On-farm Research Resources |
|
The Classic Agricultural Experiment - Randomized Complete Block Design Chris Reberg-Horton Commonly, farmers interested in a new product or production method will simply try it on a section of a field. If that section looks better at the end of the season, then he may expand the trial area next year. Is this approach of any value? Yes. Can we get more definitive answer from a different approach? Also yes. Let's imagine that we plant the same variety of corn in a field that is one hundred rows wide. Now imagine that we take yield data on the first 10 row, the second 10 row, etc. Will the yield of each 10 row section be the same? Of course not. Variations in the soil prevent them from being identical. Now pretend that we had applied a new "snake oil" to the first 10 rows and wanted to compare the results with second 10 rows. We know the yields will be different, but was the result due to the snake oil. Now imagine that we apply the snake oil to five of the plots randomly chosen. Standard practices occupy the other five. Let's average the results from the snake oil plots and compare to the average from the standard practice. Are they still different? Yes! But since we have replicated and randomized we can use statistics to test whether they are different due to the snake oil or the just the natural variation. While statistics can test these two alternatives, there is one drawback, you can never be 100% sure of your answer. The tradition in agriculture is to make sure we are 95% sure of results before recommending them to farmers. So what can we say about results that are under 95%. What if we are 90% sure or 80%. Growers can decide for themselves. They may not need as much confidence as scientists. Money may factor into the decision also. If a grower tests a treatment that only costs $2/acre, and you are 80% sure it increased yields by 5%, then he may want to at least expand the area being tested next year. How do I lay out a Randomized Complete Block Design? The first thing to consider is where you plan on putting the test. Lets say we plan on putting out six replicates or blocks. A block consists of three fertilizer options. The six blocks can be put side by side in a single field, in different fields on the same farm, or even on different farms. Choosing between these alternatives is a matter of what you hope to do with the results and how difficult it is to work in multiple locations. If you want to test these fertilizers to recommend to all the growers in your county, involving multiple farms is obviously the best choice. By testing at multiple sites, you will have more information on the effect of the fertlizer on different soil types and possibly spread with different equipment. What if you only find three growers willing to participate? That’s ok, put 2 blocks on each farm. If only one grower is interested, try to find a field that most resembles other farms in the area. Once you know where the blocks are going, you need to decide the size and shape of the plots. Long and skinny is best! If plots travel the whole length of the field, growers can use harvesting equipment on the plots easier. For hand harvested plots, this is obviously less of a concern. Now how do you orient the plots? If no slope or gradient is obvious, then it doesn't matter. However on a slope we want the plots running up and down the slope if possible. This can be a problem on severe slopes where this arrangement would cause erosion problems. If plots are oriented parallel with a slope, you may want to consider arranging plots end to end along a contour instead of side by side. Alternatively, you could settle for increasing your replication as a means of dealing with increased noise due to a non-ideal plot orientation. Ok, so the plots are laid out. Now which treatment goes where. We need to do this RANDOMLY. Random number tables are the official way to do it, but can be cumbersome. One alternative is to write the treatments on little pieces of paper, crumple them up and put them in a hat. Blindly pick a piece paper, and write that treatment on your first plot in your first block. DO NOT put that piece of paper back in the hat. Continue the process until the hat is empty. You now have one of the blocks randomized. Put all the papers back into the hat and repeat for Blocks 2, 3, etc. How do I analyze my data? To download the two spreadsheets for analyzing your data, right click on the link and click "save as" or "save link as" Instructions and graphs are included as tabs on the spreadsheet for the interpretation and visualization of your data.
|
