DBScalcio.it has been updated: the new Z-Score area is now online Over the last few days, we have worked on an important revision of the site’s rating system. The main new feature is the introduction of the new AVW_Z area, based on Z-Score normalization. New home: https://www.dbscalcio.it/ From now on, the Z-Score version becomes the main reference area of the site, while the old AVW version remains available. What is AVW_Z? The old AVW adjusted player ratings according to the average level of the league. AVW_Z goes one step further: it considers not only whether the average rating of a competition was high or low, but also how spread out the ratings were in that specific season. In practice, a 6.40 average obtained in a very strict league is not treated the same way as a 6.40 average obtained in a season where ratings were generally higher. This makes comparisons between different eras, leagues and seasons more consistent. Example player profile: https://www.dbscalcio.it/z/player_profile.php?player_id=6512 TIR_Z has also been introduced Player profiles now include a new index called TIR_Z: Team Impact Rating Z. This value does not only measure a player’s absolute performance level. It also tries to measure how much the player performed above or below the average level of his own team. For example: a player with AVW_Z 6.10 in a weak team may have a higher relative impact than a player with AVW_Z 6.20 in a very strong team. New Z-Score team pages Team squads have also been updated with AVW_Z, and the Top 11 is now calculated using the new metric. Example: https://www.dbscalcio.it/z/team_search.php?team_name=Paris+Saint+Germain&season_year=2022/23 The Top 11 no longer uses the old AVW, but the normalized Z-Score rating. New Z-Score rankings The main research areas have also been updated: Research Z: https://www.dbscalcio.it/z/ricerche.php Various Rankings Z: https://www.dbscalcio.it/z/classifica_per_ruolo.php League Rankings Z: https://www.dbscalcio.it/z/stagioni_per_campionato.php Research Match: https://www.dbscalcio.it/z/research_match.php New player and team rankings The Z-Score home page now uses dedicated Top Players and Top Teams tables based on AVW_Z, so the main rankings are consistent with the new system. Updated glossary The Feed & Glossary page has also been updated with an explanation of the new indexes: https://www.dbscalcio.it/feed_glossary.php In short: AVG = real average rating. AVW = old league-adjusted average. AVW_Z = Z-Score normalized average. TIR_Z = player impact compared to his own team. The work is not finished: the new Z-Score area will continue to be refined, but the site now has a stronger statistical basis for comparing players, teams, leagues and different historical periods.
Nice addition @vyncy - I remember having mentioned this topic to you, without seriously believing that some new calculation could be added for this purpose (you also felt it would be too difficult I think), but now you have been able to do it! I agree it should account a bit more for ratings-variance and so yeah it can be a bit more of a 'level playing field' guide for normalising ratings.
Do you have formulas for each? Or did i miss them? I can suggest some ideas. Funny enough, I use z-scores almost daily at my main job.
How is AVW calculated? What is Target STD in value for AVW_Z? Always 0,30 as in TIR_Z? It is not specified in AVW_Z formula? Also, in formula TIR_Z, z-score relative to the team performance is needlessly expressed with AVW_Zs. Simple formula with AVG player rating - AVG team rating / team STD, gives the same output. Unless this is intentionally expressed as such for some other reasons.
@vyncy That makes sense. Target STD and the TIR_Z multiplier are very important because they define the “feel” of the metrics and how accurately can results be intuitively interpreted. Imo, this should be treated as a central part of the methodology. Ideally, I would personally experiment with grounding these parameters empirically in the observed distribution of average ratings. Maybe you have already done that, but I think it would be valuable to explain the logic behind the chosen values to fans. I would also consider adding percentiles next to AVW_Z and TIR_Z values, or some kind of visual interpretation guide/cheat sheet for each. That would make the normalized values much easier to interpret at a glance.
You are absolutely right. The target standard deviation used in AVW_Z and the multiplier used in TIR_Z are not secondary details: they define the sensitivity of the model and how users perceive the final values. At the moment, AVW_Z uses a target STD designed to keep the scale close to the traditional football rating system, where 6.00 is average, values above 6.10 indicate positive performance, and values below 5.90 indicate negative performance. The idea is not to create an excessively volatile metric, but to normalize different leagues and seasons while preserving an intuitive reading. TIR_Z is intentionally more compressed. Its goal is not to replace AVW_Z, but to show whether a player overperformed or underperformed compared with his own team. That is why the multiplier is smaller: it avoids exaggerating internal team differences. That said, I agree that these parameters should be explained more clearly as part of the methodology. We are considering adding a more visual guide and possibly percentile information next to AVW_Z and TIR_Z, so users can understand not only the rating value, but also where the player stands in the distribution. For example: - AVW_Z tells how good the player was compared with the league/season environment. - TIR_Z tells how much the player stood out within his own team. - Percentiles could help translate those values into a more intuitive ranking position. This is a very useful suggestion, because the goal is not only to calculate normalized values, but also to make them understandable at a glance.