<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Vibhor Agarwal on Medium]]></title>
        <description><![CDATA[Stories by Vibhor Agarwal on Medium]]></description>
        <link>https://medium.com/@vibhormbic91?source=rss-81a4ed666fdc------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Vibhor Agarwal on Medium</title>
            <link>https://medium.com/@vibhormbic91?source=rss-81a4ed666fdc------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 07:05:51 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vibhormbic91/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Could Red Sox have scored more in the World Series?]]></title>
            <link>https://medium.com/@vibhormbic91/could-red-sox-have-scored-more-in-the-world-series-fa6fae0af56d?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/fa6fae0af56d</guid>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[baseball-analytics]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[mlb]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Tue, 22 Jan 2019 19:06:29 GMT</pubDate>
            <atom:updated>2019-01-23T03:21:06.517Z</atom:updated>
            <content:encoded><![CDATA[<p>Cricket and baseball are similar sports. Both sports involve two teams that alternate in and out using a bat, bases, and a ball. They involve throwing a ball to a batter/batsman who attempts to “bat” it away and run safely to a base, while the opponent tries to put the batter-runner out when liable (“liable to be put out” is the baseball term for unsafe). In cricket, the batting orders are flexible. The captain announces the playing XI but he doesn’t have to announce the order in which the batsman will bat. The captain and/or coach can decide which batsman would bat next depending on the match situation. In baseball, the batters come out to bat in a pre-decided order which gives rise to the question: How does a baseball team decide the batting order? Note that this is different from the question: Which position players should play the game? The intent of this article is to introduce a methodology to decide the best batting order a team can take once they’ve decided the starting lineup for a game.</p><p>For this, we shall be using the Boston Red Sox lineup from their first game against the Los Angeles Dodgers played during the 2018 world series playoff.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/416/1*On2bQPHHcc25JFHOwrSx2g.png" /><figcaption>Image courtesy mlb.com</figcaption></figure><p>There are a lot of factors to consider while deciding the batting order. One line of thought focuses on providing most batting opportunities to the best batters. This is accomplished by sorting the lineup by batting prowess. But that would mean that Sandy Leon, the worst hitter, would go last and therefore, precede Mookie Betts in subsequent innings. This might be undesirable as Mookie Betts and high home run potential might be under-utilized. In 2018, home runs drove in 40% of all runs scored in MLB. Preceding strong home run hitters by players with high on base percentages would be a sound strategy. These are often conflicting ideas which can be tricky for a human mind to evaluate.</p><p>Thankfully, math and statistics can answer such questions through simulation and optimization techniques. For this, we can start with batter’s true talents. We have observed that batter’s true talent can be identified by observing them bat often enough (statistically). A batter’s strikeout rates stabilize at around 60 plate appearances, walk rate stabilizes at around 120 plate appearances, home run rate stabilizes at around 170 plate appearances and so on. True talents for all 9 batters for all possible plate outcomes can be converted to the expected probability for a plate appearance to end in a single, double, strikeout, home run, etc.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/526/1*ZY2GnM4knw3Vyhx7laenXQ.png" /><figcaption>Image 2: Probabilities of plate outcomes for the Red Sox position players</figcaption></figure><p>Through the use of all plate appearances from the 2018 season, I also computed probabilities for how runners on base move when a batter hits a single, double or triple. For instance, out of all cases when a single was hit with a runner on second base, the runner completed the run 55.7% times and ended on third base on 44.3% cases. We can assume that these probabilities remain more or less constant for all future hits in this situation. I computed these probabilities for all 8 on-base situations for singles, double and triples. Note that I have excluded situations where the runner was out at home and have counted them under third base.</p><p>Using the probability of a plate outcome and the probabilities of the base changes, we can simulate the entire 9 innings of the Red Sox batting order. If we repeat the simulation a large number of times and average the runs scored in each iteration, we get the expectation of the runs scored for the batting order.</p><p>For this analysis, I’ve considered only singles, doubles, triples, home runs, strikeouts, field outs, force outs, walks and hit by pitches. A better approach to identifying the player skills would be to take the starting pitcher’s handedness into account. In this example, we could have replaced the probability of plate outcomes by those against left handed pitchers since Kershaw is left-handed and he’ll be pitching a big chunk of the innings. I’ve also assumed no strategic sacrifice plays, stolen bases, pinch runners, pinch hitters, etc. These strategies can be plugged into the simulation. I have also assumed that there will be no double plays in the game. Adding these complexities is very much feasible in this approach.</p><p>Now that I have the disclaimers covered, let me jump to the results. For each batting order that I inputted in the algorithm, I ran 100,000 iterations and compared the average runs scored for each combination. The original batting order that played the game led to 3.04 runs over 9 innings. I tried all batting orders that I thought were intuitively reasonable. All these batting orders resulted in runs within 0.7% of this value. This might look like a small gain. But in today’s baseball, the gap between the analytical prowess of the teams isn’t as big as it once was. Any small gains that can be identified would add up and could become big enough over the 162-game long season.</p><p>The best combination I found was one where JD Martinez moved up to 2nd after Mookie Betts and Andrew Benintendi and Steve Pearce moved down a place each. This was 0.69% better than the original batting order. Another interesting batting order was the one which was directly sorted by the On Base Percentages of the batters. That batting order was 0.46% better than the original one. In any order where I tried replacing Jackie Bradley Jr., the last batter, by one of Rafael Devers, Bogaerts or Kinsler reduced the run scoring by 0.5% or more. This could be because Jackie Bradley Jr. has a lower Home Run hitting probability. There is a 44% chance that at least 2 of the top 4 will be on bases when 5th hitter comes to bat in the first inning. A good home run hitter coming in right after them could be useful. That might explain why Xander Bogaerts batting in 5th might be the best alternative.</p><p>Another useful consideration could be the variance in the scores. The Red Sox starting pitcher in this game was Chris Sale. He is one of the best pitchers in the game and the Red Sox might expect him to not give away too many runs. They might anticipate two runs to be a winning score in this game and might decide to go with a batting order which has the least iterations resulting in less than 2 runs. Again, the algorithm is flexible enough to incorporate all these considerations.</p><p>Irrespective, this approach could prove to be a marginal gain for teams while deciding the batting orders. Adding more complexity would only make the approach better. The fact that Jackie Bradley Jr. was batting last in the actual lineup suggests that a similar approach is already being implemented by the Red Sox in some capacity.</p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fa6fae0af56d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revisiting the Pythagorean Expectations]]></title>
            <link>https://medium.com/@vibhormbic91/revisiting-the-pythagorean-expectations-61d3a4d05855?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/61d3a4d05855</guid>
            <category><![CDATA[sabermetics]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[mlb]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[nba]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Tue, 08 Jan 2019 08:17:55 GMT</pubDate>
            <atom:updated>2019-01-08T08:17:55.507Z</atom:updated>
            <content:encoded><![CDATA[<p>According to <a href="https://en.wikipedia.org/wiki/Pythagorean_expectation">Wikipedia</a>, Pythagorean expectation is a sports analytics formula used to estimate the percentage of games a baseball team “should” have won based on the number of runs scored and allowed. The assumption is that baseball teams win in proportion to their “quality”, and that their “quality” is measured by the ratio of their runs scored to their runs allowed. If this is true, the probability of wins can be rewritten as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/442/1*THJ3OPN6tFsisg-A_Oi1VA.png" /></figure><p>Slight simplifications to the formula converts it to</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/313/1*xQthVyi1aQBghq1aRDgoQg.png" /></figure><p>The same fundamental can be <a href="http://www.espn.com/nba/stats/rpi">extended to NBA</a> as well. We replace runs by points and re-tune the exponent co-efficient. Historically, it has been found to lie somewhere in the range of 13 to 17 for the NBA.</p><p>Firstly, what does a different exponent value mean?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/511/1*GEi64-p5CBVse6EpDP016w.png" /><figcaption>Table 1: Win Percentages for different score ratios and exponents</figcaption></figure><p>Table 1 shows how the win percentage changes as the exponent is changed. As you run down the table, the exponent increases. As we move away from the central column (score ratio = 1), the scoring ratio increases/decreases. As we run down the table, a small change in the scoring ratio away from 1 leads to a larger change in the win percentages. This is in line with our understanding of the two sports as well. In MLB, the variation between win percentages isn’t very high. The best teams win somewhere in the range of 105 games which is only 65%. In the NBA, the best teams win close to 67 games which is 83% of the total games in a season. The coefficient for the NBA is high also because of the nature of the two sports. While basketball teams score roughly 110 points per game, baseball teams score close to 4.5 runs per game. You still only need only one extra point to win a basketball game. Therefore, a 5% increase in scoring in basketball will lead to a lot more points and potentially, lot more wins.</p><p>The purpose of this article is to improve the win expectation formulae by going more specific, based on what we know about the quality of the divisions/conferences in the sports.</p><h3>NBA</h3><p>We know that point scoring in the <a href="https://medium.com/@vibhormbic91/understanding-the-new-nba-eae8377cd946">NBA has changed</a> a lot in the last couple of years. So, I’ll use only the previous two seasons; 2016–17 and 2017–18. Simulating the errors at different exponent values shows that the root mean square error (RMSE) minimizes itself at 5.45% when the exponent is 14.23. For reference, <a href="https://www.basketball-reference.com/about/glossary.html#wins_pyth">basketball-reference.com uses 14</a>, <a href="http://www.espn.com/nba/stats/rpi">ESPN uses 16.5</a>.</p><p>The NBA has shifted focus to a per 100 possession basis. Let us see if moving the computation to a per 100 basis would improve the win projection. I don’t expect a lot of difference because most matches end up having similar number of possessions. Still, it might be worth checking. Repeating the process using per 100 possession stats leads us to almost the same RMSE. It improved by 0.005 wins, if that counts.</p><p>Additionally, the <a href="https://www.newsweek.com/how-much-better-nba-western-conference-eastern-conference-lebron-lakers-1005284">Eastern Conference has been much weaker</a> than the Western Conference. In those two seasons, teams from the Western Conference beat their Eastern Conference counterparts in 53.44% of the games. Let us see if there is merit in having different exponents for inter-conference games. Table 2 shows the results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/319/1*VcAlQDp6Ba0IRq-AdFWMqA.png" /><figcaption>Table 2: NBA Pythagorean Expectation at a conference level</figcaption></figure><p>Breaking the pythagorean formula by conference reduces the RMSE from 5.45% to 5.18%. This is an equivalent of <strong>0.22 wins</strong>. The exponents themselves make for interesting reading. The coefficient is highest for games within the eastern conference. Since the teams in the east are more closely bunched, a small points advantage leads to more wins. The coefficient is lowest for the inter-conference games where a small increase in points doesn’t translate into as many wins because of the difference in the quality between the teams.</p><h3>MLB</h3><p>Similar to NBA, there is no consensus regarding the exponent to be used in the MLB win expectation formula. To give you an example, <a href="https://www.baseball-reference.com/bullpen/Pythagorean_Theorem_of_Baseball">Baseball reference uses 1.83</a> while <a href="http://www.espn.com/mlb/stats/rpi">ESPN uses 2</a>. To make the calculations for the exponent more granular, it would be ideal if we could somehow isolate the weaker divisions from the stronger ones. Fortunately, MLB does not have a perennially dominant division and predominantly weak division. It is hard to argue that one of the six divisions has been exceptionally strong or weak for a long time. Competitiveness for MLB teams that do not spend exorbitant amounts is cyclic in nature. Also, teams play very few games against teams from a division of the other league. Couple these two facts together and it makes sense not to use exponents at a league-division level.</p><p>My next instinct was to take a similar route as we took for the NBA. Just like we had different exponents for different NBA conferences, I tried to see if different exponents for NL and AL would improve the calculations. For this, I used the previous 9 seasons (2010–2018) of <a href="https://www.retrosheet.org/gamelogs/index.html">game logs from Retrosheet.org</a>. Over that period, the overall error is 2.47% at 1.83 and 2.63% at 2. The projections at baseball-reference work better. Table 3 shows the results</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/261/1*P_J6a-8N14oCpVVJROHkjQ.png" /><figcaption>Table 3: MLB Pythagorean expectation at a league level</figcaption></figure><p>As you can see, the ploy is unsuccessful for the MLB. We end up increasing the error. That is mostly because of the low sample sizes in the inter-league games. Only 11.7% of the games involved one NL team and one AL team. With neither of the two leagues having a consistently better quality, it is probably better that this did not work. Even if the results were preferable currently, they might not hold for the near future as the competitive balance continually swings from one side to the other.</p><p>But there is something interesting to note in table 3. <em>All of the exponents are below 1.83</em>. This might indicate that even 1.83 is a high exponent value. I re-did the entire exercise at the overall level (same as the league tables that we see) and found that the RMSE was minimized when the exponent was 1.74. Image 1 shows the exponent values for optimal RMSE every year in that period.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*tQTMhVV8JqYwAxHUwYEhPg.png" /><figcaption>Image 1: MLB Pythagorean exponent for optimal RMSE</figcaption></figure><p>The orange line represents 1.83, which is the exponent used by baseball-reference. The blue line, which represents the exponent for optimal RMSE has been below the orange line for the last 6 years now. This almost exactly coincides with the wide-spread introduction of defensive shifts in MLB. Shifts have reduced run-scoring in the MLB. That might have brought the exponent values further lower than where we thought they should be. The optimal RMSE for that period is 1.72. Changing the exponent from 1.83 to 1.74 improves the predictions by <strong>0.04 wins</strong>. This isn’t significant enough to demand a widespread change in the formula. Nevertheless, it does remind us to constantly monitor the value, especially if the sport takes a major turn some time in the near future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=61d3a4d05855" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding the new NBA — Part II]]></title>
            <link>https://medium.com/@vibhormbic91/understanding-the-new-nba-part-ii-dff6e07429ce?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/dff6e07429ce</guid>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[nba]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Sun, 06 Jan 2019 00:51:37 GMT</pubDate>
            <atom:updated>2019-01-06T01:10:18.721Z</atom:updated>
