A time scale is the duration required to obtain one individual measurement or observe a relevant outcome. It shapes how quickly experiments can produce reliable aggregate metrics: slower measurements require longer experiments. section 7.2.3 section 1.1.2
For an aggregate measurement, the required number of individual observations increases with measurement noise and with the precision demanded by practical significance: num_ind >= (2.48 * sd1_delta / prac_sig)^2. Doubling noise quadruples the required observations. section 7.2.3
Use time scale to anticipate experimental and operational constraints:
- Fast measurements, such as web-server latency, support rapid iteration. section 7.2.3
- Slow measurements, such as monthly unique users, can make experiments impractically long. section 7.2.3
- Trading strategies require predictors and risk understanding suited to their holding period, from seconds through months. section 1.1.2
Latency versus user activity
A web-server response can be measured in tens of milliseconds, while daily active users require a day-scale observation. Both can support business evaluation, but the latter constrains rapid experimental iteration. section 7.2.3
