Week 14: Aston Villa v. Watford Consequences

Here’s the quick rundown on the relegation six-pointer between Aston Villa and Watford. Aston Villa is a favorite in this match at home over fellow relegation contenders Watford, and with their performance so far this game is taking on “must-win” status early in the season.

Week 14 Villa v Watford

Three points would help Villa pick up about 1.4 points on expectations, which is especially important given they are underperforming by about 6 points so far. How much would this help in the relegation battle though?

Week 14 Aston Villa

According to my model, a loss puts Aston Villa in the relegation zone about 86% of the time, which is a big hole to dig out of so early in the season. A draw keeps their odds about the same, but makes them about 10% less likely to finish in last place, which gives them some room to maneuver but still isn’t ideal. A win however drops their relegation chances to about 70%, and drops their likelihood of finishing in last place by 20%. Still not ideal, but definitely would give them some hope. You can see in the above figure how much lighter the 20th place box gets with a win for Villa.

Meanwhile, a loss doesn’t hurt Watford too much, putting them at about 28% for relegation. A draw puts them at 20% chance for relegation, but a win drops that to 10%. Villa needs this win to get some hope and to have something to build on, Watford can significantly reduce their chances of relegation with a win of their own. A draw doesn’t really work for either of these teams, so I’d expect both of them to come out swinging.

Chelsea v. Spurs is a UCL 6 Pointer: Top 4 Implications of This Weekend’s Fixture

It’s awfully early to be thinking this way, but Chelsea and Spurs have a huge game this weekend. It’s really early in the season this sort of thing, but looking at the predicted probabilities this game could have significant implications on the race for fourth place. Here’s the heat map of predicted probabilities based on the potential results for both teams:

Week 14 Chelsea v. Spurs

It’s striking how similar the two heat maps look, the predicted probabilities are virtually identical for the two teams for each of the three results. A loss drops Spurs’s chances for the Top 4 down to 21%, while a loss for Chelsea drops them down to 18%. A tie puts Spurs at 29% while Chelsea has a 28% chance of finishing top 4, while a win gives both Spurs and Chelsea a 42% chance.

Thinking about pre-season expectations, the strategies here might be very different: Chelsea only having a 42% chance is really low but 18% (and falling) is unacceptable for them. Meanwhile, Spurs would be ecstatic with a 42% chance at this point in the season, while 21% would still likely be ahead of pre-season expectations. Emotionally, Spurs have less to lose and everything to gain so I would expect Pochettino to go for the win here. On the other side, Mourinho might be inclined to play it safe, play for a 0-0 draw (while hopefully stealing a win on a Willian free kick), and live to fight another day. Stopping Spurs from picking up the full three points could be as important as getting something from the match themselves.

Meanwhile, Liverpool has a slight preference for a draw, but their likelihood of top 4 isn’t really affected by what happens in Chelsea v. Spurs, leaving them at about 30% to qualify.

Week 14 Liverpool

The match is probably worth watching for the drama and early season implications, but I wouldn’t be surprised if it was a boring outing for neutrals.  There aren’t many games this early that could have such huge implications for the end of the season, so it’s worth keeping an eye on for Spurs and Chelsea fans alike.

Chelsea, Minimum Points needed to Make the Top 4, and Overachievement

One of the big topics on Twitter today has been whether Chelsea can still qualify for the top 4, and various folks have posted their predictions/simulations on what it will take. Michael Caley predicted 68.6 points, Colin Trainor and Simon Gleave have both agreed at 68 points, and I’m sure there are others out there. My model is a bit lower than the rest of these at 62 points, with 69 being my prediction for the third place team.

Week 13-3 Predicted Final Table

These results represent the mean points I expect each team to earn based on the remaining games and their probabilities of winning. For my predicted table I don’t do any simulations, just some simple arithmetic to get there1. Someone else on Twitter asked the question “Don’t models underpredict?” and while the right answer is no, there is some truth to this point, and that truth explains why Chelsea is probably less likely to win that fourth place spot than even the most skeptical models say (and far less than mine).

Right now I have four teams roughly tied for fourth place – Chelsea, Spurs, Liverpool, and Leicester City. Each of those teams is expected to earn between 60 and 62 points, which is a pretty tight race. But logically we’d expect a very different outcome. While we’d expect the teams to cluster around the mean, it’s likely that one team will find a purple patch, and that team will be the one who captures the 4th and final spot in the Champions League. So what do the simulations say?

