The Print Clock
A method for dating early books and prints
S. Blair Hedges, Pennsylvania State University
The print clock is a method for dating art prints and old books produced over the centuries. It is described elsewhere in two technical publications (Hedges, S. B, 2006, 2008a) and one review article (Hedges, 2008b). The purpose of this web site is to provide a general overview along with practical details for those who wish to use the method.
An early book printing shop illustrated in a copperplate print by Stradanus (1590). Copyright S. Blair Hedges.
The print clock is a method for dating undated books and prints that were produced by hand-operated presses, thus since the 15th Century. It is based on the hypothesis, backed by observations, that woodblocks and metal plates (e.g., copperplates) deteriorated in a clock-like manner during their lifespan, which was often decades. In woodblocks, the deterioration occurred mainly with aging and random cracking of the wood, resulting in line breaks in the prints. In copperplates, grooves in the metal became thinner as the surface was eroded, resulting in thinner lines in prints. The surface erosion appears to have been from the routine polishing of the plates that occurred before each print run, to remove nicks and corrosion. The time-dependency in copperplates is probably related to corrosion during storage. By measuring this change across different editions of the same print, and calibrating the rate with dated editions, it is possible to estimate the age of the undated work (see figures below).
3-D figure (left): The print clock. In these models, close-ups of a carved woodblock (above) and engraved copperplate (below) are shown, corresponding to a curved black line on the print behind it, at two time periods. Over time, cracks develop in the woodblock resulting in line breaks on prints. With copperplates, erosion of the surface produces thinner groves and thinner lines in prints. For both, the change is time-dependent. (after Hedges, 2006). Download high resolution image here (2000x3767) for educational purposes.
The print clock could be classified as a "probabilistic clock," a group that includes the radiometric clock of geologists used to date rocks and molecular clock of biologists used to date species divergences. The atomic clock used to measure global time to an accuracy of one second in 60 million years is different. It is a regular (deterministic) clock based on cyclic events at the molecular level (cesium atoms), and in that sense is similar to a mechanical clock or watch.
But how would random events such as the cracking of wood lead to clock-like change? Certainly, cracks - and corresponding line breaks in prints - do not occur at a regular rate, such as one every four months, for example. Several may occur in one year and none in the next year. That is the nature of random events. Radioisotope decay is also random, yet it is used as a clock with a relatively high degree of precision. As long as the average rate is unchanged, and a large number of events is sampled, probabilistic clocks can be excellent chronometers.
A large sample size is not usually a problem for isotopic decay, or for mutations in genomes (molecular clocks), but it could be a limiting factor for dating prints if the object is small, such as a 10 mm-wide woodblock capital letter or tailpiece in an early printed book. Although copperplate corrosion does not have the same type of limitation as discreet line breaks, there are variables unrelated to the changes that occurred in individual copper atoms. For example, both the corrosion and the subsequent polishing were unlikely to be evenly distributed across the plate. Again, importance is placed on having a large sample size, which in this case would be area of the copperplate print.
The need for calibration with dated works is a limitation of the method. However, this does not mean that it is only useful for dating different editions of the same book. Early printed books typically contained artwork in the form of printer's mark (trademarks), title page ornamentation, enlarged letters, and decorative elements (headpieces and tailpieces) and these often appeared in multiple books (authored by different persons) produced by the same printer. Thus a single surviving copy of a rare book could be dated using this method, as long as other books by the same printer were available for comparison. This was demonstrated in the original article (Hedges, 2006) using the woodblock printer's mark of an undated Italian book from the 16th Century.
Figure (right): Engraving a copperplate. A steel burin (2 mm wide) carves a triangular-shaped groove (200 micrometers in width, or one-fifth of a millimeter) in a polished copperplate, pushing up a sliver of copper about the thickness of dental floss. The burrs are scraped off before printing. Download high resolution image here (1466x1299) for educational purposes. ©S. Blair Hedges.
