First, duplicate settings
In the field experiment, the number of plots planted in the same treatment is repeated. If a piece of land is planted every time, it is called repetition; If you plant two plots at a time, it is called secondary repetition. Each unit is called a test unit.
The main function of repetition is to estimate the test error. The test error exists objectively, and we can only estimate it by dealing with the differences between different test units at the same time. If there is only one test unit for each treatment (that is, there is only one observation value for the same treatment), then the difference within the same treatment cannot be expressed and obtained, and the error cannot be estimated.
Another main function of setting repetition is to reduce the test error, thus improving the accuracy of the test. It has been proved that the test error is inversely proportional to the number of repetitions, so the more repetitions, the smaller the test error. The experimental error of four repetitions is only half of that of the same kind of experiment once repeated. But too much repetition is difficult to manage. Therefore, the number of repetitions of different experiments should be determined according to specific conditions.
Second, random arrangement
Random arrangement (random
Classification) means that without any subjective bias, each treatment in each repetition has an equal opportunity to be arranged in any experimental plot. Setting repetition can provide conditions for error estimation, but in order to obtain unbiased test error estimation, each process needs to be arranged randomly. Therefore, the combination of random arrangement and repeated setting can provide unbiased test error estimation.
Random arrangement can be made by drawing lots, computer-generated random numbers or using a random number table (table 1). The usage of the random number table will be introduced later.
Third, local control.
In the field experiment, it is difficult to control all non-treatment factors in a balanced way. But we can do local control (local
Control), that is, divide the whole test environment into several relatively consistent small environments (called block or repetition), and then set up a complete set of processing in each small ring. Non-processing factors such as test environment conditions in a small range are easy to control. Therefore, non-treatment factors such as soil differences can be controlled by regions and sections in the whole experimental environment. If the soil difference in the experimental area is obvious, it is best to block according to the fertility gradient to make the blocks relatively uniform, and then block according to the number of treatments. In this way, the only thing that can affect the test error is the subtle soil difference in a small plot, which has nothing to do with the increased soil difference caused by the repeated expansion of the test area. This arrangement is the principle of "local control" in field experiments.
To sum up, a good experimental design must be arranged reasonably and carefully according to the three basic principles of repeated setting, random arrangement and local control. Only in this way can we get the minimum test error in the experiment, and obtain the real processing effect and unbiased test error estimation.