DISCLAIMER: My model clearly overrates Chelsea so take their numbers with the largest grain of salt you can imagine. However, the model overvaluing Chelsea only means it’s going to be less likely we see these high point values for the 4th place team. 

View post on imgur.com

This density plot shows the results of 10,000 simulations, showing the number of points  each of the current contenders for the 4th place spot (ignoring those in the expected top 3) earns and the frequency of those outcomes. Chelsea has a decent chance of earning 65 points, and a low chance of earning 69 points, even at their overstated current expectations. The other three teams have a much lower likelihood of each of those outcomes. Spurs seem to have a slight but important advantage over Liverpool and Leicester, but that could change as we get more results. Leicester clearly is the least likely of these teams to make the top 4 but they’d have to be elated to even be in the conversation at this point.

View post on imgur.com

The next plot shows the likelihood of each team finishing with over 65 points. Chelsea about a 43% chance of breaking 65 points (again, see the disclaimer above), while the others are 20% or less. The joint probability of at least one of these teams finishing with over 65 points is 57%, and one of the teams other than Chelsea finishing higher than 65 is about 24%. Even given the idea that one of the teams could hit a hot streak and overachieve, or in Leicester’s case overachieve even more than they already are, we’re looking at an unlikely low point total for the 4th place team.

Finally, I look at the likelihood of finishing at 68 points or higher. Chelsea again leads the pack at about 31% (disclaimer), while the others are pretty unlikely to finish that well. Spurs are a little above 10%, Liverpool at about 7%, and Leicester at about 1%. At least one team will finish at 68 or higher 40% of the time, and a team other than Chelsea will finish at 68 or higher 14% of the time.

View post on imgur.com

My simulations indicate we’re going to see a much lower point total required for the Champions League than we’re used to, and a more competitive race for that spot than we’ve seen in years. Depending on what happens with Chelsea and if Leicester can keep up their pace, it could be a great finish to the season.



  1. I just do 3*(win prob)+ 1*(draw prob) + 0*(loss prob)

Replacing Coquelin – Prognosis and Options

Arsenal fans received some bad news for the team’s title chances when they found out Francis Coquelin is out with a potentially long-term injury. Without a lot of options in reserve, Arsenal fans are worried and have already begun to criticize Arsene Wenger for not buying a backup option for the defensive midfielder. Should Arsenal fans be worried? I want to look at the consequences of losing Coquelin for Arsenal, and potential replacements for him in the medium term and ways to strengthen the squad in the January window.

The first question is how badly will Arsenal miss Coquelin? He’s a key part of the team, and his backup is Mathieu Flamini who Arsenal fans don’t seem to rate highly at all. I broke out my Transfer Simulator to see how many points Arsenal could expect to lose with Flamini as a replacement for Coquelin, and Arsenal fans shouldn’t be too concerned.


When I substituted Flamini in for Coquelin to my model, I only see a four point loss over the course of a season, which translates to a roughly 4% decrease in win probability in any given game for Arsenal. This isn’t too bad, although 4 points could potentially be a huge different given that Arsenal’s expected final lead over Man City is currently 5 points. A 2 or 3 point drop over the time Coquelin is out of action makes the title race a lot more competitive, so there is some reason to be concerned.Week 13-1 Predicted Final Table

The next step was to find suitable replacements for Coquelin in the January transfer window. Assuming Wenger wants to buy (which is a big assumption given his history, especially in January), how many options are there? Going back to my first cut at a Points Above Replacement (PAR) measure from a few months back, Coquelin is possibly Arsenal’s most difficult player to replace, offering them a ~4 point bump over the median player in my database (more than any other player in their opening day first team). 1PAR Arsenal

This matches Wenger’s philosophy that top players are difficult to replace, so he shouldn’t necessarily spend money just for the sake of spending money. If he can find a legitimate upgrade, then he should so do, but that’s going to be difficult for Coquelin.

Are there options though? I ran some of the big names, and came up short, and honestly they’re probably unrealistic for what Wenger would be looking for/willing to pay in January anyway. Even ignoring that, most of them weren’t an upgrade over what Coquelin brings to the table in terms of offensive and defensive production simultaneously. More offensive-minded players cost Arsenal valuable defensive contributions, and a lot of pure defensive-minded holding mids fell short by a shot  per 90 minutes over Coquelin’s contribution and didn’t offer as much in the passing department. To find options, I did two things.

