Please contact Caltest by Email for finished copies of this article.
Home
Outliers - How To Treat Them
Once a calibration has been Caltest Certified, the instrument using the calibration becomes the primary, and preferential tool in the analysis of samples for the product that is represented by that calibration. The operative word here is "represented".
The software in your system uses two features to identify "outliers". One indicator is called the "Global ‘H’ and the other is the "T test". A discussion of how "Global H" and "T Test" work is beyond the scope of these instructions and can be obtained from the suppliers of the ISI/NIRS software. Suffice it to say that the magnitude of the "Global H" and "T Test" values determines whether the product you are running on your system is represented by the calibration. Global H values greater than 2.99 and/or "T Test" values that are starred or red in color, mean that the product is not represented by the Caltest Certified Calibration. Results obtained for a product not represented by the Caltest Certified Calibration should not be accepted as being accurate.
The "Outlier Record" is a worksheet that will allow you and Caltest to keep track of those samples that do not fit the current calibration model. When sufficient numbers of outliers have been identified and collected, they can be used to enhance the current calibration model, or allow for the development of new models. By developing, or enhancing a calibration model, one can significantly reduce the number of samples that flag as outliers and therefore need to be sent to a reference method for verification.
The "Outlier Record" consists of eight columns, (A - I) and thirty two rows (R001 - R032). For purposes of this procedure, there is one "odd row" (R001) and one "even row" (R002) per sample, making it possible to record outlier results for up to sixteen samples per page. All the information that is needed to keep track of outliers can be entered into this record. Each page of the "Outlier Record" can hold up to two constituents.
Global H Red, Double Starred, Or Single Starred
1. If after running a sample on your NIR you have encountered a Global H greater than 2.99 (red numbers or, 1 or 2 Stars in the G’H’ column of your computer screen) use the accompanying "Outlier Record" in this documentation package to record the results of your outlier. As in the Caltest Certification Check Sample worksheets, the constituent values entered into the "Outlier Record" should also be on a Dry Matter Basis.
2. Select the next pair of "Outlier Numbers" from column (B) that have not had constituent results recorded for them.
3. With your NIR analysis results still on the screen, record the sample number that you used when you ran the sample producing the high ‘H’ into column (A) of the "Outlier Record". On DOS versions of the ISI software, the sample number is shown in yellow characters on your computer screen. On Windows versions of the ISI software, the sample number is shown in blue characters on your computer screen.
Note: You should be using sample numbers with only a maximum of eight characters in your normal day-to-day sample analysis operation. Eight characters will allow for the addition of the four character "Outlier Number", should you encounter an outlier, or the four character Certification Check Sample number when you are prompted to run these samples. Use the ‘Sample ID’ input lines in your ISI software if you need to enter a larger sample number or you need to add additional information about your samples.
4. Record any one of the G’H’ values shown on your computer screen into the next available "odd row" of Column (C).
Note: All G’H’ values for the predicted variables should be the same. G’H’ values for the calculated variables, like moisture, are zeros.
5. Record the NIR, Dry Matter Basis results you’ve obtained for each constituent into the space provided for that constituent, in column (F), on the same row you used to record the G’H’ value.
6. Repack another sample from the same container from which the first sample was taken, into another cup and run a second analysis on the NIR. When again prompted by the computer for a sample number, use the same sample number in the ‘Sample Number’ purple box that was used and recorded in Step (3).
Very Important: But this time, add to the tail end of the eight character sample number, an R, followed by the three digit, even Outlier No. taken from column (B) of the "Outlier Record". Please refer to the ‘Instructions: Certification Check Samples - Entering Sample Numbers’ for helpful information.
7. Record any one of the G’H’ values shown on your computer screen into the "even row" of Column (C), just below the "odd row" entry made in Step (4).
8. Record the NIR, Dry Matter Basis results you’ve obtained for each constituent into the space provided for that constituent, into column (F), on the same row you used to record the G’H’ value.
9. If the Global ‘H’ of the second pack is less than 2.99, compare the NIR constituent results of the two packs for similarities. Although the sample will still need to be sent to the reference laboratory for verification, a judgement call will need to be made by you on whether to release, or hold, the batch represented by the sample. Your past experiences with NIR and with the reference laboratory, concerning similar, possibly borderline samples, will guide you in your decisions. It is hoped that when sufficient quantities of these outlier samples have been added to your data base, the frequency of making such judgement calls, in the future, will be significantly diminished.
10. Unpack the two NIR sample cups, and return their contents to the original sample container.
11. Incorporate with vigorous mixing, the NIR packs, into the material that was remaining in the sample container, seal the container and send it to the reference procedure for analysis verification.
12. Enter only your Dry Matter Basis results from your Laboratory, into the space provided on the "Outlier Record". Be careful not to cross results on the "Outlier Record. A good way to not get confused, is to write the Outlier Number, R001, R002, etc. on the sample container before sending it to the reference laboratory.
