InsightFab
Knowledge Base/Defining Confounding Structures: Choosing Resolution III, IV, and V
DOE6 min read

Defining Confounding Structures: Choosing Resolution III, IV, and V

This article addresses a common challenge in experimental design: an insufficient understanding of experimental 'Resolution'. It clearly explains the distinctions between Resolution III, IV, and V, and how each affects an experiment's ability to discriminate between factor effects and interactions. The text aims to help readers understand why their experimental outcomes might align with lower resolutions and how to prevent critical interactions from being obscured by confounding.

The Day the Yield Report Came Out, My Boss Just Said: 'Are You Sure You Understand Your Confounding Definitions?'

I remember once, a new product was about to launch, and my team and I had spent months conducting DOE. But once it went to production, the yield just wouldn't go up. The report reached my boss; he glanced at it, his face turned livid, and he just threw out a line: 'Are you sure you understand your confounding definitions?' My heart sank. I wondered if we had missed a parameter. But he immediately pointed out: 'With your Resolution V design, did you really consider those higher-order interactions? The results look like Resolution III.' To be honest, my mind was a bit muddled at that moment. Resolution III, IV, V – I knew what they were in theory, but applying them immediately to a practical problem was truly a bit challenging.

Where's the Problem? Frankly, It's Your Understanding of 'Experimental Capability'

So what exactly is Resolution III, IV, or V for confounding structures? Frankly, it's about how clearly your experimental design can 'see' or 'separate' different factor effects. Imagine you're singing in a very noisy KTV; Resolution III is like barely hearing your own lead vocal, while the background harmonies, drums, and guitars are all mixed together, making it impossible to distinguish who's singing off-key. Resolution V? That's like being in a professional recording studio, where the sound of each instrument is crystal clear, and you can focus on whichever one you want.

The key point is that an experimental design with Resolution III will confound some important Main Effects with 2-Factor Interactions, making it difficult to determine whether factor A is effective, or if it's the interaction between factor A and factor B that is effective. Resolution IV is slightly better; it can separate Main Effects from 2-Factor Interactions, but 2-Factor Interactions will still be confounded with each other. Resolution V is the best, as it can separate both Main Effects and 2-Factor Interactions, allowing you to clearly understand the influence of each factor or factor combination.

How to Apply It in Practice? It Depends on Your Level of 'Compromise'

So, in practice, how should you choose? It's simple: it depends on your level of 'compromise'.

  1. Resolution III (R=III): When your resources are limited, or you have an overwhelming number of process parameters and you just want to 'quickly screen' for a few critical factors, R=III is an option. For example, if you have 15 parameters potentially affecting yield, but you only want to quickly explore with an 8-factor, 16-run DOE, it will almost certainly be R=III. The downside of this design is that after running it, you might find factor A seems to have an effect, but it might actually be the interaction between A and B at play, and you simply cannot distinguish them. This often occurs in the early exploration phase, at the stage where Cpk is only 0.8.
  2. Resolution IV (R=IV): This is the most commonly used balance point in the industry. It allows you to separate Main Effects from 2-Factor Interactions. That is, you know whether factor A itself is effective, and whether the interaction between A and B is effective, but you cannot distinguish between an AB interaction and a CD interaction. This design typically requires more experimental runs but yields more reliable conclusions. When your product's Cpk can already reach around 1.08, and you want to improve it further, R=IV would be a good choice.
  3. Resolution V (R=V): If your goal is to optimize the process to its absolute best, achieving a Cpk above 1.33 and DPMO below 6210, and you need to precisely identify all potential interactions, then R=V is your top choice. It can separate both Main Effects and 2-Factor Interactions, ensuring they do not confound with each other. Naturally, the trade-off is the highest number of experimental runs and the highest resource requirements.

So the key is: what is your current product's pain point? What level of answer do you expect to get from the DOE? Only then can you decide how much information you are willing to 'compromise'.

The Most Common Pitfall: Your Boss's Expectations Differ from Your Work

The pitfall I most often fell into was when my boss asked me to 'identify all effects,' and to save time, I haphazardly ran an R=III design. The analysis showed that a bunch of factor effects were confounded with interactions, making it impossible to tell who was who. Eventually, my boss looked at the report and directly asked me: 'So are you certain this is the effect of A, and not the interaction between A and B?' I was stunned on the spot. Because the data told me they were confounded, I simply couldn't give him a definitive answer. At times like these, I could only grudgingly accept it, redo the DOE, and waste several weeks for nothing.

Frankly, many times we want to cut corners and think 'good enough' is fine. But DOE Resolution is like photography; if you want to capture every detail clearly, you need a high-megapixel camera. If you use a phone to photograph the starry sky, no matter how much you adjust the lighting, it's hard to achieve the effect of a professional DSLR.

One Thing You Can Do Today

Next time you design a DOE, first ask yourself: what is the core purpose of this experiment? To what extent do I need to distinguish between effects?

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools

More in this category

DOE: Not Just Running Experiments, But Hitting the Most Targets with the Fewest Bullets

A full factorial experiment with four factors and three levels each requires 81 runs. Using a Taguchi L9 orthogonal array, only 9 runs are needed, saving 89%, while simultaneously estimating the main effects of all four factors without mutual contamination.

DOE Taguchi Method: Finding Optimal Parameters in 9 Experiments, Not by Luck

A full factorial design for 4 factors at 3 levels requires 81 runs, while a one-factor-at-a-time approach, though shorter at 36 runs, misses crucial factor interactions. The Taguchi L9 method offers a highly efficient alternative, completing the analysis in just 9 experiments and achieving an 89% reduction in experimental effort.

Response Surface Method: Finding the Optimal Process Sweet Spot

After DOE identifies the significant factors, RSM (Response Surface Methodology) helps pinpoint the precise parameter combination that optimizes the outcome. This marks a crucial transition from understanding 'which factors are important' to determining 'what settings are best'.