statrefs home‎ > ‎Main‎ > ‎Fundamentals‎ > ‎Types of Data‎ > ‎

Censored Data


General Notes

Data censoring is commonly encountered in Reliability studies but may be encountered elsewhere, such as:  social statistics (recidivism); medical or clinical statistics.



Type 1, Type 2, and Type 3 censoring
  • Type 1 censoring
Singly censored data.
Terminating the test after a predetermined time or number of cycles is reached.
  • Type 2 censoring
Singly censored data. 
Terminating the test after a predetermined number of failures occurs.
  • Type 3 censoring
 Randomly censored or Progressively censored data. 
See Ibrahim Survival Lecture (attached to parent page).
 
Left, Right, and Interval censoring. 
  • Left censoring
(add info)
  • Right censoring
A unit that has not failed at the conclusion of a test is said to be right censored because it would have eventually failed, but at an unknown time.
  • Interval censoring
Monitoring of a test may require inspection of units at discrete time intervals throughout the test.  A unit that was known to be functional when it entered a time interval but was found to have failed when checked at the end of the interval.  We know that it failed sometime in the time interval, but do not know exactly when.
 
If we do not know exactly when a unit failed, how do we incorporate it into the analysis?
 
Analyses that incorporate censoring in the data are often based on Maximum Likelihood methods for parametric models, or the Kaplan-Meier method for non-parametric survival models.

 

Assumptions
 
Many methods of reliability or survival analysis make the assumption of non-informative censoring [Wikipedia].
 
 

References
 
See the StatSoft Electronic Textbook reference on Censored Observations, as well as the Glossary under "Censoring".

See section 8.1.3.1. of the NIST Engineering Statistics Handbook information on Censoring.

 
Comments