3 ms·

By similar reasoning, only about 93% of humans who ever lived went on to die.

• gwern 9y ago
  The median isn't affected by the right-censored datapoints (ie you only need to observe half of humans to find the median life expectancy, by definition), which is why it's used so much in survival analysis. Failure seems to mean 'closed' but there's no reason to expect such a short lifetime of exchanges that they would all close already (an exchange can be immortal) and even if you think that, the exact number of 48% is useful as it gives you an idea of how informative the dataset is due to right-censoring (the closer to 100%, the closer to a complete dataset in which all deaths have been observed).