QUB | Archaeology and Palaeoecology | The 14Chrono Centre
3 Principal components analysis (off) [off] Enter `0' for on, or `1' for offThis is the classic way to reduce multidimensional data to a low number of dimensions (Birks & Gordon 1985). First, the method measures the similarity between taxa (correlations or covariances), then the resulting matrix is subjected to the PCA procedure, looking for major directions of variation within the data set.
Results are printed out to the data analysis file (see menu Ec), if one is available. The output includes the basic matrix, eigenvalues, taxa loadings, and sample scores. These can, of course, be extracted and plotted in some appropriate manner. It would be usual to apply a threshold value (menu Mi), and recalculate the dataset for analysis to proportions of the sum of types included (menu Mj).
The table of eigenvalues marks, with `*', those eigenvalues considered to be interpretable by exceeding eigenvalues generated by a broken-stick model of the distribution of variance amongst the various components (Legendre & Legendre 1983; Jackson 1993).
If 0 is selected, the following submenu appears:
1 Specify the matrix to use [3] 2 Specify the number of axes for output [6] 3 Data transformation [0] 4 Modelling of pca (off) [off] 9 Leave this menu Q Return to main menuWhile PCA is running, just so that you know something is happening, some numbers are printed to the screen. During calculation of the basic matrix, these are the numbers of the current taxon being analysed. During the PCA itself, the number is the number of `sweeps' through the input matrix to achieve convergence. This number ought to increment to between 6 and 10. The output file includes a figure for the number of `jacobi rotations': this should be between 3n2 and 5n2, where n is the number of taxa in the input matrix.
Return to main menu.
Archaeology and Palaeoecology | 42 Fitzwilliam St | Belfast BT9 6AX | Northern Ireland | tel +44 28 90 97 5136