Produces statistics and graphs for checking sensory panel performance (D.I. Hedderley).

Options

Controls printed output (aovtables, graphs, summarystatistics, tables); default grap, tabl

Factor defining the different treatments that are being assessed

Factor defining the sessions on which the assessments were done

Factor defining the individual assessors

Equal scaling for x and y axes on Drift-Unreliability and Discrimination-Disagreement graphs (yes, no); default no

Extra information to print on graphs

Parameter

Variate for each attribute, containing the recorded score

Description

A trained panel of sensory assessors may test a set of products (e.g. taste a set of food samples) at several sessions, each time rating them on a range of attributes. If you have several measurements of the same samples from the same individuals, you can investigate how consistent and discriminating the individual assessors are. The scores recorded for the attributes are specified, in a list of variates, by the DATA parameter. The TREATMENTS, SESSIONS and ASSESSORS options supply factors defining the treatment, session and assessor involved with each unit of the DATA variates.

SAGRAPES presents six statistics based on analyses of variance, proposed by Schlich (1994), to describe how well individual assessors use individual attributes. These are:

the assessors' overall mean score on that attribute;

the mean standard deviation of the assessors' scores within a session;

the ratio of the root mean square residual (from a model fitting TREATMENTS and SESSIONS main effects to each assessor) to Span, i.e. what proportion of the spread in an assessor's ratings is due to changes in the relative scoring of samples in different sessions;

the ratio of the root mean square for sessions (from a model fitting TREATMENTS and SESSIONS main effects to each assessor) to span, i.e. how much an assessor's average score changes from session to session, compared to the spread of scores within a session;

the variance ratio for TREATMENTS from a model fitting TREATMENTS and SESSIONS main effects to each assessor;

an estimate of how much each assessor contributes to the variance ratio of the ASSESSOR.TREATMENTS interaction (from a model fitting ASSESSORS/SESSIONS + TREATMENTS/ASSESSORS to the whole panel).

prints a table of these statistics for each assessor for each of the attributes in DATA.

produces a composite plot of three graphs (Location against Span, Unreliability against Drift-mood, and Discrimination against Disagreement) for each attribute. The points on the plots are labelled with the labels from the ASSESSORS factor. On the plot of Discrimination against Disagreement, a star is plotted at the 5% critical values of the relevant F distributions; so ASSESSORS to the right of the star are significantly discriminating between TREATMENTS, and ASSESSORS above the star contribute significantly to the ASSESSORS.TREATMENTS interaction.

prints the panel ANOVA tables for each attribute.

prints overall summary statistics (numbers of observations, means and standard deviations) for each attribute, across the whole panel and all samples.

Unreliability and Drift-mood are measured on the same scale (multiples of Span), as are Discrimination and Disagreement (F-ratios). Setting option SCALING=yes scales the x and y axes of the Unreliability against Drift-mood and Discrimination against Disagreement graphs equally.

The DESCRIPTION option can be used to provide additional information (for instance, the name of the study) to label the graphs.

Method

Schlich (1994) proposed the procedure, and implemented it in SAS. This procedure uses the calculations given in the article to produce graphs for individual attributes. Currently it does not produce the graphs comparing different attributes which Schlich suggests.

Action with RESTRICT

Any of the DATA variates, or the TREATMENTS, SESSIONS or ASSESSORS factors, can be restricted to analyse a subset of the data units.

Reference