Please refer to the original description (Hedges, 2006) and recent article on the copperplate print clock (Hedges, 2008a) for detailed explanation of the method. Evidence of time-dependency came from analysis of 2,674 Renaissance prints. The woodblock example used in that study concerns an undated book by Bordone, the Isolario. Measurements of line breaks in dated editions permitted not only the estimation of date of the undated edition (1565), but also the date that the woodblocks were carved (1518), both questions that had been debated by historians. The copperplate examples involved five editions of a book by Porcacchi, L'isole più famose del mondo (1572-1620) and three editions of a book by Magini, Geographiae (1596-1621). In both of these cases, digital image analyses showed time-dependent change, not print dependent change. High resolution scans of copperplate prints showed how etched and engraved lines changed over time, reflecting the erosion of the surface of the copperplate, clarifying the underlying mechanism of time-dependent change. Engraved lines are triangular-shaped whereas etched lines are trough-shaped. Both narrow over time, although etched lines thin more slowly because the walls of the trough are have a steeper angle than the walls of en engraved groove. But etched troughs are also shallower than engraved grooves and therefore they will disappear more suddenly, causing "line fading" when the bottom of the trough is reached.
When you flatten a ball of pizza dough it expands in width, and copper is no different. A copper penny placed on a railroad track will become flatter AND wider after it is compressed by a train. Thus the finding that lines in copperplate prints from later editions are thinner goes against the general assumption that a plate becomes compressed over time. A compressed plate would produce wider lines, not thinner lines. Although this is intuitive, it does counter the widely held assumption regarding copperplate wear and therefore I thought it would be useful to "test the obvious" with real copperplates. The results are presented here. I used a square burin to engrave a grid of lines in two small (˜12 mm square), polished copperplates of standard thickness for engraving. Using a microscope with an ocular micrometer I measured groove widths at precise locations where the lines intersected, so that measurements could be repeated. For one copperplate, I eroded (polished) the surface with fine sandpaper. For the other, I compressed the plate by applying >6000 PSI of pressure. Then I measured the grooves in each plate at the same locations as measured before the treatment. As predicted, the eroded surface had narrower lines whereas the compressed surface had wider lines (see graph below). This further demonstrates that copperplate wear, resulting in narrow lines in later editions, was caused by erosion, not compression (see also Fig. 8 in the article).
Figure (left): Graphs showing the results of two treatments applied to engraved copperplates. Dashed lines indicate the position of data points if no change occurred (1:1 slope). After erosion (polishing) of the surface, groves are thinner (below dashed line) and this corresponds to the narrowing of lines of copperplate prints from later editions. After compression, grooves are wider (above dashed line), which does not agree with what is seen in real prints. This demonstrates that the wear of copperplates was not caused by compression - the widely held assumption - but rather by erosion. Not for reproduction (these data are unpublished).
Below are some notes on the equipment and software used in the original study and recommendations for those who wish to use this method.
There are no specific requirements because the image analysis software (ImageJ; see below) will run on Mac OS, Windows, Unix, and Linux X86. Image files use up considerable hard drive space, so this should be taken into account.
Figure (left): Generalized diagram showing change in width of an engraved triangular-shaped groove in a metal (copper) plate through time (four editions), and the printed lines produced by that groove. The cross-section of the ink-filled groove is black. Before each printing, the plate surface was polished (eroded) to remove nicks and corrosion, narrowing the width of the triangular groove. After Hedges (2006). Download high resolution image here (1800x1656) for for educational purposes.
Digital SLR (single lens reflex) camera
If you will be using a flatbed scanner with the prints, skip this and see below. A digital SLR camera is recommended to get the resolution and image quality needed. The camera should probably have at least 6 megapixels, and the standard now is ~8 mp, which would work well for this purpose. The flash is not an issue because most libraries will not permit the use of a flash. But this is why an SLR (as opposed to point-and-shoot) and tripod (see below) are important, for getting a sharp image in ambient light conditions. Also, a good quality lens will often give you more light and sharpness than the "bundled" zoom lens that often comes with digital SLRs, so consider purchasing the camera body and macro lens separately. You should have an electronic remote shutter release cable or else pressing the shutter button directly will cause some blurring. Without the cable, you could get by in a pinch by simply using the timer on your camera, but that would slow you down.
A tripod is also important to reduce blurring of the image as much as possible. I recommend a portable, lightweight tripod that fits in a small handbag or your pocket. The tiny aluminum ones (4-5") that sell for $5-10 are too small and flimsy to support an SLR camera. I use a Giottos QU-200 "U-pod", which collapses to 10 inches (240 mm), extends to 14 inches (360 mm), has a full-range ball head, and weighs only 240g.
