|Compilers' Toolbox™||- provided by Danish Food Informatics|
|Value - the content of a component in a food|
In the table metaphor (see the Main Entities), the value holds the actual numercal information about a certain component in a certain food. There are several kinds of attributes or metadata connected to a value, like unit, matrix unit, how the value was derived (value type), etc. To ensure proper use of the data, it is important that the food composition data compiler records this information very precisely.
Furthermore, there are several numerical restrictions on how a certain value can be presented.
The value documentation concerns all the metadata refering to and describing how the numerical value was derived. These issues concern units, how the value was derived (analytical, calculated, estimated, assumed zero, etc.), indication of data quality, etc. Other issues concern where the data come from (see also Reference).
In the EuroFIR context, the mandatory ducumentation that must follow a value is a component identifier, unit/matrix unit, value type, method type, acquisition type and reference.
In the FAO/INFOODS context, component identifiers (tagnames) include some of the information above, e.g. some analytical information.
For more information on the documentation of values, see Value documentation.
|The uncertainty of a value determine how the value should be presented - significant digits|
It is often a point of discussion how many decimals a certain component value should be presented with, and very often a decision is taken to use a fixed number of decimals for a certain nutrient. This is however a very dubious approach and from a scientific point of view not recommendable.
How precisely a numerical value can be presented depends on the uncertainty with which the value has been produced. The uncertainty of a component value depends among other things on the natural variation of the content of the component in food, sampling issues like representativity and number of samples, and analytical uncertainty. Summed up, the realtive uncertainty of a specific value can easily amount to 20-30%, which leaves us with values that correctly only can be represented with two significant digits at the most.
In short, the usual uncertainty of compositional values predicts
that no more than three significant digits. What this means in terms
of the numerical representation and the number of decimals is shown
in the following table:
* Round off to the nearest 10.
** Change unit, if there are many values at this level.
Please note that the rules given in the table above are practical and pragmatic rules and not strictly scientific. For example, a value of 1150 should with 3 significant digits correctly be shown as 1.15*103 (scientific notation). This is however not very readable, especially in a food composition table or database.
Instead the rounding of of the value to the nearest 10 is a pragmatic and practical solution.
Also note, that a 0 (zero) should never be given as a 0.0, 0.00 nor 0.000, which is often seen. A zero should be given as a 0. Similarly, a missing value should not be given or represented by a zero (0). Although it is frequently seen.
For more information on significant digits and numerical representation in numerical databases, see the page on Significant digits.