To Flip or not to Flip

Several decades ago, when I was a postgraduate student, I was confronted with the wellknown pitfalls of relying too heavily on coins when assigning dates to archaeological loci/contexts. I was trying to make sense of the stratigraphy of the Byzantine period at the Transjordanian site of Tell Hesban (now transliterated Tall Hisban) and in the process encountered numerous coins with findspots accurately recorded. Several archaeologists had been very generous with their use of numismatic evidence to narrow the chronological horizons in multiple excavations in Transjordan. But even in the 1970s and 1980s it was commonly accepted that the obvious chronological use of a coin was to provide a secure terminus post quem date. The reason is obvious. Numismatists can often identify the mint date of a coin if it is sufficiently legible, and this date can at times be even a particular time of a given year. It is not difficult to see how such an accurately datable tool at times became misused in a field of study that was more or less obsessed with chronology. But the date of the placement of a coin must be determined by the locus/ context where it is found; the coin does not date that locus/context. Since it is often difficult in the process of excavation to know whether the date of a coin coincides with the date of the locus/context where it was found, one must question whether it is ever safe to use coins to verify dates other than a terminus post quem. This leads to the question whether it can be shown objectively that coins provide an acceptable or unacceptable means of dating their findspots. In an attempt to answer this question, the numismatic evidence from two sites—Tell Hesban 1968–1976 and Pella 1967—were subjected to a simple statistical test, the Binomial Probability Calculator test. This test conducted in collaboration with my colleague, James K. Brower, provided a more objective measure of the reliability of using numismatic evidence as a safe indicator of findspot dating. There are many situations where a test has only two possible outcomes, true or false (yes or no). For example, in flipping a coin one would expect one of two outcomes—heads, or tails. Or, given a coin that has been excavated; it can either date its findspot accurately, or not date it accurately. Associated with such situations is a ratio which represents the expected frequencies of occurrence of each of these possible outcomes. In other words, under a given theory of operation for the situation under study, one outcome will be expected a certain percentage of the time, while the rest of the time the other outcome will be expected. This is the binomial probability. The reliability of coins as a means of dating their findspot could be determined absolutely only by excavating all coins at all sites and then comparing their mint dates with the dates of the findspot locus/context as established by other means and tallying the results. This kind of an examination of the total population of coins is, of course, impossible, and therefore an alternate method must be followed. All of the Tell Hesban data had previously been digitized and it was a relatively simple task to get a printout of the legible coins that had been excavated in the five seasons of work at the site. There were 256 legible coins. Dating their findspots by alternate means, primarily