For grayscale measurements, almost any model scanner will work, but if individual engraved lines are to be measured, it should be able to scan at 4800 dpi. Most will attach to your computer with a simple USB connection. A flatbed scanner is the optimal way to generate digital images of documents, because it produces even illumination, sharply focused images, and very little spatial distortion. And if used carefully, it will not damage most old documents, although few rare book libraries will permit flat-bed scanning. The imaging services provided (at a cost) by libraries, such as "overhead scanning," vary considerably in quality among institutions and never match the quality of a flatbed scanner, although some come close. However it is not uncommon to receive highly pixelated (grainy) images that may allow you to read some text but are useless for any image analysis. Small, hand-held scanners are not permitted at most libraries and will not give you sufficient resolution; although it seems like that technology should improve in the future. Flat-bed scanners are now light and portable enough to carry to a library and set up with a laptop computer; you would need only an extension cord (and permission from the librarian!). That would be the ideal way of acquiring images.
I used ImageJ (http://rsb.info.nih.gov/ij/) for the study and recommend it. It is free and maintained by the U.S. National Institutes of Health (NIH). Just visit the site and go to "downloads." Versions are available for almost any operating system. Other image analysis packages are available, and probably could be used as well. Although images can be thresholded and measured for gray level in Photoshop, that software is not designed for image analyses.
If photographing a print within a book yourself, place the book on a standard, V-shaped cushion (e.g., foam wedges) so that it is open to the page of interest, forming a 90-degree angle, essentially as you would position it to read. This helps keep the page from bowing out because you want it flat and perpendicular to your camera lens. weighted strings and other cushioned weights can be used in to help keep the page flat. Set up your camera on the tripod, next to the book, and use the remote shutter control to acquire the images. Having the book open at an angle allows you to point the camera lens down at a 45 degree angle instead of straight down (vertically), which would be more difficult to do with a tripod. You may find that manual focus is more reliable in these situations than automated focus. Always step back and view your camera and book to make sure that the camera lens is perpendicular or else your images will be distorted.
Figure (left). Solving two historical puzzles surrounding a Renaissance book, Bordone's Isolario. The date when Bordone's woodblocks were carved and the date of an undated edition of the book have been debated by historians. The print clock indicates that the blocks were carved in 1518, a decade before they were first used, and dates the undated book to 1565. Independent evidence supports both dates. After Hedges (2006). Download high resolution image here (1800x2147) for educational purposes.
I recommend taking multiple images (including different exposures) of the overall print and of close-ups of different sections of the print; you may need the close-ups later. Images of larger areas, such as entire prints, are subject to greater differences in lighting across the image. Even if you don't see a problem in the image, it may appear in the thresholding and analysis stage where they could show up as darkened edges or regions of the image. However, this is not as much of a concern with woodblock prints, probably because they already have good contrast.
If possible, take RAW images and convert to TIFF for maximum information. They can be converted to JPEGS later, but ideally it is best to do all analyses with TIFF files.
Scan at the highest resolution possible, with respect to your storage capacity, just in case you need more information later. For grayscale, only 300 dpi may be sufficient for a large print, but for detailed line analyses you may need 2400 or 4800 dpi. In these cases, you will probably not be able to scan, at one time, areas much larger than a square inch because of the large files generated (>100 megabytes). Therefore, if it is a large print and you wish to do detailed line analyses, it is best to choose specific areas at the onset of the study and scan the same areas in each of the prints. Again, save images as TIFF files if you have the space.
The notes and comments below are not intended to be "stand alone." They presume that the reader is familiar with image analysis terms. Please consult the ImageJ website for a description of terms and features of the software. Consistency is the most important consideration: use the same settings and conversions for all images and try to use the fewest manipulations of the images.
1. Initial settings. ImageJ will be ready to use after installation, but if you are working with JPEG files at all, reset the JPEG quality from 75% (default) to 100%. To do this, go to Edit/Options/JPEG Quality. Also, if you have sufficient memory on your PC, I suggest increasing that value as well (Edit/Options/Memory).
