Monday, January 9, 2012

The form fallacy

Any time one of the league's better players struggles for goalscoring form and then breaks his drought, you will inevitably hear the commentator point out that while his poor 'form' was temporary, his class is permanent. We aren't here to perform autopsies on former stars of previous decades to suggest their class was anything but permanent, but instead we will examine the very notion of 'form' and how this is reflected in a player's goalscoring record (and hence fantasy success).

A player is described as being 'in form' under two distinct scenarios. The first is where a (rare) commentator has actually watched all of that players' games, assessed his overall performance and concluded that he is playing up to his full potential. The second, and much more common occurrence, is where an 'expert', or more pertinently here, a fantasy manager, will assess that a player is in form because he's scored a couple of goals in the last few weeks.

We are obviously interested in the second scenario, and more precisely, whether players score more goals following a game in which they found the back of the net. We will expand the analysis to assess the likelihood of a player scoring following a scoring streak of one, two and three games.

The madness in the method
We want to ascertain how often a player follows up a goal scoring performance with another goal scoring performance, and then how this compares to their standard goal scoring rate. We will look at how often players follow one, two and three game scoring streaks to try and identify the value of 'form'.

Scoring Games (SG)
Scored in next game

Back to back SG
Scored in next game

Three in a row SG
Scored in next game

As expected, scoring last week does not have much predictive value for scoring this week (just 15%). The first lesson then is to not chase points based on a single performance alone. If a player is a recent signing or has suddenly been deployed in a new position his goalscoring may be indicative of future success, but chances are the predictive value is low.

If a player has scored in back-to-back games (or even three in a row), on first glance it appears that there is fairly strong evidence that his success will continue in the next fixture (31%). However, the proportion of goals scored by the top players is obviously disproportionately high and so you need to see the rate at which each player normally scores goals to understand the context.

The below table shows the 'streakiest' players from last year, based on the difference between the rate at which they normally score goals ('Score %') and the rate at which they scored a goal following a goal in the prior week (GW1%), having scored in back-to-back games (GW2 %) or following a scoring streak of three games (GW3%):

Not the worst list of players but aside from Meireles (who scored in 5 straight games and then never again), Adam and Kuyt the majority of the group failed to continue their 'form' across all three categories and thus we have to doubt if there's anything more to their streaks than blind luck and random chance.

Top guns
Rather than look at random players, let's turn our attention to the league's top scorers and see if any of them showed a propensity for hot streaks:

In short, there is very little correlation here. Some players had success in the games following goals, but nothing really to a degree that indicates other than a normal distribution of results. It seems that the only players who consistently showed a higher likelihood of following up a scoring performance with a goal played for Liverpool with the aforementioned Meireles, Kuyt and Adam joined by Maxi Rodriguez here. That may say as much about that team as the individual players though, and could be a separate avenue worth exploring (does a team's form or their opponents' strength dictate the result more?).

This season
Has anything changed this year? Are we seeing any different trends of predictable players? Is the world really going to end?

The top two names obviously stand out as a couple of elite players who have shown a strong tendency to follow up scoring performances with another, whether looking at a scoring streak of one, two or three games. However, Van der Vaart actually just has the one streak of five games, while Rooney has two streaks of five and three games respectively. Considering a distinct lack of 'streakiness' last year, it is therefore tough to read too much into these numbers.

The point here is not to get into individual players but to focus on the fact that, if you measure 'form' in terms of goalscoring, there is little evidence to really suggest that players consistently get 'hot'. There isn't any evidence to suggest you should shy away from players who scored last week either (indeed it is probably foolish to bench a Rooney or Aguero coming off a good game) but you should be thinking twice about bringing a player in based on a 10 point fantasy gameweek which was built on a goal from a single shot on target and the inevitable three bonus points that follow. Even two or three game scoring streaks can be fools gold if not backed up by sustainable underlying stats (shots on goal, particularly those on target).


Tony S said...

Hi Chris, great article as usual. I really thought (without statistical proof) that players were more inclined to go on 'Hot Streaks' and I always believed that a player was more likely to score the week after he just scored.
However, if a player scores 3 goals one week and then fails to score the following week, do you declare that he failed 3 times to follow up with a goal? If so then the stats (especially for All players) could be misleading.
And it doesn't really answer the important question of 'Is a player more likely to score a goal the week after he score or the week after he hasn't score'? I would like to know if Rooney is more likely to score after a blank week or after a scoring week. But I suppose that would take a lot more research and resources.
I loved the article thou, very thought povoking :)

James Richard Klien Jr said...

Love the conclusion and take away from this piece but ignoring a big piece of negative evidence - if you will indulge me here … players that never score won’t score, whereas players that do score are more likely to score. When the numbers are aggregated, a sample of all players will include, say Kenny Cunningham (0 from 331!). By taking such goal threatening ineptitude out of the sample will leave you with a stronger sample which is thus more likely to score in any given week (whether they have scored the week before or not). Hence why the percentage actually jumps upwards for BTBSG as goals from this sample are more likely. Taking it back another step and looking at the percentage of players that register scoring games (i.e. total number of games from which the 837 SG was derived) will further prove the point. The article is a point well worth making and could help avoid many costly errors of jumping on those dangerous bandwagons! Keep up the good work.

Chris Glover said...

Tony - the formula works so that a player can only failed once after scoring in 1, 2 or 3 games. Example:
(G = goal, 0 = nothing)


You'd get 5 chances to follow up on a goal scoring game (converted 3 times, 60%), 3 chances to follow up on back-to-back games (converted twice, 66%) and 2 chances to follow up on back-to-back-to-back games (converted once, 50%). On this point I think the analysis holds up.

