Biology, images, analysis, design...
Use/Abuse Principles How To Related
"It has long been an axiom of mine that the little things are infinitely the most important" (Sherlock Holmes)

Search this site



To clarify the terminology

If you are uncertain what the difference are between intra-class and inter-class variation, repeatability and reproducibility, a few words of explanation may be worthwhile.

  • All of these statistics are used to quantify measurement variation by replicating observations upon the same item, or the same set of items. These 'items' may be patients or specimens. In either case, the point of interest is not the variation between these items, but variation among observations made upon the same items.

    Given which, if, for practical reasons, a specimen has to be subdivided into a set of aliquots, steps should be taken to ensure each set of aliquots is identical - at least as far as is humanly possible. Similarly, if observations are repeatedly made upon the same item, they should be made within the shortest time possible, or the item preserved in some fashion.

  • The second important point to understand is that, although there are innumerable ways in which these observations could be made, none of these measures (or their statistical models) assume a particular process is used. In effect therefore, it does not matter whether the observations are made by observers, clinicians, clinicians using microscopes, laboratory assays, or different laboratories. What is of interest is how much the results of those observational processes can be expected to vary - when made upon the same item, or items.

  • Assuming observations are being made upon identical sets of items, in principle, there are three distinct sources of variation:

    1. Variation between different observers, machines, assays, or labs - at one point in time.
    2. Variation inherent to an observer, machine, assay, or laboratory - at one point in time.
    3. Variation among the same observers, machines, or labs over time.

    In practice, of course, these sources of variation are liable to be additive - and to interact. Therefore the crucial point of interest is to what extent these three sources are seen to apply.

    • Intra-class variation measures the precision inherent within a specific assay. In other words, a set of identical aliquots are analyzed simultaneously on the same assay plate - or using the same device. (Source ii)

    • Inter-class variation quantifies the agreement between different assays, or laboratories - either simultaneously, or over time. (Sources i & ii, or i, ii & iii)

    • Repeatability describes the variation of the same observer over time. (Sources ii & iii)

    • Reproducibility quantifies the agreement between different observers, either simultaneously, or over time. (Sources i & ii, or i, ii & iii)

  • Lastly, although these measures assume the differences between items are not of interest, to replicate an estimate of any of these measures the same procedure is often applied to a number of items - and the results pooled in some way. Notice that, in pooling these results, we are assuming observations made on different items are equally variable.