First, I looked at my list of replacements that generated the graph above. Because I have all midfielders classified in the same group, most of the replacement improvements weren’t positional fits: the model wanted Arsenal to buy Franck Ribery, presumably because his offensive contributions somehow overshadow his defensive limitations in this case. However, I did find a couple of cases that were interesting here: Jeremy Toulalan and Etienne Capoue. They’re not great options – Toulalan because of his advanced age, and Capoue because he just recently transferred to Watford and might not be seeking a second move so soon, but they’re both basically break-even compared to Coquelin.

Second, I went to my multi-dimensional scaling project to find similar players. The full method is explained in another post, but the idea is that I use game stats to create a two-dimensional placement for all players in my database. From there, I can measure the distance between each player and Coquelin, with shorter distances representing more similar players. This generated a couple of other interesting leads for players, but strangely enough most of the most similar players were wingers, indicating that Coquelin’s contribution as a defensive mid isn’t what you’d expect.

Coquelin MDS

Fer and Vujicevic are fairly close to Coquelin in terms of stats and contribution, and interestingly are incredibly similar to each other. Capoue and Toulalan are very different from Coquelin but are very similar to each other.

I like each of these players because they’re all affordable and would likely be comfortable as a rotation option for Coquelin. All of them but Toulalan are reasonably young (ranging from 25-28),  which is good and fits with what Wenger would like to do. So what is the expected point contribution of each of these players? I ran them through the transfer simulator, and here’s what I found.

Coquelin Replacements

Toulalan, Capoue, and Vujicevic are all *slight* downgrades from Coquelin, while Fer is about a five point downgrade. Not ideal, but it demonstrates the difficulty of replacing someone like Coquelin and shows what sort of position Arsenal is in. Each player has their downsides (Toulalan’s age, Capoue made a recent move already, Vujicevic would be making the transition from a less competitive Eredivisie straight into a Premier League title race, and Fer is arguably a downgrade over Flamini), but this is where qualitative scouting would come into play. Fer is the youngest of the group: can he improve quickly? Would Vujicevic’s performance transfer well into the EPL from the Eredivisie?

There are options, but I couldn’t find a silver bullet to replace Coquelin. Arsenal has a difficult series of choices ahead of them, and many “ok” options but not any great ones. It may be the best option to ride it out with Flamini, hope he can be solid in the short-term, and think about finding a better option in the summer window or even re-balancing the squad to have some more flexibility in their midfield.

  1. Arsenal fans -you’ve already beaten me up over my rating of Bellerin here so you don’t need to do a second round

Want to Make Analytics More Accessible? More Context, Less Acronyms!

One of my more popular posts in the past was my discussion of how to make analytics more accessible, so I wanted to follow up on that with some more thoughts on how to make analytics more accessible to a larger audience. These are from my own experience with Analytics TwitterTM , especially as I’ve tried to look into other sports for inspiration on different projects or different ways to do things.

  1. Use fewer acronyms, more wordsI know Twitter’s character lends itself to acronyms, and it’s difficult to be precise in 140 characters (minus someone’s username and 25 characters for an embedded image), but the acronyms imply a level of familiarity among your readers that keep analytics conversations at a level only a limited audience can understand. Here’s an example:

    To be clear, Footy in the Clouds is a great Twitter account with lots of great info for the soccer stats community that you all should follow, and I’m not singling him out here. Plenty of people do this sort of thing, and it’s a good way of communicating a lot of information in a short space. But this only works for people who understand what you’re already saying. Anyone outside the circle won’t get it, and if the goal is to expand the analytics discussion then we need to be more transparent with what we’re saying.1

  2. Offer context for any statistics you present

PDO’s a great example here: I know I’ve read about PDO in the past, and I vaguely remember that every team basically regresses to 1000 in the long-run, and I know that some analysts argue anything over 1000 is lucky while anything under 1000 is unlucky. I’ve never assessed those claims, and to be honest I can’t remember exactly what the measure is. I participate in Soccer Analytics TwitterTM regularly and read much of what people post, and I *still* don’t fully understand the measure.

Tell me what the measure is, tell me what the average is and whether a team is above or below it, tell me whether this is due to some inherent skill or whether we’d expect it to regress to the mean at some point. Numbers are useless without context, so if you want a broader audience make sure that the audience knows these things and can make proper use of the numbers.

