Testing More Than One Fertilizer

Opening Scene

Article 6 warned against changing several things at once in a single test. But sometimes there’s a genuine, well-founded interest in several factors together — a new fertilizer, a new watering schedule, and whether they interact with each other in ways neither would show up alone. Testing them one at a time, sequentially, across separate seasons, would take years. A more sophisticated trial design can test all the combinations at once, in a single season, while still statistically isolating each factor’s individual and combined effect.

In Plain English

Multivariate testing (or factorial experimentation) tests multiple variables simultaneously by creating every combination of their values as separate treatment groups — two fertilizers times two watering schedules yields four distinct groups — allowing an experimenter to measure not just each factor’s individual effect, but also whether they interact, meaning their combined effect differs from what you’d predict by simply adding their individual effects together.

The Old Way

Before formal factorial experiment design existed, testing multiple factors together was either avoided entirely or done without rigor:

Testing factors sequentially across separate trials, taking far longer and never directly measuring how they interact with each other.
Testing several factors bundled together in a single treatment, exactly the confounded design Article 6 warned against, losing the ability to isolate any individual factor’s contribution.
Guessing at interactions based on intuition, without a rigorous method for actually measuring whether two factors amplify or dampen each other’s effects.

Factorial design specifically solves the problem of testing multiple factors together without sacrificing the ability to isolate each one.

What’s Changing (and Why AI Is the Reason)

Digital experimentation platforms can now run full factorial designs automatically, managing the increased complexity of multiple simultaneous treatment groups without the manual overhead this once required.
As products grow more complex, with many interacting features and settings, understanding genuine interaction effects — not just individual factors in isolation — has become increasingly important to getting the full picture right.
Multivariate testing requires substantially larger sample sizes than a simple two-group test, connecting directly to the power considerations from Article 7, making it a deliberate tradeoff rather than a free upgrade over simple A/B testing.

The Metaphor, Fully Extended

The Field Trial	Multivariate Testing Concept
Testing two fertilizers and two watering schedules together	Testing multiple variables simultaneously
Four distinct plot combinations, covering every pairing	Four distinct treatment groups, covering every factor combination
Discovering the fertilizer works better with one watering schedule than the other	Discovering an interaction effect between two variables
A single season’s trial replacing years of sequential single-factor tests	A single, larger experiment replacing many smaller sequential tests

For Beginners: What to Actually Do

Learn the basic structure of a 2x2 factorial design before attempting anything more complex.
Practice distinguishing a main effect (one factor’s individual contribution) from an interaction effect (how two factors combine).
Recognize that multivariate testing needs meaningfully more sample size than a simple A/B test, and plan accordingly using Article 7’s power calculations.

For Practitioners and Leaders: The Deeper Layer

Reserve multivariate testing for cases where a genuine interaction is plausible and worth the added sample size cost, rather than defaulting to it for convenience.
Invest in experimentation platform capability that can manage factorial designs automatically, given their added operational complexity.
Use multivariate results to inform not just which combination wins, but which individual factors and interactions genuinely matter, for future product decisions.

Quick Recap

Multivariate testing tests several variables simultaneously by creating every combination as a separate treatment group.
It measures both each factor’s individual effect and any interaction between factors.
It requires meaningfully larger sample sizes than a simple two-group A/B test.
It’s most valuable specifically when a genuine interaction between factors is plausible and worth investigating.

Where This Fits in the Series

Article 12 covered testing several factors and their interactions together. Article 13 covers a different, subtler threat to a trial’s validity: effects that spill over between plots.