Until relatively recently, radiocarbon dating of wooden objects was the only known scientific method of dating wooden objects. Although in general it was always successful, dated produced would have a range of plus or minus 20 years at best, and at worst could span two centuries or more. And this only dated the actual rings sampled for C14 analysis; the tree may well have continued to live for decades or even a century or more afterwards. In the 1970’s a new scientific method of dating buildings was developed in this country - dendrochronology. Here was the answer to the building historian’s dream, an absolute dating process accurate to a single year, if not the season the tree was felled. The potential of this in studying the development of timber-framing was quickly realised by one of the early pioneers of the science, Dr John Fletcher, who began to investigate the medieval buildings of the Oxford region between 1974 and 1986.
The way dendrochronology works is relatively simple. As a tree grows, it puts on a new growth or tree-ring every year, just under the bark. Trees grow, and put on tree-rings, at different rates according to the weather in any given year: a wider ring in a favourable year and a narrower ring in an unfavourable year. Thus, over a long period of time (say 60 years or more) there will be a corresponding sequence of tree-rings giving a pattern of wider and narrower rings which reflect droughts, cold summers, etc. In effect, the span of years during which a tree has lived will be represented by a unique fingerprint, which can be detected in other geographically-similar tree-ring chronologies.
To obtain this fingerprint, a radial section of timber from the pith or centre of the tree out to the bark edge is required (see sampling procedures). Once prepared, they are then measured under a x10/x45 microscope using a travelling stage electronically displaying displacement to a precision of 0.01mm. Thus each ring or year is represented by its measurement which is arranged as a series of ring-width indices within a data set, with the earliest ring being placed at the beginning of the series, and the latest or outermost ring concluding the data set.
Whilst the principle behind tree-ring dating is a simple one, the determination of what is an actual match is much more involved. When an undated sample or site sequence is compared against a dated sequence, known as a reference chronology, an indication of how good the match is must be determined. Although it is almost impossible to define a visual match, computer comparisons can be accurately quantified. Whilst it may not be the best statistical indicator, Student’s (a pseudonym for W S Gosset) t-value has been widely used amongst British dendrochronologists. The cross-correlation algorithms most commonly used and published are derived from Baillie and Pilcher’s CROS programme (Baillie and Pilcher 1973).
Statistically, t-values over 3.5 should be considered to be significant, although in reality it is common to find demonstrably spurious t-values of 4 and 5 because more than one matching position is indicated. For this reason, dendrochronologists prefer to see some t-value ranges of 5, 6, or higher, and for these to be well replicated from different, independent chronologies with local and regional chronologies well represented. To give some idea of how good a match can be expected, two timbers from the same parent tree will often give a t-value of 10 or higher. Users of dates also need to assess their validity critically. They should not have great faith in a date supported by a handful of t-values of 3’s with one or two 4’s, nor should they be entirely satisfied with a single high match of 5 or 6. Examples of spurious t-values in excess of 7 have been noted, so it is essential that matches with reference chronologies are well replicated, and that this is confirmed with visual matches between the two graphs.
In reality, the probability of a particular date being valid is itself a statistical measure depending on the t-values. Consideration must also be given to the length of the sequence being dated as well as those of the reference chronologies. A sample with 30 or 40 years growth is likely to match with high t-values at varying positions, whereas a sample with 100 consecutive rings is much more likely to match significantly at only one unique position. Samples with ring counts as low as 50 may occasionally be dated, but only if the matches are very strong, clear and well replicated, with no other significant matching positions. Here, it is essential for intra-site matching when dealing with such short sequences. Consideration should also be given to evaluating the reference chronology against which the samples have been matched: those with well-replicated components which are geographically near to the sampling site are given more weight than an individual site or sample from the opposite end of the state.
It is general practice to cross-match samples from within the same phase to each other first, combining them into a site master, before comparing with the reference chronologies. This has the advantage of averaging out the ‘noise’ of individual trees and is much more likely to obtain higher t-values and stronger visual matches. After measurement, the ring-width series for each sample was plotted as a graph of width against year on log-linear graph paper or similar graphic display. The graphs or curves for each of the samples in the phase under study are then compared visually at the positions indicated by the computer matching and, if found satisfactory and consistent, are averaged to form a mean curve for the site or phase. This mean curve and any unmatched individual sequences are compared against dated reference chronologies to obtain an absolute calendar date for each sequence. Sometimes, especially in urban situations, timbers may have come from different sources and fail to match each other, thus making the compilation of a site master difficult. In this situation samples must then be compared individually with the reference chronologies.
Therefore, when cross-matching samples between each other, or against reference chronologies, a combination of both visual matching and a process of qualified statistical comparison by computer is used. The ring-width series were compared on an IBM compatible computer for statistical cross-matching using a variant of the Belfast CROS program (Baillie and Pilcher 1973). A version of this and other programmes were written in BASIC by D Haddon-Reece, and latterly re-written in Microsoft Visual Basic by M R Allwright and P A Parker.
Of course, a precise felling date is only achievable when the last growth ring beneath the bark is present. However, when some of the outer sapwood rings have been removed through decay or conversion, it is still possible to ascribe a felling-date range within a 95% confidence limit. For the south, a figure of 9-41 is used, while 11-41 is used in Wales and the border counties and 12-45 for the north of the country. This is discussed more fully in the Interpretation, Presentation, and Use of Tree-Ring Dates, and is currently the subject of a doctoral thesis here at Oxford.
When even some of the outer heartwood rings as well as the sapwood has been removed from the sample, then only a terminus post quem or ‘felled after’ date can be given. This is the least useful interpretation of a dated sample and is calculated by adding the minimum number of sapwood rings to the last measured or counted heartwood ring.
It should be noted that, for a variety of reasons, tree-ring dating can not always be successfully applied. For example, timbers may be too fast-grown with fewer, complacent rings, or from slower grown trees with distorted ring sequences due to pollarding or other damage. In neither case are they generally datable. Most historic roofs and timber-framed buildings were made from oak, which is the most suitable species for successful tree-ring analysis. Although elm and beech can also be dated and matched with the oak chronologies, elm in particular can often contain distorted and fast-grown rings making routine dating difficult.
Finally, dendrochronology only gives us the date or date-range when the tree was felled. However, we know from numerous documentary sources that oak was used ‘green’, or unseasoned. This means that construction probably took place in the year of felling or within a year or two thereafter. If a group of five or ten timbers from a building, preferably of different types, all were felled within the same year, then it is most likely that the actual construction of the building would have taken place during that or the following year.