How to recode data to take Level of Detection / Level of Quantitation into consideration

Question

I'm still learning a bit about JMP, but I created a model where the output was oxygen level.  My instrumentation gave me a precise number (down to hundreds of percent), but we determined that the accuracy is not appropriate at these levels, so I would like to recode the data for all values less than the set threshold.  Most of the data actually falls below this threshold value.

How should I approach that so that I don't lose the continuous nature of this data?  Is there a way that I can recode the column to show the values at < LOQ, but still have something that will be significant when I try to build a model from this?

paulp · Answer

Perform your modeling in the Analyze>Fit Model >Personality: Parametric Survival -- it will add a dialog box to indicate which column contains the censor indicator values.  You can choose the distribution type as Normal (I think it defaults to Weibull) --  examine your residuals to see if your selected distribution is appropriate.  If you are unfamiliar with these Survival distributions, I would get in touch with JMP Support.

tmartinot · Answer

Thank you.  That's very helpful.

How do I then port that feature when building a model or trying to fit a model with these data?

paulp · Answer

You have data that can be handled in the Qualtiy and Reliability platform.  Your < LOQ data is what's called 'left-censored'.  For these results, your value of LOQ really means LOQ OR LESS.  You need to add another column called "Censor" beside your results.  This is an indicator column, place a 1 next to all LOQ values, and a 0 beside the remaining values that all represent actual precision measurements.  See the link below for a 1-page summary to get a better idea.

http://www.jmp.com/academic/pdf/learning/10_distribution_fitting.pdf