Problems in chemical analysis
Thread Content
We have doubts about the company’s test results, so we took some samples and sent them to qualified laboratories for external testing. The results are as follows: Sequence, External Test Result, Internal Test Result, Difference, Pass/Fail. 1: 40.79, 43.16, -2.37, Fail; 2: 41.86, 43.46, -1.6, Pass; 3: 41.28, 42.09, -0.81, Pass; 4: 41.98, 43.36, -1.38, Pass; 5: 40.54, 42.63, -2.09, Fail; 6: 41.08, 42.93, -1.85, Fail; 7: 40.96, 43.16, -2.2, Fail; 8: 40.01, 41.53, -1.52, Pass; 9: 40.79, 41.77, -0.98, Pass; 10: 40.75, 42.73, -1.98, Fail. The pass rate is 50%. The allowable error is 1.8%, but it is clear that the internal test results are all lower than the external test results. How should we analyze this situation and how should we prepare a description of it? Or, for similar situations, I would appreciate advice from experts on how to conduct statistical analysis.
I’m not sure what the units are for analyzing the samples and the results; if the units are PPB, then it’s considered normal. Also, it’s not clear whether the results of the external testing are an average from several testing agencies or just from one agency; if the results come from only one agency, they don’t hold much significance. Because two identical instruments in the same laboratory, using exactly the same method, can also yield such results. It's not only that the instrument conditions cannot be guaranteed to be 100% identical; there are also errors in instrument calibration and sample preparation. There are many reasons for the differences in results; I think that if we know how 1.8% was obtained, we should be able to figure out how to deal with the two sets of data. To ensure the accuracy of the data, it is not sufficient to simply compare two sets of results. When testing samples, it is also necessary to examine certain control samples; for example, conducting verifications using standard samples of known concentration that originate from a different source than those used for calibration, adding a known amount of a certain component to the sample (as a matrix spike or surrogate) in order to determine the accuracy of the detection, and so on, all in order to prove the accuracy of the test results. Personal opinion, for reference only.
I agree with what was said upstairs. I’ve encountered this problem in my own experiments before, and the most effective solution is to have the people conducting the experiments go to the institution that is performing the tests, so they can observe their testing methods and the reagents used. Alternatively, two people can conduct the experiments simultaneously, which allows for a comparison – it’s important to determine whether the issue lies in the method or in the reagents.
The above errors are mainly caused by several factors: instrument errors, operational errors, and reagent errors. By addressing each of these one by one, it is believed that correct results can be obtained
Your response was very professional; we carried out the external inspection at a qualified facility. We learned all our techniques from their organization. The results of the previous set of tests were the sulfur content in the ore. I’m not in the analysis field; it’s just that this current situation is affecting our judgments regarding production. The results are showing a deviation in one particular direction, and I’m not sure whether this is normal or if there’s some problem. I have consulted them, and their testing methods are exactly the same. The same goes for instrument selection.
Sample preparation of ore is very important; there is a lot of knowledge involved in ensuring that the samples are representative. Samples must be prepared using methods such as the binary or quartering method for analysis. As a result, many testing agencies state that they are only responsible for the samples provided, not for the entire batch of products. Moreover, the allowable error range for ore product analysis should be relatively large, so there’s no need to make a fuss over it. One should be more cautious when making judgments at the limit values (whether it is qualified or not).
I suggest: check the other party’s sampling method, as an unrepresentative sample can lead to significant errors.
The sample was divided into two parts, and both samples were taken by us.
I also agree with your view. In the past, the pass rate of our random inspections was around 80%, but this time it was only 50%. As a result, our former leader even doubted the previous inspection results, suspecting that we had cheated. I’m not sure if there are any analysis methods for such test results. Analyze it from a mathematical perspective.
Agree with the view from the 3rd floor. The most effective way is to have those who are supposed to analyze it do it together. Because even if the methods are the same and the instrument models are identical, different instruments will still yield different results.
Discuss this with the other party; have your staff go to their facility to conduct tests or even actual operations, to see whether there is a significant difference in results using the same instruments. This will help determine whether the issue lies with human factors or the instruments. You can also turn to the quality control authorities to have your equipment inspected and calibrated again.