2. TIFF or JPEG? TIFF is the best format for any digital analysis because some data are discarded using JPEG compression. However, TIFF files are much larger, sometimes creating storage and memory problems. The two types of files (TIFF and JPEG) will give different gray values for the same image, and different levels of JPEG compression will also result in different gray values. Therefore, the critical thing is to be consistent. Never compare TIFF measurements of one image with JPEG measurements of another image. If JPEGS must be used, always use JPEG-High (usually "12" if numbers are given). The highest JPEG has the least amount of compression. Never convert from JPEG to TIFF because you will not regain data that have already been lost through compression. If any files in your analysis are JPEGS, you should probably convert all to JPEGS for consistency. Also, you will probably need to save TIFF files in 8 bits/channel instead of 16/bits per channel to analyze in ImageJ and other software or else you may run into problems of compatibility.
Figure (right). Time-dependent change in copperplate prints. Close-ups of a small section (1/2 inch wide) of prints from two editions of an Italian Renaissance book by Porcacchi illustrate time-dependent image fading, useful for dating books and prints. The graph shows mean gray-level data for two prints, Cuba and Hispaniola. After Hedges (2006). Download high resolution image here (1800x2593) for educational purposes.
3. Digital enhancement of images. It is possible to alter images in many ways, to improve their appearance (e.g., sharpen, smooth, eraser, auto-contrast, brighten, subtract background, etc.). However, it is best to avoid all of these unless you know for a fact that it will not bias your analysis. Most WILL bias your analysis! For example, if one print being compared has manuscript notes on it, removing the notes would be better than leaving them. However, digitally erasing the notes, creating white blotches (erased areas) against an off-white (toned) paper color, may bias the automated thresholding. Instead, digitally cutting and pasting a region of the blank toned paper over top of the manuscript notes would eliminate that bias and be a better alternative than using the eraser.
4. Image size. The larger the image, the more precise the measurements. However, again, consistency is the most important thing. Determine the size of the smallest image that you have to compare, and resize others to that size. But remember that even resizing can bias your measurements, so make certain that aspect ratios are maintained when resizing.
5. Thresholding. Almost all color and grayscale images will need to be converted to binary images (black and white pixels) before analyses are made. This "thresholding" can be done manually, by moving the boundary between black and white pixel categories, or by using automated thresholding. For consistency, I suggest using auto-thresholding, although keep in mind that subtle differences in the background (e.g., white) or objects (e.g., black), such as the presence of water stains on the paper, can affect the boundary selected by auto-thresholding and hence all further analyses (e.g., gray level).
6. Image focus. If an image is not in focus, it will affect the gray scale measurements, and probably other measurements as well. This may seem obvious, but I mention it here as a separate item because I believe that it is one of the most significant variables that can bias your results. Flat-bed scans can virtually eliminate this potential problem, but if you are taking digital photos I strongly recommend using a tripod and remote shutter control to reduce vibrations and getting a sharp image.
1. Skeletonizing images. Because of the nature of how woodblock prints were made, by inking a surface with relief, it was not as easy to control the amount of ink applied as it was with intaglio printing (grooves). Therefore lines in woodblock prints vary in width from print to print (of the same woodblock and same edition). This can create problems with almost any type of image analysis, so it is advantageous to standardize line width before beginning analyses. I found that skeletonizing an image (Process/Binary/Skeletonize in ImageJ) is a simple and effective way to accomplish this. Skeletonization reduces lines to a single pixel width. If an image is large and has few lines, skeletonization may reduce black pixels to a low level that is awkward to measure, in which case you can amplify the lines using the "dilate" feature in ImageJ (Binary/Threshold/Dilate). All of the type of analyses detailed below are simply different ways of measuring the same thing: line breaks. Because they are inter-related, this should be kept in mind when doing and interpreting statistical analyses. You may choose to use only one or two methods, or even a different method. In the study (Hedges, 2006), I used all six of them and then combined them into a single "deterioration index" because I thought it would reduce the noise evident in each data set (measurements were scaled from 0-1, slopes were all made positive, and measurements averaged). Further research on the best method or methods for woodblock analysis is needed.
