Although Harry Woolf’s great collective volume Quantification mostly overlooked biology, Thomas Kuhn’s chapter there on the role of quantitative measurement within the physical sciences maps quite well onto the forms of reasoning that actually persuade us as biologists 50 years later. Kuhn distinguished between two contexts, that of producing quantitative anomalies and that of resolving them. The implied form of reasoning is actually C. S. Peirce’s abduction or inference to the best explanation: “The surprising fact C is observed; but if (...) A were true, C would be a matter of course; hence there is reason to suspect that A is true.” This article reviews abduction and the Kuhnian dichotomy in a range of classic examples where quantitative reasoning has ended arguments in the natural sciences. Included are John Snow’s discovery of the cause of cholera, Jean Perrin’s proof that atoms exist, the discovery of the double helix, the Alvarezes’ explanation of the extraterrestrial origin of the Cretaceous-Tertiary extinction, and current examples in passive smoking, ulcers, and the anthropogenicity of global warming. Modern biology is a quantitative science to the extent that we operate by “strong inference,” the insistence that our data are surprising on everybody else’s hypotheses but follow as a matter of course from our own, and that we demand numerical consilience whenever we infer across levels of analysis or across disciplines. (shrink)

The current literature that attempts to bridge between geometric morphometrics (GMM) and finite element analyses (FEA) of CT-derived data from bones of living animals and fossils appears to lack a sound biotheoretical foundation. To supply the missing rigor, the present article demonstrates a new rhetoric of quantitative inference across the GMM–FEA bridge—a rhetoric bridging form to function when both have been quantified so stringently. The suggested approach is founded on diverse standard textbook examples of the relation between forms and the (...) way strains in them are produced by stresses imposed upon them. One potentially cogent approach to the explanatory purposes driving studies of this class arises from a close scrutiny of the way in which computations in both domains, shape and strain, can be couched as minimizations of a scalar quantity. For GMM, this is ordinary Procrustes shape distance; in FEA, it is the potential energy that is stored in the deformed configuration of the solid form. A hybrid statistical method is introduced requiring that all forms be subjected to the same detailed loading designs (the same “probes”) in a manner careful to accommodate the variations of those same forms before they were stressed. The proper role of GMM is argued to be the construction of regressions for strain energy density on the largest-scale relative warps in order that biological explanations may proceed in terms of the residuals from those regressions: the local residual features of strain energy density. The method, evidently a hierarchical one, might be intuitively apprehended as a geometrical approach to a formal allometric analysis of strain. The essay closes with an exhortation. (shrink)

Suspicion of “physics envy” surrounds the standard statistical toolbox used in the empirical sciences, from biology to psychology. Mainstream methods in these fields, various lines of criticism point out, often fall short of the basic requirements of measurement. Quantitative scales are applied to variables that can hardly be treated as measurable magnitudes, like preferences or happiness; hypotheses are tested by comparing data with conventional significance thresholds that hardly mention effect sizes. This article discusses what I call “shmeasurement.” To “shmeasure” is (...) to fail to apply quantitative tools to quantitative questions. We “shmeasure” when we try to measure what cannot be measured, or, conversely, when we ask binary questions of continuous measurements. Following the critics of standard statistical tools, it is argued that our statistical toolbox is indeed less concerned with the measurement of magnitudes than we take it to be. This article adds, however, that measurement is not all there is to scientific activity. Most techniques of proof do not resemble measurement as much as voting—a practice that makes frequent use of numbers, figures, or measurements, yet is not chiefly concerned with assessing quantities. Measurement is only one among three functions of the scientific toolbox, the other two being collating observations and deciding which hypotheses to relinquish. I thus make a plea for “shmeasurement”: the mismeasure of things starts to make more sense once we take into account the nonquantitative side of scientific practice. (shrink)

Suspicion of “physics envy” surrounds the standard statistical toolbox used in the empirical sciences, from biology to psychology. Mainstream methods in these fields, various lines of criticism point out, often fall short of the basic requirements of measurement. Quantitative scales are applied to variables that can hardly be treated as measurable magnitudes, like preferences or happiness; hypotheses are tested by comparing data with conventional significance thresholds that hardly mention effect sizes. This article discusses what I call “shmeasurement.” To “shmeasure” is (...) to fail to apply quantitative tools to quantitative questions. We “shmeasure” when we try to measure what cannot be measured, or, conversely, when we ask binary questions of continuous measurements. Following the critics of standard statistical tools, it is argued that our statistical toolbox is indeed less concerned with the measurement of magnitudes than we take it to be. This article adds, however, that measurement is not all there is to scientific activity. Most techniques of proof do not resemble measurement as much as voting—a practice that makes frequent use of numbers, figures, or measurements, yet is not chiefly concerned with assessing quantities. Measurement is only one among three functions of the scientific toolbox, the other two being collating observations and deciding which hypotheses to relinquish. I thus make a plea for “shmeasurement”: the mismeasure of things starts to make more sense once we take into account the nonquantitative side of scientific practice. (shrink)

Bioscientists generate far more data than their minds can handle, and this trend is likely to continue. With the aid of a small set of versatile tools for mathematical modeling and statistical assessment, bioscientists can explore their real-world systems without experiencing data overflow. This article outlines an approach for combining modern high-throughput, low-cost, but non-selective biospectroscopy measurements with soft, multivariate biochemometrics data modeling to overview complex systems, test hypotheses, and making new discoveries. From preliminary, broad hypotheses and goals, many relevant (...) samples are selected and measured with respect to many informative variables. The resulting tables represent a “cacophony” of data. From these, the most relevant and reliable “underlying harmonies and rhythms” are extracted and tested statistically, displayed for interpretation, and used for prediction. Outliers are detected automatically. Interesting subsets of samples can then be chosen for in-depth analyses in subsequent research cycles. This pragmatic, top-down approach takes advantage of developments in both “soft,” data-driven modeling and “hard,” knowledge-driven cultures. Data analytical examples show how information-rich biospectroscopy can be used for characterizing and quantifying known and unknown chemical constituents and physical phenomena in intact biosamples. This is based on a combination of deductive “hard” and inductive “soft” modeling. The examples represent NIR spectra of biochemical mixtures, FTIR spectra of a microbiological fermentation process, and FTIR analysis of fatty acids in milk for functional genomics. (shrink)