Page:Sm all cc.pdf/116

 7) include replicate measurements in your design, if possible. Normally it is unnecessary to replicate every measurement. Replicating perhaps 5% of measurements, on randomly chosen samples and standards, gives a good estimate of overall precision.

8) in the experimental design, never change more than one variable or experimental aspect at the same time (unless you are using a factorial design).

9) list the entire planned experimental procedure. Do not simply try to visualize the entire procedure in your head.
 * Calculate how many measurements or samples will be needed.
 * Visualize every step, imagine what could go wrong, and seek a way of avoiding the potential problem. Possibly the experimental procedure needs revision, or perhaps all that is needed is increased caution at key points. What affects precision at each step? What affects accuracy? Which steps must be done painstakingly, and which are more robust? The ideal of doing every step of every experiment as painstakingly as possible is not only unrealistic; it is a recipe for scientific progress that is so slow and inefficient that the scientist will have trouble keeping a job.
 * Seek ways of streamlining the list of steps without jeopardizing the integrity and reliability of the experiment. Could some steps be done more efficiently as batch process rather than in series? Are there long wait times during which other steps can be done? Where are the natural break points for each day’s work?

10) do a gedanken experiment, a thought experiment, before the actual experiment. Try to predict all of the possible outcomes of the experiment, how you would interpret them, what the weaknesses would be, and what the alternative explanations would be for each interpretation. Can the experimental design be changed to provide a more diagnostic, less ambiguous interpretation of each result? Pretend that you are a reviewer, intent on finding a fatal flaw because the results are contrary to prevailing theory. Remember that the gedanken experiment takes up a small fraction of the time of the actual experiment. When writing up results, we usually wish that we had done some part of the experiment differently; often a more careful gedanken experiment would have helped.

11) avoid last-minute changes to the experiment, unless you have carefully thought through all of their possible implications.