2. Line breaks. In large and uncomplicated woodblock prints, line breaks can be counted by eye without digital image analysis. However, care must be taken not to be score lines that are poorly inked or over-inked (this is different from normal variation in inking). Normally inked lines will be solid black and have a sharp edge. Poorly inked lines will be stippled and look fuzzy, sometimes having many irregular breaks in one region. The stippling surrounding the break will give it away as caused by poor inking. Identify such lines before skeletonizing images to avoid measuring them. A real break caused by a gap in the woodblock relief will have a sharply-defined edge on either side of the gap. Over-inked lines will appear blotched and show joining of parallel lines that are close together.
Dave Gants brought to my attention a little-known aspect of early printing relevant here: "dabbing." Headpieces and ornamented initial letters were sometimes copied from wood to metal in a process called dabbing, which essentially extended their useful lifespan and permitted multiple copies to be used at one time (Mosley, 2006). Of course, they would not be expected to change with time, so would not be informative for dating. They may be hard to identify, but sometimes the resulting prints have a smudge of ink surrounding the illustration, resulting from contact by edges of the thin metal plate (see illustration in Mosley, 2006).
An edge-finding tool in ImageJ (Process/Find edges) can help to locate line breaks within the image, for direct counting by eye. It is probably best applied after converting the grayscale image to a binary image with auto-threshold (see above).
3. Gray value. These range from 0 (black) to 255 (white). Thus a lower number means a darker image with a greater percentage of black pixels. Measure gray value (Analyze/Measure) after thresholding the image. The point here is that line breaks are gaps in black lines and hence they are missing black pixels. If breaks accrue over time, one should expect to see an increase in gray value over time.
4. Fractal dimension. Fractal dimension (D) is most often measured as the negative slope of a log-log box count plot (see ImageJ online docs). The box count tool in ImageJ (Analyze/Tools/Fractal Box Count) counts the number of boxes of an increasing size needed to cover a one-pixel binary object boundary, and produces the plot and fractal dimension calculation. It is one way of measuring complexity, and since line breaks will affect the complexity of a print, it seems an appropriate measurement, although I have found it to be closely correlated with gray value.
5. Particle measurements. I decided to use these because, as more line breaks occur, lines are broken into increasing smaller particles which can be measured automatically using ImageJ (Analyze/Analyze Particles). Although it turned out to be useful, it was not as straightforward as I had hoped. Some complex images can trap smaller particles within larger particles, and the particle counting procedure only detects outer boundaries. One way around this problem is to disrupt the image with random noise, before making particle measurements, by using other tools such as Gaussian blur (Process/Filter/Gaussian Blur) and variance (Process/Filter/Variance). But if this is done, it is critical to apply the same procedure to all prints. I found three related measurements to be useful: large particle count, large particle size, and particle perimeter. As expected, I found that all three measurements decreased with time as line breaks increased, producing smaller and smaller particles (more particles, smaller particles, and smaller perimeters).
An early copperplate printing shop illustrated in a copperplate print by Stradanus (1590). Copyright S. Blair Hedges
1. Retouched lines. Before making measurements it is important to determine that the prints being compared have not been retouched (recut). Sometimes this is easy to see, because the artwork will look noticeably different, or lines will look unusually dark and bold. But often you must look very closely to see alterations, even under high magnification, especially if the retouching was of high quality. This is critical because retouching will alter the timeline of the print clock. However, it is still possible to use a print that has been retouched, especially if the retouching was in one area of the print, in which case that region can be avoided in all comparisons. Also, if the retouching was throughout the print but some original lines remain, it might be possible to take line width measurements (see below) of those original lines, for analysis.
2. Gray value. See above for description of this under "woodblock analyses." With copperplate prints, this is a good way to get a quantitative estimate of the fading seen in prints across different editions (the fading is from thinning of lines). As noted above, uniform illumination of the print is critical for accurate measurement. A flat-bed scanner will give you this but be cautious with other imaging equipment such as a digital camera or even overhead "scanners" in libraries because a slight unevenness in illumination, whether from position of lights or arching of the print being photographed, can cause problems for thresholding. If you are working with images from digital cameras, it will probably be best to identify selected areas within the print (e.g., 50 mm in width) to analyze because smaller areas will have greater uniformity in illumination. But if doing so, always be consistent and use that same area and settings.
