Experiment!: Planning, Implementing and Interpreting

11.5 The Data Collection Plan

We have now decided what to measure, developed a measurement system for it, corrected for bias, and identified and quantified important sources of error. Before proceeding with the actual measurements we should now formulate a data collection plan. As this plan should be based on the particulars of your measurement system it will be unique for each experiment. It is simply a written description of how the measurements are to be carried out in order to obtain representative data of good quality. As the input–output diagram lists noises and background factors it is a good starting point for the plan. The purpose is to specify practical procedures for diminishing such influences.

Background factors can sometimes be controlled, for example by isolating the measurement system from them. In other cases we can make corrections, for example by calibration. If they vary predictably we can attend to them by blocking. If they are expected to drift slowly with time we may have to calibrate at regular intervals during the measurements. It is more problematic if they vary quickly in an unpredictable way. In such situations we may measure the background factor and record it as a covariate, which can then be used for corrections of the data during the analysis phase. Remember that some background factors will remain unknown; this is why it is recommended that the run order of the measurements is randomized.

Although external noises are random effects they can often ...