Understanding the ancient Celts and their art: pre-Roman Celtic Society, part eight

"There are three kinds of lies: lies, damned lies, and statistics". Mark Twain said it, but attributed it to Disraeli. Exhaustive research has recently offered a more likely candidate for its origin.

Celtic numismatics is not kindly to statistics: the biggest problem being that statistics is its most honest when applied to very large numbers and when applying statistics obtained through large numbers to a specific case, or a case consisting of very small numbers you might just as well ask a psychic or toss a coin.  Any doctor who applies statistics to your personal health should be replaced immediately.

Algernon: The doctors found out that Bunbury could not live... so Bunbury died.
.
Lady Bracknell:  He seems to have had great confidence in the opinion of his physicians. I am glad, however, that he made up his mind at the last to some definite course of action, and acted under proper medical advice. 

Oscar Wilde, The Importance of Being Earnest.

The first time I had a big problem with statistics in Celtic numismatics it was to do with an alleged reduction of the silver content by about 2% for several classes of Coriosolite coins. This was flawed on more than one front: the modern classification system; the belief that 2% had any meaning to the moneyers of that time and place; the difficulty in applying it if it did; and the failure to consider Gresham's Law.

Colin Haselgrove, who is aware of the pitfalls surrounding insufficient data suggests that if applying statistics does not deliver a useful product then the question should be reframed to be able to include much larger numbers. (Iron Age coinage and archaeology, BAR, British series No 222, 1992, p. 126). Sound advice. Of course, in doing so, specificity must suffer and the results might not be as useful as was first hoped.

I have found that looking at clustering patterns is often  a better approach than the statistical method especially when the primary data is flawed.Through this method, I was able to determine that the primary action in decreasing the intrinsic value of British L (Whaddon Chase) gold staters was by increasing the copper content because the types clustered far better through this method than by sorting them by decreases in the gold, or increases in the silver content. Although I found errors in the dataset I was using, these errors were of a wide variety of types and could not have delivered such a focused result as that which I found through looking at the clustering patterns. It was just that I could present any absolute numbers with any degree of certainty.

Clustering patterns are fairly useless, however, when you have very small numbers. Coin types consisting of only a few specimens such as less than about 80 (a number I found to be valid with Coriosolite hoards divided into six classes) do not present valid or useful data. Such a hoard of only 25 coins will be utterly useless. Looking for information about distribution patterns from coin types or series with only a few known find spots is pointless as a number of the coins could easily have entered the broader area as a single lot and were them subsequently diffused over a wider area. About all you could tell with about 80 examples (if you are lucky) is the approximate location of their arrival to that region. A far more reliable method is to look for clustering patterns in the artistic motifs using a far wider collection of types. If you find that the motifs do not exist, at all, in the region of the finds, then it is far more likely that the coins were given to a person in the region but that they originated elsewhere, This can also be checked by looking at the type of alloy used: if the alloy type matched the focus of motifs on the larger set then it is virtually certain that the finds are "outliers".

I see frequent reattributions being made because of find spot data. Even worse, the reasons for reattribution often do not travel with the reattribution. With local products such as brooches, these can be valid (if there are enough examples), but with coins, and with rare examples of high status goods, you might just as well contact that psychic...

More on Monday. Have a statistic-free weekend.


John's Coydog Community page