Briefly |
Scrambled collections help you run some kinds of simulations and statistical tests. |
Scrambling for independence
The scrambled collection is a new collection like the source collection. Most attributes are the same, but one has scrambled values. Therefore in the scrambled collection, the scrambled attribute is independent of the others.
You define a measure of association yourself as a meaure. Then hook up your scrambled collection to a measures collection. By repeatedly scrambling you build up a distribution of those measures under conditions of independence, that is, with the null hypothesis true.
You can use the distribution of measures to establish P-values for accepting or rejecting a null hypothesis. That is, if your test statistic is huge compared to the artificially-independent, scrambled data, it's hard to account for it by chance.
When you collect the measures, the scrambled collection does the scrambling as many times as it says in the inspector of the measures collection. One hundred is plenty. Each time it scrambles, Fathom recomputes the measure--your statistic for the scrambled collection. You define the formula for the statistic. It puts those measures in the measures collection, and you can look at them on a graph.