censoring provide one example

#1
Hi there,

Censored variables come up usually in survival analysis (also in reliability) where the time to event is censored. Thus, usually (or I should say always to my knowledge), the censorings take place in positive random variables (since the time is the variable censored).

Is anyone aware of any setting that the variable subject to censoring can also take negative values? Or can anyone think of such a setting?

Thanx in advance for any answers!
 

fed1

TS Contributor
#2
I think the vast majority of outcomes in Biology are reported as strictly positive because they are usually things like concentrations or blood pressure and so forth.
 

Link

Ninja say what!?!
#3
you should be careful to accept that censoring happens randomly in studies. Informative censoring happens all the time.
 
#4
Thanx fed1,

So apart from Biology, I am trying to think of a setting where one would want to fit for example a normal distribution (or any other not necessarily positive supported distribution) to a sample of censored data..

Thanx Link, I am aware of (non) informative censoring, just wondering for a setting when a random variable that is not necesarily positive may be censored.

Thanx again for any answers!
 
#5
For survival analysis observations are censored because no event occured. You could apply this to other applications if observations didn't match some criteria. This would basically be sub-setting.

How to plot that to indicate that censoring occurred would depend on what you are doing.

Is there a specific problem you are trying to address?
 

Dason

Ambassador to the humans
#6
Thanx fed1,

So apart from Biology, I am trying to think of a setting where one would want to fit for example a normal distribution (or any other not necessarily positive supported distribution) to a sample of censored data..
You could be taking measurements and the tool you're using to measure has some sort of upper or lower limit. Maybe your thermometer can only measure down to -40 degrees celsius and anything below that is just marked as -40. Is this what you're looking for?
 
#7
Thanx Dason,

That's a good thought! The most common type of random censoring is the right censoring.
In the setting you described there could be no random censoring. It is like censoring on a cutpoint (whatever below the cutpoint (or above a cutpoint if you had a maximum temperature and negative temperatures are permitted also) is being censored).

This setting is like censoring at the end of a survival study where all event free subjects are being censored at that time.

Are you aware of any literature that deals with a data set that contains a variable that takes values in the real line and is subject to censoring?

Maybe the censoring takes place only for positive values, but still the variable can also take negative values (that may be always observed exactly). That would also be a nice example for me!

Thanx again Dason!
 

bugman

Super Moderator
#8
So apart from Biology, I am trying to think of a setting where one would want to fit for example a normal distribution (or any other not necessarily positive supported distribution) to a sample of censored data..

Thanks for the interesting post leo nidas:

Just some thoughts reffering to above:

how about surveys relating to salaries:

eg. <20, 20-30; 30 -60 60< etc...

where the measurements can be made but are not becasue they contain sensitive information.

Age or test scores might be another.

I have attached a couple papers that you might find interesting.


P