Statistical Analysis: Acquiring Knowledge from the Full Scale Printing Trials, 1996 International Printing & Graphic Arts Conference Proceedings

Frøydis Bjerke

The Norwegian Pulp and Paper Research Institute (PFI)

In full scale printing trials huge amounts of data from various sources are generated and stored in a database. The main groups of variables in this database are: 1) Laboratory tests on unprinted and printed samples 2) Data sets from image analysis of printed samples 3) Visual assessments on printed samples, made by a test panel. A proper analysis of these data requires ample statis-tical knowledge in different branches of multivariate statistics, as well as paper and printing knowledge to formulate relevant problems and interpret the statistical results. The paper briefly presents some important statistical methods used to analyse the full scale printing trials, connected with some examples. The methods are a) Multidimensional scaling (MDS) with preference mapping, used on difference data and preference data from the visual assessments. b) Principal Component Analysis (PCA), used on subgroups of variables to classify the objects (samples), and Partial Least Squares (alias Projec-tion on Latent Structures, PLS), which explains variation in a set of dependent variables by a set of independent variables using linear models. c) MDS combined with PLS and other multivariate methods, used to explain the visual assessments by variables from group 1 and 2 above. The paper emphasizes the application of the methods, and not their theoretical framework.