| Popis: |
Large “hierarchical” datasets, that is a dataset that consists of identifiable subsamples of a population, offer the possibility to calculate a statistic per subsample. Together they form a sample distribution of that statistic. This distribution may be compared to some null hypothesis distribution and tested for deviations. To determine whether two variables are related, some type of correlation statistic usually is calculated. If the assumption for a normal distribution is not true necessarily the nonparametric Spearman rank correlation coefficient is better suited to estimate the existence of a relation than the Pearson product-moment correlation coefficient. The Spearman rank correlation coefficient thus has more general applicability at the cost of only a marginal loss in efficiency in the situation of normality. The dBase program SPEARMEN calculates Spearman rank correlation coefficients for all the subsamples of a “hierarchical” dataset and tests the sample distribution of this statistic against the null hypothesis H o : r s = 0 by using the Kolmogorov-Smirnov one sample test. This approach has the advantage that it is sensitive to deviations for the entire range of observations and not only at critical tail-end values. For subsample sizes under 14 the theoretical null hypothesis permutation distribution is used, whereas above this value the distribution is approximated by a Student- t distribution. The program uns under dBaseIII + , dBaseIV, Foxbase + , and Clipper and can be used for virtually unlimited datasets. |
