World Statistics Day was yesterday, but we’re celebrating all week long! This celebration means acknowledging the impact statistics has on our world. Who is your favorite statistician? Share with us who they are and why they top your favorites list.
Choose Language Hide Translation Bar
Staff (Retired)
A simple designed experiment with multimillion-dollar results

On Saturday (or Sept. 34), we marked the 25th birthday of JMP, a product I have been using since version 2. Until 2006, I was a JMP customer, even attending the first Discovery Summit conference for JMP users back in 1996. This birthday has made me nostalgic, and I wanted to share a story from my years as a JMP customer working in the chemical industry.

In 1991 at my previous employer, I completed one of my first designed experiments using JMP. That experiment was the first of many major breakthroughs enabled by using JMP and the DOE approach to scientific experimentation.

JMP Distributions

The Distributions (rerun using JMP 11) from the first DOE I did using JMP in 1991. Note that the highest-yield product was obtained when the dibromo was high, which is nonintuitive.

Our research laboratory had identified a new-generation compound. This new material would provide substantial advantages over our previous material and would give us a significant edge over one of our key competitors.

The synthesis of this compound required the use of a brominated intermediate. Although this intermediate would provide the desired cycle time in the subsequent chemical step, it was not easy to isolate using any of the conventional manufacturing isolation techniques. In addition, even when we were able to isolate it, the purified brominated intermediate produced an undesirable dimer impurity in the subsequent chemical step. This impurity caused our customer to have major mechanical difficulties in their process. To avoid causing these problems for our customer, we had to incorporate an additional purification step to produce a dimer-free product. The purification was effective, but it required the use of a solvent that was not friendly to the environment.

We decided to set up a DOE investigation where the brominated intermediate would be carried on in the process “as is” without an isolation step. The crude intermediate reaction mixture would be subjected to the second step without prior isolation and purification. This would overcome the isolation difficulties of the brominated intermediate, afford cycle time and environmental solvent saving benefits and would generate knowledge about the sensitivity of the investigated inputs on the process conversion.

We explored three factors in this non-isolation design. They were residual water and amount of brominating agent in the first step, and the reactant amount in the second step of the conversion.  This seems like a simple design in retrospect, but let’s see the ramifications of this “simple” design. The results from this investigation were both unexpected and nonintuitive. As it turns out, we obtained the best overall conversion of the final product when the intermediate was at a suboptimal level. That was a big surprise. The Prediction Profiler from that experiment is shown below. As you can see, we got optimal results when we over-brominated the starting material by 10%! Not only did those settings optimize the overall conversion (>95%), but it also totally shut down the troublesome competitive dimer reaction and provided the desired product at the highest yield.

Prediction Profiler in JMP

The Prediction Profiler shows that we got optimal results when we over-brominated the intermediate by 10%.

When we initially set up the design, the primary goal was to understand the sensitivity of the non-isolation process to varying amounts of the inputs. We were hoping for a procedure to avoid isolating the brominated intermediate because of its undesirable physical characteristics.

What we obtained, however, was much better.  We delivered a process that had a higher yield and produced a dimer-free product that eliminated the need for the purification as well. This breakthrough allowed us to save millions of dollars each year and win corporate recognition for the environmental gains that resulted.


JMP 3D graph

Thanks to our DOE using JMP, we delivered a process that allowed us to save millions of dollars a year and make environmental gains.

So here’s to the 25th birthday of JMP! I have enjoyed the wonderful journey of the evolution of this product firsthand. (And as it so happens, Sept. 34 is my birthday, too!)

Article Labels

    There are no labels assigned to this post.


Emil Friedman wrote:

Was the prediction profiler available in version 2?

Staff (Retired)

Louis Valente wrote:

No Emil, The Profiler was available in version 4 as I recall. The archived data have been visualized in this blog using JMP 11.


Peter Bartell wrote:

Great story Lou! Probably my most impactful JMP/DOE story came about back in the 90's too. Our team was trying to optimize image quality (things like solid ink density and dot gain for your printing industry types) in the flexographic printing process for 4 color process printing on flexible white opaque polyethylene material...think bread bags or frozen vegetable bags. Got to the verification stage and ran a trial to confirm our process optimization work...always the last step before issuing orders to production. Well our results were a million miles away from our optimal quality characteristics. The DOE naysayers on the team thought they finally had their last laugh...they had been suspect that DOE was black magic and wouldn't work in the flexo process which was 99% art and 1% science. Root cause of the disappointing results was an operator who had taken it upon himself to NOT set some key process input factor levels at the prescribed settings because, "We NEVER run at those conditions...they must have made a mistake with these settings, I'll put them where we usually run...we'll get 'good' results there." Once this issue was discovered, I trotted out the JMP Prediction Profiler live for our team...put in the REAL settings that the operator had used, and voila, the model predicted the exact lousy quality we obtained. It was at that point even the naysayers were convinced that there was something too this DOE black magic...and the JMP DOE and modeling functions made it uber easy to discover the breakthrough non intuitive process results that were observed.

Staff (Retired)

Louis Valente wrote:

Thanks for sharing your story Peter!