In my previous post on Fritts and Bradley, I observed that Bradley’s so-called ”seminal” textbook had copied 12 of the first 13 figures in its dendro chapter from Fritts 1976, together with verbatim or near-verbatim caption (with a little more examining, this is now 17 of the first 19 figures in the textbook.) By focusing on the figures in this post, I didn’t mean to suggest that the comparison of running text in the two textbooks did not provide anything worth discussing. On the contrary. There is much of interest in this comparison, some of which I’ll discuss today.
I started with the documentation of the extensive copying of figures for two reasons.
First, the figures are relatively easy to match between the two texts and provide a sort of road map for comparing running text, since both textbooks comment on the figures. I’ll discuss further results on running text today.
Second, I was a little surprised by the sheer extent of Bradley’s re-use of the graphics from Fritts’ text. In saying this, I’m not moralizing or offering an opinion as I’m not familiar with conventions of textbook publishing and do not plan to offer an opinion without examining such conventions. If nothing else, the sheer extent of Bradley’s re-use of Fritts’ material repudiates Deep Climate’s assertion that Bradley’s textbook provided a “seminal” description of tree rings– an assertion that Deep Climate and others should withdraw.
Bradley’s re-use of Fritts’ material is not limited to his re-use of Fritts’ figures and captions. While I haven’t yet commented on the running text, given that much of Bradley’s dendro chapter comments on the 17 figures copied from Fritts 1976, Bradley’s language inevitably tracks Fritts’ to a varying degree. Here is the sort of parallel that one sees:
Bradley 1985, 346. Once the regression coefficients have been calculated, the eigenvectors incorporated in the regression equation are mathematically transformed into a new set of n coefficients corresponding to the original (intercorrelated) set of n variables. These new coefficients are termed weights or elements of the response function and are analogous to the stepwise regression coefficients discussed earlier
Fritts 353. Once the regression coefficients for the selected set of orthogonal variables have been calculated, they may be mathematically transformed into a new set of coefficients which correspond to the original correlated set of variables. These new coefficients (sometimes referred to as weights or elements of the response function) are analogous to the stepwise regression coefficients described in the previous section…
In addition, you can also distinguish two quite distinct sentence and paragraph styles in Bradley 1985, which I’ll illustrate through the following two parts of the first paragraph of section 10.2.1:
In conventional dendroclimatological studies, where ring-width variations are the source of climatic information, trees are sampled in sites where they are under stress; commonly, this involves selection of trees that are growing close to their extreme ecological range. In such situations, climatic variations will greatly influence annual growth increments and the trees are said to be sensitive. In more beneficent situations, perhaps nearer the middle of a species range, or in a site where the tree has access to abundant groundwater, tree growth may not be noticeably influenced by climate, and this will be reflected in the low interannual variability of ring widths (Fig. 10.2). Such tree rings are said to be complacent. There is thus a spectrum of possible sampling situations, ranging from those where trees are extremely sensitive to climate to those where trees are virtually unaffected by interannual climatic variations. Clearly, for useful dendroclimatic reconstructions, samples close to the sensitive end of the spectrum are favored as these would contain the strongest climatic signal.
However, it is now clear that climatic information may also be obtained from trees which are not under obvious climatic stress, providing the climatic signal common to all the samples can be successfully isolated (Lamarche, 1982). For example, ring widths of New England deciduous and coniferous trees have been used to reconstruct the history of drought in the area since ADd1700 (Cook and Jacoby 1977) and, recently, reasonably good paleoclimatic reconstructions have been achieved using Tasmanian mesic forest trees (Lamarche and Pittock 1983). For isotope dendroclimatic studies (Section 10.6), the sensitivity requirement is not critical and it would, in fact, be preferable to use complacent tree rings for analysis (Gray and Thompson, 1978).
There’s an obvious stylistic difference between the two parts of the paragraph. The first part contains no academic references, while, in the second part, all the points are referenced in academic style. (I’m not moralizing about this – I’m just observing a style distinction.
This style distinction is pervasive in the Bradley dendro chapter. Except for a few isolated subparagraphs, Bradley 1985 pages 331 to 353 are in the unannotated style. The only exceptions are the subparagraph shown above at the bottom of page 332, four sentences at the bottom of page 336, a sentence at the top of page 341, one reference at the bottom of page 343 (the other references at the bottom of page 343 derive from Fritts 1976) and the discussion of Cook and Jacoby 1979 at the top of page 351.
The distinction between referenced and unreferenced style corresponds almost exactly with whether or not the material is derived from Fritts 1976. For example, here is a section from Bradley 1985 in unannotated style:
Bradley 1985, 346. In the case of Figure 10.11a, the regression equation with only one variable (amplitudes of eigenvector 1) accounts for 36% (R2100) of the variance of ring-width indices during the period of instrumental records. This first eigenvector represents a climatic condition in which tree growth is associated with below average temperatures in all months leading up to and including those in the growth season, and, above average precipitation in all months. Note that the 95% confidence limits on these weights are small since they are based on only one variable. Figure 10.11b shows the response function weights resulting from an equation utilizing three eigenvector variables; these account for 67% of the tree growth variance. Using this equation, ring width indices are inversely related to temperature in most months, but positively correlated with precipitation, May of the growth year and September of the previous year are the only two months for which temperature is positively and significantly related to growth.
