One type of post I want to try and do in these pages is to try and make analytics more accessible to more people. The tired “analytics versus grass” debate is boring and will hopefully get left with the gaggle of trolls, goons, and grifters who are stubbornly holding their domain on the hilariously failing platform-formerly-known-as-Twitter. Analytics already “won” in the sense that every professional football club has a team of analysts generating gigabytes of data on a daily basis to measure, assess, and improve performance, from how often players should press, to where to stand when defending a corner, to what players should eat immediately after a game.
Likewise in FPL, analytics has already changed the game and when folks cackle and cry foul every time a team posts a higher xG than their opponent but loses, they aren’t striking a blow back for the “old days”, they are just making themselves sound “old”. The variance in the game means that a pure model may not yet be able to consistently win it all, but that doesn’t undermine the place basic statistics and deeper analysis have in the game.
All that said, yelling at people that they are wrong, or, I don’t know, writing a blog post saying they sound “old” because they continue to push back on analytics’ role in the game doesn’t really help the situation (oops). My personal interest in analytics came from the extremely cliched starting point of reading Moneyball, primarily because I was a Michael Lewis fan and had a vague interest in baseball having recently moved to Canada from the UK for a gap year. I quickly became enamored with this new way of digesting what can be a somewhat slow game to outsiders and really rode the crest of the analytics wave that had started in the late 70s and 80s but really rose to prominence in the 2000s.
At some point I plan to write a larger piece comparing the (in my opinion, similar) arcs of the proliferation of stats and analytics across baseball and football, but for now I want to draw on a single point. There was a point in baseball - both in front offices, the media, and, yes, in fantasy sports - where a clear divide emerged, and you were either “for” or “against” stats: On Base Percentage versus Batting Average. Wins Above Replacement versus The Will To Win. Analytics versus Grass. It was likely never that simple though, and it certainly isn’t now. Stats need to be contextualized, and it’s by engaging with the game itself that you find the questions which statistical analysis can help answer.
So in the spirit of new BlueSky thinking, let’s actually engage with a question and apply logic and statistics, but try and keep things grounded and non-snarky (pre-apologies for failing on the latter, it’ll take some time to fully drop the snark which ruled the old Twitter days).
Is Selecting Three Defensive Players from the Same Team Too Risky?
This question arose most recently with Andy’s (Let’s Talk FPL) selection of a third Arsenal defender in Gameweek 12, but it’s a question that has certainly come up before and is generally not well answered, even when I think most folks who are analytically inclined inherently understand the correct answer.
Some Thought Experiments
A couple of thought experiments to set some parameters here as to where the answer might lie.
1. As with many pseudo-Freakonomics / Thinking Fast and Slow inspired blog posts, let’s start very simply with a coin toss to really get to the root of the problem (I understand footballer is chaotic and full of random events, but let’s start simple and build up). In this game we have an unfair coin that comes up Heads (H) 66% of the time, and you know this information ahead of time. We are going to toss the coin three times and guess the three outcomes, and you can select HHH, TTT, THH, HHT etc. What combination would you like to bet on? Loading up on Heads or “spreading the risk” by adding in some Tails?
2. Extending to football, imagine that in a given week, with the below fixtures you are given the option of either (a) captaining one of your players, or (b) adding any other player from the player pool regardless of price. You own Salah, and the best alternatives are Haaland and Palmer:
Salah vs LEI (H), projected points 8.1, betting odds 60% goal
Haaland vs EVE (H), projected points 6.7, betting odds 66% goal
Palmer vs BRE (H), projected points 7.1, betting odds 49% goal
Do you want to “put your eggs in one basket” and double Salah’s score, or do you want to “diversify” and add Haaland or Palmer’s score for the week?
We’ll get to some of the caveats we might raise in a minute, but in these basic hypothetical cases, I think we’d mostly agree that we’re backing the three Heads and captaining Salah, right? In each case you have three events. Using the coin example, in event one your faced with a 66:34 decision so pick Heads. In event two, it’s a 66:34 decision so you pick Heads, and again with the third. With Salah, you can ignore whether you already own him and simply decide between the three options, presumably picking Salah based on the model, goal and assist potential, better fixture, or whatever would normally lead you to conclude Salah was the best play this week.
Okay, but a defensive triple up isn’t the same thing is it?
At its basic level, it sort of is the same thing, though. If in the coming gameweek you have an open defender slot and Arsenal have the best chance of a clean sheet, then all else being equal they are the best option to pick.
There is a temptation to talk about “diversification” here and perhaps comparisons with an investment portfolio, but I humbly submit that’s a flawed argument. First, the point of diversifying your portfolio is to hedge against a myriad of risks in the world economy. You want some equities in a range of sectors to capitalize on economic growth, some fixed income investments perhaps for some income security and as a hedge against interest rate shifts, and then some diversified, exotic hedge fund pieces to use complex instruments to mitigate potential losses from a range of very specific risks. You are not gambling on which equities will go up, or whether interest rates will fall, you are trying to win regardless of what happens. Second, while football results are certainty subject to variance and a ton of statistical noise, they are not as complex as trying to forecast the world economy. We can reasonably project the odds of a team keeping a clean sheet, at least to an extend that works over a reasonable sample of games.
