Dear Colleagues,
I have conducted a sensory evaluation for 8 biscuit samples using 7 point hedonic scale with 30 semi-trained panelists. The data obtained from the sensory evaluation was analyzed according to the Friedman non parametric method followed by the Mann-Whitney U test.
Can anyone suggest a suitable graphical representation to clearly visualize the sensory profile of these 8 products based on the sensory panel performance?.
Thank you.
Bar graphs could be used considering that the scores are continuous numeric on y axis and having the treatments (biscuits) on the x axis. In terms of design, however, I would have gone for a randomized complete block design (RCBD) to eliminate the variability in the data due to the panelists. It would also be great if mixed model could be used, taking the serving orders as a random variable. These could have given you the best estimate of the differences among the 8 samples. I hope this could help for future experiments involving sensory assessments. For professional graphics, SigmaPlot is preferable. Good Luck!
Dear Prof. Tadesse Fikre Teferra, Thank you very much for your valuable response.
Check https://www.researchgate.net/publication/232546822_Visualization_of_sensory_profiling_data_for_performance_monitoring
Also check https://www.researchgate.net/publication/305982998_Trained_sensory_panel_performance_monitoring_using_bias_matrix_estimation
Dear Dr. Chinaza Godswill Awuchi, Thank you very much for your valuable responses.
Heshani Rathnayake
The Mann-Whitney test isn't a suitable test to use for a post-hoc test for a Friedman test. It really doesn't make sense to use a test that assumes independent observations as a post-hoc test for test that assumes that the observations are dependent.
There are several post-hoc tests available for the Friedman test. For a list, see the following document, in the index, all the tests that start with "frdAllPairs" or "frdManyOne": https://cran.r-project.org/web/packages/PMCMRplus/PMCMRplus.pdf . (Conover and Nemenyi are common; be sure to find the test baring these names that's appropriate for Friedman).
If you really want to do pairwise tests, you could use Friedman with two groups, or maybe something like a two-sample sign test. But really those post-hoc tests designed for this situation are better.
As to the plot, because the effect of the rater is taken into account for each group (product), it's possible to just plot the medians and data for each group, perhaps with a box plot ( https://i1.wp.com/statisticsbyjim.com/wp-content/uploads/2019/01/boxplot_teaching.png ) or as a dot plot (like this, but probably add the median: https://i1.wp.com/statisticsbyjim.com/wp-content/uploads/2019/01/IVP_strength.png ).
These plots don't really represent what the test is doing (because it's based on the ranks of the data, and the effect of the rater is taken into account). But in most cases, these plots will be a fair representation of the data.
Tadesse Fikre Teferra , the Friedman test uses data arranged in non-replicated complete block design. It essentially takes the effect of each rater into effect by ranking the treatments (groups) within each rater.
- Badges
- Science topic
- Similar topics
- Computer Science
- Computer Communications (Networks)
- Internet