Eli
Vibeke Olsen, DMRI

To
estimate the content of some substances you need a measuring method, but very
often the substances can only very hardly be measured directly. Instead, you can
measure some correlated substances or characteristics.

However,
you still need to measure the substance of interest to be able to calibrate your
indirect method. If your methods do not produce an outcome, which is known
generally and comparable with other methods, you cannot use the results as
information to the surrounding world. If the substance is the amount of lean
meat – or other characteristics related to meat like water holding capacity,
tenderness etc – a lot of technical measuring problems arise. The aim of this
text is to suggest a standard to document measuring methods within pig carcass
classification. However, the postulate is that the principles can be expanded to
nearly all types of measuring methods within meat technology.

The
proposal is established as a conclusion based on the experiences from the
EUPIGCLASS project.

*Uncertainty
*

Test
results obtained from measurements using equipment will always be more or less
defective. In general tests performed on presumably identical materials in
presumably identical circumstances do not yield identical results. This
un-avoidable uncertainty is influenced by several factors and only some of them
can be controlled.

The
uncertainty or error can either be systematic or random. In this context
systematic error (bias) is the possible deviation from a “true” value taken
the arithmetic mean of a large number of test results.

*Accuracy –
definition (International standards: ISO 5725 / GUM)
*

For
many years these problems have been studied for chemical laboratory methods or
similar methods in the laboratory. The International Organization for
Standardization (ISO) has carried out the so-called ISO 5725 standard: Accuracy
(trueness and precision) of measurement methods and results (1995).

The
basic definitions can be explained by a statistical model, which describe the
test results obtained from a measurement performed on a sample item having a
known true value:

Y_{ij}
- m
= d
+ B_{i} + e_{ij}

m
is the “true” value,

d
express possible deviation from the true value

B_{i
}express the influence from uncontrolled factors like: operator,
equipment, calibration, environment (temperature, humidity, air pollution,
etc.), time elapsed

- this random contribution is assumed to follow a Gauss distribution N(0,s_{L}^{2})

e_{ij}
express unexplained error attached tests performed on identical materials
in identical circumstances (same method, same operator, same equipment within
short intervals of time) – this random error is assumed to follow a Gauss
distribution N(0,s_{r}^{2}) independently of B_{i}.

The **trueness**
is the closeness of agreement between a test result and the accepted reference
value, i.e. in statistical terms we want to test the hypothesis

H_{0}:
d=0

This
test will be performed using the **reproducibility
standard deviation ****s _{R}**,
which is defined by the square root of

**s ^{2}_{R
} = s^{2}_{L}
+ s^{2}_{r}
**

and **s _{r}**
is called the

In
general, B_{i} can be decomposed in several different factors inclusive
interactions between the factors.

The
Guide to the expression of Uncertainty in Measurement (GUM) focus on this
decomposition. The aim is to evaluate the importance of different factors, which
can be used to monitor the method in detail or possibly to improve the methods
most efficiently. The method uses the “law of propagation of uncertainty”
i.e. express the **reproducibility variance**
as a **sum of uncertainties** attached
the influencing factors. However, it is a big problem is to estimate all
contributions - actually, from a “GUM point of view” the best guess made by
an expert can be accepted as an estimate.

*Accuracy –
problems, when measuring meat characteristics
*

*“Identical test
items”
*

A *general* problem when measuring characteristics attached pig
carcasses and pig meat is to obtain identical test items. The time of sampling
is important because the meat will change in a time interval of at least 24
hours after killing. The changes are caused by the glycolysis and other
biochemical processes dependent on the energy level just before killing and the
environment at the slaughterhouse – the chilling for instance.

If the
measuring method includes the whole carcass it is impossible to obtain identical
items – perhaps cloning can solve this problem. If only a half carcass is used,
the left and right side are obvious duplicates. However, pigs are not completely
symmetric and probably, the measuring method and/or the manual handling can
depend on the side.

The
appearance of the whole carcass depends on the slaughtering process – in
Denmark for instance the surface of the skin is processed very intensively to be
used as crackling pork rind.

Repeated
measurements using methods based on insertion of probes into a pig carcass are
impossible. A second measurement at the same site will differ because the probe
has cut the muscle fibres in the first measurement.

*Time intervals
*

The
dissection method is time consuming, but the time used to dissect one half
carcass differ from one butcher to another, which can imply different weight
loss. Furthermore, only one dissection per day implies that repeated measurement
on identical materials within a short period of time is impossible.

*Accepted reference
value
*

The
ISO standard only concerns certified reference materials or materials whose
properties have been established by measurements using an alternative method
whose bias is known to be negligible. Possible uncertainty attached to the
reference is neglected too.

In pig
carcass classification the dissection method is the “accepted reference method”,
even though the “quality” of the reference was unknown before the EUPIGCLASS
project.

**Accuracy
in pig carcass classification
**

There
are two main problems:

*How
accurate is the reference? Is it reproducible from one country to another?
*

*How
accurate are on-line measurements? Is it reproducible from one abattoir to
another?
*

It is
obvious to try to use the ISO standard to answer the questions above. However
some modifications must be considered. By way of introduction the definition of **repeatability**
will be considered. The ISO standard define the concept by a standard deviation
describing the distribution of replicates under certain conditions. Another
point of view is to consider the repeatability by the **correlation
between replicates**, which has the advantage of being without unit. This
estimate can be interpreted as a “signal to noise” estimate i.e. it is
possible to evaluate the noise attached to the measuring method in relation to
the phenomenon, we want to describe. In other contexts this correlation is
described as the **reliability** of the
method. It is proposed to use this term in pig carcass classification.