5. Discussion
We have introduced an approach for generating effective SIDs and illustrated that verifiable clustering results can be obtained therefrom. These SIDs strive to make the products considered by each panelist as different as possible, based on the sensory profiles, while also maintaining position balance. The fact that nested designs can be easily produced makes possible a verification step that is relatively rare in cluster analyses. We wish to emphasize that the work herein should be viewed as sort of blueprint or para- digm for how incomplete-block designs should be constructed and analyzed when used in liking studies. Depending on the applica- tion, various steps could be changed along the way. For example, simple Euclidean distance was used to find the distance between products, whereas, in another study, a weighted distance might be desirable.
In the white bread analysis, we observed that the clusters gen- erally corresponded to higher scorers versus lower scorers. There are a few things that could be done about this. For one, each cluster could be investigated for possible heterogeneity therein, leading to a sort of hierarchical clustering procedure. Another approach would be to collect data as ranks, rather than on a hedonic scale. This would, of course, bring its own problems, including having to deal with ties. One thing that we cannot do is standardize pan- elists’ results within the SID framework because doing so would require that every panelist saw the same products. In fact, a nice illustration of why standardization will not work here can be seen by looking at the nested designs.
Acknowledgements
This work was supported by a grant-in-aid from Compusense Inc. and by a Collaborative Research and Development grant from the Natural Sciences and Engineering Research Council of Canada.
5. DiscussionWe have introduced an approach for generating effective SIDs and illustrated that verifiable clustering results can be obtained therefrom. These SIDs strive to make the products considered by each panelist as different as possible, based on the sensory profiles, while also maintaining position balance. The fact that nested designs can be easily produced makes possible a verification step that is relatively rare in cluster analyses. We wish to emphasize that the work herein should be viewed as sort of blueprint or para- digm for how incomplete-block designs should be constructed and analyzed when used in liking studies. Depending on the applica- tion, various steps could be changed along the way. For example, simple Euclidean distance was used to find the distance between products, whereas, in another study, a weighted distance might be desirable.In the white bread analysis, we observed that the clusters gen- erally corresponded to higher scorers versus lower scorers. There are a few things that could be done about this. For one, each cluster could be investigated for possible heterogeneity therein, leading to a sort of hierarchical clustering procedure. Another approach would be to collect data as ranks, rather than on a hedonic scale. This would, of course, bring its own problems, including having to deal with ties. One thing that we cannot do is standardize pan- elists’ results within the SID framework because doing so would require that every panelist saw the same products. In fact, a nice illustration of why standardization will not work here can be seen by looking at the nested designs.AcknowledgementsThis work was supported by a grant-in-aid from Compusense Inc. and by a Collaborative Research and Development grant from the Natural Sciences and Engineering Research Council of Canada.
การแปล กรุณารอสักครู่..
