Stats are NOT sampled to be able to represent uncommon cases hence, stats are ALWAYS recorded unless  dropped.
The reasoning behind this is that outlier statistics such as “99th latency event” occur rarely but also occur unpredictably. If our recording system has a low sampling, we are most likely to miss recording such events. In addition, unlike spans, recording measurements only requires updating aggregated data (view data) rather than creating new objects, so it is relatively light-weight even always sampled. Hence this is the reason why stats are NOT sampled but are ALWAYS recorded.
This video below “How NOT to Measure Latency” by Gil Tene might help with understanding this decision.
Measurements are dropped if the user is not aggregating them using Views