|
Rolla, ND | With Fisher's least significan difference, usually abbreviated, LSD. Any results that do not vary by more than the number are not considered different. So if the LSD is 15.7 the two varieties need to be different by 15.7 bu to be considered to be different in yield.
There should be a probability stated as well. If the two are different at the 1% confidence level it's pretty solid. I think a lot of university results use 5%. But if you wanted you could use say 50%.
You really don't know what you are looking at if it isn't stated.
I think a lot of people input the data into a program and run it today and don't even know what confidence level they have set.
Nonreplicated trials never bother listing an LSD because they never have any significant results. Which is somewhat interesting when you consider how many people use them to decide variety.
It's not about data, it's about sales.
There are some other statistical methods to tell is a difference is a real difference or if they are within natural variability that are probably better than LSD, but in general they are understood even less and I don't think SAS includes them in their basic academic software.
I trust the test plots with a check planted repeatedly throughout and that used to determine variability and such. I'm not sure I like the corrected plots where if your plot is next to a check plot that they adjust yours based on the closest check so you can go up or down quite a few bushels. | |
|