I am in awe of Dr Elisabeth Bik and her amazing ability and dedication to spotting duplications in images.
Follow the thread to see the duplications she finds in this photograph. I usually struggle to find duplications until they are pointed out.
But not always.
Take, for example, this diatom calibration set.
The lower two assemblages are, within the resolution of the PDF, identical. Having spotted that, I wanted to know whether any of the other assemblages were identical, so I scraped the data out of the PDF. All the other assemblages are distinct.
But what about the next figure which shows the down-core stratigraphy?
While the upper part of the diagram is spiky as expected, the lower part of the diagram is step like with nine pairs of apparently identical assemblages.
These identical assemblages in the calibration set and the down-core stratigraphy are unexpected. I asked the authors if they could explain what happened, but they haven’t replied.
The paper also includes a transfer function that I cannot reproduce with the assemblage data I scraped from this paper and the environmental data I scraped from another paper on the same calibration set lakes. The reported cross-validation performance is better than the apparent performance I get, but it is possible that the environmental data had more problems than just using the wrong unit prefix. If my model is correct, then the actual cross-validation performance is very weak (R2 = 0.03), then the reconstruction shown in the paper is bogus and the paper should be corrected (or retracted).
I’m not holding my breath.