With the current desktop computers, large data sets and data bases can be easily manipulated. The speed and ease of use of personal computers can lead to projects being tackled without prior planning. Files can be manipulated easily. Matrices can be inverted. Corrections can be made on the fly and a new version of a file produced to answer any new question that arises or to feed further analysis that could be done, though may be should not. When the project is revisited, for verification or the production of later papers, it can be very difficult to identify the file versions or most recent update even if a record of file production was kept. Therefore it is always wise to keep file production to a minimum and expand the documentation.

A good analysis does require forethought and careful planning. We report here on a project that required only one set of data entry, each participant response being entered into a record similar to an Excel record with the variable values reading across horizontally and being identified by the study number of the participant, the participant?s group (exerciser or control), and the number designating the time point in the study. This matrix sufficed for most of the reliability analysis, and for the calculation of intra-class correlation coefficients. These records could then be merged horizontally by study number to do repeated measures analysis and to calculate Cohen?s kappa. Responsiveness and coefficient of variability could be assessed using aggregate statistics from the same matrix design.

It is relatively easy to design questionnaires, but to be useful they should be validated. Validation is a long process often taking at least a year and therefore is not undertaken lightly. There is a vast literature including complete volumes dedicated to the validation of questionnaires and health scales (a scale is a set of questions, whose answers can be combined in some way to produce a meaningful number which measures a state of health) [1,2,3]. A scale usually measures one clinical outcome, such as pain, disability or fatigue. After the questions are designed, constructed, ordered and tested on people unconnected with the design but representative of the target population, then the developers often consider how to assess validity. Validity means not just labeling a set of questions a ?pain scale? or ?a headache problem questionnaire? but assessing if the questions ?do? what they are supposed to do; do they assess pain or headaches? Usually this means assessing association with simple questions or the taking of appropriate medications, or other already validated measures etc.