Hay growers and buyers can avoid disputes over forage quality by having one lab test three eight-core samples from the same stack. At least that's the hope of Dan Undersander, the National Forage Testing Association and the National Hay Association.
The two organizations have endorsed Undersander's idea of a replicated sample program that shows forage analysis results as estimates and not absolutes.
“It's important for us to remember that, when we get an analysis back, we can't assume that it's the absolute God's truth,” says the University of Wisconsin forage specialist. “It is just an estimate and there is a certain error curve around that estimate.
“I hear about hay dealers who sell lots of hay and somebody comes back and says, ‘Well, I got five fewer RFV points and I want a price adjustment.’ Or, if they're from California, they'll say, “I got a tenth of a point less TDN and I want a price adjustment.' That's not realistic.”
Usually, a grower or seller sends a lab one 20-core sample per stack or load. The buyer then retests the hay and his results may vary from the seller's — hence possible buyer-seller disputes.
By sending labs multiple samples from the same stack, he says, and getting the average of the test results (along with a standard deviation that indicates variation in sampling or lot), growers and buyers get an estimate of reliability. If a buyer sends his own three samples to another lab, it's more likely his numbers will fit in the range of a three-sample test than one 20-core test.
An additional reason Undersander advocates three samples is the concern that labs are not grinding entire 20-core samples because of sample size and the amount of time it takes, he says.
The three replicated, eight-core samples are essentially taking the place of a 20-core sample, says Aron Quist, president of Stanworth Crop Consultants, a forage-testing lab in Blythe, CA.
“Let's say you took a normal (20-core) sample and it came out at 20% protein. If you tested three smaller samples, you might have an 18%, a 20% and a 22% protein result. The average of those three cores was 20%, but the variability between those three samples would have been 18% and 22%.”
Replicated sampling helps hay buyers and sellers realize there is variation in analysis that cannot be avoided, says Dave Mertens, dairy scientist with the U.S. Dairy Forage Research Center at Madison, WI.
Variation in a stack is caused by differences in the crop from one part of a field to another. Because of variation, 20-core samples have been recommended to provide good estimates of a stack's average analysis.
But sellers and buyers tend to assume a 20-core sample's result is the composition of the hay, says Mertens, also a driving force behind NFTA's lab certification program that works to improve among-lab repeatability.
So when a buyer's analysis differs from a seller's, labs may be blamed. Yet there are differences among labs just as there are differences among samples, says Mertens.
“One of the things Dan's idea would do is let everyone know that they're not going to get identical numbers (with forage analysis),” Mertens says.
But smaller samples are more variable because they're a smaller representation of a hay stack, he says.
“If you wanted to minimize a problem with a client coming up with different numbers because he used a different lab, take three 20-core samples, send them to three labs, average those numbers and know the variation,” he says. The results would contain variation from sampling and labs, but the average and range would be the best estimate of what to expect from repeated samples sent to different labs.
Quist's lab will offer discounts on three eight-core samples. They shouldn't be “split samples;” each eight-core sample should be taken from various places in a stack and put into one of three bags. Those bags should then be placed in another bag to show they're from the same stack.
Other NFTA-certified labs also support offering discounts on replicated samples, Quist says.
Undersander will monitor standard deviations. Expected ranges of standard deviations for a three-sample analysis are: Dry matter (for hay), 0.5-1.5%; crude protein, 1.2-1.8%; NDF, 1.5-2.5%; NDF digestibility, 1.5-2.2%; ADF, 0.6-1.2%; Ash, 0.2-0.5% and RFQ, 10-14 points.
Each forage analysis is an individual estimate of the true average of a forage lot, and every one will be slightly different, says Dave Mertens, dairy scientist with the U.S. Dairy Forage Research Center.
But Mertens has summarized data on hundreds of forage analyses from NFTA-certified labs to show the amount of variation or standard error in forage tests for NDF. He measured two sources of variation: one on how samples are taken — the other on the differences among A, B and C labs doing single or duplicate analyses. (To determine if your lab is NFTA-certified and has a grade, visit foragetesting.org. If it does, call the lab and ask for its grade — A, B or C.)
NFTA's certification program determines how close a lab's results on six samples are to a reference average value. Grades measure how variable results are: An A lab has a standard error of plus or minus (±) 0.6% NDF; a B lab, ±1.5%; and a C lab, ±2.5%. “Statisticians call it ‘error’ but it is not a mistake, just a measurement of variation,” he says.
“If an A lab reports an NDF analysis of 40%, the standard error tells that, about 66% of the time, a repeated analysis by A labs would fall between 39.4% and 40.6%. Two standard errors encompass 95% of the results (38.8% and 41.2%).”
In Mertens' table, variations should be squared, then the square root taken to combine sources of variation.
Say a grower submitted a five-core hay sample (±2.15%) to an A lab (0±.6%). The variation expected is the square root of 2.15 × 2.15 + 0.6 ×0.6 = 4.98, which is ±2.23% NDF. If five-core samples were submitted to 20 labs, results of 19 (95%) would fall between two standard errors of the average. If the average was 50% NDF, then 95% of results would be between 45.5% and 54.5%.
If a 20-core sample (±1.07%) of the same hay was submitted to a B lab (±1.50%), total variation in results would be 1.07 ×1.07 + 1.50 ×1.50 = 3.39, or ±1.84% NDF. Nineteen of 20 results should fall between 46.3% and 53.7% NDF.
“These results demonstrate how unrealistic it is to assume that two analytical results will agree with one another exactly, and illustrate the real variation you should expect between analytical results,” he says.
|Lab Analysis||Standard Error (%)|
|Average A Lab Single Analysis||0.60|
|Average B Lab Single Analysis||1.50|
|Average C Lab Single Analysis||2.50|
|Hay Sampling||Standard Error (%)|
|Cores from 5 bales||2.15|
|Cores from 10 bales||1.52|
|Cores from 15 bales||1.24|
|Cores from 20 bales||1.07|
|5 grab samples from chopped hay||3.00|
|10 grab samples from chopped hay||2.12|
|15 grab samples from chopped hay||1.73|
|20 grab samples from chopped hay||1.50|
|Silage Sampling||Standard Error (%)|
|Samples from 5 harvest loads||2.46|
|Samples from 10 harvest loads||1.74|
|Samples from 15 harvest loads||1.42|
|Samples from 20 harvest loads||1.23|
|5 daily or location grab samples||2.37|
|10 daily or location grab samples||1.68|
|15 daily or location grab samples||1.37|
|20 daily or location grab samples||1.19|