
Possession-adjusted defensive stats try to put recorded actions into an opportunity context. A player whose team rarely has the ball may encounter more situations in which to tackle or intercept than a player on a possession-dominant side. Raw counts alone cannot distinguish workload from effectiveness.
An adjustment can help frame that comparison, but it is not a direct measurement of defensive quality. Its result depends on the provider’s events, playing-time basis and adjustment method. The adjusted figure should remain beside the underlying count and context.
Understand what the adjustment is trying to do
Hudl Statsbomb’s radar explanation describes per 90 figures as a way to standardise playing time and possession adjustment as a way to account for opportunity to perform defensive actions. Those are two different normalisations.
The first asks how frequently events occurred per unit of playing time. The second tries to address differences in the conditions under which the player had chances to make them. Neither removes every contextual difference between players.
Check the product’s actual formula. A provider may use a nonlinear transformation rather than simply dividing by the percentage of time without possession. Do not label a homemade calculation as the provider’s published metric.
Also check which actions are adjusted. Tackles, interceptions, pressures, blocks and clearances are not interchangeable events. A product’s rule for one field does not automatically establish its treatment of another.
Separate the raw event from the transformed number
A raw count is an observation under an event definition. An adjusted figure is the output of a calculation applied to observations and context. A value such as 2.4 adjusted tackles does not mean that the player literally made a fraction of a tackle during the match.
The word adjusted should therefore remain in the label. Removing it can make a model-derived comparison look like an unmodified count. Keep units and time basis visible in tables, charts and accompanying explanations.
If one report shows a season total and another shows adjusted events per 90, the numbers cannot be ranked directly. First identify whether the underlying events, minutes and transformation are compatible.
That check matters even when both reports use the same player name and season. Transfers, missing fixtures and different competition coverage can produce different samples before any mathematical adjustment is applied.
A simple illustration of opportunity and ranking
The following is an invented teaching calculation, not Statsbomb’s formula or a standard possession-adjusted metric. It uses observed minutes without the ball solely to illustrate how changing exposure can change a rate.
Player A records two tackles while his team spends 27 observed minutes out of possession. Player B records three while his team spends 54 such minutes. B has the larger raw count.
Expressed per 30 out-of-possession minutes in this illustrative system, A has 2÷27×30, approximately 2.22. B has 3÷54×30, approximately 1.67. The ranking reverses because the calculation uses a different exposure basis.
This does not prove A is the better defender. The illustration says only that A’s recorded event frequency per chosen unit is higher. The difficulty, location and outcome of the situations remain unmeasured by that simple rate.
Do not substitute team possession percentage multiplied by total match time without understanding the timing definitions. Providers may measure possession through different event or clock conventions. The invented example assumes the relevant durations are known.
Time without the ball is not equal opportunity
A centre-back can spend several minutes without possession while the opponent passes harmlessly far away. A full-back can face repeated direct attacks in a shorter spell. Both are out of possession, but their opportunities to intervene are different.
Position, formation and assigned responsibility matter. A player instructed to hold a passing lane may contribute without making the event counted in the selected column. Another may produce a high tackle count after repeatedly being forced into recovery situations.
A possession adjustment cannot automatically identify which of these explanations applies. Use it to ask a more precise question, then inspect events, locations and the team’s defensive structure.
A low count is not automatically evidence of passive defending, and a high adjusted rate is not automatically evidence of successful defending. The label describes a calculation, not a complete judgement of the player’s decisions.
Check success and consequences separately
Event frequency and effectiveness need distinct evidence. If the dataset distinguishes attempts from successful interventions, retain that distinction. Do not combine a successful-tackle count from one source with all attempts from another and present a common success rate.
Inspect what followed the intervention where the event data or video allow it. Did the team recover possession, concede a dangerous continuation, give away a foul or simply move the problem elsewhere?
Those follow-up questions do not imply that every action can be reduced to a single definitive value. They make the limits of the selected count clearer. An adjusted total can describe frequency without establishing the importance of each event.
Likewise, defensive work without a recorded ball event can be substantial. Tracking information may offer another perspective, but do not invent that information when only ordinary event data are available.
Game state can alter both possession and workload
A side protecting a lead may choose to defend deeper. A trailing opponent may increase territorial pressure. A dismissal can change both teams’ opportunity structures abruptly. Whole-match summaries can hide those different phases.
Split periods when the data support a meaningful comparison. Preserve the number of players, score and duration for each part. A small filtered sample can clarify the situation while leaving the resulting rate unstable.
The field-tilt guide explains why territorial share and chance quality should not be treated as substitutes. Defensive workload needs comparable care: knowing where a side spent time does not fully establish what happened there.
Venue and opponents also matter. The home-and-away comparison guide shows why an apparently simple split can conceal unequal schedules. Possession adjustment does not make those differences disappear.
Use playing-time and sample checks together
A per 90 rate based on a brief cameo can be large without supplying much evidence about a player’s longer-term role. Show minutes and raw events alongside any normalised number.
For example, one invented action in 15 minutes becomes six per 90 before a possession adjustment is even considered. That extrapolated rate does not establish that six actions would have occurred if the player had stayed on for the whole game.
Do not let several layers of transformation conceal that small starting sample. A sophisticated-looking adjusted result can still be based on very few events. Retain a minimum sample policy if the comparison uses one, and state how it was chosen.
When comparing periods, preserve coverage rather than silently dropping matches with unavailable data. Missing observations should remain missing. They are not evidence that the player had zero events or zero defensive workload.
Do not average incompatible adjustments
Suppose one product publishes an adjusted per 90 rate and another publishes a transformed season total. Averaging them creates a number without a coherent unit. Similar chart headings do not make that operation valid.
Even within one product, a formula update can affect comparability. Keep the method version or documentation where available. If historical values were recalculated, distinguish a revised series from an older screenshot.
For a pooled comparison, follow the provider’s documented aggregation method. A simple average of match-level adjusted rates can give short appearances the same influence as full games and can differ from an adjustment applied to compatible combined data.
If the necessary inputs are unavailable, narrow the claim. Reporting the available provider figure with its definition is more honest than inventing an undocumented conversion to make two dashboards agree.
Read a radar as a comparison, not a verdict
A radar may express the adjusted rate relative to a reference population. The population, role grouping, sample threshold and time period affect that visual ranking. A long spoke cannot be interpreted fully without those choices.
Check whether a high displayed value means a higher raw rate, a better percentile or some other transformed comparison. Those are different representations. Percentile distance is not the same as the numerical gap in underlying event frequency.
A useful review therefore states the raw count, minutes, provider adjustment, role and reference group. It then discusses what the events achieved and what the data do not observe.
Possession adjustment earns its place when it improves the question being asked. It should help distinguish unequal workloads, not convert a narrow frequency measure into a claim that one player is unconditionally superior or that a future match outcome is predictable.
For adults aged 18 and over. Numerical examples are hypothetical teaching calculations, not a replication of a provider formula or actual OptaTips performance. Provider references do not imply a commercial data relationship.
댓글 0
아직 댓글이 없습니다.