Flavour and texture profiling data obtained from an experienced, well-
trained, homogeneous panel (N = 20) were analysed by ANOVA and multiva
riate techniques for the full panel as well as for several smaller pan
els (N = 10 and N = 5) which were drawn, a posteriori and at random,fr
om the original panel. Homogeneity of variance was maintained over all
the panels. These data had been collected originally to investigate f
lavour-base interactions in vanilla ice cream. The sample relationship
s were unaffected by panel size, but there was considerable loss of in
formation with the smaller panels when trying to explain treatment eff
ects (ice cream processing factors) in terms of descriptors. When the
panel size was reduced to half, only 67% of the descriptors having sig
nificant treatment effects were retained; for interaction effects ther
e was only 34% retention. Reducing the panel to one-quarter of its ori
ginal size allowed retention of only 34 and 9% of the significant desc
riptors for these effects respectively. The loss of information (which
cannot be predicted beforehand unless the same set of samples has bee
n evaluated previously) iis discussed in relation to the reduction in
degrees of freedom inherent for smaller panels.