Paul Dion asks for information regarding more "objective", i.e. laboratory analysis, and less "subjective" organoleptic analysis.

He should be more exact in the use of terms. The correct term for tests that uses human senses is 'sensory', not 'organoleptic'. 'Organoleptic' refers to the properties of a material capable of stimulating the senses. Sensory tests can be objective or subjective. (Ref: ISO 5492:1992, Glossary of terms relating to sensory analysis of food). An objective test should be used for grading for quality or for measuring a defined property.

There are no existing laboratory tests that can in principle or in practice be used for grading the quality of fish at the dockside. (See: Olafsdottir, G. et al., 1998, Methods to determine the freshness of fish in research and industry. Proceedings of the final meeting of the concerted action "Evaluation of fish freshness" AIR3CT94 2283, International Institute of Refrigeration, Paris. ISBN 2-9033633959). I believe it is extremely unlikely that a non-sensory test will be developed that will outperform sensory assessment for grading of raw fish. Sensory tests for assessing the 'quality' of raw fish, when properly carried out, are objective, rapid- about 15 seconds a fish, non-destructive, and highly associated with the judgements consumers use for judging quality. What more do you want? Throughout the world, fish is bought at docksides or landing sites on the basis of sensory evaluation, and I am not aware of any marketing system that uses non-sensory methods. At its simplest, the grading is just buy or
not buy, or to be more exact, buy or not buy at the particular price on offer. What Paul is referring to is a more sophisticated system in which fish are graded according to some criteria of quality, and I assume the price paid will depend on the quality grade. There are a variety of difficulties in grading fish, some of which are dependent on the properties of the product and the objectives of the task, and are quite independent of the measurement procedure being used; some do depend on the procedure. There are further difficulties associated with sampling and decision rules if 100% testing is not being used.

For a start, fish do not reside in a quality grade then jump completely into the next grade. Whatever the criteria being used there will be a continuum in these criteria. Take freshness for example. When a fish is first caught it will be in grade A for freshness, and as it is stored it will become grade B. Somewhere along the line it will be exactly on the borderline. In a batch of fish caught at the same time and handled identically at some point in the storage, there will be some variation in the rate of loss of freshness and there will be a time when half the fish are in grade A and half in grade B. This will be the case whatever measuring system is used. A perfect instrument will allocate half the fish to A and half to B. What grade is the batch in? The problem remains even if every fish is tested. When the fish are at the boundary of the criterion, the instrument will randomly allocate them to grades A and B. What is more, if the same set of fish are tested again, half the of them will be allocated different grades from the first time round. As a batch of fish approaches the boundary between grades, a increasing proportion of the fish will be in the lower, less fresh, grade. If the grades are very narrow compared with the natural variation in properties, fish within the batch might even be in more than two grades. This is not a consideration with 100% testing, but presents a difficulty when sampling. What proportion of out-of-grade fish can the quality controller accept in a nominated grade, and how does the sampling scheme and decision rules predict this proportion, i.e what are the operating characteristics of the sampling scheme.

The measurement procedure acts to increase the variation in allocation of grade over that imposed by natural variation. This is true whether a sensory or a non-sensory procedure is used. If the measuring method measures the properties that define the grades, which is the usual approach, then the error is the error of the method. This will include the within instrument error and the between instrument error. It is a general assumption, which I accept, that these errors are lower for non-sensory methods that for sensory methods. However, quality grades are typically not defined in the parameters measured by the non-sensory methods, and the non-sensory method is used as a predictor of the sensory outcome. The prediction error tends to be high and cannot be less than the error in the sensory method. The calibration equation predicts the mean value of the spread of sensory results at that value of the predictor variable, plus the error of the prediction equation.

Going back to the example, H&G fish, up to 5 days in ice, packed in 70l boxes. Say, there is 50kg of fish in a box, then a consignment of 100 tonnes would contain 20 000 boxes. If on average the fish weigh 1 kg, there are 100 000 fish in the consignment. Grading this amount of fish is quite an undertaking. Paul does not expand on the criteria for quality, but reference to date stamping suggest the criteria will be weighted towards properties that changes with storage time. At most the fish will span 5 days of storage - not allowing for time taken to/from the fishing ground. How may grades does Paul expect there to be in this span? Even with just two grades, the natural variability among fish will mean that most boxes in the consignment will contain fish in both grades. Is there to be 100% testing or will the QA system allow sampling? If so what will be the sampling plan? All this before the measurement procedure is decided on.

I would suggest that Paul first evaluates the objectives of the grading scheme, the natural variability in the criteria being used in define the grades, and the desired accuracy of the grading system. He might then decide that it is pointless trying to grade fish as fresh as he has described. At this level of quality, and over this short range, if quality
differentiation is required, the QA system should insist that boxes are tagged with batch codes, a batch being the fish from the one haul.