1
$\begingroup$

I am working with a subset of a radar image of the ocean, where every pixel has one single value that represents the amplitude. The probability density function (PDF) of the sea pixels can be estimated using a normal distribution. However, due to the possible presence of boats and other objects on the surface that will negatively affect the distribution with their strong amplitude, I removed the 5% highest amplitude values, and therefore pixels of the image, before calculating the mean and the sigma. This obviously introduces a bias in the calculation of the parameters. Is there a way to take into account this bias? is this a problem of right-censoring?

enter image description here

$\endgroup$

1 Answer 1

1
$\begingroup$

One option is to analyze the trimmed data as-is using a normal model. Another option would be to exponentiate the observations and analyze them like a time-to-event analysis using a lognormal distribution. You can create a Kaplan-Meier plot with all of the data points and identify the top observations you would like to administratively censor. You would replace these highest values with an administrative censoring value and flag them as censored. Using this censoring flag you would feed this data into a procedure like SAS Proc LifeReg requesting a lognormal distribution. The intercept in this model is the estimate of the mean in the original scale, accounting for censoring.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.