            <content:encoded><![CDATA[<h3>Understanding the new NBA — Part II</h3><p>In <a href="https://medium.com/@vibhormbic91/understanding-the-new-nba-eae8377cd946">part I</a> of this series, we saw that the advent of the three-point revolution has made a lot of changes to the NBA. These changes had otherwise gone almost unnoticed. Today, I’ll explain how these changes have impacted the roles of the players playing in different positions.</p><h3>Shot Attempts</h3><p>We know that teams have recently been attempting a lot more three pointers. The next question is, “Players from which position are taking more shots?” Image 1 shows the percentage of 3 point attempts for a position as a proportion of total 3 point attempts. The black dashed line is the 20% line. If all five players take equal three point attempts, this is where they would end.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*vy4UUdkfI9S0zs4LHEY0XQ.png" /><figcaption>Image 1: Proportion of total 3-point attempts</figcaption></figure><p>Other than the orange line for Shooting Guards, all other lines are converging towards the 20% line. This shows that coaches are looking to spread the three point shooting equally among all players on the court. Even the centers, who were traditionally expected to stay close to the rim, are now attempting almost 7% of the team’s three-pointers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*xFQz2cqEJSKIbAFEqA3ZFg.png" /><figcaption>Image 2: 3 pointers as a proportion of the player’s attempts</figcaption></figure><p>Excluding centers, players from all other positions are attempting 35-40% 3 point shots. For a position like Power Forward, that is a 25% increase over the last 5 years. But some of that is due to a decline in shooting frequency from power forwards. Image 3 is similar to image 1 and shows how teams are distributing their two point attempts among the different player positions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*rKWiTgd4UDwM-X9vEuZGvA.png" /><figcaption>Image 3: Proportion of total 2-point attempts</figcaption></figure><p>As one would expect, two point attempts were historically much more closely bunched across all positions. Contrary to what we saw for the three pointers, it is now diverging away from the 20% line. Forwards are seeing a significant dip in their 2 point attempts.</p><h3>Shooting Accuracy</h3><p>In part I of this article, we concluded that players hadn’t improved in their 3-point shooting. Image 4 shows that it isn’t completely true.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*ELht4kEF9DcfFl3CH0coWQ.png" /><figcaption>Image 4: 3 point shot conversion</figcaption></figure><p>While all other positions have been fairly bunched up in terms of their 3-point shot conversion, centers have been converting more of their 3 point shots than they were historically used to. With more 3-point attempts going their way, they have improved their shot accuracy to justify the increased frequency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*nKGcujbFA4d4mmCAV17szA.png" /><figcaption>Image 5: 2 point shot conversion</figcaption></figure><p>Image 5 shows the 2 point shot conversion rates for players in different positions. I have changed the scale of the y-axis to emphasize the increase in shot conversion which wasn’t so clearly visible without the scale change. But have the players really improved their 2-point shooting ability? We already know that the NBA has collectively reduced the frequency of 2 point shots from outside the restricted area. Because teams are taking better shots, they would be expected to convert more. This might explain the increase in 2-point shot conversions. Of course, Gregg Popovich isn’t a firm believer as the <a href="https://medium.com/@vibhormbic91/the-spurs-are-differently-efficient-d9dababd05de">Spurs are playing a different game</a>. To ensure that there is no improvement in the shooting ability, a deeper analysis would be needed. I’ll leave that for later.</p><h3>Free Throws</h3><p>In part I of the series, we saw that free throws have decreased because of reduced perimeter penetration. Image 3 showed that the forwards are seeing fewer two point attempts. If we are correct, they should have a reduced proportion of free throws as well. Let us see if that adds up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*du7XEsuCWlJk01n3af62RQ.png" /><figcaption>Image 6: Proportion of total free throws</figcaption></figure><p>I can’t blame you if you feel image 3 and image 6 look like replicas of each other. Trust me, they aren’t! This is testament to the absoluteness of the relationship between 2 point attempts and free throw attempts. Forwards, who were getting a lower proportion of 2 point attempts, are also getting a lower proportion of free throws.</p><p>I won’t continue with the barrage of line charts with another one to demonstrate the change in free throw conversion. The only evolution in that regard is the improvement in the free throw conversion from centers. They are now getting more free throws and have, therefore, improved their conversion rates as well.</p><h3>Rebounds</h3><p>As observed in the first part of the series, the game is moving towards fewer offensive rebounds because of the increased perimeter play. Since the forwards have lost most of their 2-point shot attempts, they have also seen the reduction in the rebounds. If you don’t drive into the area enough, you don’t get a lot of free throws and you don’t get a lot of rebounds either. Center players, as expected, are taking up most of the rebounds.</p><p>Defensive rebounds are also following the same trend. Opposition forwards are driving inside the perimeter less often. Hence, defending forwards are not getting close to the rim often enough to get as many rebounds as they used to. Image 7 shows this in graphical form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*SuN_GImiHJAls2kPNaDhDQ.png" /><figcaption>Image 7: Proportion of rebounds by position</figcaption></figure><h3>Ball Handling</h3><p>Point guards used to be the primary ball handlers of the team. Their role was to create better shooting opportunities for their teammates. That still remains true. But other players are starting to help them out more often. Centers, especially, have seen a significant increase in their assist production.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*gCCHDEMlnbpdnIROru5IRQ.png" /><figcaption>Image 8: Proportion of assists by position</figcaption></figure><p>In general, assist to turnover ratios have been increasing all this while. Centers and Power Forwards have led this revolution with 40% increases over the last 7 years. Other positions have stayed fairly stagnant in this regard.</p><h3>Conclusion</h3><p>The role of a center is becoming way more demanding now. They are everywhere. They are attempting more threes, getting better at shooting 3s and free throws, creating more for their teammates and grabbing a higher proportion of the rebounds. If it continues this way, the value of an all-round big will soon reach astronomical levels. Remember, we’re only talking about an average big in the NBA. Anthony Davis is no average big. He’s absurdly good. If he wants out of New Orleans, he will have a lot of suitors.</p><p>Power forwards and small forwards are attempting fewer 2s which has led to fewer rushes to the free throw line and fewer rebounds as well. Power forwards have improved their assist rates, though. All in all, if you’re not a center, you’d better be a good 3-point shooter and a good shot creator to have a good NBA career. Of course, a good defense also helps.</p><p><em>Data courtesy basketball-reference.com</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dff6e07429ce" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Spurs are differently efficient]]></title>
            <link>https://medium.com/@vibhormbic91/the-spurs-are-differently-efficient-d9dababd05de?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9dababd05de</guid>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[san-antonio-spurs]]></category>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[nba]]></category>
            <category><![CDATA[sports]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Wed, 02 Jan 2019 23:15:19 GMT</pubDate>
            <atom:updated>2019-01-04T02:11:43.872Z</atom:updated>
            <content:encoded><![CDATA[<p>In the <a href="https://medium.com/@vibhormbic91/understanding-the-new-nba-eae8377cd946">previous article</a>, I explored the evolution of the NBA in the last few years. Today, I’ll analyze one of the greatest teams of the modern NBA, the San Antonio Spurs. This season, the Spurs rank 5th in offensive efficiency. Let us see their mantra behind achieving that efficiency.</p><h3>Pop hates the 3</h3><p>Prior to their game against the Chicago Bulls towards the end of November, their boss <a href="https://bleacherreport.com/articles/2808516-gregg-popovich-says-hes-hated-the-3-point-shot-for-20-years">Gregg Popovich commented that he hated the 3-point shot</a>. The numbers back that up as well. In the ongoing 2018–19 season, only 27.4% of the Spurs shot attempts have been three-point attempts. This puts them rock bottom in 30th place in the NBA. For context, the league average has been 35%. This is not a new trend either. Last season, the league average was 33.5% and the Spurs attempted only 28% 3-pointers.</p><h3>Few restricted area shots</h3><p>The single most efficient shot in the NBA currently is one taken from inside the restricted area. With 24.4% of their shots coming from the restricted area, the Spurs rank 30th in the frequency of these shot attempts as well. The Spurs have managed to be the fifth most efficient offense in the league even though they rank dead last in the frequency with which they take the two most efficient shots in the NBA.</p><h3>Not many free throws either</h3><p>In my previous article, we saw that free throws could compensate a portion of the decrease in the points expectation created because of teams taking fewer three pointers. The Spurs have gotten to the line 21.5 times on every 100 possessions. That puts them in 23rd position for the season. They were 17th with 22 free throws free throws last season.