# Estimate values below the limit of detection from a left-censored dataset

I have 32/180 determinations of a laboratory marker below the limit of detection (LOD) and would like to estimate them by maximum likelihood estimation. By using Minitab's reliability / survival analysis I was able to find a well-fitting model (lognormal) and to estimate mean, SD, and percentiles. How can I estimate the single missing values that are below the LOD? Is it OK to just use the percentiles for the first 32 values (up to the 17.77th percentile)?

• Without additional information (that is, covariates for each determination) it is impossible to estimate specific values. That is because your estimation method assumes the data are independent. That implies there is no basis to assign different values to different less-than-LOD results. – whuber Jul 4 '13 at 16:07
• Thanks for your comment, now I ran multiple imputation to determine the missing below-LOD-values. – MedFellow Jul 7 '13 at 22:14