What is the best way to deal with top-coded age while doing regression analysis.
If you are examining the presence or absence of a disease as your outcome, and a number of independent variables in the model how do you control for age if you have a continuous age variable available for all adults aged 60-80 years, and then everyone aged 80 years and older has been top coded as 80 years old only?
Hello Lauren,
Censoring data (or Winsorizing the upper tail only) is usually done because people worry about outliers, and censoring/Winsorizing induces somewhat less bias than outright deletion of such cases. There is bias, however. So, how to handle it in your case?
1. Only use instances of age 60-79, since you would presume that (other than coding errors or participant giving incorrect values) that these are accurate as is.
2. Use the censored data set, but acknowledge that your confidence about the accuracy of the function relating age to outcome is necessarily lower at the upper end of the age distribution (and possibly overall) because of the way age was recorded.
Your situation is a good example of why it's important to consider how each variable in a study is to be quantified before data are collected.
Good luck with your work.
Hello Lauren,
Censoring data (or Winsorizing the upper tail only) is usually done because people worry about outliers, and censoring/Winsorizing induces somewhat less bias than outright deletion of such cases. There is bias, however. So, how to handle it in your case?
1. Only use instances of age 60-79, since you would presume that (other than coding errors or participant giving incorrect values) that these are accurate as is.
2. Use the censored data set, but acknowledge that your confidence about the accuracy of the function relating age to outcome is necessarily lower at the upper end of the age distribution (and possibly overall) because of the way age was recorded.
Your situation is a good example of why it's important to consider how each variable in a study is to be quantified before data are collected.
Good luck with your work.
Thom Baguley I am only aware of ways to censor the response, not a predictor. Could you give me a hint?
I confess I skimmed the and was thinking of censored outcome data, but I do believe there are models for this - I just don't know much about them and they might no be appropriate here.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS