Censored data exists when the value of an observation is not fully known but is not missing either! The most common example in our fields is in follow-up data. A number of participants will have been recruited and perhaps the study aspires to know how long they spend in open ended therapy and is designed to collect data until 200 new clients have agreed to participate and each one has attended at least two therapy sessions. At that point many participants will have terminated therapy and their durations of therapy are known and not censored. However, another set of participants will be still in therapy the durations of their therapies are only known to be at least the durations for them between starting therapy and that point: their durations are “right censored”.
Details #
This matters because if we ignore the censoring we have introduced a bias into our estimation of general durations of therapies for that study. However, there are statistical methods that handle censoring and will avoid that bias (given some sensible assumptions mostly that the pattern of duration hasn’t actually changed across the period of the study). See survival analysis for probably the biggest class of such methods.
Try also #
Frailty analysis
Interval censoring
Left censoring
Right censoring
Survival analysis
Type I censoring
Type II censoring
Chapters #
Not covered in the OMbook.
Online resources #
None currently.
Dates #
First created 8.xii.24, links tweaked 15.xii.24.