A great example of this is Mike Goodman’s most recent ESPN column, an excerpt of which I’ll post here:

German teams tackle more, intercept more and generally contest their opponents more aggressively further up the field. In Germany, if an attack has progressed to the point where a player might consider shooting, they’ve already accomplished a lot of the hard work. In England, a player at a similar point is more likely to have the defense still set in front of him. It might be easier to get into the final third in England, but it’s harder to get a shot on target once you do. This might also explain why English teams play more passes in the final third (117 per game), than their German counterparts (97)

Mike Goodman does this better than anyone – finds numbers, thinks about what they mean, and writes them up in a sophisticated, yet transparent way. This is why he’s one of the best writers out there and is so widely respected. We should all try to emulate him, whether we’re trying to expand the audience of analytics work or not.

Don’t assume people know what you’re saying, and don’t assume that your point is self-evident. Maybe this is why we need more long-form blogs to supplement the Twitter conversation – it’s hard to explain and offer context in 140 characters, but it’s crucial if we want to expand the reach of analytics work.

  1. There is a legitimate discussion to be had about audiences here: I get into this with Neil Charles and Simon Gleave on a semi-regular basis and there is no right answer. It’s all about what individuals want from their work – if one wants to appeal to the niche high-level analytics audience, that’s ok with me. If one wants a larger audience, then that requires a broader communication style.

The Importance of Community and Sharing the Work of Others

Disclaimer: This blog won’t have any analysis/stats/fancy charts. It’s just my thoughts on how to build a greater community in the analytics world, encourage more long-form blog posts, and maybe move the ball forward by decreasing the barriers to entry. 

Like most of the people who read my blog, I follow the same handful of “big accounts” everyone else follows. Most of them are “big” for a reason: they post interesting, well-written work and have built up a significant audience over time. But it’s a logical fallacy to think that everyone who posts interesting, well-written work has a big audience while people who have small followings aren’t doing that. I’m fortunate that my work was discovered and embraced fairly quickly by the analytics community, and I’ve quickly developed a decent-sized following on Twitter for my work. I’m also impressed by the quality of my followers: they’re smart, engaged people who are interested in talking about soccer stats on a higher level than the “average fanTM.”

I’ve written about it before, but in my day job I’m a political science professor and I have a Twitter account where I talk academic politics. I have a small following there (~350 followers), but my followers are almost all high quality folks: current/former students, reporters from national and international media, professors, and major practitioners. It has opened up any number of professional opportunities to me, and I feel very lucky to have been able to take advantage of them as they presented themselves to me. But more than anything, I’m grateful to the folks who initially followed me back when I only had a few (< 10) followers. They followed me early when no one else was, and shared my tweets so other people could discover that I had an account and potentially follow me too and without them I wouldn’t have been able to build up even the modest follower count I have now. Because my soccer account has gained some quick visibility, I’d like to maybe pay that forward a little bit and try and draw some attention to writers you may never have read or heard of.

There is so much interesting work being done out there, most of which I probably never see. So Sunday I tweeted the following:

I found a few really interesting articles from this, which I’ll share here before I go any further:

All of those articles are interesting, and worth a read when you finish here. And follow their authors on Twitter. I never would have seen them if I hadn’t put out the open call. I’ve benefited from “big accounts” sharing my work in the past, and always appreciate when someone with high visibility tweets something I spent some time on and am proud of. I’m not a big account in the community, but hopefully I can share some underappreciated work and help some people in the way that others have helped me in the past.

So from now on, once a week (probably Tuesdays) I’m going to put out an open call for bloggers to send me their articles and I’ll share them with my audience. And I’d ask that anyone reading this takes a few minutes to do the same. Liking someone else’s work is costless, and retweeting one or two pieces a day is virtually costless as well and can help someone out. If you can tweet your own pieces three or four times (or retweet everyone who says something positive about your work), why not cut back to two or three self-promoting tweets and use one of those to promote someone else? If we really want to build a community and expand analytics rather than just promoting our own brilliance, this seems like a better way to do it.


Leicester City is Performing as Well as Chelsea is Poorly

With another decent upset today, combined with losses by the two league leaders, Leicester City moved to the top of the EPL table today for the week. This is particularly impressive given that a year ago today they were at the bottom of the table, and preparing for a relegation fight they would barely survive. Because Chelsea managed to beat Norwich City at Stamford Bridge, we were given a respite from “Chelsea in crisis” stories and are finally talking about Leicester City. But how well are they performing?

