Regression with censored values of the dependent variable.

In geology experiments have been made by Mother Nature. We are rarely in the position to repeat such experiments under controlled conditions. Example: we wish to measure how much HC charge made it to a trap. If the trap is filled to spillpoint we can only conclude that the HC in the trap are a minimum value. May be there was more charge but what could not be stored travelled further updip to another trap. Another example relates to the top seal. An underfilled trap may due to lack of HC charge, but also to a limited topseal capacity. Therefore, only after careful geological reasoning can we assign codes to values as minimum, real and maximum.
As a normal regression analysis would require "real" values, a different regression technique is required when some of the Y-values are censored. There are quite a few programs that can handle this situation for a multiple regression. What happens is that the real values are estimated in an iterative procedure. Heavily censored values contribute less than one degree of freedom to the regression. The benefit is that significant parts of the body of data can be used that otherwise would be useless. Taking only the real data would possibly result in a biased regression equation. By the way, a fair proportion of real values for Y is required for a successful iteration and meaningful regression result.

This technique was used for gaeapas for HC charge volume correlation and the top seal study by Nederlof & Mohler (1981).