Why blocking matters

Blocking is a fundamental principle in Design of Experiments (DoE) that helps ensure the quality and reliability of your experimental data. This blog post illustrates what happens when you do not use blocking even though you should have.

Experiment Background

We tested how the use of a reactive diluent affects the properties of an epoxy coating. We had one EP base resin and two different hardeners: a fast-reacting one and a slow-reacting one. The response variable was the pot life of the epoxy mixture, which is the time it remains usable before it starts to harden.

Analysis and Findings

Main Effect of Hardener

The left plot below shows the main effect of the hardener on pot life. A higher pot life means the reaction is proceeding more slowly. As observed, Hardener 2 reacts significantly faster than Hardener 1.

Main Effect of Reactive Diluent Concentration

The middle plot depicts the main effect of the concentration of the reactive diluent. Surprisingly, increasing the concentration seems to decrease the pot life. This finding is odd because the pot life without reactive diluent is 10 minutes, making it illogical that increasing the diluent concentration reduces the pot life.

Interaction Plot

The interaction plot reveals a more nuanced story. The decrease in pot life with increasing concentration is primarily observed with Hardener 1. For Hardener 2, there is no significant change in pot life with varying diluent concentrations.

What Went Wrong?

Upon closer inspection, it was discovered that the different batches of Hardener 1 that were used had varying reactivities due to a production error. By not grouping the experiments based on the batch of hardener used, we allowed an external variable (batch-to-batch variability) to confound our results. Fortunately this error was discovered in time, but in a more comprehensive design with more factors and levels, it might have gone unnoticed.

Lesson Learned

Therefore, always consider potential sources of variability such as different batches, machines, or operators and apply blocking to control them. It will improve the robustness and reliability of your experiments. In the best case, blocking might not be needed if the batches do not differ. In the worst case, you know that something went wrong and can counteract accordingly.

How do I perform blocking?

Fractional designs are well-suited for incorporating blocking. You can add it as an additional variable without having to perform more experiments.