</p><h3>Great shooting</h3><p>Clearly, the Spurs are not focusing on taking efficient shots. They aren’t getting to the line either. Surely, they have to be great at shooting the shots that they take. Image 1 agrees.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*RN_qy93kpUpyOz62THxuqA.png" /><figcaption>Image 1: Spurs Field Goal Conversion Percentages against League average</figcaption></figure><p>The Spurs are an above average shooting team from each location on the court. They have been the best in the league from the paint (excluding restricted area shots) and from above the break three point attempts. On converting corner threes, they are second behind the Warriors. From the restricted area and from mid-range shot attempts, they are 8th and 6th respectively. That is too good. Even when it comes to converting from the free throw line, the Spurs are the best in the NBA at 82.3%. Surely, the Spurs must be doing something different which makes these theoretically bad shots really good.</p><h3>Players playing to their strengths</h3><p>In sports, we often say that teams should play to their strengths. Gregg Popovich has set his team up such that each of their players are playing to their strengths. Players are shooting almost exclusively from parts of the court where they are their own best selves. DeMar DeRozan and Le Marcus Aldridge have a 28% career success rate from 3 point ranges. Therefore, they have barely attempted 3-point shots. Only 5.6% and 2.3% of their respective shots have been three point range. Similarly, more than half of the shot attempts from Bryn Forbes, Patty Mills, Marco Belinelli and Davis Bertans have been three-pointers. From any range, the bulk of the shooting is being carried out by the best shooters. You can view this in my Tableau visualization <a href="https://public.tableau.com/views/SpursShotTaking2018-19/Spurs2018-19?:embed=y&amp;:display_count=yes&amp;publish=yes">here</a>.</p><p>The teams right above and below the Spurs in terms of offensive efficiency are the Pelicans and the Clippers. Looking at the strengths of their players and where they take their shots from shows how unique this strategy from the Spurs is. On those teams, players who are below average shooters from any given range take a significant proportion of their shots from those ranges. Gregg Popovich has somehow managed to take the adage of ‘players playing to their strength’ to the extreme. Soon, the league may figure out what the Spurs are doing and follow suit. Alternatively, they may start to realize this and re-organize their defenses to block this strategy from the Spurs. Either way, the next few months will be pretty exciting.</p><p><em>Data used is from stats.nba.com as of 12/31/2018</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9dababd05de" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding the ‘new NBA’]]></title>
            <link>https://medium.com/@vibhormbic91/understanding-the-new-nba-eae8377cd946?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/eae8377cd946</guid>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[nba]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Mon, 31 Dec 2018 00:51:51 GMT</pubDate>
            <atom:updated>2019-01-06T00:53:57.355Z</atom:updated>
            <content:encoded><![CDATA[<h3>Understanding the ‘new NBA’ — Part I</h3><p>Almost 7 seasons ago, a young <a href="http://grantland.com/the-triangle/title-tk-the-james-harden-kevin-martin-shot-chart-comparison/">James Harden was revolutionizing the NBA</a> with his frequency of 3-point shot attempts and shot attempts at the rim. Today, we all have become accustomed to the ‘new NBA’ where 30 three point attempts per game is a norm for a good team. This glamorous three point revolution was accompanied by a lot of other changes which haven’t really come to the forefront of the public eye. This two part series focuses on identifying the changes that have happened.</p><h3>Focus on each possession</h3><p>In 2003, <a href="https://twitter.com/DeanO_Lytics">Dean Oliver</a> published Basketball on Paper. In that book, he noted that possessions are the currency of a basketball game. <a href="http://philbirnbaum.com/wagesofwinsreview.pdf">Oliver notes that</a> for basketball, it’s points per possession, and not points per game, that’s important. He suggested that a team scoring 100 points in 100 possessions was better than another team that scored 102 points in 105 possessions. This is because the opposition team would also have the same number of possessions. Efficient use of each possession was the key to success in an NBA game. This gave rise to the metric ‘points per 100 possessions’, which is the holy grail for NBA analysts.</p><h3>Pace</h3><p>This idea also brings us to the concept of pace in a basketball game. Pace is simply the average number of possessions a team plays in a 48 minute basketball game. Image 1 shows the evolution of pace and points per 100 possessions in basketball.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*5JwKkXh8I0-ZUa3vIBykwQ.png" /><figcaption>Image 1: Evolution of Pace (number of possessions per 48 mins) and points per 100 possessions in the NBA</figcaption></figure><p><em>Note: Free throw points are not included in this image.</em></p><p>As you can see, the points per possession had been rising even before the three-point revolution. One may argue that it is more accelerated now. More strikingly, pace has also seen a continual increase since the beginning of the revolution in 2012. Neither Dean Oliver’s theory nor the increase in the frequency of three point shots explains the increase in pace, though. So, why did the pace increase?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*TofKUr3bOBuLoQo3RMpb5g.png" /><figcaption>Image 2: Field goal conversion by time left on the shot clock</figcaption></figure><p>Image 2 shows the variation in the field goal conversion percentages with the change in time remaining on the clock. Clearly, shots taken early in the possession have a higher conversion rate. This is mostly because of the inability of defenses to get back to their defensive positions in such a short time span. This knowledge led to teams focusing more on fast breaks and transition plays. The reduction in time per possession explains the increase in the number of possessions per game (or pace) as well.</p><h3>Shot conversion rates</h3><p>This change in pace of the game should have directly increased the field goal conversion rates. Let us see which range of shots has seen this increase.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/526/1*0PcQhwk2tZc3ZKYzpDb7uA.png" /><figcaption>Image 3: Field Goal Conversion % in NBA (1996–2018)</figcaption></figure><p>Image 3 reveals that the NBA players have not gotten better at making shots. The increased points per possession should be solely explained by the three point revolution, then. Image 4 confirms that.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/526/1*mKPBQncOvMAaehjqM2WcjQ.png" /><figcaption>Image 4: % of Field goal attempts in NBA (1996–2018)</figcaption></figure><p>As you can see, the yellow color of the mid range shots has been constantly decreasing and the three pointers have replaced those shots.</p><h3>Free Throws</h3><p>Remember, our discussion on points per possession has not covered free throws. Intuitively, one would believe that the increase in three point attempts would lead to reduced contact between players and therefore, fewer players would be fouled in the offensive half. This may have offset some or all of the increase in field goal points. Let us see if that is the case.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*pM3phApq0tzRVOigd1jhXg.png" /><figcaption>Image 5: Free Throw attempts and conversion rates in the NBA (1996–2018)</figcaption></figure><p>Image 5 confirms our hypothesis. Careful examination of the orange line which represents free throw attempts per 100 possessions shows that free throw attempts have seen two significant drops in the past decade. The first drop is approximately worth 2 free throws and coincides perfectly with the beginning of the three point revolution in 2012. The second drop of 1.7 free throw attempts per 100 possessions came in the previous completed season. I checked the corresponding numbers for this season and found that half the deficit has been recovered this season. This recovery might well be attributed to the new ‘freedom of movement’ related rule changes introduced in the NBA this season.</p><p>To see if the reduction in free throw points has offset the increase in points per 100 possessions (which, remember, did not include free throws), let us create another metric which adds them up. I call it total points per 100 possessions. We’ll compare total points per possession for two 7-year periods, 2004–2011 2011–2018.