Replication, Randomization and Blocking in DoE
High quality data is accurate, reliable and free from biases. In DoE there is the concept of replication, randomization and blocking that ensures that you record high quality data.
- Replication: Repeating the same experiment multiple times. This repetition helps to quantify the natural variability in the data. We make sure that observed effects are consistent and not random.
- Randomization: Means to conduct your experiments in a random order. This technique mitigates the impact of uncontrolled disturbance variables and reduces bias.
- Blocking: is used to control for the variability introduced by known but irrelevant variables. It involves grouping similar experimental units into blocks, and then randomizing treatments within these blocks. Blocking helps to isolate the effect of the primary factors of interest by accounting for the block-to-block variability.
Example
Louise, an R&D manager at a tire company, faces the task of verifying a new rubber formula that could significantly reduce braking distances - a potential breakthrough in vehicle safety. She considers different experimental plans to demonstrate the efficacy of this new formula.
Without replication, randomization and blocking
The simplest approach involves just two tests: one with Tire A (a competitor's product) and one with Tire B (the new formula). While this method is quick, its reliability is questionable. This is because the driver's reaction time could vary between tests. For instance, if the driver reacts slightly slower in applying the brakes during the test of Tire B, this could lead to an inaccurate conclusion about the tire’s performance.
With replication but no randomization and blocking
Recognizing the need to counteract the variability in human reaction time, Louise decides to repeat each braking test 10 times, aiming for an average measurement that is more reliable. Only problem though, they won‘t be able to finish all the experiments in one day and splitting the tests over two days introduces new variables, such as weather conditions. Assuming, Louise decides to split the experiments by tire, so that Tire A is tested on day 1 and Tire B is tested on day 2, it could unfairly impact the result if it rains on one of the test days. The reason is that wet conditions typically increase the braking distance.
With replications and randomized but no blocking
To mitigate the weather issue, Louise considers randomizing the tests - mixing Tire A and Tire B tests across both days. This way, both tires are tested under varying conditions, ensuring a fairer comparison. However, this approach introduces high variability within the dataset since the braking distances on wet and dry roads differ significantly, affecting the precision of the mean braking distance for each tire.
With replications, randomized and blocking
Louise realizes that the optimal solution is to also use blocking. By grouping the experiments by day (block), she can control for the day-to-day variation (like weather conditions). This method allows for more accurate comparisons between the tires while still accommodating intra-day weather changes through randomization within each day.
Summary
In summary Replications help to address uncertainties inherent in the experimental set-up, like a driver's reaction time. Randomization deals with uncontrollable, changing variables during the experimental period (such as weather conditions that change during the day). Blocking is used to control known disturbance variables that aren’t the focus of the experiment (like day-to-day weather conditions).
Blocking can be particularly useful in scenarios where:
- Experiments are conducted on different machines.
- Tests are performed by different people, especially when subjective evaluation criteria are involved.
- Different batches of raw materials are used in the experiments.
