Randomization strategies for statistical analysis are based on repeatedly drawing thousands of new subsamples from the original sample. Long known to be useful, they have only become feasible since the development of powerful and inexpensive personal computers.
Randomization statistics are generally safer and better than traditional statistical formulas because they do not assume a normal distribution or an underlying continuous distribution of scores, and they are equally good with matched samples (such as the same subjects in both groups) and with single-case designs (such as comparisons of subsets within a lengthy dream journal).
There are five experimental randomization programs available through dreamresearch.net for comparing indicators from the Hall/Van de Castle system of dream content analysis. At the moment, the programs are only able to access a few Hall/Van de Castle data sets -- such as the Male & Female Norms and the "Barb Sanders" baseline sample -- but they are useful for exploring the potential of randomization statistics. (In the future, we hope to allow randomization analysis of additional Hall/Van de Castle data sets, possibly with an "upload" feature for third-party data.)
Because the randomization programs are so processor-intensive, we can't make them widely available or they'd bring our server to a screeching halt. If you have a genuine scientific interest and would like to have access or more information, please contact us at dreams [at] dreamresearch [dot] net for more information.
Go back to the "Resources for scientists" page.