1.6 Guidelines for designing an experiment

1.

Recognition and statement of the problem. Write down the objectives. A clear statement often makes the design obvious; a vague one guarantees an experiment that answers nothing in particular.

2.

Choice of factors, levels and ranges. Distinguish the design factors from the nuisance factors, and decide how each nuisance factor is to be handled: blocked if it can be controlled, measured and adjusted for if it can only be observed, randomised over if it is unknown.

3.

Selection of the response variable. It should measure what the objectives are about, and be measurable with known precision.

4.

Choice of experimental design. This includes the number of replicates, the run order and any blocking, and should be made with the analysis in mind.

5.

Performing the experiment. Monitor the process to make sure everything is done according to plan; errors in procedure at this stage usually destroy the validity of the experiment.

6.

Statistical analysis of the data. Estimates, tests and confidence intervals, together with residual checks of the model’s assumptions.

7.

Conclusions and recommendations. Practical conclusions, followed where necessary by confirmation runs.

Steps 1–3 are often called pre-experimental planning. Experimentation is also sequential: a first small screening experiment identifies the important factors, and later experiments study those factors in more detail.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.