#792 · Squad comparison, a new view: Attributes & output · round 2
The last of the three new views, after Game time & form (#788) and Contracts & terms (#790). Ross: "i like all 3 ideas and would like to see all of them". This one asks whether a player's attributes show on the pitch - which attributes go with what our players produce, and how strongly.
What to look at, round 2 - and what measuring it changed
Ross, on round 1: "Own tab, hold back the judging, write it up". Round 1 also marked each player above or below a line through our squad, beside the chart and in a table of its own, Over and under. Both are held back: positions sit at different levels, so one line marks a striker down for being a striker, and a line per position needs ten players in each. Round 1 is in this file's history for when the save has them.
The pitch was wrong, and measuring it first is what showed it. The two examples offered for this view fail, in two different ways. Passing against pass completion has no link at all - pass completion is set by role, centre backs completing simple passes and creators risking hard ones - and neither do others that look as obvious: …. Finishing against conversion swings with who is counted - see below. So the view does not fix a pairing and judge players against it. It measures every pairing on the save, within the positions it is for, and says how strongly each links.
A tab of its own, Attributes & output, between Statistics & metrics and Game time & form, since it joins the two tabs before it. It reads the latest capture. It could have been a section of #784's Graphs; Ross chose its own tab.
Pairings, one chart at a time, as #784's Graphs: an attribute along the bottom, what it should produce up the side. Every pairing is listed on the left by area, each with how strongly the two go together across our squad - a rank correlation, ρ, across our players over the card's minutes bar. A pairing with little or no link, or too few players to be sure, is listed faint, and its chart says which.
Each pairing is for the positions whose job its output is - interceptions and tackles for defenders and midfielders, xG and key passes for midfielders and attackers, crossing for defenders and attackers, the rest for all outfield players - by the position each is filed under (#495). It is measured among those alone: a striker who intercepts little is doing his job. Anyone else picked is drawn hollow and grey.
ρ needs ten players: with fewer it swings with who is in it, so the pairing is listed faint with how many it rests on, and it firms up as the save grows.
Beside the chart, those picked, by position: most of the output first, with the attribute beside it - like with like, and nobody marked. With more than six picked, the dots are coloured by position too.
Hollow: a player the pairing is not for, and a player under the minutes bar (his rates are the card's greyed ones). Not on the chart: a player out on loan (his figures are another league's), keepers - see below - and anyone with no statistics.
Real: every attribute and metric of the December save's latest capture, December 25/26, and the minutes behind them.
Found building it - computed on this page from the December save:
What shows now: ….
Few players make ρ swing.… So ρ needs ten players to rest on, and a pairing with fewer is listed faint. In December that holds back every shooting, chance and defending pairing - though some are real, and will show as the save grows: ….
A link across the squad can be positions, not the attribute. Round 1's first cut judged Benjamin Šeško short on interceptions, because its line ran through defenders. … So each pairing is measured only among the positions whose job it is.
Even a pairing for everyone has a level that differs by position. Strikers contest harder headers, and win fewer at the same Heading than defenders do - so round 1 still put Šeško below the headers line, partly for being a striker. The honest fix is a line per position, and that needs ten players in each - more than one half season holds. So the judging is held back, and the view shows each pairing and how strongly it links.
Few of our players are over the bar.… So a chart states its ρ and how many it rests on, and a weak one says so.
Keepers cannot be measured.… Keepers are left out of this view.
The league capture cannot widen it. The Premier League capture in the December copy holds 365 players and 64 metrics, and no attributes - FM's league view does not show them. So a pairing can only be measured on our own players, and it firms up as the save grows.