Design of experiments
An experimental design means that the samples to study are chosen according to a well considered design. When optimising experimental work, the samples and setups can be chosen according to a scheme where all variables of interest are varied simultaneously but in a controlled way. Then we can get orthogonal variable directions (Figure 1), which facilitate the use and interpretation of multivariate statistic models. The significant influences from single variables and interactions between variables can subsequently be calculated.
|
Figure 1: The red dots shows the experiments to do in a “full factorial design” with three variables (represented by the dashed arrows). |
In a corresponding way, we may pick samples for a study. It is of course important to include samples holding all the variation that we want to study -representative samples. For example, when modelling a disease we have to include both sick and healthy patient samples in our study. It is however not very useful to study very similar samples, e.g. two samples from the same patient (if we don’t want to study repeatability or reproducibility). Further, the sample collection may need to be limited not to include a lot of variation between samples that we don’t want to study. This is often very difficult working with biological samples, which include a large biological variation. For example, the study might have to be limited to one sex or a certain range of age. However, parameters that we can not control or are unaware of have to be included, preferably in all possible combinations and ranges. Then we have the possibility to ignore these parameters if they are random between samples, or detect them as significant if they will help to describe our response (e.g. state of disease). This is done by multivariate data analysis.
A well designed study has improved chances to show significant results. Also, optimisation of the laboratory work by a design of experiment will in almost all cases lead to a laboratory procedure and an instrumental setup better than if it was chosen by “trial and error”. Read more about this subject in the links below.
