Okay, this really is sad. And from my old alma mater.

The high-tech planter purports to add 9 bu/ac over the low-tech planter yield. But the reported LSD is 21 bu/ac, meaning they found no significant difference.

Case closed? Well, not so fast!

First, I really dislike tests that use Least Significant Difference. They rely on a pooled estimate for variance -- which may be valid, or may not. You see this especially where there aren't many measurements and a properly done t-test would report wide confidence intervals. Well, there's a reason why t-tests will do that...when you don't have enough data! Fisher's LSD test was one of the first for this kind of analysis, but is no longer recommended. Curiously, I see it only in ag data, from universities and seed companies.

Next, looking at the fine print, you will note that they used an "unprotected" LSD test. A "protected" LSD test stops when the analysis of variance (ANOVA) fails to pass the significance test, which is almost always set to a p value of 0.05. Anything above that and the null hypothesis cannot be rejected and the common conclusion is that there is no difference between the sample means. An unprotected LSD sallies forth regardless of a high p value. There is no correction for multiple comparisons and no recognition of a failed ANOVA.

Finally, look at their p value: 0.10. Okay, I have used that when brainstorming my own farm data...but I would never post results for anything over p = 0.05, let alone put it on a university Web site. It's as if Ohio State did everything it could to find a difference between the two planters...and still failed.

Now, I'm just a simple farmer, not a statistician, not a PhD. But I can recognize sloppy work when I see it. This little study should never have seen the light of day.