QUB | Archaeology and Palaeoecology | The 14Chrono Centre
Options to control modelling: 0 Switch on/off (off) [off] 1 Number of models to generate (10) [10] 2 Random number generator to use (1) [1] 3 Seed for random number generator [TIMER] 9 Leave this menu Q Return to main menu Enter number <Enter>The results of zonation depend to some extent on the reliability of the input data. However, if it is possible to obtain any kind of confidence interval for the results, I am unaware of it, although the broken stick model (discussed above) helps. One solution is to `model' the data, using the uncertainties associated with the input data (which can be expressed as confidence intervals) to generate alternative datasets, and carrying out zonation on those. I have implemented this in a simple way for proportional data. Such data are distributed binomially, so I draw random numbers from a binomial distribution, which requires knowledge of the observed proportion for a given type in a given sample, and the sum on which that calculation was based. In order to carry out the modelling, therefore, a sum must be available (marked by `*' at the beginning of the `taxon' name: see Data preparation), which must follow in the dataset all types included within the sum.
Types selected for inclusion in the sum are recalculated, and the sum itself reduced in proportion to the total abundance of the types excluded. Then the zonation is run a number of times which should be as many as possible (e.g., 100): the default is 10 (menu Mc.7.1). Results are then accumulated and printed to the data analysis file (which must be available). A particular number of zones must be designated via menu Mc.1 (i.e., the number of zones cannot be set to 0). Output varies depending on whether zonation is binary or optimal splitting: CONISS output resembles that from binary splitting. For binary splitting, the output shows which zones boundaries are indicated and at what proportion of the number of models, in order of occurrence. For optimal splitting, the output shows which complete patterns of splits occurred, and the proportion of the models that these were found with. In all cases, the output concludes with results of zonation from the actual data for comparison.
Zonation is expensive in computer time anyway: running many models may take an inordinately long time. Run this option overnight, or use a mainframe.
Menu Mc.7.2 allows choice of random number generator, and menu 2.7.3 allows choice of the seed used. These submenus are identical to those described under menu N3, and the same variables are used to store the choice of random number generator and seed.
Return to menu Mc.
Archaeology and Palaeoecology | 42 Fitzwilliam St | Belfast BT9 6AX | Northern Ireland | tel +44 28 90 97 5136