13. If your laboratory reports results only on an As Is, or Moisture Basis, you must compute the Dry Matter Basis of your Laboratory results before entering those results into the Record.
Example:
Protein (Dry Matter Basis) = Protein (Moisture Basis) ÷ [(100-Moisture) ÷ 100]
14. Fax your tables to Tony Pietrantonio at (518) 383-5024 only after you have entered the results of at least sixteen samples into your "Outlier Record".
15. Caltest will incorporate your new data into the database and expand your calibration after a sufficient number of outliers have been processed through this procedure (usually 10% of your current data base).
16. Refer to "The Proactive Use of Global H" included in the "Notes & Information" section of your documentation package for additional helpful information.
‘T’ Red, Double Starred, Or Single Starred
Unlike high Global H values, it is not uncommon to find only one starred ‘T’ value for only one constituent in what may seem to be an ordinary NIR analysis. High, or starred T’s usually mean that the constituent that has been flagged has produced a result that is outside the calibration range. Since NIR can be a poor extrapolator, if a constituent goes outside it’s calibration range, the result returned by the NIR, is most likely, erroneous. Although watching Global ‘H’ is important, watching the T’s is very important.
17. If after running a sample on your NIR you find that some of your constituents have red ‘T’ numbers or T’s with 1 or 2 Stars, use the accompanying "Outlier Record" in this documentation package to record the results of your outlier. As in the Caltest Certification Check Sample worksheets, the constituent values entered into the "Outlier Record" should also be on a Dry Matter Basis.
18. Select the next pair of "Outlier Numbers" from column (B) that have not had constituent results recorded for them.
19. With your NIR analysis results still on the screen, record the sample number that you used when you ran the sample producing the high ‘T’ into column (A) of the "Outlier Record". On DOS versions of the ISI software, the sample number is shown in yellow characters on your computer screen. On Windows versions of the ISI software, the sample number is shown in blue characters on your computer screen.
Note: You should be using sample numbers with only a maximum of eight characters in your normal day-to-day sample analysis operation. Eight characters will allow for the addition of the four character "Outlier Number", should you encounter an outlier, or the four character Certification Check Sample number when you are prompted to run these samples. Use the ‘Sample ID’ input lines in your ISI software if you need to enter a larger sample number or you need to add additional information about your samples.
20. Record the ‘T’ values shown on your computer screen into the next available "odd row" of Column (E). Record the ‘T’ values for all of the constituents even though some of the ‘T’ values are not red or have red stars on them.
Note: All ‘T’ values for the predicted variables will be different.
21. Record the NIR, Dry Matter Basis results you’ve obtained for each constituent into the space provided for that constituent , on the same row as the ‘T’ value
22. Repack another sample from the same container from which the first sample was taken, into another cup and run a second analysis on the NIR. When again prompted by the computer for a sample number, use the same sample number in the ‘Sample Number’ purple box that was used and recorded in Step (19).
Very Important: But this time, add to the tail end of the eight character sample number, an R, followed by the three digit, even Outlier No. taken from column (B) of the "Outlier Record". Please refer to the ‘Instructions: Certification Check Samples - Entering Sample Numbers’ for helpful information.
23. Record the ‘T’ values shown on your computer screen into the next available "even row" of Column (E) just below the "odd row" entry you made in Step (20). Record the ‘T’ values for all of the constituents even though some of the ‘T’ values are not red or have red stars on them.
24. You can release your results for all constituents that did not produce a red or starred ‘T’ and/or a Global H greater than 2.99.
25. Unpack the two NIR sample cups, and return their contents to the original sample container.
26. Incorporate with vigorous mixing, the NIR packs, into the material that was remaining in the sample container, seal the container and send it to the reference procedure for analysis verification. It is important that analysis verification be performed for all of the constituents, even for those constituents that did not produce red or starred T’s.
27. Enter only your Dry Matter Basis results from your Laboratory, into the space provided on the "Outlier Record". Be careful not to cross results on the "Outlier Record. A good way to not get confused, is to write the Outlier Number, R001,
28. If your laboratory reports results only on an As Is, or Moisture Basis, you must compute the Dry Matter Basis of your Laboratory results before entering those results into the Record.
Example:
Protein (Dry Matter Basis) = Protein (Moisture Basis) ÷ [(100-Moisture) ÷ 100]
29. Fax your tables to Tony Pietrantonio at (518) 383-5024 only after you have entered the results of at least sixteen samples into your "Outlier Record".
30. Caltest will incorporate your new data into the database and expand your calibration after a sufficient number of outliers have been processed through this procedure (usually 10% of your current data base).
31. Refer to "The Proactive Use of Global H" included in the "Notes & Information" section of your documentation package for additional helpful information.
To receive copies of the work sheets described in this document, contact Caltest and request the "Outlier Worksheet" by Email. There is no charge for this service.
Copyright Caltest, LLC 1999 All rights reserved.