Figure (right): Characteristics of engraved and etched lines. Below each are cross-sections of corresponding copperplate grooves. Engraving is done with a steel burin that leaves a triangular-shaped groove and produces tapered lines with sharp points and gentle curves. Etching is done with acid and usually results in lines of even width and blunt ends, and which often make sharp bends. Download high resolution image here (1400x4000) for educational purposes.
3. Line width. A copperplate print may have thousands of individual lines and each might be expected to undergo thinning with time as the plate surface is worn down from polishing (before each print run) and grooves become narrower. This is true for engraved and etched lines (see figure for differences), as was determined in the study (Hedges, 2006, 2008a). Thus in theory, each line could be a separate print clock, useful for dating the print. However, in reality we must consider sources of variation, and the measurements derived from one line are unlikely to have much precision. This is because the polishing of the copperplate was unlikely to be perfectly uniform across the surface, or even if uniform it may not have been applied to a depth proportional to time elapsed. Yet another variable could be differential inking of the plate, which might have left less ink in some grooves. For these reasons, it is best to measure many lines from around the print to increase the precision of your overall estimate of line thinning for a particular print. In ImageJ, this can be done, using a thresholded high-resolution image (e.g., 4800 dpi scan) by moving the cursor (straight line) across the width of an enlarged printed line on your monitor and measuring (Analyze/Measure). To convert from pixels to micrometers, you'll just need to calibrate by taking a direct measurement of the print. I found that small etched lines thinned at the same rate as large etched lines, but that small engraved lines thinned more slowly than large engraved lines (Hedges, 2006, 2008a). This is likely because a lozenge burin, with narrower (60 degrees) angle, was used more often for fine lines so that they lasted longer (i.e., were deeper into the metal). A square burin (90 degree angle) was used more often for wider lines. But because different angle change the rate of line thinning, caution must be used in making comparisons of engraved line thinning between prints. The best way to avoid any bias would be to measure exactly the same lines, and same positions along lines, in all of the prints being compared.
Frequently Asked Questions (FAQs)
The printing presses exerted great pressure on woodblocks and copperplates, so why are you claiming that the wear on those blocks and plates occurred between printing runs instead of during printing runs?
This is based on observations. The deterioration of the blocks and plates can be seen in prints, and the evidence shows that this occurred between printing events. Yes, it goes against the common assumption that wear occurred during printing.
Is the print clock useful only for dating different editions of books?
No. Artwork contained in early books, such as the printer's mark and title page ornamentation, often appeared in other books by the same printer (but by different authors), providing a means of dating a rare book that was only printed once.
Why can't undated books be dated simply by their watermarks?
Some books can be dated by watermarks, and it is one of several useful methods. Paper molds usually had a short lifespan (about two years, by some estimates) and paper stock was usually consumed within 5-10 years, affording some degree of precision. However, there are limitations to the use of watermarks. Current watermark atlases and databases contain only a small percentage of the total watermarks that exist in books, manuscripts, and prints, which may be more than a million. Also, exact matches are required for dating, because very similar watermarks (e.g., anchors) were often used for decades and even centuries, and many of the drawings in watermark atlases do not have the precision for accurate dating. Furthermore, because of the way that paper was folded and bound in books, many quartos, octavos, and smaller prints have no watermarks, except for occasional fragments of countermarks. For example, half of Rembrandt's etchings have no watermarks (Ash & Fletcher, 1998, p. 21).
If different prints and different lines within a print change at different rates, isn't that a problem for this method?
No. Different genes also evolve at different rates, which is an advantage for molecular clocks because it permits dating at different timescales. For the same reason that clocks have a second hand, minute hand, and hour hand, different rates of change can be useful in any problem involving time estimation.
Hasn't the deterioration of woodblocks and copperplates been used to date old documents previously?
Yes. But in those cases, the defects observed in prints have been used only to bracket an undated document in time (e.g., between two dated editions of a book), not as a quantitative measure of time that affords more precision.
Did woodblock relief become worn down with time just like the copperplate surface?
There is no evidence of this so far, and it is not predicted. The copperplate surface became worn down apparently from polishing by the printer before each print run, to remove nicks and corrosion, not from wear by the press. It is certain the woodblocks cracked as they aged, because the cracks can be seen as line breaks in prints. Some breaks probably were produced by collisions with objects. But there is no record that printers sanded down the woodblock relief, and no reason why they would do that, although more study of the change in woodblocks is needed.
