Selecting variables

Any of the options for selecting samples are also available for selecting variables, e.g. selecting by variable numbers or by levels of an indicator, the latter as seen in the example of the previous section, in which the Tasmanian copepods of ‘Undetermined taxa’ were excluded. There is a similar construction of selecting variables with no Missing! entries across the full set of samples. Note that if the selection option of (•No missing values) is chosen for both samples and variables, the order in which these are taken will affect the outcome. In practice, if it is required to form a complete matrix (and this is now less essential than in previous versions of PRIMER since all resemblance measures are now defined under pairwise-elimination of missing values, Section 5), a more careful manual deselection of the array rows and columns is likely to be preferable, utilising knowledge of which are the most important samples or variables to attempt to retain. Alternatively, where the data can be approximated by multivariate normality, missing entries can sometimes be successfully estimated by the EM algorithm – see the Tools>Missing menu, in Section 12.