Some Observations on the Correlation Determinant

摘要

The lower 5% point of the correlation determinant in the null case, that is with zero parental correlations in the multivariate, normally distributed, data set, are presented for sample sizes up to 30. Repeated Monte Carlo simulation suggests that the limits are correct to +/-2 units in the third place of decimals. Thus the limits permit a test of hypothesis for the mutual independence of the variates involved. For sample sizes greater than 30, an asymptotic approximation based on the chi-squared distribution, as proposed by Morrison (2005), is shown to be quite reliable.