Your second point is spot on though and this is the next step I will take as 'form' should mean more than just a streak, say, 3 goals in 5 games. I need to work out how to code this in Excel but I don't think it will be too hard so I should get there soon. This should be be more informative.

Thanks for the input.

Chris Glover said...

James - players who don't score don't show up in the analysis at all as to get in you need to score, and then we'll assess your chance of scoring in the next game. You're right though, players who score the odd goal will distort the numbers and that's why you'd like to find a player who shows positive tendencies in each category before you concluded that form was a major factor for him.

For fantasy though even the random 1 goal men's data has value. The fact that players as a whole don't tend to follow up with another goal is important given the countless transfers wasted on fantasy teams chasing random defenders who happen to score a flukey goal off their knee. But the more important point is for the big name forwards, hence why I highlighted them rather than the mixed bag of players who have shown some streaky tendencies.

The analysis needs work for sure and all input is welcome but I think, as a general point the basic conclusion stands up.

SFQuakes said...

I wonder why you don't pay more attention to team value? You spend a lot of time on looking for player value, but with better funding you could upgrade your Moses' and Gowers of the world to the likes of a Dempsey or a Mata. Sometimes chasing player value can hurt for the short term, but over the course of a season an extra 6-8M in the bank is huge. Great blog, thanks.


James Richard Klien Jr said...

Chris - apologies if it came across as a critique, it was mere musing (rambling) on my part. I agree the conclusion is valid and even more so when you factor in key indicators / stats mentioned in previous posts.

Chris Glover said...

James - not all James, sorry if my reply seemed offended - I wasn't! It's always good to get some input as there's a pretty good change I have messed something up or missed a key piece of info.

SFQuakes - it's a valid strategy but to be honest I've just never been very good at making money in the transfer market (and thus can't really advise anyone else). Be definition, to make money you need to buy the popular players (whose price will rise) and thus unless you're onboard first, you just end up with a homogeneous team. As an experiment I am going to track my team from today, making transfers purely in the interest of making money. At the end of the year I will see a) how successful I was in making money and b) how much better/worse this team did than my actual team.

My concern is that while an extra 8m would be nice, in reality this is only going to achieved by a select few managers and even then, by the time you have all that extra money (say GW35), you barely have any time to take advantage.

If anyone has any particular money making schemes, let me know and we'll see how much money we can make by the year end.

vanilla said...

Something interesting... i have not looked this season, but in prior seasons it has been possible to view the most valuable teams in the competition. There are many teams out there who are in leagues competiting by team value rather than points.
I've always been interested to know what the "ideal" season is... this would require a large spreadsheet with day-by-day player price data throughout an entire season (not to mention higher than average excel/solver knowhow), but it would give great examples of instances when multiple transfers were ideal. Hindsight is the only way to truly know when that wildcard "should" have been spent!

vanilla said...

"to make money you need to buy the popular players (whose price will rise) and thus unless you're onboard first, you just end up with a homogeneous team."

This is (sometimes) the cost of considering value/schedule/market instead of only value/schedule. My main league is only 8 friends, and it will make the difference.
Consider the upside gained when Aguero is benched one week and your autosub is Sessegon/Dempsey instead of Gower/Moses.
In a season of rotating elites, the strength of your squad's depth is a consideration. Yes, I've spent 20-30 points on extra transfers. No, I'm not #1 (i'm third place by 50pts).. but my budget is ~4.5m higher than the #1 team in my league. YMMV, but i'm very optimistic.

Rutwik said...

My current team
Friedel, Ruddy
Kompany, Jagielka, Enrique, jones, Taylor
Larsson, Bale, Silva, Fox, Fergusson
Adebayor, RVP, Yakubu

I am using my wildcard this week. any suggestions ?

Top Bro said...

Congratulations on your guest spot over at Fantasy Football Scout, Chris! You definitely deserve all the attention you're getting - great articles as usual.

I'm sure this has been mentioned a few times, but I'd love to hear how your own personal FF team does each gameweek; what transfers you made, team selection dilemmas, highs/lows ect... Would be great to see your thinking throughout the season, even if it's just a paragraph tacked onto a regular post.

All the best!

Grounderz said...

Hi Chris, I heard you've been brought in as a guest writer at FantasyFootballScout. Congratulations, and I hope to see more articles from you there.

Of course, I hope you do the same here.

Ben said...

Hey Chris, delighted to hear you'll be contributing to FFS, fantastic news! My two favourite sources working together.

stooshermadness said...

Chris - congratulations on the FFS guest column. Your blog is a massive, massive influence on my rotation selections in defense and for the armband. As happy as I am for your success, I'm a little bummed as this blog seemed a nice little place for the cognoscenti. Hope the commenters keep throwing out good stuff like Tony, James and Vanilla and we don't see a million RMTs which has sort of ruined the commenters section of FFS . . . all the best !

vanilla said...

Well done, Chris! Thank you for putting the work into this blog, and may the FFS people appreciate you as much as we do here.

Thanks, Stoosher, for encouraging what feels sometimes like madman ramblings. I think more traffic here will make for lively debate and the fresh blood will bring some new angles to consider (and to slander!)

Gummi at said...

I'm late to the party (have had a lot of work) but my congratulations on your guest spot.

Your blog was the major reason I started my own (admittedly inferior) Fantasy blog, so you were a massive influence.

You continue to be the bedrock of my Fantasy decisions.