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/385/1*fSrFMCZj509eZ3LAZflnaA.png" /><figcaption>Image 6: Impact of free throws (per 100 possessions)</figcaption></figure><p>Image 6 shows that the field goal points have increased by 2.51 points per 100 possessions and free throw points have decreased by 2.49 points per 100 possessions. Contrary to popular belief, this chart shows that the <em>three-point revolution hasn’t increased point scoring at the NBA</em>. This doesn’t seem right. If it were not beneficial, so many analytics heavy teams would not be focusing on the three point shots. Surely, there is something more to it that we are missing.</p><p>A deeper look at image 4 shows that the proportion of three point shots had another major increase in the 2016-17 season. In the 2015–16 NBA season, 28.3% of all shots were 3-pointers. This figure rose to 31.4% in 2016–17 and 33.5% in 2017–18. Reproducing image 6 for the previous two seasons answers the mystery:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/388/1*CnfcipogAt_IS9JkIqx66w.png" /><figcaption>Image 6b: Impact of free throws (per 100 possessions)</figcaption></figure><p>The points earned on free throws have gone done by 0.5 while the field goal points have increased by more than 3.3 points per 100 possessions. The total points per 100 possessions has increased to almost 108. This could be because not all teams took to the three-point revolution immediately. Most teams joined the three-point bandwagon in the last couple of years which has led to the increased numbers that we are now seeing.</p><h3>Rebounds</h3><p>With the increased focus on three point shots which have a slightly lower conversion rate, we should expect more rebounding opportunities. Image 7 explores that dimension of the game.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*mcQ0sgqcPh5pNFXoT1LdFQ.png" /><figcaption>Image 7: Rebounds in the NBA (1996–2018)</figcaption></figure><p>The three-point revolution has contributed to a very slight increase of almost 0.5 rebounds per 100 possessions. The decreasing height of the blue bar is telling, though. The offensive rebounding rate has been consistently decreasing. This should be explained by the fact that the offensive players are farther from the rim. This is confirmed by another statistic; Offensive rebounds as a proportion of total rebounds have a direct negative correlation (-0.91) with the proportion of three point field goal attempts.</p><h3>Turnovers</h3><p>I don’t expect a change in the turnover counts because of the three-point revolution. The increased pace of play might have an impact, though. In the quest to play faster, offenses might make more mistakes which could manifest itself in turnover counts. On the contrary, turnover rates might be directly correlated with the length of a possession. Therefore, lesser time might also mean fewer turnovers. Image 8 shows that the second theory plays itself out. Turnover rates have gone down almost every year since 2012.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*pVIrfPOXoNVSP3-j2sEk0A.png" /><figcaption>Image 8: Turnovers per 100 possessions in the NBA</figcaption></figure><p>Turnover rates have a strong negative correlation (-0.76) with pace which gives more substance to the theory layed out above.</p><h3>Conclusion</h3><p>The heavily touted three-point revolution more has led to a decrease in free throws, an increase in defensive rebounds and a decrease in offensive rebounds. The cumulative effect of these changes has increased the shooting efficiency of the teams. Another significant change in the ‘new NBA’ is the increase in the pace. It has led to increased turnovers. For success in the ‘new NBA’, the key differentiation is the adaptability to an increased pace and being better than the rest of the pack at one or more of the ‘four factors’ (effective field goal percentage, rebounding, turnovers or free throws), both offensively as well as defensively. Being good at three-pointers alone might not be enough.</p><p><a href="https://medium.com/@vibhormbic91/understanding-the-new-nba-part-ii-dff6e07429ce">Click here</a> for part 2 of this series where I explain the evolution of the role of different player positions in the NBA.</p><p><em>Data courtesy stats.nba.com. All statistics are for the regular season only.</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eae8377cd946" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Value of a possession in football]]></title>
            <link>https://medium.com/@vibhormbic91/value-of-a-possession-in-football-4e54aaec71d4?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e54aaec71d4</guid>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[soccer]]></category>
            <category><![CDATA[football]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[soccer-analytics]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Sat, 29 Dec 2018 04:21:10 GMT</pubDate>
            <atom:updated>2018-12-30T02:54:21.889Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Note: The article has been edited on 12/29 such that a chain of a possession is counted as a single possession</em></p><p>Scoring more goals than the opposition: That is the objective of a football (or soccer, depending on where you are) team in each game. Therefore, the roles and responsibilities of most players can be broken down to a combination of a) scoring goals, and b) stopping goals. Some actions can be directly linked to one of these two objectives. For example, shots, forward passes, crosses, dribbles, etc. are played with an objective of scoring goals whereas blocks, clearances, tackles, etc. are played with an objective of stopping goals. But how about a ball recovery like a tackle or an interception? How about lateral passes and backward passes? The objective of these actions is to either regain possession or to maintain possession. Therefore, this question arises: what is the value of these actions towards scoring or stopping goals? In different words, <em>what is the value of a possession</em>?</p><p>To answer this question, we’ll use the data made <a href="https://github.com/statsbomb/open-data">publicly available by StatsBomb </a>(Thank You, StatsBomb!). This data represents the FIFA World Cup 2018, FA Women’s Super League 2018/19 and NWSL 2018.</p><p>Firstly, let us see what an average shot in each of these tournaments was worth. We’ve already seen that <a href="https://medium.com/@vibhormbic91/why-do-we-need-xg-c37b228f7155">xG is a better measure for predicting goals</a>. Hence, we’ll use xG for this purpose. Image 1 shows the average xG per shot for these three tournaments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/632/1*OXCdLAK6q0djAJtbA3PQ3w.png" /><figcaption>Image 1: Average xG per shot</figcaption></figure><p>The image shows us that the chances of a shot being turned into a goal range from 9.7% to 10.8%. The three tournaments combined yield 10.4% as the xG/Shot. For purposes of this article, we’ll assume that this is the actual figure.</p><p>Next, let us find out what proportion of possessions convert to shots. A little more slicing and dicing tells us that 19.2% of all possessions end in shots. The variation across the three competitions is not too high either. Simple multiplication of these two figures (10.4% * 19.2%) will give us the xG per possession. This equates to <strong>0.0199 goals per possession</strong>. So, each time a player loses possession, their team loses roughly 2% of a goal. Also, the opposing teams gains roughly 2% of a goal. So, <strong>every possession costs approximately 4% of a goal</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/135/1*fdexY91s5keJuJuEtg9WwQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/235/1*OO1lL_hOdJro32iWtNWJYg.png" /><figcaption>Image 2: Possessions, shots and xG per game</figcaption></figure><p>This information could be useful. We can use this to <em>compare players based on their pass success rates</em>. For instance, let us say that player A has a 70% pass success rate. Out of every 100 passes that player A attempts, he/she concedes possession 30 times. He/She costs the team 1.2 (30 * 4%) goals on every 100 pass attempts. Similarly, <em>every interception and/or tackle that wins possession can be considered as a 4% swing in chance of a goal</em>.</p><p>Before we get too far ahead of ourselves, let me clarify that this is just a framework that can be used to compare the contribution of the players on all different aspects of the beautiful game of football. In further studies, we can incorporate the position of the ball on the pitch. A ball lost in the defending team’s third might cost more than 4%. Similarly, a ball lost in the midfield that triggers a counter attack for the opposition would be worse than 4%. We can incorporate all of these factors in the form of position on the pitch, players of the defending/attacking team behind/ ahead of the ball, type of possession, etc.</p><p>Let us look at image 3 which shows how much the type of possession impacts its value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*se6kxHwIjpiVnFmJPQecqQ.png" /><figcaption>Image 3: Value of different kinds of possessions</figcaption></figure><p>This seems counter-intuitive to me. Data suggests that counter-attacks are the least productive attack form. This is inexplicable to me. If anybody has any rationale for why that may be the case, please put it in the comments. Similarly, free kicks are equally productive as regular plays. Of course, a lot of regular plays culminate in free kicks, corner kicks, etc. which are currently included in the regular play section itself. These would be masking the true value of a free kick or a corner.