First, sorry this community has been kind of dead. I've been pretty preoccupied with work, and blowing of steam shitposting memes

One thing we often encounter in reclamation, if you're a consultant, is clients wanting research done to confirm their methods are feasible and will work when it comes to close a site. As such, they often want to look at all the things, and get the most data for the cheapest price. This leads them to wanting to have a bunch of treatments, often using a factorial design (i.e. split-split plots), where you'll have 20 plot (for example) and each plot has 2 or more treatments within it.

The problem with this is that by nature, you're limited to a low number of reps, since adding one extra rep can significantly impact the amount of money spent on analysis. The thing, though, is that reps serve to smooth out highly variable data (like soil!), and by having a bunch of treatments all smushed together, you get a lot of confoundment going on in your data sets. Further, even when you're militant about controlling variability, you essentially answer many questions poorly and end up needing to do more research to answer them all. You get a 'well kinda' answer.

Alternatively, if you design your experiment in stages, you can better answer questions, and can have the flexibility to adjust in between experimental phases.

For instance, say I want to look at the effect of two subsoil decompaction methods, two amendments, and two planting prescriptions. You could design this easily with a factorial approach, and get data all at once.

Alternatively, if you look at one of each type of treatment (e.g., 1 decompaction method, 1 soil amendment) and 2 planting prescriptions with more reps, you'll have stronger statistical power, and be able to answer questions better. It's more defensible. It's usable data. The kind that gets you another budget to find out the other half of the experiment. In round 2 you look at the other configuration.

Ex//

Round 1
Decompact A x Amend A x plant A
Decompact A x Amend A x plant B
Decompact A x Amend B X Plant A
Decompact A x Amend B x Plant B

Round 2
Decompact B x Amend B x plant A
Decompact B x Amend B x plant B
Decompact B x Amend A X Plant A
Decompact B x Amend A x Plant B

In a factorial, you'd have something like:

Decompact A x Amend A (half plot) X Amend B (half plot) X Plant A (half Plot) x Plant (B) (half Plot) in this case, there's generally too much spatial overlap/noise.

While this approach is a little more expensive in the long run, it's generally cheaper in the short term, and more palatable to clients, particularly when you get solid answers rather than non-answers.

This applies to all field trials, not just Reclamation. Simple experimental designs are elegant. Think of it as a field of vision. If you use a factorial you have a broad field but narrow depth. More elegant approaches? More depth less field. Each has their merits, but reserve factorials for occasions where you aren't sure what is important or aren't trying to prove something

E: some minor spelling mistakes that my phone didn't catch. tweaked design to include overlapping treatments my 11 pm brain didn't catch. Principles remain the same.