|
INTRODUCTION
PRODUCT PROFILE techniques are used to answer to the following points:
|
Assistance in product development with product identity sensory cards according to their different visual, olfactory or gustatory attributes...
|
|
Assistance in decision making for raw materials or process selection by defining specified and required sensory characteristics.
|
|
Follow up to analyse the sensory attribute deviation during ageing or comparing with previous studies.
|
|
"Market strategy" by giving the product a sensory image when compared to its competitors.
|
SESSION DESIGN PREPARATION - PRODUCT PROFILE
You can choose different experimental design (complete, by rotation, by blocks, ...) in order to avoid bias or for better evaluations.
In this example, the code design suggests comparing seven products in seven blocks each containing three products (i.e. three products for each assessor).
TEST PREPARATION - PRODUCT PROFILE
Test preparation defines the required information to prepare a session and create the forms used for data entry. The parameters can be for example:
|
The evaluation mode (monadic or compared)
|
|
The scale: continuous (i.e. low to high) or structured (i.e. 1 to 10)
|
|
The random code creation, to prempt product recognition.
|
|
The code design and the way to present the samples
|
|
The attribute names
|
|
The suggested comments to assist the comment data entry,
|
|
Or to define the help information screen.
|
|
DATA ENTRY - PRODUCT PROFILE
Once the test preparation has been done, you can print the forms directly under WORDİ or amend them very easily to allow automatic scanner data collection. Here is an example of a scanner form with a continuous scale:
Depending of the TASTEL system configuration the data entry can be more or less automatic:
|
Direct data entry on a PC
|
|
Scanner data entry by an automatic form reading
|
|
Network data entry with PCs, or PalmPilots directly in the sensory booths, and
|
|
via the Internet directly into the assessor site.
|
|
ANALYSIS - REPORTS - PRODUCT PROFILE
Different calculation parameters can then be considered before the analysis. For example:
|
Test addition or comparison to regroup or not the different selected tests.
|
|
Test level to use in statistical tests or result scale to be used
|
|
Standardisation to centre and reduce automatically the values, assessor by assessor
|
|
Interval scale separation to separate scale to display the data distribution
|
| Product, attribute, or assessor selection to take into account in the next calculations or also
|
| Variable selection to take into account the variable order and variable number of the different analysed tests.
|
Other options can also be displayed in case of missing values,...
Many graphs are then available, for example to describe results in histogram, in spider graph, in pair comparisons, by product, by product with tolerance threshold, ... But also on assessors, assessor and panel, on distribution, and with different suggested tests such as ranking tests with Friedman test, Student tests, Q of Cochran,... Various multidimensional analysis can also be conducted, Analysis of Variance, Principal Components Analysis, Discriminant Analysis, STATIS Analysis (close to General Procustean Analysis), Correspondence Factorial Analysis, Text analysis is also available to analyse free and suggested comments on products with several analysis steps such as: Word filter, text descriptive analysis, synonym dictionary, and context definitions and results.
GRAPHS
This graph distinguishes easily three different strawberry origins (France, Poland and Mexico) and you can see the Mexican product is ripe, fruity, but a little bit less sweet and more acidic than the Polish product which is sweeter with a higher sweet note, but oxidised and maybe harvested too late. On the other hand, the French product seems to have missed some sun.
These first graphs can be specified with tests based on means (Student test, Analysis of variance) or with tests based on ranks (Willcoxon or Friedman test) permitting to see if the observed differences are significant or not.
In our example, vegetable, fruity note and astringent cannot been taken into account to distinguish the products. You can amend easily the graphs by selecting to change the font, font size, colours...
Graphs can also be created directly under the EXCEL format with 'macros' available in the TASTEL system to make changes to the document presentation
Various NONPARAMETRIC TESTS are also available:
|
On ranking, by transformation of intensity values in ranks, then with a FRIEDMAN test on it,
|
|
On normalisation validation, with Chi-square test and KOLMOGOROV-SMIRNOV tests and,t
|
|
On data distribution with value groups, then COCHRAN Q test...
|
ANALYSIS OF VARIANCE
This analysis can be carried out with two, three, even four factors with the totality of linked interactions. Analysis of normality is provided on the basis of the chi square test, then different mean comparison tests are available: LEAST SQUARE DIFFERENCE - NEWMAN & KEULS test - DUNCAN test. Different reports can be created to get synoptic and detailed results as well as roundups by segments of meaning. Remark: In case of unbalanced factors, the information on these elements is displayed before the standard results - results are then calculated on the number of minimum factors when possible. Otherwise tests of ASPIN WELCH take into account the totality of the available information.
Summary: product factor
MULTIDIMENSIONAL ANALYSIS
Then, many multidimensional techniques can be conducted, two examples, P.C.A. and STATIS are shown as example.
Principal components analysis
Statis analysis
This technique computes transformations on data as similarities, translations and rotations before calculating the results. Two methods are available either on products or on assessors. Then, it is possible to display compromises with or without the representation of an equivalent PCA. The details (or trajectories) allow the user to review the notation consensus. Variables or attributes can also be displayed.
Compromises often allow a more robust solution to the previous mathematical fitting operations.
Syntaxic analysis (comment analysis)
Synonym dictionary operation defines keywords and link different words with the same meaning. You can also open synonym files to link automatically words from previous studies. Then, keywords can be transformed into textual qualitative variables and frequency results by product can be displayed in histogram and even with synoptic calculations in Correspondence Factorial Analysis.
|
|