Sure enough, there is a very closely matching subparagraph in Fritts:
Fritts, 366. The regression coefficient is multiplied by the weights of eigenvector 1 to obtain a response function that accounts for 36% of the growth variance (Fig 7.13). This response function can be interpreted as representing the condition of yearly climate which leads to above average growth, characterized by below average temperature for any month and above average precipitation for any month. The vertical line in Fig 7.13 indicate the 0.95 confidence intervals used to test variable significance. … The second response function shown in Fig 7.13 is the result of using stepwise multiple regression analysis on the three most significant regression coefficients. … This response function assumes a more complicated shape and the variance reduced is 67% of the total growth variance. It can be interpreted by the significant elements to indicate that temperatures in May are the only temperature variables that are directly related to growth while temperatures in seven of 14 months are significant and inversely related to growth. Variables for precipitation are significant and directly related to growth in nine of the 14 months. Precipitation for the July prior to growth is the only precipitation variable that is inversely related to growth at this step.
Similar instances abound. Here the parallel is obvious. In other cases, Bradley’s paraphrase is more thorough, but you can track what Mosher calls “high entropy” words through the paraphrase – words like “polynomial” or “eigenvector” or even “discarded”, which, in combination with the figures, provide a quite precise to locating the Fritts text from which the corresponding Bradley text has been derived.
Despite the obvious parallel in the sentences shown above, there is no direct reference to Fritts 1976 associated with these sentences. In fact, Fritts 1976 is mentioned only four times in the 24 pages of running text of Bradley 1985 pages 330 to 353 and only once in the running text Bradley 1999 – although, as noted in my previous post, Fritts 1976 is mentioned in the captions to seven figures in Bradley 1985 (Bradley 1999 – four).
The following three references in Bradley 1985 to Fritts 1976 were removed in Bradley 1999:
Although much work has been carried out since these early pioneering studies, the greatest strides in dendroclimatology have been made in the last 10-15 years, largely as a result of the work of HC Fritts and associates at the laboratory of Tree Ring Research at the University of Arizona, Tucson; much of this work has been documented at length in the excellent book by Fritts (1976). 330.
In practice, additional coefficients are not included unless they reduce the variance of the ring width data by at least a further 5% (Fritts 1976) though this cut-off point is quite arbitrary.
[in section 10.2.4 Calibration] For a more exhaustive treatment of the statistics involved and more examples of how they have been used, the reader is referred to Fritts (1976 chapter 7).
The sole surviving reference to Fritts 1976 in the running text of Bradley 1999 is:
From the point of view of paleoclimatology, it is perhaps useful to consider the tree as a filter or transducer which, through various physiological processes, converts a given climatic input signal into a certain ring width output that is stored and can be studied in detail, even thousands of years later (e.g. Fritts, 1976; Schweingruber, 1988, 1996).
As readers are aware, the text of Wegman section 2.1 Background on Paleoclimate Temperature Reconstruction is derivative from Bradley 1999. (As noted previously, this section does not mention or discuss the papers that Wegman was asked to review, MBH98-99 and the McMc critiques.) It is possible that Bradley’s failure to clearly reference Fritts 1976 resulted in Wegman not being aware of the actual amount of dependence of Bradley 1999 on its predecessor. As compared to the single mention of Fritts 1976 in the running text of Bradley 1999 in an incidental context (down from four mentions in more extended use of Fritts 1976 in the 23 relevant pages in Bradley 1985), Wegman mentioned Bradley 1999 three times in his running text of only a couple of pages.
Table 1 based on Bradley (1999) illustrates the wide variety of these natural phenomena that may be used as proxies. Some proxies measure very low frequency (slowly varying) climatic variables and thus are not useful for measuring average annual temperature changes. Table 2 found in Bradley (1999), which was reproduced from Bradley and Eddy
(1991) summarizes a variety of proxies and also indicates their minimum sampling interval as well as the range of years for which they could reasonably be used for temperature reconstruction.
See Bradley (1999) for a discussion of the fitting and calibration process for dendritic-based temperature reconstruction
Reader John McManus observes:
Bradley , as you have proved, will never have to worry about being falsely accused of plagiarism.
A point on which other readers may agree, though not necessarily for the same reasons.
His other point – that I’ve “documented [Bradley's] meticulous adherence to scholarly convention” is one that would be undoubtedly be very reassuring to Wegman. If Bradley 1985 (and Bradley 1999) can be taken to represent “community standards”, these standards do not seem to preclude the referencing practices in the Wegman Report that have been criticized both here and elsewhere.