ABSTRACT
Data on the times to the event and the times of censoring (as well as any information about the censoring mechanisms) from such populations may be analyzed to provide meaningful information about parameters of the underlying population distributions. However, the data structure is different from that of complete (no censored observations) data or data reflecting only the occurrence or nonoccurence of the event (e.g. normally or binomially-distributed data), and therefore require special statistical analysis methods. The primary difference is that times-to-event are positive and are generally skewed to the right. Another key difference is that time-to-event is censored for some patients. For these patients, we only know their times-to-event are greater than those recorded, but the exact survival times (if the event is death) cannot be directly observed. We again need special methods to model this type of censoring, which in general gave rise to the statistical field called survival analysis.
