Hello,
I'm attempting to construct a general method for dealing with left censored
laboratory data ("nondetects"). I have consulted texts like Helsel's
"Nondetects and Data Analysis" but these approaches don't really do what I
want. I want to be able to use a wide variety of methods to analyze the
imputed datasets, since this problem appears in almost all of my data.
Multiple imputation seems to be a workable solution to my problem.
Since I'm the sole statistician at my workplace, I was hoping to get some
input on my ideas and pointers on places to go for software implementations
for multiple imputation that can directly impute left censored data. My
problem is similar to the one described in:
Multiple Imputation for Multivariate Data with Missing and Below-Threshold
Measurements: Time-Series Concentrations of Pollutants in the Arctic
Biometrics 57 (1), 22-33.
I have often time series data with "nondetects" and am looking to
incorporate both aspects into an imputation method. I have some code for
this method from the authors but it would be a little difficult to implement
so I'm looking for other solutions.
Another idea I had is to use the Amelia package in R and to use the priors
argument to supply an interval for the censored observation, like
(0,threshold), with some high probability. Is this a valid way to
incorporate censored data into the method?
Based on the papers I've read, it seems like I need to include all the
covariates in my imputation model for any analysis I want to perform after
the imputation. Is this the case? I'm often only investigating one
analyte, but the laboratory analysis performed for that one analyte gives
values for several related analytes. It makes sense to use these related
analytes to perform the imputations because the concentrations are often
correlated. Would it be problematic to only use these extra analytes to
impute the censored values and then to use the data in, say, a regression
model with covariates that did not appear in the imputation model?
Thanks for any help you can give me,
Jason Law
Biostatistician
City of Portland, Bureau of Environmental Services
Water Pollution Control Laboratory
6543 N Burlington Avenue
Portland, OR 97203 -5452
[email protected]
