Do you have a nagging agronomy question you want answered? Follow these steps to set up an on-farm trial and put that question to the test.
Adam Gurr wanted to know if he could cut his canola-seeding rate and not sacrifice yield or extend maturity. So the Wagga Wagga farmer set up a three-year trial to answer the question.
Gurr compared his usual seeding rate to one that was 20 per cent less and one that was 50 per cent less. He ran the trial on four different sites — two in year one and a single site in each of the next two years. These sites included four different fields and three different soil types.
“We concluded that indeed we could cut our rates by 20 to 50 per cent depending on TKW (thousand kernel weight), and in the process we developed a different way of determining our seeding rate,” he says. “We now take our TKW in grams and subtract 20 to 30 per cent to come up with a seeding rate in kilos per acre.”
“In the trial, we also noticed that our seed mortality rates are really low,” Gurr says. “We are confident that our seed survival will be 80 per cent or more, and ultimately that low mortality rate allows us to cut our seeding rates.”
Gurr’s whole canola-seeding operation is based on achieving a final stand of six to eight plants per square foot. “After our on-farm trial, now we target seven to eight plants per square foot for our seeding rate as we are confident the final plant stand will meet our target”
This is one example of the value a trial-driven decision can provide. The key is to create an effective trial and keep the results in perspective. Confidence in on-farm trials depends on the following 10 steps.
10 steps to useful on-farm trials
These few steps will take on-farm experiments to another level, making them fair to all treatments and putting results in proper perspective.
1. Commit to finishing the trials. If the trial is worth doing and will provide information that will benefit the farm’s profitability, then set a goal to take it to harvest.
2. Keep it simple. It doesn’t have to be complicated. Start with a question you want answered, such as: Will 50 per cent more nitrogen increase my canola yield AND profitability? Does a second in-crop herbicide application pay? Does boron applied at flowering reduce flower abortion and increase overall yield? One variable makes for an easier trial setup and more straightforward statistical analysis. Simplicity also makes it easier to accomplish Step 1.
3. Avoid “confounding.” The only difference between the two strips should be the treatment in question — otherwise the result will be confounded. An example of a confounded study would be comparing the effect of boron added to fungicide versus a check strip with neither one. In that situation, you can’t tell if the effect is due to fungicide or boron. If you want to test boron at flowering, use fungicide in both treatments. If you want to test fungicide, leave boron out or have it in both treatments.
4. Arrange the strips fairly. Place two strips beside each other in an area of the field where they cover similar slopes and soil quality. Multi-year yield maps can help identify good locations. Have at least 500 feet of length per strip and make them wide enough for a windrow to fit well within the boundaries.
5. Replicate the strips. More strips mean more data points, which will increase statistical relevance of the result. Include three or four treatment/no-treatment pairs through the trial area. Also, if treatment strips are 100 feet wide, this allows for two windrows within each strip. Each windrow could be combined and weighed separately, providing more information for statistical analysis.
6. Weigh the strips. An accurate scale increases confidence in the results. A weigh-bin is ideal, but keep it parked in one level location and don’t move it between strips.
Yield monitors can be accurate if calibrated against the specific crop condition. The risk here is that if treatments show clear differences, the combine yield monitor should be calibrated between each strip. For this reason, a weigh-bin remains the simpler and more accurate option. Ask your DuPont Pioneer Area Manager if you can borrow his weigh-bin.
7. Take time to make in-season observations. Keep a list of all field and treatment information, including rainfall. Try to walk the full length of plots a few times during the growing season to check for differences in drown-out spots, plant densities and calendar dates for bolting and first flower, for example. Identify and record conditions that may have influenced the final result. If the treatment is going on after these differences were already observed, it may be difficult to determine whether any yield difference between strips was the result of the treatment or the pre-existing conditions.”
8. Go beyond averages. In addition to calculating the average yields of two strip treatments over many sites, estimate the probability that the difference is due to chance. (see below)
9. Do statistical analysis of the results. This allows you to sort out whether the differences in yield are a result of the treatment applied or simply a result of natural variability. The danger without statistical analysis is that you make changes to your system based on differences that are not ‘real’.
10. Share your results. Collaborate with other growers to learn from others and share your results. That way, you don’t have to make all the discoveries and mistakes yourself. Communicate results with research and extension personnel. Sharing results from the same trial repeated on other fields with a mix of soil types and weather conditions provides a result that you can more confidently extrapolate to most situations.
A quick statistics test
Paired T Test: Results can be plugged into the Paired T-test calculator (You can find the calculator here) that shows mean, probability and LSD results.
Mean: This is the average of the yields for each treatment.
Probability of this result: Measures whether your result could have occurred by chance alone. The lower the better. When analysing field data, we generally have more confidence in a result when the probability is five per cent or lower.
The Indian Head Agricultural Research Foundation (IHARF) data analysis tool, provides an instant analysis of least significant difference (LSD).
LSDs allow data to be eyeballed, without having training in statistics first. LSD is a simple calculation that allows the means of two or more pre-determined varieties to be compared. At a glance, the probability that the difference between the means is the result of chance can be evaluated and confidence is gained that the inference(s) drawn from the data are correct.
As the IHARF tool says: “A p-value of 0.05 is chosen most frequently in scientific experiments. However, 0.10 is sometimes recommended for large-scale field trials to account for increased overall variability relative to small-plot studies. At p=0.05, there is a five per cent probability of either (1) concluding there is a difference between two treatments when, in actuality, there is no difference, or (2) concluding there is no difference between the two treatments when a difference actually existed. At p=0.10, the probability that one of the previous two errors will occur is 10 per cent.
It is for this reason that Pioneer has a minimum of 10 sites for any published data to ensure the data published is statistically relevant.