How do I cite this web site?
This web site was intended to supplement what is already published in Royal Society article (see top of page). Therefore, that article may be cited. However, if there is some information presented here, uniquely, that you wish to cite, the normal convention for citing a web resource would be something like this: Hedges, S. B. 2006. The print clock: a method for dating early books and prints. (http://www.printclock.org). Pennsylvania State University, University Park, Pennsylvania 16802-5301, USA.
FAQs regarding the study (Hedges, 2006, 2008)
Why were line breaks scored by eye in one case and by digital image analysis in another?
As explained in the article, the Bordone Isolario prints are large, simple, and have few lines. A large number (2,576 prints) could be scored by eye, quickly. The printer's mark woodblock print, on the other hand, was small and complex, and contained many line breaks, necessitating digital analyses.
How confident are you in the date (1565) estimated, using the print clock, for the undated book (Isolario) by Benedetto Bordone?
This same date (1565) was obtained using two independent print clocks: One was based the prints contained within the Isolario, produced by blocks carved by Bordone himself (or an assistant) and the other estimate used the printer's mark of Francesco de Leno, produced by a woodblock carved by him (or an assistant) decades later in the mid-1500s. Both have a standard error of about 1 year. This is good evidence that the book was printed in the mid-1560s.
Isn't the fact that the woodblock printer's mark of de Leno only known from 1559-1570 evidence by itself that the undated Isolario was printed during that time interval?
No. See the electronic appendix to the paper. Books published by Francesco de Leno are known starting in 1542, and even though the early dates are problematic, they cannot be discounted a priori.
Can extrapolation of time be trusted, compared with interpolation?
Yes. In figure 2 of the article, time estimates are extrapolated forward to predict the undated Isolario and backward (x-intercept) to predict the carving of the woodblocks. While it is true that interpolation (e.g., figure 4) will have a lower variance for the estimate, extrapolation is a valid and widely practiced method in all fields of science. The particular situation at hand will dictate which method is used.
Why didn't you simply bracket the date of the undated Isolario using damage in the printer's mark?
Although the use of bracketing damage in the printer's mark shows that the book was probably printed during the 1560s, the use of quantitative change (print clock) provided greater precision. Also, in this case, there were books published by the printer almost every year, providing abundant information for timing. In other cases, different editions may be separated by decades and the bracketing method would not afford much precision compared with the quantitative method. In addition, the bracketing method assumes that the deterioration is cumulative and completely observable, with no reversals. In real prints, some line breaks appear and disappear through time as a result of differential inking. Date bracketing is certainly still useful, but these considerations should be taken into account.
Was the undated Isolario book included in the regression analysis?
No. It was only plotted on the graph based on the mean number of line breaks, to show its position. It was not used in the regression, or in calculation of standard errors.
References (see Hedges, 2006, 2008a, and 2008b for other references)
- Ash, N., and S. Fletcher. 1998. Watermarks in Rembrandt's prints. Washington, DC: National Gallery of Art.
- Hedges, S. B, 2006. A method for dating early books and prints using image analysis. Proc. R. Soc. A : Mathematical, Physical, and Engineering Sciences 462:3555-3573. Download PDF here. Also see Electronic appendix PDF.
- Hedges, S. B. 2008a. Image analysis of Renaissance copperplate prints. Proc. SPIE 6810: 681009. 20 pages. E-print.
- Hedges, S. B. 2008b. Dating old maps with the print clock. The Portolan 72: 25-33. E-print.
- Mosley, J. 2006. Dabbing, Abklatschen, clichage... Typefoundry (http://typefoundry.blogspot.com/), 13 January, 2006. And a recent article about wormholes in prints:
- Hedges, S. B. 2012. Wormholes record species history in space and time. (main article). Biology Letters dx.doi.org/10.1098/rsbl.2012.0926. E-print.
- Hedges, S. B. 2012. Wormholes record species history in space and time. (supplement). Biology Letters dx.doi.org/10.1098/rsbl.2012.0926. E-print.
Please direct any questions to sbh1 @ psu.edu | www.hedgeslab.org