In a slight moment of bragging, I would like to say that my model tipped Leicester to have a pretty good season, picking them to finish 8th in my pre-season analysis. Clearly my SVM (which I’m calling “Model of the Same old Nonsense”, or “MOTSON”) saw something in their team at the beginning of the season, despite a bad TAM Coefficient from the season before.Preseason Predicted Final Table

However, even with my lofty predictions, they’ve managed to over-perform to the tune of a +10 over expectations, which is remarkable given the already high prediction my model had for them. Compare this to Chelsea’s disastrous season where they are 12 points below expectations, and it’s evident that Leicester is in the middle of a special season here.

Week 13-1 Deviation

Analytics TwitterTM  pointed out that their next 6 games are particularly difficult: Man United (h), Swansea (a), Chelsea (h), Everton (a), Liverpool (a), and Man City (h). But the good news for Leicester City is that I only have them expected to pick up 6.42 points out of these 6 fixtures, or basically 1 point a game. So even if their pace cools off quite a bit (which you’d expect), they don’t have to do much to stay at the level they’re at through the first set of fixtures, which has them challenging for a European place and starting to separate from the pack a little bit,  especially if/when Chelsea drops off a little bit more. The Leicester/Chelsea fixture could be particularly important – I have them as a *slight* favorite actually, expected to win 1.39 points compared to 1.2 for Chelsea. Who would have thought in the pre-season that this could be a Europa League 6 pointer?

Week 12 Heat Map

I’ll leave the “Can Leicester maintain their form?” posts to others, but for my purposes it’s important to note that MOTSON hasn’t updated its predictions with Leicester’s form. Even if they only perform at pre-season levels they’ll be in contention for a European spot. It will be interesting to see what happens over the next 6 games, but their expected points from the next fixtures are low enough that they shouldn’t have much of a problem meeting them and staying up where they are as we go into the return fixtures.

The Importance of Qualitative Scouting: A Response to the Analytics FC Response

@BobbyGardiner posted a response to my “Sell your overrated striker” post over at the Analytics FC blog, and it’s a good read for anyone interested in the topic. You should read the whole thing (and everything the @AnalyticsFC crew does), but his main point of disagreement with my post was the idea that Stephan el Shaarawy overachieved for Milan back in 2012. Bobby’s argument was that El Shaarawy wasn’t particularly spectacular that year, converting at a rate of 15% and scoring many of his goals from good positions. His shot chart from that season is below:

The alternative hypothesis is that el Shaarawy’s decline in production was based on a series of injuries rather than overachievement in one year and a regression to the mean in others.

I think these sorts of debates are interesting and important, and I appreciate Bobby’s response to my post. One of the things I was taught in graduate school was the highest compliment someone can pay is to engage with your work, and the worst is to ignore it. So I’m posting this response to engage further rather and maybe advance this discussion, not as some sort of attempt to “win.”

To that end, I have a couple of thoughts: one is sort of technical while the other is posting the next step of the argument. The first is that I picked il Faraone because, as a lifelong Milan supporter, his was the first name that came to mind. Milan’s biggest mistakes in recent years have been to sell low: Pato, Zlatan, and el Shaarawy have all been the subject of massive bids only to be sold a year or two later well below their original market value. He’s not necessarily the best case study, and Bobby’s choice of Alexandre Lacazette may very well be a better one. El Shaarawy also isn’t a good choice because I originally was aiming at talking about “selling” clubs, which Milan hasn’t historically been, and Lacazette may be a better option there as well. The case study wasn’t necessarily the point, the argument was that knowing when to sell overachieving players can make clubs a significant profit in the long-run.

More importantly, I think this points out the need for qualitative scouting to complement analytics approaches. Let’s use Lacazette as our example for a minute: he performed spectacularly last year, and Bobby’s argument is that he may very well have over-performed significantly. He also points out, correctly, that we don’t know this just from looking at stats. This is an area where Milan supposedly shined, but evidence seems to show that their methods were imperfect at best.

The vaunted Milan Lab describes itself as “a high tech interdisciplinary scientific research centre.” One of its big claims to fame back in the day was finding older players who could play at a world-class level far later into their careers than any other club, letting them buy 30+ year olds at a bargain price. This was demonstrably false when they let Andrea Pirlo go on a free transfer to rivals Juventus a few years back, and he put together some of the best years of his incredible career for perhaps their biggest rival. It’s other big claim to fame was maximizing training techniques to minimize injuries/to know which players were more susceptible to injuries. Pato proved this false on a regular basis, and more importantly to my point, they should have recognized el Shaarawy’s seemingly injury prone nature. That would be another inefficiency that could be exploited – knowing that el Sha would be likely to see significant injuries would be another way to get more money than his value from Manchester City.

