This article is part of our series of original articles on emerging featured topics. Please check here to see other articles in this series.
The fact that the nutrient content of feeds varies is not news to anyone involved with dairy nutrition. However, many nutritionists do not consider why two samples of the same feed may have different composition. Variation in the composition of a feed can be partitioned into three major sources:
- true or real variation
Analytical variation is determined by repeatedly analyzing subsamples from the same ground sample of a feed by a single lab. For most chemical assays (protein, fiber, etc.) conducted by a lab following good laboratory practice, this variation is small (Table 1). Lab-to-lab variation is another form of analytical variation and is greater than within-lab variation, but this can be avoided by using a single lab.
|Analytical Variation||Sampling Variation|
|Corn silage||Hay silage||Corn silage||Hay silage|
|1 nd = not determined.|
Sampling variation is measured by repeated sampling of a specific population of feed. For example, if a pen of cows was fed 1 ton of corn silage today and you took five samples from that 1-ton lot, variation among those five samples would be sampling variation (plus any analytical variation). Sampling variation is dependent on the type of feed and the care with which the sample is taken. Feeds that contain homogenous particles with respect to size and chemical composition (for example, soybean meal) usually have a small sampling error, but feeds that have heterogeneous particles can have an extremely large sampling error. Corn silage has particles comprised of stem, leaves, grain, and cob, and the nutrient composition of those plant parts varies widely. Grain has a high concentration of starch and a low concentration of fiber so that a sample with a few extra grain pieces relative to the average will have a greater starch concentration and less fiber concentration than the actual silage. If you took several samples, chances are that some samples will contain more grain pieces and some samples will contain fewer grain pieces than what the silage actually contains. Because of variation among samples in the amount of grain (or cobs or stem), the starch concentration also will vary among the samples. If only one sample was taken and it contained more grain than the real average for the silage, a nutritionist would mistakenly assume the silage was higher in starch and lower in NDF than it actually was, and the resulting diet based on that assumption would not be correct.
True or real variation is what we are trying to capture when collecting and analyzing feed samples. Composition of feeds can truly change because of changes in the plants (e.g., different hybrids or cuttings), changes in growing or harvesting conditions, or differences in manufacturing processes (e.g., differences among distilleries).
Often people assume that any change in feed composition represents a real change, and the new data are used to reformulate the diet without regard to any previous data. If the starch concentration in the corn silage you sampled this week was 25% but two weeks ago it was 30%, the diet might be reformulated to contain more corn grain to make up for the apparent reduced starch concentration in the corn silage. However, in reality, the change from 30% to 25% starch may represent a true change in the silage, or it may be caused by analytical variation, sampling variation, or some combination of all three sources of variation. If the silage really did decrease in starch concentration, reformulation is warranted and will keep the diet consistent with respect to total starch concentration. But if the silage really did not change (i.e., the difference between 30% and 25% starch was caused by sampling variation), then when the diet is reformulated, the nutritionist increases dietary variation because he or she changed the composition of the total diet by adding more corn grain.
To Minimize Sampling Variation, Take Multiple Samples
Sampling error is a major source of variation in feed composition (Table 1). It can only be calculated by repeated sampling of the same material (e.g., from a pile of silage that will be fed on a given day), but because of cost, this is rarely done in the field. Occasionally taking repeated samples from a given feed can help evaluate your sampling technique because excessively large sample-to-sample differences can indicate poor sampling technique. We found that for corn silage, sampling variation could differ by a factor of two between people taking the samples.
A practical, but not perfect, alternative to repeated sampling of the same material on a given day is repeated sampling of a feed over time. Variation among samples over time represents both sampling and real variation.
Reacting to sampling variation by unnecessarily reformulating a diet or, conversely, ignoring a real change in composition could result in diet changes that are detrimental to the cow or to profitability. For example, if the NDF concentration of a silage really did decrease but you did not reformulate the diet, cows may suffer from ruminal acidosis. On the other hand, if the NDF concentration appeared to decrease when it really did not (i.e., sampling variation) and the diet was reformulated with increased concentrations of silage, the higher fiber diet may reduce feed intake and milk yields.
The way to prevent overreacting to a supposed change in feed composition is to average lab results over time, but excessive averaging reduces sensitivity and you may miss a real change in composition. From studies we have conducted on repeated sampling of silages (both corn silage and hay crop silage) from commercial dairy farms, we determined that an average of three samples within about a two-week period usually accurately estimated the true composition of the silages (DM, CP, NDF, and starch). In practice, you should take two or three samples over a period of a few days, average the composition you get from the lab, and use that average to formulate the diet. When you obtain analytical data from a new sample of that feed, average the new value with the value you used previously to formulate the diet. The new average replaces the old average, and the process is continued. For example, if a new silo of corn silage was sampled three times over a one-week period and the samples had 42%, 38%, and 44% NDF, the average to use for ration formulation would be 41.3%. If another sample was taken three weeks later and it was 46% NDF, the new value to use in formulation would be the average of 41.3% and 46%, or 43.7%. This approach is adequate to smooth the data to avoid overreacting and still be sensitive enough to capture a real change. This running average would continue until you know or think that the feed may have truly changed. If you were feeding from a silage bag that had been marked where fields changed, you might start a new running average when you get to the mark. To avoid missing a real change with this approach, one needs to have good inventory management and records. A nutritionist should also look at all new analytical data relative to the last one or two analyses. If a nutrient or nutrients markedly changed, another sample should be taken immediately and if those results are similar to the previous, this suggests a real change, and a new running average should be started.
- Sampling variation can be substantial for silages and is often mistaken as a real change in forage composition.
- Good sampling techniques reduce sampling error but will not eliminate it. The most important sampling technique is to thoroughly mix the feed before sampling. Instead of taking a few handfuls of silage from the face of a bunker (which not only results in a poor sample but is also dangerous), have the feeder remove a few loader bucket amounts of silage from across the face, mix as much as possible, and sample. The best option would be to use a TMR mixer to mix the silage (with no other ingredients added), unload the mixed silage, and sample.
- Examine all new data in the context of the past two or three samples.
- Average the most recent analytical information with the value used previously to formulate the diet. Do not rely on data from a single sample.
- When composition appears to have markedly changed, take another sample immediately and determine whether the change in composition was a real change or a result caused by sampling variation before diet reformulation.
The Ohio State University