Attribute importance by position in FM27: a test on my own tactic

I replayed FM27's real match engine outside the game (a harness injected into the game) to check whether the value of an attribute depends on the position, in my tactic. Other testers found no clear position differences with small +5 tests; I wanted enough power to see one.

Setup​

Brentford (mid-table), a saved 38-match Premier League season with real opponents. My 3-4-2-1: 3 CB, 2 WB, 2 DM, 2 AM, 1 ST. Goalkeeper and opponents never change. My team uses the game's assistant for substitutions (about 3-4 a match).

Method​

1. In every simulation, every attribute of every outfield player gets its real level plus a random shift of -5 to +5 (clamped 1-20): 10 players x 50 attributes change at once.
2. 50,000 such configurations x 4 seeds = 200,000 simulated matches; the 38 fixtures rotate.
3. For each configuration I sum the level changes per position group and regress the goal difference on them (fixed effect per fixture, errors clustered by configuration). Each coefficient is the value of +1 level of one attribute in ONE player of that position, in goals per season. Points and xPts were stored too.
4. ***** = significant (|z| > 2) with the same sign in both random halves of the data. Anything else should be read as "not shown".
5. All 250 attribute x position effects come from the same runs: about 20x more precise than a one-attribute-at-a-time sweep of the same cost.
6. Checks: fabricated data with a known effect was recovered (5.2% false alarms for 5% expected, no false "confirmed"). On 1,800 new configurations the weights predict points on about the right scale (slope 0.8-0.9). Effects add up across players (no diminishing returns, all |z| < 2); Jumping Reach is convex (worth more at high levels).

Results​

Goals per season for +1 level in one player (x0.7 is roughly real points). * = confirmed. 95% margin about +-0.09 ST, +-0.06 AM/WB/DM, +-0.05 CB. "Team" = one coefficient for the 10 outfield players. † = differs by position (p < 0.001). Sorted by team effect; all 50 attributes shown.

Do positions differ?​

Yes, for some attributes (heterogeneity test over the 5 positions): p < 0.001 for Pace, Jumping Reach, Acceleration, Anticipation, Stamina, Finishing, Strength; p < 0.01 for Concentration, Balance, Heading. 13 of 50 attributes reach p < 0.05, against about 2.5 expected by chance.

• The ST is worth the most: Pace +0.90 (about 2x an AM or WB), Jumping Reach +0.73 (AM: +0.05, CB: +0.35). Finishing counts for the ST and, less, the AM.
• Pace beats Acceleration for ST, AM and WB; the reverse for CB and DM. For CBs, Acceleration and Jumping Reach come before Pace. For DMs only Acceleration stands out.
• Most technical attributes show nothing (Passing, Technique, Tackling, Marking, Positioning, Crossing, Vision, Decisions, Composure, First Touch). Crossing for my wing-backs: +0.05 +-0.06. Dribbling is the exception (confirmed for AM).

Goals vs xPts​

xPts (from xG) are far more precise but blind to conversion. Of the 30 confirmed effects, 27 are also significant in real points and in xPts, same sign. The other 3: Finishing ST and AM (pure conversion: ST goals +0.27 vs xPts 0.00; xPts z -0.2) and Determination CB (same sign everywhere, just under the line in points, z 1.9). For the ST, Heading (+0.15 vs -0.03 xPts) and Strength (+0.21 vs +0.06) also act through conversion. About 25 effects show in xPts but not in goals (e.g. Concentration for WB): probably real but small. A combined estimator (xPts + significant conversion) predicted no better on the new configurations (z -0.7), so the table stays on goals.

Limits​

One club, one tactic, mostly levels 8-18, single matches only (no Natural Fitness or injury effects over a season). ST estimates are the noisiest (one player): trust the big values (Pace, Jumping Reach, Acceleration). Goalkeeper not covered. FM27 build of 12 Nov 2026.

Awesome to see! We should re-do this with other setups but the differences are pretty clear!

Surprising ones in there:

Low stamina weight for DMs

CB jumping reach lower than ST (it makes sense for those 'box players' to have higher weight than other positions)

DMs generally seem to be the least important outfield position?

This is also my feeling when playing the game itself: my DMs don't do much during matches, they mainly bring balance.
I should specify the roles in the tactic: 3 BCBs, 2 WBs, 2 BBMs, 2 AMs, and 1 CFD.
The idea is to run the test with different tactics to see the changes and interactions that can occur between positions, roles, attributes, and instructions.
I'm also surprised long shots don't yield anything with DMs, as they are often in shooting positions outside the box and the patience instruction is set on standard.

