The term "experimental design" refers to a family of techniques that are used by scientists to produce data that can be converted into knowledge by mathematical analysis. The particular (peculiar?) type of math we are discussing here is called "statistics". More precisely, we are discussing parametric Fisherian statistics.

Yes, there is another complete system of statistics. It is called Bayesian. It actually predated Fisherian statistics, and there is currently a move in some quarters to return to it. There is also a sort of a hybrid of the two, called non-parametric statistics, that is especially useful when you are working with small data sets (N of less than about 30 or so).

No matter which experimental design you use, nor what sort of fancy name it has, they all have in common a careful, precise definition of:

• Dependent variables (what you are trying to influence, e.g. the blood sugar levels of diabetic patients)
• Independent variables (what you are going to do to try to change the dependent variable(s), e.g., your new patient education program)
• Hypothesized relationships between independent and dependent variables (e.g., the new education program will make the blood sugar levels come down)
• Measurement Procedures (the tricky part - read on, please)

Measurement Procedures - The Key

If you take a blood sugar reading on someone, and it is above the normal range, several questions will pop into your mind, such as: Has this person been diagnosed as diabetic? If so, what type? What has he/she eaten lately? How much? How long ago?

These questions aren't all that difficult to answer, and they also have a pretty well-defined average and variability. That is, these are variables that can be measured with a considerable degree of precision and accuracy.

But what influences the time since the last meal, what was consumed, and how much? This is not as easy to answer, and it can even be extremely variable from person to person. The answers to these questions, and the subsequent classification of persons into groups who are similar on key characteristics, can be crucial in the design of an experiment. These characteristics, and the procedures you use to try to influence the time between meals and the amount and type of food consumed, must be precisely specified and controlled for your experiment to make any sense.

Experimental design is therefore nothing more than carefully specifying the conditions under which two or more measurements are taken so that you are able to pinpoint the reason for any differences observed in the measurements. Remember the discussion of extraneous factors?