A look at the unexpected results from the first four Gameweeks
by @FantasianPL
We are four weeks into the season and with an international break, it seems as good a time as any to look back at the xElo Experiment model and see how it is doing. Below is the predicted and actual offensive xG for each game for each team.

As you can see, some games went exactly as planned, most were higher or lower than the predictions. Let’s put that aside for a moment though. One thing you will notice is that the range of predicted xG is approximately between 0.75 and 2.25. This comes down to the relative strength of the teams. Even when comparing the worst and best teams, the model isn’t going to predict a very low or high xG score. Back to the reasons for differences. One key reason is variance due to the sample size. With only one event, almost any single outcome is possible. But once you start combining two, three, five, or ten events, these “random” events become less significant. Assuming we could get variance under control, the other reason for differences is changes in team performance. The model was built to account for this. A team’s strength will fluctuate as they play better or worse.
What I want to do is look at these results and see if we can potentially see the starts of some patterns that may indicate a change in team performance. But first:
SMALL SAMPLE SIZE!
You must remember this. We can make speculation and guesses based on these, but one needs to be careful reading too much into single results.
Red Cards
First, I am going to ignore the games with a red card. Playing with a man disadvantage can change the result of the game. For example, here is the xG plot (from understat.com) for the Chelsea vs Liverpool game.

The model predicted Liverpool to have the fourth lowest xG that week, and the first half result agrees with this prediction. But after the red card, things started going LIverpool’s way (not necessarily fully as a direct result). Now the interesting thing is that, even with a red card, some games finished almost exactly as predicted for some teams.

Moving on, how are the promoted teams doing and were their preseason rating appropriate?
Note: because the offensive xG is plotted on the graph, a low value on “defensive” points indicates the opponent underperforming/defense over performing.
Fulham

Five of the eight games had Fulham performing below expectations. This isn’t a statistically significant pattern, but their offense is one to watch. Not because they seem to be on fire, but the opposite. It is possible that their preseason offensive rating was a bit high, so keep an eye on this moving forward.
Leeds

While the attack has been all over, Leeds has consistently outperformed their predicted defensive results. Based on other analysis I’ve seen, this makes sense. I would expect the predictions for Leeds defense to be a bit pessimistic for a little while as the model adjusts.
West Brom

The West Brom results are not looking too good. Their offense and defense are both generally underperforming. That said, the red card vs EVE obviously impacts 25% of the sample size. I wouldn’t write off West Brom yet, but maybe wait to see what happens next.
Finally, some of the biggest unexpected results. Some may indicate the start of a new trend, for those risk hungry folks, or else be prime examples of variance in practice. You may notice that the most surprising results are all over performances, this is in part to the limits of the model with a high number of predictions clustered around an xG of 1.0, there is more room to over perform than to underperform.

Aston Villa vs Liverpool – GW4 (Predicted: 0.8, Actual 3.0)
Easily one of the biggest surprises of the season. While the score can be chalked up to some lucky bounces, the xG shows Aston Villa deserved many of those goals. Aston Villa may be showing an uptick, but it is only two weeks of beating expectations. As for the Liverpool defense, they have been slightly worse than predicted over the first four gameweeks, but probably too early to write them off.
West Ham United vs Wolverhampton – GW3 (Pred: 0.9, Act: 3.0)
West Ham United vs Leicester – GW4 (Pred: 1.1, Act: 2.0)
The second from the bottom for predicted xG in GW3, West Ham showed up and destroyed Wolves both with xG and goals. The Wolves were middle of the league for xG that week, but barely managed to get any.
Now add in the fact that West Ham slowed down Leister in Gameweek 4 and also overperformed their predicted xG (both on the surprise results list), and we MAY have the start of a trend. With Tottenham, Manchester City, and Liverpool up next, West Ham are a team to watch. Intriguing now, they could be even more tempting come gameweek 8?
Leeds vs Manchester City – GW4 (Pred: 0.6, Act: 2.6)
Leeds came to play and put up some great numbers. While they only managed a draw, this game is interesting for the outlook on Leeds. I’m not going to be making FPL decisions off this game, but it does have me watching to see how Leeds carry on.
Brighton vs Manchester United – GW3 (Pred: 0.8, Act: 2.4)
Manchester United have seem to have gotten off to a slow start. Taking away penalties and the time they have 10 men, they are still averaging roughly 2 xGC per 90. Much of this number comes from the result Brighton got against them in Gameweek 3. Brighton has the lowest predicted xG that week, but ended up fifth on actual xG behind Westham, Tottenham, Liverpool, and Chelsea.
Tottenham vs Newcastle – GW3 (Pred: 1.7, act: 3.0)
Speaking of Tottenham, they also make this list. They got solid xG numbers versus Newcastle in Gameweek 3 following a similar positive surprise vs Southampton the previous week. They did manage only 1 goal vs Newcastle, but destroyed Southampton (defensive setup issues?). With the two top FPL point getters, Tottenham may be more than a team to watch, but, as stated before, this is still a small sample size, so act on it based on your risk appetite.
Tottenham vs Southampton – GW2 (Pred: 1.3, Act: 2.3)
West Brom Albion vs Southampton – GW4 (Pred: 1.3, act: 0.2)
Burnley vs Southampton- GW 3 (Pred: 1.4, act: 0.4)
While the Southampton defense may not have been up to the challenge of Spurs in Gameweek 2, they were well up for Burnley and West Brom in the next two Gameweeks. There seems to be some “football” type thinking on this, but I’ll leave that up to the experts. I could see this going either way. The good news is that based on thier price, the defensive options aren’t a huge risk.
Leicester vs West Ham – GW4 (Pred: 1.7, Act: 0.5)
Wolves vs West Ham – GW3 (Pred:1.4, Act: 0.4)
The interesting aspect is that both these games are on this list for the attacks of both teams. However, all four actual results favour West Ham. While this is a bandwagon I would be interested to join, the next 3 fixtures make me hesitant to jump on them defensively. If they continue to outperform their predictions, it seems a no brainer to jump on in gameweek 8.
Arsenal vs Sheffield United – GW4 (Pred: 1.7, Act: 0.7)
Sheffield United vs Arsenal – GW4 (Pred: 1.1, Act: 0.1)
Finally, the game and underwhelmed. The interesting result where both teams were 1.0 under their predicted xG. Is there a reason? Were they related? Don’t know.
That brings us to the end of this. In summary, there is not too much we can take away, but a couple of potential teams to target are Aston Villa, Tottenham, and West Ham. Leeds may be an option defensively going forward, but just because they are outperforming their predictions, doesn’t mean that they will outperform other, higher rated, teams. I also will reiterate that this is only three to four weeks of data and, much like the Southampton defense, everything could look much different in two weeks time. Thanks for reading and good luck in gameweek 5 and beyond.