In FPL, we aren’t building a sensible portfolio. We are making 11 individual bets each week and should look to maximize the chance of winning each one. We typically don’t want James Ward Prowse’s 4 points every week, we want a player with a good chance at 6 points this week, not caring that he’s projected for 3 points next week, because we can sell or bench him without most of the costs that are associated with selling real world investments. Of course, at some point we balance risk/reward and don’t select explosive players who might not play, but generally we are looking to score points with some upside, some just generate median returns.
There is, of course, some potential elements of elevated risk by tripling up on a defense.
You can only hold three players from a team, so holding three defenders locks you out of anyone else from that team. In this case, the main issue is of course Bukayo Saka, a terrific asset, fresh off his third double-digit haul of the season, but also to a lesser extent someone like Kai Havert who is a useful forward if the fixtures are right. While agreeing in principle - as a very satisfied owner! - that not being able to get Saka isn’t ideal, the fact that you are barred from doing so because of squad rules doesn’t seem any different than not being able to own Saka because you think Salah, Palmer, and Haaland offer better value for your team, and you can’t own all four. Being forced to use two transfers to bring Saka back into your team should Palmer get injured would be annoying, but that’s also the case for most people who own Salah but not Haaland, or anyone who wants Alexander-Arnold rather than a fourth pricier midfielder.
By backing three defenders from one team, you are potentially exposed to more negative variance or “bad luck” as one event can mess up three very good defensive performances. Chelsea gave up 0.3 npxG against Leicester in GW12, and were on for a clean sheet until the 95th minute penalty. If three of your defenders all play for one team then they all suffer the same underperformance against expected points. Over the course of a season or more, we’d expect that to somewhat even out with some other games where your team concedes 2.1 xG but keeps a clean sheet, just as Spurs did this week at the Etihad. However, if you own this triple up for just a few gameweeks, these events are quite likely not to necessarily equal out and so you could end up with some noisier returns compared to the expected data by being leveraged into one team. I don’t necessarily see this as a reason to avoid the strategy but I think I’m right in saying we expect higher variance here. Of course, that variance can swing both ways though, so this isn’t really an argument against a triple up, just a statement about the range of likely outcomes being spikier.
This one for me is very minor and not really worth considering, but it is technically true that by picking players from one team, you are more at risk of a game being postponed. However, having two defenders and a midfielder would barely be any better in such a situation, and no one is arguing that a Raya / Gabriel / Saka combination is too risky.
Similarly, although for me slightly more relevant is the risk that a key player gets injured which then undermines several of your assets at once. This could be someone like Rodri being sorely missed by Man City, or perhaps a top-level goalie having a much weaker replacement (Everton defenders’ value if Pickford was injured, for example). This wouldn’t be determinative for me, but there is a shred of truth in this factor, even if it’s hard to really pinpoint too many players who could cause one to go from considering a team’s defense to be so good that you are considering three of their assets, to now so bad you don’t even want one of them. I guess William Saliba is potentially one, which would make owning him doubly painful if he’s now injured and your other two defenders are now less valuable.
Backing the same team three times perhaps puts some additional stress on your forecasts, as if you are wrong about Arsenal’s potential defensive prowess then you are making three bad bets rather than one. Again, there is some logic in this one, but projecting clean sheets isn’t the most complicated projection to make and we have plenty of robust models making predictions, who will generally all converge on the same teams offering the best chance at a clean sheet each week.
There are also other factors which might mean you personally don’t think an Arsenal triple up is a good move. Maybe you think Liverpool’s defense is comparably good and you want to also enjoy Alexander-Arnold’s attacking threat. Perhaps you think Timber and Calafiori cannot be relied upon for minutes and don’t like premium goalkeepers. Maybe you think Arsenal’s fixtures actually aren’t that great and you fancy Man Utd, Fulham, and Palace to all nick a goal in the next five gameweeks. Or maybe you think a five-man rotation of budget defenders can deliver comparable value, and funds can be better used elsewhere. These are all very fine arguments why you don’t want to pick Timber, or Raya, or Gabriel, or indeed any Arsenal defender. However, they are not really convincing reasons as to why you are comfortable with two Arsenal defenders but not three.
Summary
I think there is maybe a terminology gap driving some of the discourse here. Owning three defenders from a single team potentially increases the variance we might see over a very small sample of games - as a bad (or good) break can impact more of your opportunities to score points - but isn’t necessarily a big “risk” as that variance can swing both ways. Indeed, many analytics inclined managers will specifically target a defensive triple up on a free hit to take advantage of this fact.
Closing your easiest to route to Saka isn’t ideal either but again I don’t see this as necessarily a “risk” so much as just another decision made as to how best to allocate your scarce resources. The ability to roll transfers makes it easier than ever to re-allocate your team value, so you are never really more than a week or two away from making pretty seismic shifts to your team structure. Willfully giving up points to maybe save yourself one transfer down the line doesn’t seem optimal unless you think Arsenal are only marginally better than other options, in which case, there is no argument for a triple up anyway.
The one factor we haven’t discussed is the mental anguish from suffering a triple clean sheet wipeout. Generally, this should be irrelevant to your decision making, unless (a) you remember that you are playing FPL for fun and know that this will upset you to the point that your real life is impacted and so want to avoid such a scenario, or (b) foresee that having a bad beat while tripled up will cause you to make other decisions on tilt, making poor moves you otherwise wouldn’t have made. Again, these are valid concerns but don’t really speak to the value or risk of the strategy, so much as whether it just isn’t something you are personally interested in. We should all “play our own way” but that doesn’t mean that some decisions are just empirically more logical than others.