The Relative Eating Quality of Individual Muscles in the Ovine Carcase

Summary

This experiment was conducted to establish the protocols and finalise design issues associated with the sensory testing of sheep meats.

Experiment 1.1

A cutting, packing, transport protocol has been written and is attached. This specifies the cutting lines, sample number requirements and packing details for samples to be transported.

In addition the sample preparation protocols have also been written for grill and roast samples. However subsequent analysis of sensory roasts data suggested that treatment effects in the roasted samples were more difficult to detect for roasts than for grills. It is therefore suggested that the roast protocol be reviewed prior to undertaking more testing.

Experiment 1.2

A sensory experiment was undertaken to generate a data set to examine design issues, the potential for indicator cuts and the relationship between sensory and objective samples. The design was aimed at generating variation in palatability by use of different age classes and the use of stimulation.

The design comprised a total of 18 animals allocated to a 2x2 factorial design with 2 age classes (1st cross full mouth ewes and carry over lambs) and 2 stimulation treatments (high voltage stimulation and a non-stimulated control). As part of the tasting protocol consumers were asked to score tenderness, juiciness, flavour strength, flavour like and overall satisfaction of each of the samples.

Design Issues

The current latin square design (which has fixed the total number of samples that are tasted in a pick, ie 108), uses a total of 10 consumers to taste every sample. Given this constraint there was opportunity to examine the effect of both a decrease in the number of tasters per sample with a concomitant increase in the number of samples tested. .

As a preliminary analysis, the homogeneity of consumer score variance within samples was examined for all sensory dimensions. Analysis of the log transformed variances showed that the consumer data for the samples was relatively homogeneous both for the different treatment effects and also between animals, for all sensory dimensions. The magnitude of the within sample variance was similar to sensory scores for beef.

A simulation data set of 10800 tastings was generated based on the variance structure from the grills. This data set was randomly sampled to assess the effect of different design structures, where the total number of tastings per pick was kept constant, but as the number of tastings per sample decreased (from 10 to 2) the number of animals increased (18 to 90). The simulation data set was structured so that it included between animal effects (both fixed and random) and a cross classified effect (ie muscle within animal).

Based on the results of 600 simulations it was concluded that the most efficient design (ie the one with the highest F ratios and therefore the greatest level of significance for cross classified and fixed effects) occurred with a design that contained 10 tastings per sample. The significance of the random animal term changed little over the range of numbers of tastings per sample.

Also on design issues, the effect of clipping outlying consumer scores from the data before meaning to obtained sample scores was examined. Analysis of the lamb data showed that clipping the data reduced the ability to pick up design effects. Therefore it is recommended that in the future lamb sensory data not be clipped.

There was also a design issue in terms of what questions to ask the consumer to score on each sample. The previous MLA Sheep meat eating quality experiment asked a sample of the consumers to score flavour strength, in addition to the normal 4 sensory dimensions (ie tenderness, juiciness, like flavour and overall likeness). A preliminary analysis of this small data did not find any justification for inclusion of the extra sensory dimension in the consumer score sheet. The present data set allowed an analysis of the flavour strength dimension on a larger data set. Our results showed that flavour strength had both the lowest correlation with other sensory dimensions and had the smallest range between rate categories. In addition a discriminant analysis showed that flavour strength contributed little to a linear function to allocate samples to rate categories. Therefore it was recommended that flavour strength be dropped from future the consumer sensory score sheets.

The present data set confirmed the transportability of the weightings for the different sensory dimensions to be combined into an overall palatability score (SEQ). Using the present data the combined SEQ score resulted in 67% correct classification for rate, which was similar to using the four individual sensory dimensions.

Sheep meat consumption tends to vary widely between countries and ethnic groups. Therefore the importance of demographic effects as a potential source of bias in sensory panels was investigated. This analysis concluded that socio economic factors did not appear to be a source of bias for any of the sensory dimensions when prepared by grill or roast techniques. In addition, analysis of consumer sensory traits found large differences between tasters, although even in the presence of these large taster effects, animal effects were highly significant for grills, although less significant for roasts.

Lamb Indicator Cuts for Sensory Testing

Given the high cost of sensory testing and the constraint on muscle size in sheep carcasses the use of indicator cuts was investigated. This problem was investigated using several data sets

(a) The data for the 5 muscles which were both grilled and roasted was used to identify the sub-set of muscles which gave similar discrimination between fixed effects as the full set of 5 muscles. The results showed that the LD was the best indicator cut, followed by the SV (particularly for discrimination on cooking effects). Given the dominance of LD as an indicator muscle this same analysis was re-run with the LD excluded. The results showed that in the absence of the LD the SV was the next best indicator cut.

(b) Data from the 6 muscles that were grilled were also subjected to the same analysis. Again the LD and SV proved to be useful indicator cuts for the age class effect.

(c) Data from the 6 muscles that were roasted were subjected to the discriminant analysis. Because treatment effects were less apparent in the roast data the usefulness of any cuts as indicators was reduced with the LD identified as a useful discriminator of age class effects.

Correlations were calculated as error correlations were calculated on mean sample scores after adjustment for the main design variables. These error correlations between muscles were generally low suggesting that after adjustment for treatment effects changes in one muscle will only effect a small change in other muscles.

The relationship between sensory and objective measurements was examined for the LD muscle. The results showed a curvilinear relationship between peak force and grilled SEQ scores. The relationship between sensory and objective was much weaker for roasted samples.​​​​