Today, I continue my series of post-season research articles with a detailed look at the final top 10,000 teams.
Top 10K History Pages
Who are all these people in top 10K? Was it beginner’s luck or years of experience that helped them get there? To answer this question, let’s take a look at their history pages which show us each team’s historical record since season 06/07 (7 previous seasons in total).
|Seasons played before||891||1,412||1,696||1,328||1,231||1,131||1,009||1,302||10,000|
|Previous top 10K finishes||7,925||1,308||482||182||72||21||9||1||10,000|
|Previous top1K finishes||9,536||384||64||14||2||0||0||0||10,000|
- Turns out that successful novices accounted for only 9% of the final top 10K; the other 91% had some experience behind them.
- For almost 80% it was a first top 10K finish; 13% finished there once before; for the other 7% a top 10K finish is perhaps nothing but a routine.
- For 867 teams in the top 10K it was a second consecutive top 10K finish; only 1 team has managed 8 consecutive top 10K finishes. Which one? Bring on the next section.
Top 10K Heroes
- Bonos heroes from Sweden is the one with eight consecutive top 10K finishes; no other FPL team could achieve such consistently high results.
- Among other impressive history pages, I’d like to single out:
- 3 previous FPL winners (since 06/07) managed a top 10K finish this season:
Road to Success
a. Getting there for the first time
At what point of season did the final 10K teams get into top 10K? The diagram below shows what share of the final top 10K teams:
- were in top 10K after any particular gameweek (dark bars);
- had already been in top 10K at least once after a particular gameweek (light bars).
- Teams that finished the season in top 10K got there gradually. Only 5% of them were in top 10K after GW5; 16% were there after one of the 5 starting gameweeks. A great start wouldn’t hurt, but it’s far from necessary.
- By the end of the first half of the season, 50% of the teams had already been in top 10K at some point; the other 50% got there for the first time only in the second half of the season.
- 2,129 teams got there only in the last 2 gameweeks of the season. 305 teams achieved their first top 10K rank of the season after GW38 seizing the last opportunity.
b. Overall Ranks
How did their ranks change throughout the season? Below, I calculate average overall ranks for different percentiles of the final top 10K:
- the median, or 50%, shows where was the ‘middle team’ of the sample, i.e. so that 50% of the final top 10K were above and 50% of the final top 10K were below this team;
- the 75th percentile corresponds to the current overall rank so that 25% of the final teams were below it;
- the 95th percentile corresponds to the current overall rank so that 5% of the final teams were below it;
Because of scaling issues, I divide this diagram into two, one for each half of the season. I don’t include the worst team’s overall rank for the same reason.
- Overall ranks of the final top 10K teams improved gradually. After the first 5 gameweeks 50% of them were below the 276K mark; 5% were outside top 1.3 million teams.
- By the middle of the season, 50% were in top 20K; while 25% were outside 50K.
c. Overall Points
During the season, a lot of us ask ourselves a question: “I’m X points behind top 10K, can I still make it?”. The diagram below shows how a distance from the current top 10K mark changed during the season for:
- the median team of the final top 10K teams, i.e. so that 50% of the final top 10K teams had less points at that moment;
- the 75th percentile;
- the 95th percentile; and
- the currently worst team of the final top 10K.
- For the first 8 gameweeks, the final top 10K teams were mostly falling behind; after GW8 50% were at least 40 points behind the top 10K mark of that time.
- The biggest gap that was closed this season was 217 points behind the top 10K after GW16. Hakuna matata was the hero.
- TEAM BANCROFT covered a 132-point gap in last 7 gameweeks, breaking into the top 10K after being ranked at 178K before GW32.
- 84 points could be covered in the last 3 gameweeks. Loken Lions YEE xD managed it.
Transfers and Point Hits
a. How many transfers did a top 10K team make this season?
Too many transfers transform into point hits. The diagram below shows the distribution of points spent on transfers. Only 337 teams made no point hits at all. Globo Gym set the highest value here too: 216 points spent on transfers.
b. Is there a link in the top 10K between the overall rank and the number of transfers?
I plot the number of transfers for each team in the graph below; steep vertical lines for lower ranks reflect the fact that FPL use total number of transfers to rank teams which have the same amount of points.
- The graph above suggests that there was hardly any correlation between the number of transfers and final position in the top 10K. A simple linear regression built based on these data has a very low and statistically not very significant coefficient. There was a 1 transfer difference between an average team at the top and an average team at the bottom at best.
- Even though the average number of transfers made is roughly the same for any big subsample of the final top 10K, variance of the number of transfers increases from teams at the top to teams ranked lower. Standard deviations of the number of transfers for teams in the top 1K and for teams ranked from 1K to 10K were 7.8 and 9.25 respectively, i.e. the numbers for the top 1K were concentrated closer to the average value.
What about point hits?
- Well, it’s easier to find out some correlation here, although it’s not a very strong one. Average points spent on transfers are almost 5 points lower in the case of teams at the top. A linear regression has a positive and statistically significant coefficient. Looks like taking less hits was one of the factors that actually helped some teams get above the others, but there was only a 1 hit difference between an average team at the top and an average team at the bottom.
- Just like in the case of transfers, standard deviation of the amount of hits taken also increases from teams at the top to teams at the bottom, from 28.4 for the top 1K teams to 34.4 for the rest of the final top 10K sample, suggesting that values were less spread out for teams ranked higher.