I work for a large agricultural company, in my part of it we sell fertilizer, chemical, and agronomic services. As part of this, we end up putting out a lot of trials so we can actually say something true instead of “buy our stuff it’s great I promise ;)”

One of my favorite slides is when we compiled dozens of trials on something that’s basically a nitrogen fertilizer, which as much of a guaranteed positive effect as you can get in agriculture. When compared in a graph most of the trials show an overwhelming effect on increasing yield over an untreated check, however there’s always a portion of the trials where the yield decreases compared to the (untreated) check.

Real life is extremely noisy for a multitude of circumstantial reasons that are either not practical or possible to control for, so a single trial is generally worth fuckall. It takes a lot of testing to see a consistent trend across them.