</p><p>In my opinion, a framework like this could be most useful in evaluating midfielders. Passes, interceptions, tackles, 50/50 balls, aerial duels, etc. can all be assigned a weightage towards goals. We can incorporate a lot of game context into the framework. The more we add, the better we may become in evaluating players. We may be able to help coaches and players with strategic in-game decisions as well. We can say exactly how effective time wasting near the corner flag towards the end of the game could be. We may also be able to say whether clearances sent out of bounds are better than clearances played down field. We have often seen that offensive players chase over-hit crosses and prevent them from going out for an opposition throw in and instead give it away to the opposition team. If this data is anything to by, coaches should players to not prevent it from going out unless they can retain possession of the ball. Such benefits, although miniscule in nature might add up to create a significant statistical advantage for teams who are interested in analytics.</p><p><em>All data updated till December 29, 2018</em></p><p><em>Data courtesy StatsBomb (</em><a href="https://github.com/statsbomb/open-data"><em>https://github.com/statsbomb/open-data</em></a><em>)</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e54aaec71d4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Will PSG join the Invincibles list?]]></title>
            <link>https://medium.com/@vibhormbic91/will-psg-join-the-invincibles-list-793d9e587d9a?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/793d9e587d9a</guid>
            <category><![CDATA[juventus]]></category>
            <category><![CDATA[football]]></category>
            <category><![CDATA[paris-saint-germain]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[liverpool]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Mon, 24 Dec 2018 22:15:29 GMT</pubDate>
            <atom:updated>2018-12-24T22:15:29.448Z</atom:updated>
            <content:encoded><![CDATA[<p>In the world of European soccer leagues, ‘Invincibles’ is a term given to a team that plays out an entire season of league football without losing a single game. Let me give you a sense of the rarity of this event; this has happened thrice in the big five European Leagues (discounting occurrences before the second world war).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/537/1*icJBZF7-p-4TJbqkzioFmQ.png" /><figcaption>Invincibles</figcaption></figure><h3>Is it possible this season?</h3><p>At Christmas this season, three teams are unbeaten and are potentially heading for this remarkable achievement. Each of Juventus, Liverpool and Paris Saint Germain have had great seasons. Even though each of them are league toppers right now, my match by match predictions put Liverpool at second place at the end of the season. In any other season, their astronomical performance would have put them beyond the rest of the league. This season, they have Manchester City for company. Even though Liverpool currently have a 4 point lead, <a href="https://public.tableau.com/profile/vibhor.agarwal#!/vizhome/Schedule-adjustedexpectedpointstables/Schedule-adjustedexpectedpointstables">my model</a> still picks Man City for the title. You can view my model’s end of season projections <a href="https://public.tableau.com/profile/vibhor.agarwal#!/vizhome/Schedule-adjustedexpectedpointstables/Schedule-adjustedexpectedpointstables">here</a>.</p><p>Coming back from the digression to the original question, is there a possibility of an invincible season? I decided to use my models to see what odds each of these teams have right now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/545/1*rtbZmH7PrBnWzrV4pokOAQ.png" /><figcaption>Chances of 2018–19 Invincible seasons</figcaption></figure><p>The results are mind-boggling. The model predicts Liverpool to have a 10% chance, Juventus to have a 6% chance and PSG to have a 58% chance. <strong>58%</strong>. That is huge. Let me put these numbers into perspective. Assuming equal probabilities of all three outcomes (Win, Draw, Loss), the chance a team would go unbeaten for 21 games comes out to be 0.02%. Keeping that in mind, even the 6% for Juventus sounds huge. 58% for PSG is outlandish. In the social media age, if somebody were to tell me that a team has a better chance of remaining unbeaten than for being unbeaten, I would block that person and classify him/her as a lunatic. But here I am, claiming that it is actually the case.</p><p>Let me go a little deeper to see why that is the case. One of the models has shown that in the current season, if PSG were to face an average team at home, they would score 2.66 goals and concede 0.97 goals (7.4% better than average). The next best for scoring at home is Rennes (1.81 expected goals). More remarkably, PSG is topping both of these numbers in away games. They are 58% and 17% better than the next-best teams in these categories. That is absolute domination. When a team’s weakness is 7.4% better than average, that team is, loosely speaking, <em>unbeatable</em>.</p><h3>Key Games</h3><p>Let us have a quick glance at the key games that these teams need to navigate to make this possibility more realistic.</p><p>Starting with Liverpool, their toughest game is awaiting them in the beginning of the New Year. They face Manchester City at the Etihad on January 3. My model puts Manchester City at 55.5% to win the game. If Liverpool can come out of that unscathed, it will be a significant boost to their chances. Excluding that game, Liverpool’s chances of an invincible season increase to 22.8%.</p><p>Juventus has played the easiest schedule in Serie A so far. Their schedule was 9.5% better than average. They are yet to play a lot of tough away games at Napoli, Inter Milan, Lazio and Atalanta. Their next fixture on Boxing Day is against Atalanta at Stadio Atleti Azzurri d’Italia. My model is a fan of the 9th placed Atalanta team because of their performance in the games. Going by their expected goals in the games, they could have been fourth on the table. At home, they are particularly outstanding. My model puts them as the fourth best defense and 6th best offense at home games. Atalanta has a 35% chance of ending Juventus’ streak on Boxing Day. Keeping in mind that this number does not account for the absence of a certain Christiano Ronaldo makes this clash even more important in Turin’s bid for invincibility.</p><p>As I said before, the models love PSG. The only match where PSG’s opposition is pipped at a greater than 10% chance of winning is Strasbourg away. PSG play them on April 6. Interestingly enough, PSG’s second-toughest game is away at LOSC Lille on April 13. Sandwiched between these two games is the first leg of the Champions League Quarter Final. If PSG go past Manchester United, that second week of April could be pretty scary for their fans. Thomas Tuchel might decide to rest his star players for one of the away trips at Strasbourg or Lille. That could bring their chances of success down. As of now, they are in line to thrash all records and finish at an unprecedented 104 points for the season. Could they get themselves a golden trophy?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=793d9e587d9a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Schedule-adjusted expected points tables]]></title>
            <link>https://medium.com/@vibhormbic91/schedule-adjusted-expected-points-tables-5869832410cd?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/5869832410cd</guid>
            <category><![CDATA[soccer-analytics]]></category>
            <category><![CDATA[soccer]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[football]]></category>
            <category><![CDATA[sports]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Sat, 22 Dec 2018 23:57:31 GMT</pubDate>
