Stat-Ease Blog

This is part 3 of an adaption from Mark Anderson’s 2023 YouTube webinar, "Do's & Don'ts for Screening Process Factors."

Resolution IV designs that detect two-factor interactions

In our training courses, we teach screening using a 9-factor, 20-run example for an arc-welding experiment. This case deploys a Resolution IV Minimum-Run Screening design. The significant effects are shown below:

This experiment reveals a significant two-factor interaction term (AB). The half-normal plot also indicates significance for both parents of AB—factors A and B. A resolution IV design aliases main effects only with three-factor or higher order effects that rarely occur, thus it is save to conclude in this case that factors A and B do indeed create main effects. But you must remain wary about the AB term. Here’s the catch: with resolution IV designs, two-factor interactions are aliased with other two-factor interactions. As shown in Figure 2, in this case, AB is aliased with eight other two-factor interactions.

The reason AB is listed on the half-normal plot instead of the other options is simply because the list is alphabetical. Perhaps it makes sense to the process experts that AB is indeed the proper choice. Or perhaps you feel confident that because both A and B were correctly identified, the interaction between the two is the most logical choice. But it is possible to have an interaction present even when the main effect terms do not show up as significant. Anyone who has ever encountered our software’s hierarchy warning has seen this phenomenon in action.

The consequence of getting this wrong in a screening design is that you may miss factors that are indeed consequential. For example, in the above list if CH was the correct interaction rather than AB, then you would miss carrying C and H forward from the screening design.

You can use a semifold design to validate the AB alias. This will add 50% more runs but will more confidently ensure the likelihood that you’re not missing something important from your screening effort. Figure 3 shows how to do so with Stat-Ease software.

In the next menu, selecting either A or B to fold on will be best given the goal ofresolving the AB interaction. In this case, a good choice will be J+ Edge Prep at the high level), because this factor at the plus setting generated significantly higher tensile strength for the welds. Figure 4 shows these entries in the semifold dialog box.

Inspecting the resulting alias structure shown in Figure 5, the semifolded design cleanly identifies not only main effects, but all the two-factor interactions involving factor A, including the AB term of interest.

The semifold increased the original 9 factor, 20 run design to 30 runs. Another option would be to use a resolution V design from the start to ensure all two-factor interactions could be estimated clear of any troublesome aliasing, for example, the minimum-run resolution V characterization design with 46 runs. So, for screening, there is a clear efficiency advantage to using resolution IV designs, even if there is an interaction term worthy of investigating further using a semifold augmentation.

For all DOE’s, it is important to evaluate a design for power–the ability of the design to identify factors impacting responses by a selected magnitude. This is mainly driven by the number of runs. For example. If you’re running a seven-factor resolution IV screening design to look for factor effects of 1.5 standard deviations, you will need at least 19 runs to have acceptable power. The standard geometric resolution IV design has only 16 runs and the minimum run screening design has only 14 runs. Both approaches will require adding a few more runs to satisfy the power requirements.

For fractional factorial designs–especially screening designs–it is also important to evaluate the design for aliasing. The lower the resolution, the more consequential the aliasing. The augmentation approaches discussed in the post can be helpful in addressing tricky aliasing issues that could otherwise limit the success of your screening effort.

When things don't go as planned, whether you've inherited a resolution III design or uncovered a suspicious interaction term, augmentation strategies like the foldover and Semifold offer practical paths forward without starting from scratch. Yes, these repairs cost additional runs, but they cost far less than drawing the wrong conclusions and carrying the wrong factors into your next phase of experimentation. A little upfront diligence in design selection, paired with a willingness to augment when the data demands it, is the surest route to a screening effort that sets your entire experimental program up for success.

This is part 2 of an adaption from Mark Anderson’s 2023 YouTube webinar, "Do's & Don'ts for Screening Process Factors."

Salvaging a Resolution III design

The first situation is when you violate the advice in our previous blog post and run a resolution III design for screening. Recall that a key assumption for screening is that you anticipate important factors will present a statistically significant main effect. But with a resolution III design, each main effect is aliased with one or more two-factor interactions. So if there are actually interactions present in the system, some or all of the main effects identified will be incorrect. Remember that screening with a resolution IV design keeps us away from this problem by aliasing main effects only with highly unlikely three-factor interactions.

If you didn’t know better and deployed a resolution III design for screening: no worries, resolve the issue by using a foldover design. This design augmentation doubles the original run count in a way that de-aliases the main effects from two factor interactions. The second block of data reverses every factor level (plus to minus and minus to plus) from the original design.

Let’s look at an 11-factor, resolution III, 16-run design for example. The alias structure is shown in Table 1 (interactions involving three factors or more not shown).

Notice that each main effect is aliased with multiple two-factor interactions.

By augmenting this bad design with a foldover (see Figure 1), you can de-alias the main effects from the two-factor interactions.

The new runs are put in a second block. This structure removes any shift in response that may occur from when you ran the first experiment, such as an increase or decrease due to differing ambient conditions.

Table 2 shows the new, improved, alias structure.

Note that it took 32 runs to get to the point where it’s confident that the main effects are properly assessed. It would have been far better off to start with a minimum-run screening design for 11 factors, requiring only 24 runs (including 2 runs to bolster it against outliers). However, in our scenario, this would be wishful thinking, since we cannot go back in time. Consider the foldover in this case to be a design repair to achieve the resolution IV needed to safely screen factors down to a vital few for further investigation.

Foldover augmentation also works to de-alias Plackett-Burman designs. Handy!

Adapted from Mark Anderson's 2023 webinar, "Do's & Don'ts for Screening Process Factors."

Over the years working with process development engineers on scale-up and manufacturing troubleshooting, we've noticed a pattern: the factors that experts think drive their process are rarely the whole story. There are often other variables at play that nobody anticipates. The best way to uncover these is by using screening designs: broad, shallow experiments that help you uncover previously unknown factors. Done right, a well-designed screening study can transform your understanding of a process and point you directly to the vital few factors worth exploring in depth.

First, let’s make sure we understand what a screening design does in the overall arc of process optimization. Screening designs exist to help ensure we are working with the right factors in subsequent optimization studies. Figure 1 explains the overall strategy. Note that interactions – often the key to process improvement – are not identified until the subsequent step. But a good screening design can shed some light on whether or not there are interactions to further pursue.

With this strategy in view, here are the core do's and don'ts on screening designs. Consider this your field guide for avoiding the most common (and costly) mistakes.

DON'T: Include Factors You Already Know Will Affect the Process

This one surprises a lot of people, and has been the topic of heated discussions within our team. Why would you exclude a known important factor?

The answer is strategic focus and efficiency. By setting known factors aside during screening, you can concentrate on previously unknown factors: ones that might derail your process in unexpected ways. A broad and shallow two-level screening design lets you quickly identify the "vital few" from the "trivial many." In our experience, roughly 20% of factors you didn't expect to matter, matter! The known factors can be merged back in during the next phase of experimentation.

DON'T: Use Low Resolution Designs for Screening

This is our biggest pet peeve. Two types of designs fall into this trap: regular fractional factorials at Resolution III (shown as "red" designs in Stat-Ease software[MA3.1]), which alias main effects directly with two-factor interactions, and Plackett-Burman designs with even worse aliasing. That's a fatal flaw for screening, because if any factors interact (and in real processes, they often do) your main effect estimates are corrupted. You simply cannot trust what the analysis is telling you.

We're particularly troubled by how often Plackett-Burman designs get recommended for screening. Even the NIST Engineering Statistics Handbook suggests using them, while simultaneously noting that “main effects are in general heavily confounded with two-factor interactions.” To us, that's an oxymoron. If main effects are confounded with two-factor interactions, how exactly is this a screening design? You can’t screen anything out!

To illustrate the danger, we ran a simulation using the classic filtration rate dataset from Doug Montgomery's textbook Design and Analysis of Experiments. The full factorial result was clear: factors A (temperature), C (concentration), and D (stirring rate) were significant, along with strong AC and AD interactions.

When we re-ran the same underlying model through a 12-run Plackett-Burman simulation, the results were alarming. The AC and AD interactions got “smeared out” across multiple dummy factors. In particular, a fake factor E appeared significant when it was actually picking up aliased pieces of AC and AD. Meanwhile, the real main effect of D was undercut by its aliasing with one-third of AC, causing a cancellation. The result? Only factor A was correctly identified. Factors C and D were missed entirely.

DOE pioneer George Box once said that running Resolution III or PB designs are "like kicking the TV to make it work." Sometimes you're desperate enough to try it, but there’s no guarantee you’ll get a usable result.

A Case Study in What NOT to Do

One of our users, a pharmaceutical process developer, sent in his design results hoping we could help salvage them. He had seven factors (time, temperature, and related process variables) and chose a Resolution III design with seven factors in eight runs. This is known as a ‘saturated’ design—the most factors that can be crammed into a given number of runs in a regular fractional factorial. Then, apparently recognizing the power would be low, he replicated the design, giving him 16 runs total, still at Resolution III.

As Ronald Fisher put it, a statistician is more like a pathologist than a medical doctor. We can tell you what killed the patient, but we can't bring it back to life. We wish this researcher had contacted us before running the design. The 16-run Resolution IV option for seven factors was right there in the software, highlighted in yellow (indicating a design more suitable for screening) It would have given him both the power and the resolution he needed. Instead, he replicated a bad design, which is a bit like making a photocopy of a photocopy.

The power calculations for these two designs are the clincher. One replicate of eight runs gave only 50% power to detect his specified signal-to-noise ratio of 1.67. Two replicates (still Resolution III) pushed that to about 87%: good power, terrible resolution. The unreplicated Resolution IV design in 16 runs also reached about 83% power, while giving him a design that could actually distinguish main effects from interactions.

DO: Start with a Resolution IV Design

As stated above, Resolution IV is the “Goldilocks” choice for screening. Main effects are aliased only with three-factor interactions, which are rarely active. That means that any significant main effects detected are almost certainly real. While two-factor interactions in a Res IV design may be murky, you'll know to investigate these further.

In Stat-Ease software, these are the yellow designs in the Regular Two-Level design builder. For up to eight factors, these medium resolution designs work beautifully. For nine or more factors, Stat-Ease’s proprietary, optimally templated, Minimum-Run screening design provides an excellent option when the standard design alternatives get too big.

Summary: The Screening Do's and Don'ts

To recap: hold known factors aside during screening and focus on the unknowns. Known factors will be studied together with the survivors of the screening design in the next round of experimentation when characterizing two-factor interactions with high-resolution designs. Avoid low-resolution designs: the red standard ones or Plackett-Burmans. Instead, go with medium Resolution IV or minimum run screening design from the start.

All Stat-Ease software licensees have access to our DOE experts. We encourage you to contact us before making a big mistake in your design of experiments. Don’t hesitate to reach out: do your screening right the first time.

Like the blog? Never miss a post - sign up for our blog post mailing list.

Stat-Ease software provides powerful tools for design of experiments (DOE) with a great deal of intelligence baked in. Here are 10 “smart” features that make DOE easy for our users. From bottom to top (ordered by DOE phase: design, modeling, optimization, and confirmation), every one of them provides great value.

Here we go—the countdown begins!

Finally, one bonus feature in Stat-Ease software that will make you more intelligent: screen tips via the lightbulb icon (click the >> chevron if showing) next to the Help bubble. This will show interesting information about each feature on the screen for you to understand the underlying statistics.

Email me your favorite “they thought of everything” quality aspect of Stat-Ease software, and I will add it to my list for my next ‘shout out’ on intelligent features.

Next in our 40th anniversary “Ask an Expert” blog series is Leonard “Len” Rubinstein, a Distinguished Scientist at Merck. He has over 3 decades of experience in the pharmaceutical industry, with a background in immunology. Len has spent the last couple of decades working on bioanalytical development, supporting bioprocess and clinical assay endpoints. He’s also a decades-long proponent of design of experiments (DOE), so we reached out to learn what he has to say!

When did you first learn about DOE? What convinced you to try it?

I first learned about DOE in 1996. I enrolled in a six-day training course to better understand the benefits of this approach in my assay development.

What convinced you to stick with DOE, rather than going back to one-factor-at-a-time (OFAT) designs?

Once I started using the DOE approach, I was able to shorten development time but, more importantly, gained insights into understanding interactions and modeling the results to predict optimal parameters that provided the most robust and least variable bioanalytical methods. Afterward, I could never go back to OFAT!

How do you currently use & promote DOE at your company?

DOE has been used in many areas across the company for years, but it has not been explicitly used for the analytical methods supporting clinical studies. I raised awareness through presentations and some brief training sessions. Afterward, after my management adopted it, I started sponsoring the training. Since 2018, I have sponsored four in-person training sessions, each with 20 participants.

Some examples of where we used DOE can be found at the end of this interview.

What’s been your approach for spreading the word about how beneficial DOE is?

Convincing others to use DOE is about allowing them to experience the benefits and see how it’s more productive than using an OFAT approach. They get a better understanding of the boundaries of the levels of their factors to have little effect on the result and, more importantly, sometimes discard what they thought was an important factor(s) in favor of those that truly influenced their desired outcome.

Is there anything else you’d like to share to further the cause of DOE?

It would be beneficial if our scientists were exposed to DOE approaches in secondary education, be it a BA/BS, MA/MS, or PhD program. Having an introduction better prepares those who go on to develop the foundation and a desire to continue using the DOE approach and honing their skills with this type of statistical design in their method development.

And there you have it! We appreciate Len’s perspective and hope you’re able to follow in his footsteps for experimental success. If you’re a secondary education teacher and want to take Len’s advice about introducing DOE to your students, send us a note: we have “course-in-a-box” options for qualified instructors, and we offer discounts to all academics who want to use Stat-Ease software or learn DOE from us.

Len’s published research:

Whiteman, M.C., Bogardus, L., Giacone, D.G., Rubinstein, L.J., Antonello, J.M., Sun, D., Daijogo, S. and K.B. Gurney. 2018. Virus reduction neutralization test: A single-cell imaging high-throughput virus neutralization assay for Dengue. American Journal of Tropical Medicine and Hygiene. 99(6):1430-1439.

Sun, D., Hsu, A., Bogardus, L., Rubinstein, L.J., Antonello, J.M., Gurney, K.B., Whiteman, M.C. and S. Dellatore. 2021. Development and qualification of a fast, high-throughput and robust imaging-based neutralization assay for respiratory syncytial virus. Journal of Immunological Methods. 494:113054

Marchese, R.D., Puchalski, D., Miller, P., Antonello, J., Hammond, O., Green, T., Rubinstein, L.J., Caulfield, M.J. and D. Sikkema. 2009. Optimization and validation of a multiplex, electrochemiluminescence-based detection assay for the quantitation of immunoglobulin G serotype-specific anti-pneumococcal antibodies in human serum. Clinical and Vaccine Immunology. 16(3):387-396.