Last Monday took place the 17th meeting of Working Group 41 (WG41) that is dealing on the new standard for Instrumental Odour Monitoring Systems (IOMS). This meeting was focusing on the statistics behind the Quality Assurance Level 1 (QAL1) of IOMSs. There are some aspects that make IOMS special as compared to other Automatic Measuring Systems (AMS) and one of these aspects is the cost of comparing the electronic signal of the IOMS with that of the reference method EN 13725.
One of the key issues addressed many times during the WG41 meetings is that of the number of samples that should be taken in order to make QAL1 checks viable for IOMS manufacturers. Dynamic Olfactometry comes at a cost and odour is not a substance, it is a class. The question is: how many samples should be taken in order to have a representative picture of the IOMS performance?
Every year in Europe a few hundreds of AMS makers undergo QAL1 procedures like these of Germany. The idea is to evaluate the performance of the devices with a procedure that can be taken as a reference for further checks on QAL2, QAL3 and AST. However, life is easy when you have a stable reference material or an electronic reference method that can be used at a low cost per sample. Unfortunately, there is no reference material for ambient odours. Also, a sample that has to be analysed within 30 hours cannot be considered that stable.
The more samples you take, the better confidence interval you get for your IOMS, but the cost of in this case grows quickly at range of 100-250 € per sample.
This is when economics come at stake in the standardization world. It is also when statistics come to help economics.
The Bland–Altman statistic approach is great to tackle the limitations that a correlation coefficient have, by looking at the differences of the results at different odour concentrations. However, this statistical approach assumes a normal distribution of the results. If there is a reduced number of samples, which is the scenario that the WG41 is targeting, then the power of the test to check if the data is really normal is very low, and therefore it is difficult to be certain that the data is gaussian-distributed.
Chebyshev's inequality does not require a normal distribution of the data, so there is no need to have a large amount of data to check IOMS performance, and that is why the group decided to choose Chebyshev's inequality. Unfortunately that decision comes at a painful cost as it brings a wider confidence interval.
A few other important issues were discussed during this meeting, but there is no space in this humble website to describe everything discussed in CEN working groups in detail, nor is the aim of this short text. Should you be interested in joining in the group and learning more about the future developments of this standard, we encourage you to contact your national standardization body to learn the procedure to be part of the group. Everyone is welcomed!
Picture source: wikipedia
If you find this article interesting, you might also be interested in these articles:
- First Meeting of the CEN group dealing with gas sensors and pattern recognition.
- Calibration of o-sensors? Is that possible? Some inputs from the last meeting in Berlin, Germany
- WG41 meeting on Instrumental Odour Monitoring Systems in Bilbao, Spain
- WG41 last meeting of 2019 in Delft
- New meeting of WG41 dealing with IOMSs