using the correlated gaussian method. The luck factor has nothing to do with the rating calculation, but a team that is very lucky (positive numbers) will tend to be rated lower by my system than their record would suggest.</p>\n</blockquote>\n\n<p><a href=\"https://kenpom.com/blog/ratings-explanation/\">https://kenpom.com/blog/ratings-explanation/</a></p>\n\n<p>The \"correlated gaussian method\" text is hyperlinked on his blog, but I believe that the destination is a typo on Ken's part. I think this is the page he meant to link to: <a href=\"http://www.rawbw.com/~deano/helpscrn/corrgauss.html\">http://www.rawbw.com/~deano/helpscrn/corrgauss.html</a></p>",
      "rawMarkdown": "Looks like the residuals of a gaussian model fit to some basic stats around points scored/allowed.\n\n&gt; ...The easiest one to understand is Luck, which is the deviation in winning percentage between a team’s actual record and their expected record using the correlated gaussian method. The luck factor has nothing to do with the rating calculation, but a team that is very lucky (positive numbers) will tend to be rated lower by my system than their record would suggest.\n\nhttps://kenpom.com/blog/ratings-explanation/\n\nThe \"correlated gaussian method\" text is hyperlinked on his blog, but I believe that the destination is a typo on Ken's part. I think this is the page he meant to link to: http://www.rawbw.com/~deano/helpscrn/corrgauss.html",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 750771,
      "author_name": "btesch",
      "author_url": "",
      "post_date": "02/19/2020 17:16:57",
      "content": "<p>Interesting question about luck. I did a Google search (because I couldn't find anything on the site even though I'm a subscriber) and found the following:</p>\n\n<p>Luck rating (Luck): KenPom defines this as a measure of the deviation between a team's actual winning percentage and what one would expect from its game-by-game efficiencies. Essentially, a team involved in a lot of close games should not win (or lose) all of them. Those that do will be viewed as lucky (or unlucky).</p>\n\n<p>There are a lot of ways to quantify luck:\n-You could look at Pythagorean Expectation (<a href=\"https://en.wikipedia.org/wiki/Pythagorean_expectation\">https://en.wikipedia.org/wiki/Pythagorean_expectation</a>)  vs actual winning pct. \n-I also looked at the pct of games won that were decided by less than x points and tried different x's and then compared to actual win pct. \n-You could even calculate in game win probabilities and define luck as any team that won a game with less than a 5% - 10% win probability.  </p>\n\n<p>I'm not aware of any place where the actual formulas are available. Maybe a lot of Google sleuthing will get you there.</p>\n\n<p>To your second question, I believe the answer is that 2018-19 refers to the 2019 season.</p>",
      "votes": null,
      "replies": [
        {
          "id": 750808,
          "author_name": "joshstone",
          "author_url": "",
          "post_date": "02/19/2020 17:50:41",
          "content": "<p>Looks like the residuals of a gaussian model fit to some basic stats around points scored/allowed.</p>\n\n<blockquote>\n  <p>...The easiest one to understand is Luck, which is the deviation in winning percentage between a team’s actual record and their expected record using the correlated gaussian method. The luck factor has nothing to do with the rating calculation, but a team that is very lucky (positive numbers) will tend to be rated lower by my system than their record would suggest.</p>\n</blockquote>\n\n<p><a href=\"https://kenpom.com/blog/ratings-explanation/\">https://kenpom.com/blog/ratings-explanation/</a></p>\n\n<p>The \"correlated gaussian method\" text is hyperlinked on his blog, but I believe that the destination is a typo on Ken's part. I think this is the page he meant to link to: <a href=\"http://www.rawbw.com/~deano/helpscrn/corrgauss.html\">http://www.rawbw.com/~deano/helpscrn/corrgauss.html</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "750749": "Hello,\n\nFor the widely used kenpom statistics, is there any publicly available documentation on how each variable is calculated / the exact definition? Most of the information appears to be behind a paywall. In particular, how is the 'Luck' factor calculated.\n\nAlso, do kenpom years correspond to seasons in this dataset? ie: is 2019 in kenpom data actually referring to the 2018-2019 season as in this dataset, or does it just mean the calendar year?\n\nThanks",
    "750771": "Interesting question about luck. I did a Google search (because I couldn't find anything on the site even though I'm a subscriber) and found the following:\n\nLuck rating (Luck): KenPom defines this as a measure of the deviation between a team's actual winning percentage and what one would expect from its game-by-game efficiencies. Essentially, a team involved in a lot of close games should not win (or lose) all of them. Those that do will be viewed as lucky (or unlucky).\n\nThere are a lot of ways to quantify luck:\n-You could look at Pythagorean Expectation (https://en.wikipedia.org/wiki/Pythagorean_expectation)  vs actual winning pct. \n-I also looked at the pct of games won that were decided by less than x points and tried different x's and then compared to actual win pct. \n-You could even calculate in game win probabilities and define luck as any team that won a game with less than a 5% - 10% win probability.  \n\nI'm not aware of any place where the actual formulas are available. Maybe a lot of Google sleuthing will get you there.\n\nTo your second question, I believe the answer is that 2018-19 refers to the 2019 season.",
    "750808": "Looks like the residuals of a gaussian model fit to some basic stats around points scored/allowed.\n\n&gt; ...The easiest one to understand is Luck, which is the deviation in winning percentage between a team’s actual record and their expected record using the correlated gaussian method. The luck factor has nothing to do with the rating calculation, but a team that is very lucky (positive numbers) will tend to be rated lower by my system than their record would suggest.\n\nhttps://kenpom.com/blog/ratings-explanation/\n\nThe \"correlated gaussian method\" text is hyperlinked on his blog, but I believe that the destination is a typo on Ken's part. I think this is the page he meant to link to: http://www.rawbw.com/~deano/helpscrn/corrgauss.html"
  },
  "source": "meta"
}