When a well-developed vocabulary list, or preferably, a well-structured list of categories , has been developed and is applied to more than one text source, it is entirely possible that the matrices of joint word frequencies produced by HAMLET joint frequencies contain items for which there are no collocations in the current text input, but there are good grounds for regarding the 'missing' items as, in some sense, 'latent', rather than simply absent.
In this case, it is still possible to submit the matrix in question to multimensional scaling by MINISSA, by temporarily ignoring the zero entries, but restoring them later to the resultant configuration as 'latent categories', which can be recalled when reviewing MINISSA results graphically, and carried forward into comparisons of different text sources to which the same search list has been applied, using Procrustean Individual Differences Scaling.
Where PINDIS detects variables which have been treated in this way among the input configurations to be considered, it draws this to the attention of the user, and simply admits the entries concerned as extreme outliers in carrying out its usual analysis, taking the first configuration as the reference for comparisons rather than the centroid, so that the outliers remain clearly visible at all times in the results.
Where a large number of items are missing in particular texts, the more limited procedure Individual Differences Scaling (INDSCAL) has been found to be more robust for this purpose.