I am using Survival Analysis to build an attrition model. I have a data set with ~15,000 people that spans the past 5 years. Approx. 2,000 left in the last 5 years. The remaining people stayed at the organization. Technically, are the 13,000 people who stayed observations that should be right centered? If so, is there anything I need to do or does the model understand that they are still there because the left/not left dummy variable distinguishes who left and who stayed?
Approx. 2,000 left in the last 5 years. The remaining people stayed at the organization. Technically, are the 13,000 people who stayed observations that should be right cen[so]red?
Right censoring in this context means that you only have a lower limit for the time elapsing until the event of interest. Much thus depends on your choice of a time origin from which to measure attrition.
If your interest is in time-to-attrition starting from 5 years ago, then you are correct. You have exact times to attrition for 2000 cases, and a lower limit for the 13000 others. The "left/not left" indicator will be the same as the censoring indicator.
If, however, you are interested in attrition time from the time of hiring, then some of the 2000 might also need to be coded as right censored. That would be the case if your data only go back 5 years and you don't know original hiring dates. If someone left 2 years into your study in that case, all you have is a lower limit to the attrition time: it's at least 2 years, but could be much longer depending on the date of hiring. If this represents your situation, a "left/not left" indicator isn't the same as the censoring indicator, and you need to code the cases with only lower limits for time-to-attrition as censored even if you know they left.