Phil Williams
Canadian Grain Commission
Grain Research Laboratory

The change will improve to farmers (in particular) the probity of receiving protein premiums. Under the previous (0.5 % increment) system, many farmers were disappointed to learn that the protein content of their wheat was 13.4, or 12.9 %, thereby missing out on the higher premiums.

But the new system will focus renewed attention onto the accuracy of protein-testing. Practically all CWRS wheat in western Canada is tested for protein content by a Near-InfraRed (NIR) instrument, either at country delivery points, or at terminal elevators in Vancouver or Thunder Bay. The reference method used by the Canadian Grain Commission (CGC) for calibration and monitoring is the Dumas method, or Combustion Analysis. The sample is burned in oxygen, and the protein nitrogen measured directly.

Table 1 gives typical results of testing a sample of wheat with different degrees of error. This set of data was compiled to provide a mean protein content of 13.6 % (13.5 % moisture basis), which is the long-term average protein content of western Canadian CWRS wheat. Notice that at an error of 0.30 %, the range in protein content (of 12 samples) was over 1.0 %. These ranges in individual results are characteristic of the spread obtained with the errors stated.

Table 1. Typical results of protein testing with different degrees of error

There is more to this than meets the eye. Unfortunately we can’t escape mathematics! Mathematically only about 67 % of all tests on the same sample will lie between plus or minus the error from the mean. In this example in 100 tests on the same sample, an error of 0.15 would place 67 % of the results between 13.6 + 0.15 (13.75) and 13.6 - 0.15 (13.45), which rounds off to a range of 13.8 to 13.4 %.

It also means that the other 33 % of the results will be outside this range. About 95 % of all results would be within twice the error from the mean, and 98 % of the results would fall within 3 X the error from the mean. Table 2 gives the range you can expect for the results of testing the same sample 100 times with the five levels of error in Table 1.

Table 2. Ranges in protein results at different error levels

The "dollar value per tonne" figures are based on an arbitrary spread of about $42.00 from 11.9 to 15.0 % protein, and give the value of a positive or negative difference from the "true" protein result. These values can be expected to vary considerably, depending on the availability of protein. For example, premiums are higher in low protein years when protein is in short supply, and lower in high protein years. A grade differential between 1 CWRS and 2 CWRS usually means only about $3 - 4.00 per tonne.

These results are the "Real World" - this is what is bound to happen if a sample of wheat is tested 100 times with these different levels of error. The other 2 results (100-98) will be even further away from the mean. Unfortunately the laws of mathematics are absolute - so if 2 results of greater than +/- 0.75 from the mean do not show up in the first 100, there will be 4 such results in the next 100 tests, and so on.

Since 1981 the Grain Research Laboratory (GRL) of the CGC has provided the Elevator companies of western Canada with a free check sample service for both CWRS and CWAD wheat classes. The purpose is to help operators of company NIR instruments to determine how accurate and consistent their instruments are, relative to the reference method. About 10 Kg of CWRS or CWAD wheat are thoroughly blended, by GRL staff. Duplicate sub-samples are taken, ground on a U-D Cyclone grinder, well-blended, and tested for protein and moisture content by reference methods (the Dumas method for protein, and the 130^o C oven method for moisture contents). The results are reported on the constant moisture basis of 13.5 %. The bulk sample is divided into 300g sub-samples, which are sent out to the head offices of the western Canadian Elevator companies. Extra samples are sent to companies on request. This free service has been in place for nearly 20 years.

A report accompanies the sample, with the results of the previous month’s sample. The protein content of the present sample appears on the envelope, as a guideline to the operator. Elevator companies are coded, so that each can recognize their own results. Over 300 sets of results are submitted to the GRL every month. The average error for most of the companies is 0.150 %. In a typical month the range between highest and lowest results has been 0.52 %. This has varied between 0.3 and 0.8 %. These tests are presumably carried out by the most competent operators in each company. The error of the Dumas test, the reference test for protein content at the GRL is 0.1 (actually 0.09) %. This is very strictly monitored.

Sampling itself is responsible for a third of the total error in protein-testing. The sample has to be representative of the delivery. This can be achieved at the farm by taking a handful of wheat every few minutes during the whole time the truck is being loaded. The handfuls should be put into a pail, then when the truck is finished loading the contents of the pail must be very thoroughly mixed, and a bag-full taken as the sample to be used as a reference for protein-testing. During harvest-time this method is not practicable, since farm trucks may be loaded directly from the combine, sometimes while the combine is still moving. The sample can also be drawn from the truck after loading, by using a sampling probe at at least six locations covering the full depth of grain in the truck. These six sub-samples must then be thoroughly blended, as before.

At the elevator the elevator staff should sample the truck continuously during unloading. A good sample can be taken by diverting a handful every half-minute or so into a pail, or by probing the truck, as mentioned above. The sample should be very well mixed before carrying out the protein test. Cleaning the sample may not be sufficient to give a good blend, and the sample should be well-mixed after cleaning if it is to be used for a protein test.

Accuracy can be changed in most NIR instruments by changing the intercept of the calibration equation. So the actual Accuracy of the instrument can be set so that it reads the true (reference) result, or a higher or lower value than the true result.

Precision is a measure of the instrument performance, and can not be changed in the same way. It is affected by both the instrument itself (including the calibration), and the operator. Only once is the instrument calibrated to provide accurate results, but the precision affects the instrument performance every time it is used. It gives a measure of the ability of the instrument to produce the same result for the same sample consistently.

Accuracy and precision should be checked every day, the accuracy by comparing the results with those of certified reference values, and the precision by determination of the variability of the instrument in testing the same samples over a period of at least a few days. The monthly check sample service supplied by the CGC/GRL enables the operators to do both of these tests. The mean result of several tests gives a measure of the accuracy, and the variability from day to day gives the precision. The results of both are sent to the elevator companies each month.

Table 3. Influence of grinding and cell-loading on precision of Reference and NIR testing of wheat

The error was much lower (precision was better) when the NIR instrument was used to test the sample without re-grinding or re-loading, which showed that the instrument itself was very precise, and most of the error came from the operator. Whole-grain analyzers, such as the Tecator InfraTec and Perten Model 9100 instruments, eliminate both the grinding and cell-loading errors in NIR protein-testing, and are recommended for the best results.

Since the normal error of the test is about 0.2 %, adjustment in payment should only be made if the CGC result differs from the original result by more than 0.4 %

Question No. 1. Why does protein change in the bin?
Question No. 2. Why do protein testing results vary so much even when testing the same sample?
Question No. 3. Why doesn't the CGC grade on the basis of protein?

A future article in planned for this web-site to answer these, and other on-going questions.