Beyond this, regular qualitative scouting is important. I watched virtually all of Milan’s games that year (back when they were streaming on ESPN3’s app), and while he had an amazing season, el Shaarawy is a winger who was forced to play centrally by a combination of injuries and the sale of main striker Zlatan Ibrahimovic. He had a great season, but he was playing far too centrally far too often in the new Milan lineup. He was also a 20  year old kid being asked to lead the line for one of Europe’s biggest clubs, and there was plenty of talk at the time that this was too much pressure and too many minutes for him. I don’t watch enough of Lacazette to know for sure, but Lyon should be thinking whether he can repeat last year’s achievements or if they should max out now and sell him at his peak.

As with everything, the numbers can lead you where to look. El Shaarawy had an above-average season, just like Lacazette. Their performances were significantly more than one would expect from a young attacker, so a red flag should be going up from the analytics department. Then let the qualitative folks watch their performances: are they likely to improve, sustain, or decline? Is it worth selling now to cash in at the max rather than going through a couple drought years and then selling at a fraction of what you would have made originally? Numbers alone won’t get you there, they can only narrow down the range of possibilities you are expected to look at. Qualitative approaches are the perfect supplement for questions that numbers can’t answer alone.

Moneyball, but for Selling: Using xG/Goals Ratio to Profit

One of the most overused clichés in all of the Internet is “Moneyball, but for…”1. Moneyball, correctly applied, is the idea of using undervalued stats to figure out which players to sign at a bargain price. By exploiting an information asymmetry, small clubs were able to find value in players that bigger teams who focused on traditional stats were missing. However, soccer has a version of this that American sports don’t have: the ability to sell players for a profit rather than simply signing or trading for them. This, along with sophisticated analytics work, can help teams identify inefficiencies in the transfer market as a way to profit or make money to reinvest in multiple players.

Continue reading Moneyball, but for Selling: Using xG/Goals Ratio to Profit

  1. “Uber, but for…” Has to be a close second

Game Theory: Alan Shearer Isn’t Entirely Wrong

Alan Shearer riled up Analytics TwitterTM today with this comment:

Analytics TwitterTM hates these sorts of arguments, best summarized by @MessiSeconds response:

I’m confident Joel’s video response will be well-written, well-argued, and well-produced like the rest of his videos, but I think the answer is much simpler than this: at the end of the day, points are all that matter in the relegation fight, but we’re nowhere near the end of the day yet.

However, I’m going to depart slightly from what I assume the rest of Analytics TwitterTM is probably going to say. Alan Shearer is right: Three points for Newcastle over Bournemouth was a significant result, and changed my predicted final table in a significant way given it is only one result. Here’s the weekly heat map results from my model for the two teams:

Week 12 New Bou

The win moved Newcastle’s relegation probability down a few points, and moved Bournemouth a few percentage points closer to relegation. Regardless of the stats, this was a bad week for Bournemouth. Playing well isn’t much comfort in this case.

However, the stats tell a more nuanced version of Shearer’s point, ultimately leading him to the wrong conclusion. Newcastle doesn’t care about winning ugly this week, but in the long-run they do care that they won a game they in all likelihood shouldn’t have because that means they can expect to earn fewer points across the season. In a one-shot game, Newcastle is happy to have taken the three points from a relegation rival and they don’t have to give those back just because they were out-played. However, across a 38 game season, if they keep playing like this, the law of large numbers will probably catch up with them and they’ll be down in the Championship.1 It’s the reason Las Vegas casinos make billions of dollars: if the fundamental statistics are in your favor, in the long-term you can’t possibly lose.

Three points today are a victory today and no xG map can change that. Newcastle’s probability of staying in the EPL next year increased, and the points they earned stay on the table. The opposite is true for Bournemout: their probability of staying decreased, and the points they missed can never be made up. But in the long-run Newcastle’s manager knows that they got lucky and will have to improve if they want to avoid the drop.  Alan Shearer likely knows that too, so we shouldn’t feed the troll.


  1. 38 is hardly what the law of large numbers folks had in mind as a “large number” but it’s close enough for our purposes. If you’re not satisfied with this explanation, then extrapolate it to multiple seasons until you get to a number large enough for your taste.