U of I Professor Details How To Conduct an On-Farm Strip Trial

Minimizing variability is key to obtaining accurate results in on-farm trials.

Published on: May 5, 2014

By Emerson Nafziger

This article is reprinted from the University of Illinois' The Bulletin. For the original version, click here.

On April 16 I posted an article on the use of nitrogen fertilizer on soybean, mentioning at the end of the article that this would be a good thing to test in on-farm trials. Here I'll provide a little background and a brief description of how to go about doing such a trial.

If there were no variability in yields (or soils) going across a field, a single strip of N fertilizer (or without N while applying N to the rest of the field) would measure the effect. There is no such field: every strip will yield at least a little more or less than every other strip in every field.

U of I Professor Details How To Conduct an On-Farm Strip Trial
U of I Professor Details How To Conduct an On-Farm Strip Trial

So a single strip where we might plan to apply N fertilizer will always yield differently than the strip next to it, and we can never be sure if any "difference" we see in yield was due to treatment or just to this "random" variability, with one or another treatment just "lucky" in getting assigned by chance to higher-yielding strips.

One way we deal with this variability is to locate trials in the most uniform part of the field we can find. This can never eliminate variability between strips, but it minimizes it. Using a yield map from a previous year can help find areas in the field where there is good uniformity.

If a map of a given field doesn't show an area large enough for a trial with good uniformity, consider moving the trial to a different field. If strips need to be shorter than the field in order to avoid a variable area, it's OK to do that, as long as good yield measurements are possible.

The only way we can measure and deal with the strip-to-strip variability in a field is to replicate treatments – that is, to apply the same treatments to a number of different strips. The variability among yields of strips treated the same tells us how variable yields are, and we then use stats to help us figure out if yield differences between treated and untreated strips are probably due to treatment or could easily be due to random variability and chance placement of treatments.

If we can't be sure that a difference is due to treatment and not just to random variability, then we have to conclude that we haven't proven that the difference is due to treatment, and we say the difference is "not significant." This doesn't mean the treatment did nothing, it just means that we haven't proven to our satisfaction (that is, to a probability level of 90 or 95%) that it did.

By the same token, a treatment that does nothing at all can get "lucky" and end up on higher-yielding strips by chance, producing a "significant" difference that was really no difference at all.

Running a strip trial
Laying out a trial like one to test N fertilizer on soybean is reasonably simple:

Find a uniform part of a field large enough to accommodate 12 strips wide enough to apply N fertilizer to, and large enough to get accurate harvest yields with the combine. Unless N can be dropped very precisely to strips exactly as wide as the combine will harvest, N strips will need to be wider than the combine.

They should be long enough to get a good yield estimate, whether that's with a yield monitor or a weigh wagon. Record GPS coordinates for the 4 corners of the area, soil type, previous crop and its yield, planting date, seeding rate, variety, herbicides, application date, harvest date, and anything else that you think might have affected the crop.

Assign treatments to strips randomly within each pair of strips. Here is how this might look:
Strip 1  No N
Strip 2  +N
Strip 3  +N
Strip 4  No N
Strip 5  +N
Strip 6  No N
Strip 7  No N
Strip 8  +N
Strip 9  No N
Strip 10 +N
Strip 11 +N
Strip 12 No N

Apply in-season N to the strips where it was assigned. Timing and form are not fixed, but I suggest using 50 to 80 lb of actual N (120 to 200 lb urea 45-0-0 or similar; protecting with a urease inhibitor and using slow-release urea in the mix are good options) applied between stage R4 (full flower) and R6 (full seed). Liquid (UAN) can be injected, but dry forms are most common and easiest to apply in plants this size. If you drive down strips to apply by ground, and there's a possibility that this will do some damage, you'll want to drive (with applicator off) down the "No N" strips as well so all strips experience the same thing. N can be applied by air, but strips will need to be wide enough so fertilizer doesn't fall into no-N strips. Make certain, either with GPS or with flags (PVC lengths installed using a soil probe work well and are visible), that you know exactly where the N went on and where it didn't, so you can harvest correctly.

Harvest and record yields for each strip. Be sure that the width harvested is the same for each strip, and trim the ends if using yield monitor data.

Average yields with and without N to see if there are differences. Do a statistical analysis, or work with an adviser to do so. If you wish, you can send me yields on a spreadsheet and I'll return it with stats run and a note about the outcome. If you do your own analysis, I'd much appreciate getting the data so we can look at this across all sites.

Doing such trials in different soil types – especially in soils with different textures and amounts of organic matter – will help show whether or not an N response is affected by soil. Some may want to do trials in more than one field, or to cooperate with neighbors to do this. For questions like this, the more data, the better.

Nafziger is a crop sciences professor with U of I.