A counterintuitive approach to captaincy {LONG THREAD}

19/20 was a very disappointing season for me when it came to captaincy. So now during the break I thought it was time to make a thorough analysis of what went wrong.

#FPL #FPLCommunity

I always aim to find clear guidelines to go for, since this will make my findings both easier to retain for readers but also easier for myself to employ successfully.

With the help of a visualized "decision map", I will summarize the handful rules I personally will use this upcoming season when making my captaincy decision.

Another strategy I found feasible was this one https://twitter.com/fpl_gents/status/1277221501557788673

To get an understanding of which players were the best captains to choose from, I took a look at last 2 seasons' data for the biggest hitters for the respective seasons.

First I want to give you an overview of the different probabilities for home and away games and versus hard or easy opponents. I chose the bottom 8 as a metric for easy opponents.

The problem with counting bottom 8 is that it’s hard to know during the season which 8 teams that are going to be the worst.

To solve this for future captaincy decisions I will look at bookies' predictions of which the worse teams are, lending some extra help when needed from...

other fixture difficulty tools. I will use the term “conversion rate” to describe when the player got 5 points or more. I also took a look at 10+ pointers (hauls) and minor returns (5-9 points) but won’t mention the differences in kinds of returns too much.

These were the conversion rates in order of best to worst:

Home games vs easy teams: 78%

Home games (vs easy or hard teams): 68%

Games vs Easy teams (home or away): 68%

Away vs easy teams: 60%

Home vs hard teams: 59%

Away vs hard teams: 26%

If we go by this then it’s clear that there is one instance when you should never captain a player and it’s away games vs hard teams. And there is one instance where you should always captain and that is the exact opposite scenario.

The second take away is that playing easy opponents is equally attractive as playing at home. So away games vs easy opponents are equally attractive as home games to harder opponents. This helps us set up our first rules:

1: Always captain a premium home player.

2: Always captain a premium player vs an easy opponent.

Both strategies should work equally well by themselves, but they're naturally even more effective when combined.

Now comes the part where we dive deeper. I wanted to include form in the calculation so I looked at how players were affected by their most recent event. To see if the outcome of the most recent game affected the player's likelihood of scoring in his next game.

However, since hard fixtures and easy fixtures differ so much I chose to only focus on easy games. The sequences I looked at was either an easy game followed by a difficult game and I called this easy game an isolated fixture.

Another instance of an isolated fixture was if the player had a rest or injury in the previous game, then there was no record of the previous game.

The other sequence I looked at was when the player had two easy games in a row.

My findings should be taken with a pinch of salt since sample sizes were only two seasons, ranging from 20 to 50 matches depending on how unusual the events were, but: As I said before you have 78% chance of a return if your player has a home game vs an easy opponent.

However there is a 5% edge you can potentially gain if you also considered how the player did in his last game, jumping from 75% hit rate to 80%. This is where I think regression to the mean plays a part because what you will read next will feel counterintuitive.

(check out this thread to refresh yourself on the concept) https://twitter.com/fpl_gents/status/1285538374951075845

In the instances where a player scored a return and then afterwards had an easy home fixture he had a 75% chance to score in the next fixture. In the instances where the player had an isolated event or had blanked in the previous game he had a 80% hit rate.

I think we all believe in form so that is why this result is counterintuitive. A statistical explanation of this could be the aforementioned regression to the mean. The player most likely experienced above average luck when he scored his return,

and (except for considering form) we can only expect him to experience average luck the next game.

However, even though the player was more likely to blank after a return, he was also more likely to haul.

Here is a comparison to illustrate if there’s a way to take advantage of these tendencies:

Example 1:
When a player has a home game that is either isolated or he blanked vs an easy team in the previous fixture:
Blanks: 20% Returns: 40% Hauls: 40%

Example 2:  
When a player has scored a return in an easy fixture and has an easy home game next:
Blanks: 25% Returns: 25% Hauls: 50%

So example 2 has more blanks but also more hauls.

Let me make an easy calculation for you so we can see which of these two we prefer if there's 10 games and the averages will be like the examples. Let's give returns a value of 7 and Hauls a value of 12.

Example 1 would give you a total of 152 captain points, example 2 would give you 155 points, so basically they are equal. This result along with the rest of my findings from looking at these premiums has helped me believe that the following is true:

A player with worse form isn't necessarily a better captaincy pick than one with better form. However we seem to overestimate the chance of a return thanks to a player's good form.

It also might be down to the player having blanked he gives extra effort next time.

There's another important sequence where we can gain an edge by looking at the history. These are even smaller sample sizes so I think I would need at least another season of data.

I looked at away games vs easy opponents (remember that away games vs difficult opposition is never a good idea for captaincy as a rule of thumb, with the exception of maybe KDB who gets to play more forward in the tougher games).

I made a differentiation between away games that followed an easy game and those that are isolated. And there’s a clear difference! In away games where the player had just scored one return in the previous game he was now 58% likely to return in the next one.

If the game was an isolated away game it increased to a 68% chance to score. This again shows an indication that regression to the mean might be a factor.

One last note is that I am emphasizing that these findings should be applied foremost in regards to premium players.

I have a hunch that form is not as an important factor when it comes to super premiums. These are players who gets returns in 65% returns in games,

and so their bad form is harder to spot because they rarely go on long streaks of blanks. And when they do they seem to have been unlucky. Even if they have bad form they still play in teams where they need to be either wasteful or unlucky to blank in let's say 3 games straight.

If you’ve understood all of this then you are ready for my decision tree that I will try out for next season, unless Messi joins Manchester City because then he'll be captained nearly every week

It is influenced by my growing realization that a lot of trends and probabilities seem counterintuitive at first glance, so following it and going against the grain so to speak will hopefully give us an edge.

END OF THREAD!

This content was originally posted on Patreon but have been repurposed for Twitter. Patreon members receive early access to these types of articles as well as more in depth analysis that adds context. https://www.patreon.com/fplgents 

I want to say thank you to @DhillonAjit who convinced me to post this thread no matter how long.

Also to @FPL__Raptor who indirectly, through his articles, advocates the importance of not keeping info to yourself when you think the community might benefit from the insights.