            <atom:updated>2018-12-22T23:57:31.913Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://understat.com/">Models are already available online</a> where league standings for the big five European Leagues are adjusted for the actual quality exhibited by the teams in the game. As explained in my <a href="https://medium.com/@vibhormbic91/why-do-we-need-xg-c37b228f7155">previous article</a>, these standings more aptly indicate future performance. In that article, I introduced a model that better predicted the probability of wins and hence, points. The objective in this article is to extend those models to also include an adjustment for the toughness of schedule faced so far.</p><p>This technique is inspired by <a href="https://twitter.com/cchappas">Constantinos Chappas</a> and his <a href="https://statsbomb.com/2018/11/introducing-the-schedule-adjusted-league-table/">article published on StatsBomb</a> last month. In that article, he used current points from each game and the strength of schedule to adjust the points scored so far. I’m replacing points gained by expected goals scored and conceded. This will give me the schedule adjusted xG and xG against. Post that, I would use my model from last week and arrive at the schedule adjusted expected points.</p><p>Note: If you’re not interested in the intricacies of the technology, you can jump to the end of the article where you can see the schedule adjusted league tables with expected points for the big five leagues.</p><h3>Schedule Adjusted xG and xGA</h3><p>Firstly, we create the 190 pairs (153 for Bundesliga) of teams. Each pair of teams will have played some common games. For instance, Arsenal and Bournemouth have played three home games in common Everton, Huddersfield, Leicester and Liverpool. In these four games, Arsenal’s expected goals averaged at 1.87 as against 1.54 for Bournemouth. This results in an expected goal difference of 0.33 per game. Once we compute such figures for all 190 team pairs, we can implement a non-linear optimization function which generates the optimal xG values for all 20 teams such that the Root Mean Square Error (RMSE) is minimized. In simple terms, the xG values obtained as an output from the optimization algorithm can be perceived as the xG for the teams against an average team from their league.</p><p>I repeated the above process to obtain these values for xG at away games, xG conceded at home games and xG conceded at away games. The RMSE for each of these ranged from 0.33 to 0.41. Each of these 80 values can be used as a proxy to assess the difficulty of the game. For instance, the difficulty for Chelsea when they host Leicester tomorrow can be defined using Leicester City’s expected away goals of 1.18 per game and expected away goals conceded of 1.42.</p><h3>Fixture difficulty using xG and xGA</h3><p>In my previous article, I introduced a model to compute win probabilities using xG and xG against. Feeding the xG scored and xG conceded values for Leicester’s away games into the model, we can determine the probability of an average EPL team beating Leicester at home. We can also convert this into an expected points for the home team. Once we have this for all Chelsea games, we can average them up to determine the average points per game that an average team would have scored in those games. This figure, after scaling it around 1, becomes the difficulty of the games that Chelsea has faced so far. The higher the value, the easier the games.</p><p>We can then divide the expected points of the team so far by the fixture difficulty to determine schedule-adjusted expected points.</p><h3>Results</h3><p>The results of this exercise are encouraging. One one extreme was Eintracht from the Bundesliga whose schedule so far has been 7.5% harder than an average Bundesliga team. On the other extreme were teams like Atletico Madrid, Nantes and Mainz 05 whose schedule so far have been 6.5% easier than an average team in their respective leagues.</p><p>Combining this with the expected points from each match yields interesting findings. Let us have a look at the top of the tables. In the Premier League, Pep Guardiola’s Manchester City posts astronomical expected points figures and are pushed to the top even though both teams have faced an easier-than-average schedule. Valencia are the biggest movers. Their expected points and schedule adjustment combined has put them up to 3rd in the league table. That makes their 14th table in the league look really bad. I expect their performances to start converting to points sooner rather than later. In Germany, Borussia Dortmund has been pulled down to 5th. But then, the xG models have never been able to figure out Lucien Favre. I can’t comment on that till the football analytics community figures his teams out. AS Monaco have been unlucky and the models put them on 9th instead of their actual 19th position. Expect their performances to also convert to points soon.</p><p>For the complete tables, please visit <a href="https://public.tableau.com/views/Schedule-adjustedexpectedpointstables/Schedule-adjustedexpectedpointstables?:embed=y&amp;:display_count=yes">my Tableau Public page</a>.</p><p><em>All data sourced from </em><a href="http://www.understat.com/"><em>www.understat.com</em></a><em> and is updated till December 20, 2018</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5869832410cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why do we need xG?]]></title>
            <link>https://medium.com/@vibhormbic91/why-do-we-need-xg-c37b228f7155?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/c37b228f7155</guid>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[soccer-analytics]]></category>
            <category><![CDATA[football]]></category>
            <category><![CDATA[soccer]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Sun, 16 Dec 2018 23:45:52 GMT</pubDate>
            <atom:updated>2018-12-16T23:45:52.538Z</atom:updated>
            <content:encoded><![CDATA[<p>The 2015–16 season of the English Premier League was a huge anomaly. <a href="https://www.cbsnews.com/news/leicester-city-5000-1-longshot-eyes-english-premier-league-title/">Leicester City defied odds of 5000-1</a> to win the premier league title, perhaps the single biggest upset in the history of sports. But, another almost unprecedented phenomenon occurred that season. Chelsea, who were the title holders from 2014–15 finished 10th. It is rare to see the title holder perform so poorly. At the halfway stage of the tournament, Chelsea languished at <a href="https://maheshsenni.github.io/dataviz-epl-table-2015-2016/">14th on the table</a> at just 20 points in 19 games. At that stage, a lot of pundits believed that Chelsea were simply unlucky and that they would improve on their performance. They did improve to 50 points and finished 10th. But was there any evidence to suggest that they would improve? Yes, there was..</p><h3>Chelsea 2015/16</h3><p>The evidence is called xG. By the grace of some good souls, xG is publicly available at <a href="https://understat.com/">https://understat.com/</a>. We scraped the data through a Python query for this analysis. At the halfway stage that season, Chelsea’s performance looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/458/1*uyLREiKFmMsgvzvXHsK0rg.png" /><figcaption>Image 1: Chelsea 2015/16 after 19 games</figcaption></figure><p>As you can see, Chelsea had a negative goal differential which is indicative of a below average team. Incorporating their xG figures, this is how it looked:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/586/1*Fzm3bJUpoGsddhoR7Ng8Aw.png" /><figcaption>Image 2: Chelsea 2015/16 after 19 games with xG</figcaption></figure><p>Image 2 shows that Chelsea had underperformed their expected goal difference by 8.88 goals in 19 games. For those of you that may not be aware about xG, please <a href="https://www.optasportspro.com/about/optapro-blog/posts/2017/blog-expected-goals-explained/">read this explanation by Opta</a>. In most cases, this would reflect bad luck and that regression towards xG should be expected. Taking historical outcomes for the xG difference on match by match basis, understat suggests that Chelsea should have scored 28 points by then. This concrete evidence led the pundits to believe that a turnaround for Chelsea is imminent.</p><h3>Does this hold true in general?</h3><p>Yes, it does. For this, we look at data from the big 5 European leagues from the previous four seasons, 2014–15 to 2017–18. We want to see whether expected points is a better indicator of future performance. We check correlations of two metrics with the points scored by a team in the second half of the season. The two metrics are a) points scored, and b) expected points scored by the team in the first half of the season. By definition, correlation values lie between -1 and 1. The closer it is to 1, the stronger is the relationship.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/443/1*VSjvysFcSXMCA4UPJSzK5Q.png" /><figcaption>Image 3: Correlation with points scored in the second half of the season</figcaption></figure><p>As you can see, expected points have a stronger relationship with second half points for each of the big 5 leagues. This is clear indication that expected points in the first half are a better indicator of future performance than actual points.</p><h3>Can we improve the expected points model?</h3><p>As discussed previously, expected points are a reflection of historical match outcomes based on xG difference on a match by match basis. Having scraped the data for the previous 5 seasons from understat, we leveraged R to model outcome probabilities based on xG figures from 2014–2018. Using these probabilities, we computed expected points for the ongoing 2018–19 season. The objective was to arrive at a model which could improve the expected points model provided on the understat website. For this, we computed the Root Mean Square Error (RMSE) from our model and compared it against that from the model on the understat website. After spending three days on this and trying 20+ iterations of different modeling techniques, here are the results comparing the performance of our best model with the understat model:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/319/1*VZwgtNywAiOJeiSNMOba3w.png" /><figcaption>Image 4: Comparing RMSEs</figcaption></figure><p>Since RMSE is a measure of error, lower values are better. As is obvious in image 4, our model performs better than that at understat with an overall RMSE improvement of 0.014. For some reason, my model for the Serie A doesn’t perform as well, though.