Abstract:

The main objectives of this paper are to demonstrate the relevance and value of data fusion, as a technique, in the Media research area, by reference to two case studies; and to discuss the variety of opinions and difficulties attaching to the assessment of the 'quality' of data fusion results. Subsequent to this introduction, we first briefly recapitulate the decisions to be taken in relation to any specific fusion. We then review, rather more fully, the criteria which are or which should be (in our opinion) employed, post hoc, to test data fusion output. An account of our case studies follows, with emphasis on the particular. Problems they presented and the solutions adopted. We conclude this introduction with a brief word on terminology. We shall use the term fusion' to imply the merging of data from two initially discrete files, whether or not the transfer of data is in both directions, A > B and B > A, or only one- way, whilst recognizing that others may prefer to label only the former 'fusion' and to term the latter, unidirectional transfer 'ascription' or 'injection'. We shall denote any variable common to both files as X and recognize three sub-sets of common variables: critical Variables, on which the values of a data-donor and a data recipient must match exactly, if information is to be transferred between them; matching variables, which are common and employed in the pairing of donors and recipients, but nonuritical; and (rarely) control variables, which are common but held out from the matching process, for testing purposes. We shall employ Y to denote any variable initially unique either to the donor file or to the receiver file. Yd will stand for any donor variable; Yf for a donor variable, after fusion and then present, therefore, in the receiver file; and Yr any variable unique to a receiver file. For both X's and Y's, subscripts will only be added when it is necessary to distinguish between two or more variables in any group.