Oswald No-Till Farm Cleghorn, IA | I just spent 20 minutes tapping the statistical elements only to have my iPad lock up as it often does through Safari on Ag Talk wiping out my work.
Here's the short version.
First a person assumes no difference which is called the null hypothesis. To negate the null hypothesis, you essentially look at how much overlap to the curve distribution of the check vs. treatment data.
When you have wide variance in things treated the same, you likely have substantial overlap of those distributions (curves) making it difficult to be confident it isn't some random result.
If your things treated the same have tight (tall narrow) curves it gets easier. For example, if the things treated the same vary +\- 5 bu/ac and the treatment means are 10 bu/ac different, then you likely have some confidence there is a difference.
This doesn't say that 10 bu/ac IS the difference but rather the null hypothesis iis likely negated. Too often some people assume the LSD IS the difference.
There are lots of ways to take the same set of numbers, pair them differently, and generate a different statistical result even with the same means and difference in the means.
Very seldom in research and statistics will you hear absolute language. Too often people presenting data like to imply much more.
Simply stated, look at the things treated the same. If they vary a lot, it takes a big difference to defend difference.
Enough statistical knowledge to be dangerous.
Tom |
1st 2014 TrackTill Plot Data