</p><p>Of course, the difference isn’t that huge. Our model might be better because we may be using more recent data to build the model. Understat might have refreshed the model slightly earlier. It might also be that our model only <em>appears</em> better because we are testing it on just half a season of data. More data for testing might make it look worse. Lastly, we have just 2 digits after the decimal from the understat model. With such fine margins, the models might be closer if we take more exact figures from their model.</p><p>Nevertheless, the importance of xG is clear. We’ll leave you with a chart showing the variation in win/loss probabilities from our model. Notice how beautifully the model has captured home team advantage as well. Data science is amazing, isn’t it?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/981/1*qUeZ-oyDaMFTjnVHEmqFyA.png" /><figcaption>Image 5: Outcome probability variations</figcaption></figure><p><em>All data updated till December 10, 2018</em></p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c37b228f7155" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Group stage of the UEFA Champions League? — Nah, Boring!]]></title>
            <link>https://medium.com/@vibhormbic91/group-stage-of-the-uefa-champions-league-nah-boring-d527c3e380dc?source=rss-81a4ed666fdc------2</link>
            <guid isPermaLink="false">https://medium.com/p/d527c3e380dc</guid>
            <category><![CDATA[uefa-champions-league]]></category>
            <category><![CDATA[football]]></category>
            <category><![CDATA[soccer]]></category>
            <category><![CDATA[uefa]]></category>
            <category><![CDATA[champions-league]]></category>
            <dc:creator><![CDATA[Vibhor Agarwal]]></dc:creator>
            <pubDate>Wed, 12 Dec 2018 22:48:21 GMT</pubDate>
            <atom:updated>2018-12-12T22:55:13.731Z</atom:updated>
            <content:encoded><![CDATA[<h3>Group stage of the UEFA Champions League? — Nah, Boring!</h3><p>If you are a football fan (or a soccer fan, depending on what you prefer), chances are that you believe that the UEFA Champions League is the pinnacle tournament of the club football landscape. After all, the best football clubs from Europe assemble at one stage and fight it out to be crowned the best of the best, champions of the champions, <em>the best football club in Europe</em>. Such a tournament ought to be really exciting, isn’t it? Of course.</p><p>The tournament starts with the group stage comprising 8 groups. Each group has 4 teams who play each other twice, home and away. The top 2 from each group qualify for the round of 16. For fans like me, the tournament is really exciting.. and the excitement starts from the group stage. As the recently concluded last round of the group stages of the 2018/19 Champions League have shown, it is pretty darn interesting. Even in the last game of the group stage, major powerhouses of European Football like Tottenham, Inter Milan, PSG, Liverpool and Napoli were battling to make it to the round of 16. Pretty interesting, right? Right. I used to feel the same way till I heard <a href="https://statsbomb.com/author/tknutso/">Ted Knutson</a> and <a href="https://statsbomb.com/author/jamesyorke/">James Yorke</a> from StatsBomb discuss something similar during the <a href="https://soundcloud.com/statsbomb-pod/statsbomb-podcast-december-2018-2-pl-review-development-transfer-tales-and-more">StatsBomb podcast</a> yesterday. There was this lingering thought that provoked me to test what they said.</p><p>Year in year out, more than 10 of the 16 teams to progress beyond the group stage hail from the big 5 leagues (EPL, Bundesliga, La Liga, Serie A and Ligue 1). This means that we more or less know which teams are to qualify from a group. This was disconcerting to me and I decided to check how intense this situation was. Let us first check the trends. Below is the chart showing the number of big 5 teams that progressed to the round of 16 in each of the last 10 seasons.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*SQHZX8jilqkaiHghHyF8kQ.png" /><figcaption>Image 1: Contribution of the ‘Big 5’ leagues to Champions League R16</figcaption></figure><p>An average of 13 (12.8, to be precise) Big 5 clubs have qualified to the elimination stages over the last 10 seasons. That is a staggering 75% of all Big 5 clubs that participate in the tournament. While that is a very high number, 25% of the big 5 clubs getting eliminated doesn’t look so bad for the competition. Another way to look at this is that the elimination stages comprise of 80% of the teams from the Big 5 leagues. 80% domination sounds good enough. But, wait, the domination is probably more than that. Some Big 5 teams that are going out of the group stages are actually going out because there are two more Big 5 teams in their group. If a group has three big 5 teams, one has to bow out. That’s the nature of the rules itself. Let us change the denominator of the percentage to the maximum number of big 5 teams that could have qualified.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/534/1*GkkvvY_wElm3hBy2bpMHTQ.png" /><figcaption>Image 2: Proportion of Big 5 League teams qualifying for R16</figcaption></figure><p>The orange line in Image 2 shows the proportion of teams that qualified out of the maximum that could have qualified. This season, the two groups which saw one of the non-big 5 teams go through did not have 2 big 5 teams. Those two lucky teams were Ajax and Porto. That’s 100% conversion. In the last 10 seasons, 91% of the Big 5 teams that could have qualified beyond the group stages did. In other words, non-Big 5 league teams qualified above a big-5 league team only 9% of the times. That is just 1 in 11. That doesn’t sound so good.</p><p>It doesn’t improve in the knockout stages either. In the last 9 seasons (2018/19 excluded for obvious reasons), only 7 non-big 5 teams have progressed to the quarter finals. Only 4 of those had beaten a big-5 team to go through; Galatasaray beat Schalke 4–3 in 2012, APOEL beat Lyon in penalties in 2011, Shaktar beat Roma in 2010 and CSKA Moskva beat Sevilla 3–2 in 2009. In the last 5 editions of the Champions League knockout stages, no Big 5 team has lost to a non-big 5 league team. If this is not domination, I don’t know what is.</p><p>Most of the knockout stage games involve two teams from the big 5 leagues and are intense, high quality matches. That is what the fans want. Fans don’t want to know the result before the start of a match. The knockout stages deliver that. The group stages, on the contrary, are hardly unpredictable.</p><p>The major source of unpredictability in group stages comes from the groups where more than 2 Big 5 league teams are drawn together. In groups with 2 big 5 teams, two games (the ones involving them both) mostly decide the order of the top 2. Out of the 80 Champions League groups in the last 10 years, 26 had more than 2 Big 5 League teams drawn together. That averages to be 2.6 per season. There were 5 such groups this season. This increase is attributable to the increase in the quota for the big 5 leagues. The 5 unlucky teams to go out in such situations this season were Monaco (against Dortmund and Atletico), Hoffenheim (against Man City and Lyon), Valencia (against Juventus and Man United) and the two Italian clubs; Napoli and Inter. Napoli and Inter can claim to be truly hard done by because they would have had a strong claim to the knockout stages had they been in other groups. Had they not been in the groups that they were part of, the Champions League group stages would have been more predictable, if not boring. The 5 teams that I mentioned in the beginning of this article happen to come from these two groups. That these two groups will be ultra competitive was known as soon as the chits were drawn.</p><p>Personally, I love this format of the Champions League as it builds up in intensity throughout the year and rewards a meritorious winner in the process. While I am not a big supporter, I can understand why there are <a href="https://www.bbc.com/sport/football/46078651">rumours of a Euro Super League</a>. If those teams are providing all the excitement, it makes sense for them to want to play amongst themselves and keep the fans excited for prolonged periods. I can also understand why UEFA has decided to come up with a <a href="https://www.bbc.com/sport/football/45483586">third European club competition</a>. That could reward the better teams from other countries who are currently being battered by their big 5 counterparts.</p><p><em>Penned by Vibhor Agarwal — Graduate Student, Sports Management at Columbia University</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d527c3e380dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>