[javert]

Hi all.

I work at an analytical laboratory. Although we run the required quality control measures like blanks, fortified samples, etc. we run into another problem: sometimes there are some samples with big values that our clients are inevitably going to complain about. Given the amount of samples we analyze each day, it is unfeasible to detect and investigate every sample result. It is not merely a matter of highlighting values beyond a certain threshold, since each clients has a usual range where the majority of the samples fall on (we can retrieve the results for each client and each site from the database), so I need to put some statistical control on the results so the suspect ones can get highlighted and checked accordingly. Problem is, most analytes follow an asymmetric distribution: no negative values, the bulk of the values around the mean and a single tail to the right for data that may be outliers.
I have tested the usual quality control that assumes normal distribution and it behaves poorly, using median and median absolute deviation improves it but I'm looking for something more specific. I want to use an asymmetrical distribution as the basis for the distribution of the results have looked in several books regarding statistics and quality control and so far I have found little in the way of models and parameter estimation for asymmetrical distributions.

Have any of you been in this kind of situation? Do you know any sources I can consult for this issue?

Thanks in advance.

[curiouscat]

I'm still confused about the goal: Outlier detection? Control charts?  6 sigma?

What's your target?

[Illuminatus]

Hello javert,

I assume your goal is create a working statistical range to use to knock out outliers. I don't know why you have asymmetric distributions, but I'm sure that by now you've realized you cannot feasibly model skewed distributions for each analyte. In my opinion, your only course of action is to produce more replicates of samples (increase your n value). This will normalize your observed sample distribution and allow you to work with standard deviation a lot more accurately.

Note: If you work with liquid analyte samples I have seen skewed distributions occur due to carryover when pipette tips are not changed; this could be applicable to measuring out solids as well. To maintain accuracy always use clean equipment to create your samples!

Enjoy