Amazing job.

But we talk about Goals, what about Difference, or winning ?

Purity said:

For each configuration I sum the level changes per position group and regress the goal difference on them

Click to expand...

Falbravv said:

But we talk about Goals, what about Difference, or winning ?

Click to expand...

It's goal difference I think, not goals scored

correct, this is goal difference. Sorry if it was misleading. I can also give you points and xPoints tables :

Real points per season, per +1 level in one player​

Typical 95% margin (median over attributes): Team (10 outfield) ±0.02, ST ±0.07, AM ±0.05, WB ±0.05, CB ±0.04, DM ±0.05

Expected points (xPoints) per season, per +1 level in one player​

Typical 95% margin (median over attributes): Team (10 outfield) ±0.01, ST ±0.03, AM ±0.02, WB ±0.02, CB ±0.02, DM ±0.02

Goalkeepers​

Source: a dedicated campaign of 200,000 simulated matches (50,000 configurations x 4 seeds), same setup as the outfield one (Brentford, 3-4-2-1, real 38 fixtures, assistant substitutions, tactical effectiveness forced to 100). The 48 visible attributes of the goalkeeper each get their real level plus a random shift of -5 to +5 (larger shifts in a third of the configurations); the 10 outfield players never change. Same regression as outfield (fixture fixed effects, errors clustered by configuration).
Unit: +1 level of ONE goalkeeper attribute, over a 38-match season. "Conceded saved" = goals conceded less. */Confirmed = |z| > 2 with the same sign in both random halves.

(14 strongest of 48 shown, sorted by points.)

Value by level range (points per season for +1 level;​

The other 43 attributes are not confirmed (read as "no effect shown"; Acceleration, Anticipation and Command of Area come closest, z 2.6-3.1, but in each of them one random half stays below z 2).

Reading: Agility and Reflexes dominate (about 0.5 points per season per level, i.e. ~0.65 goals conceded less); Aerial Reach is worth about a third of that; Concentration and Important Matches are small but real. Everything else, including Positioning, One on Ones, Rushing Out and Command of Area, is not distinguishable from zero at this sample size (margin about ±0.07 goals per level).

Personnally, I'll only consider, in importance order : Agility, reflexes, aerial reach, concentration, important matches, acceleration, command of area, handling and anticipation.
If we want to cut it more, consider : Agility, reflexes, aerial reach, concentration, important matches.
Even more, we only keep the 2 kings : Agility and Reflexes

ST Val
Pace 100
Jumping Reach 81
Acceleration 78
Anticipation 41
Stamina 38
Finishing 30
Concentration 29
Balance 27
Strength 23
Heading 17
Important Matches 17
Dribbling 14
Pressure 14
First Touch 11
Determination 10

AM Val
Pace 100
Acceleration 77
Strength 67
Stamina 51
Anticipation 38
Finishing 36
Dribbling 36
Balance 31
Long Shots 23
Concentration 21
Determination 18
Heading 18
Composure 15
Jumping Reach 13
Important Matches 13

WB Val
Pace 100
Acceleration 83
Stamina 70
Anticipation 38
Strength 38
Dribbling 38
Jumping Reach 25
Balance 23
Important Matches 15
Finishing 15
Concentration 13
Long Shots 13
Crossing 13
Pressure 13
First Touch 10
Composure 10
Vision 10

CB Val
Jumping Reach 100
Acceleration 97
Pace 66
Stamina 40
Anticipation 40
Concentration 29
Balance 23
Pressure 23
Strength 20
Determination 20
Work Rate 17
Vision 17
Important Matches 11
Dribbling 11

DM Val
Acceleration 100
Pace 48
Jumping Reach 39
Stamina 39
Strength 35
Dribbling 35
First Touch 35
Pressure 30
Important Matches 30
Composure 30
Concentration 26
Work Rate 22
Determination 22
Agility 22
Finishing 17

Converting values for Genie Scout. I’ve been using this since yesterday and it works great—give it a try and let me know!

GK Val
Agility 100
Reflexes 90
Aerial Reach 38
Concentration 26
Important Matches 18
Acceleration 14
Command of Area 14
Handling 14
Anticipation 12
Adaptability 10
Work Rate 10