Data Weighting
Data weighting is the practice of giving individual measurements different influence in an analysis, based on recency, measurement quality or data source.
Definition
“Data weighting” is the practice of giving individual measurements different amounts of influence in an analysis. Instead of treating every data point equally, each value receives a factor that determines how much it contributes to the result. The usual criteria are recency, measurement quality and the origin of the data.
A weighted mean is calculated as:
weighted mean = Σ (value_i × weight_i) / Σ weight_iWeighting appears throughout training analysis: in load models that count recent sessions more heavily, in rolling HRV baselines, and when combining values from lab diagnostics with field-test estimates.
Why it matters
Without weighting, old or unreliable values distort the picture. A training load from six weeks ago says little about today’s fatigue, and a single HRV outlier should not flip a trend. Weighting ensures the model reflects reality rather than noise.
Weighting by source is just as important. A threshold measured in a metabolic test is a different quality of information than an FTP estimate from a 20-minute test. Treating both equally dilutes the better information with the worse.
| Criterion | Question | Typical implementation |
|---|---|---|
| Recency | How old is the value? | Exponential decay of weight over time |
| Measurement quality | How clean is the recording? | Low weight for artefacts, gaps, outliers |
| Source | Measured or estimated? | Diagnostics over field test over device estimate |
| Context | Comparable conditions? | Separate by sport, position, time of day |
How obseed measures it
obseed does not report data weighting as a metric of its own; it is a principle behind the analyses. Load models such as fitness and fatigue use time-decaying weights, so the most recent sessions have the greatest influence. Rolling HRV baselines represent trends across several days rather than single readings.
For performance parameters the rule is: objectively measured thresholds from a diagnostic test take precedence over estimates from field tests or devices. See performance parameters.
Example
An athlete has three morning RMSSD readings: 62 ms (three days ago), 58 ms (the day before yesterday) and 41 ms (today, after a short night with alcohol the evening before). The unweighted mean is 53.7 ms.
Weighted by recency (weights 1, 2, 3):
(62 × 1 + 58 × 2 + 41 × 3) / (1 + 2 + 3) = (62 + 116 + 123) / 6 = 50.2 msThe current value pulls the result down more strongly, which better reflects the present state of recovery. If today’s value were additionally given half weight (1.5) because of known confounders, the result would be 53.6 ms: the baseline stays stable and the outlier does not become a trend.
Categories
- Datenanalyse
- Technische Begriffe