INTRODUCTION:
A common question for all analytical techniques is, “How precise are the results?”.
There are two methods to describe the precision of an analytical technique – repeatability and reproducibility. The Wikipedia definitions for both are below:
Repeatability or test-retest reliability is the closeness of the agreement between the results of successive measurements of the same measure when carried out under the same conditions of measurement.
Reproducibility, also known as replicability and repeatability, is a major principle under pinning the scientific method. For the findings of a study to be reproducible means that results obtained by an experiment or an observational study , or in a statistical analysis of a data set should be achieved again with a high degree of reliability when the study is replicated.
For NIR analyzers, these terms have the following significance:
- Repeatability is generally repeated measurements on the same instrument under the same conditions. In general, higher-quality NIR instruments are highly repeatable, and this error is typically a small fraction of the SECV or SEP, two measurements of the prediction model accuracy.
- Reproducibility describes how reliable measurement is under conditions found in normal analytical conditions, including different instruments, technicians, and locations. Reproducibility is a good estimate of the performance one can expect when implementing the analytical technique in routine and normal use.
Reproducibility is essential when deploying fleets of instruments or enterprise implementations, as it ensures that results across devices and locations can be directly compared and trusted.
At NeoSpectra by Si-Ware, we recently had a customer who purchased many NeoSpectra Scanners for deployment in small fleets at multiple remote locations. To provide an indication of reproducibility, we set 40 of the instruments up in our regional office and analyzed a corn sample of whole kernel corn on each of the NeoSpectra scanners using a calibration from one of our LabStore partners.
EXPERIMENT CONDITIONS:
- 40 NeoSpectra Scanners
- Samples were analyzed sequentially across two days
- Samples were scanned in contact with the NeoSpectra Scanner
- Five measurements were used with manual movement between to calculate the average prediction
- Reference scan was taken immediately before each sample measurement
- Whole kernel unground corn calibration
Whole Kernel Corn Calibration Statistics:
RESULTS:
The results from all NeoSpectra scanners were very consistent across the group, and all results were well within the calibration error from the population mean predictions.
NIR Results by NeoSpectra Scanner Serial Number:
NIR Results Summary:
DISCUSSION:
The prediction error RMSEP is the propagation of the laboratory error, instrument error, and sample presentation error. The standard deviation (SD) of the predictions for protein and fat were very low across all instruments, while the moisture was slightly higher. The moisture calibration had a higher RMSEP and a higher range, which lead to results with a larger SD. Also, it was noticed that the results had a slight increase and then decrease across the sample scanning times, likely indicating a change in the sample’s moisture due to the humidity in the environment (July, Midwest United States) and/or repeated sample testing. This slow change in the sample moisture increased the SD across the results. Still, the SD observed in the test was well below the calibration error and highly reproducible across all 40 NeoSpectra Scanners. This test demonstrates that a new NeoSpectra Scanner and a validated whole corn calibration from our NeoSpectra LabStore partner work reliably out of the box with highly reproducible results for all parameters across a fleet of analyzers.
