arXiv subjects
Publications and source records attributed to Michael B. Weissman.
A much-cited 2022 paper by Pekar et al. claimed that Bayesian analysis of the molecular phylogeny of early SARS-CoV-2 cases indicated that it was more likely that two successful introductions to humans had occurred than that just one had. Here I show that after correcting a fundamental error in Bayesian reasoning the results in that paper give larger likelihood for a single introduction than for two.
A recent paper evaluating a new rubric-based graduate admissions approach using generic methods tentatively suggested that its decisions differed noticeably from the previous approach in an unspecified way. Using prior knowledge that the often-stated specific goal was to open a path to increased diversity by reducing barriers to admission of applicants with low undergraduate grade point averages and GRE scores allows simple statistical tests of changes in the distributions of the metrics. The simple tests confirm with good statistical confidence that the barrier-reduction changes were achieved. Nevertheless, the paper's argument that the de-emphasized test and grades are not predictive of graduate outcomes is not supported by the prior literature. On technique, although a method used in some of the analysis for dropping data points before running the machine algorithm is likely to bias those results, it helps to clarify why models of the rubric-based system were only weakly predictive.
Over 1100 students over four semesters were given the option of taking an introductory undergraduate statistics class either by in-person attendance in lectures or by taking exactly the same class (same instructor, recorded lectures, homework, blind grading, website, etc.) without the in-person lectures. Roughly equal numbers of students chose each option. The online lectures were available to all. Attendance by online students was rare. The online students did slightly better on computer-graded exams. The causal effect of choosing only online lectures was estimated by adjusting for measured confounders, of which the incoming ACT math scores turned out to be most important, using four standard methods. The four nearly identical point estimates remained positive but were small and not statistically significant at the 95% confidence level. Sensitivity analysis indicated that unmeasured confounding was unlikely to be large but might plausibly reduce the point estimate to zero. No statistically significant differences were found in preliminary comparisons of effects on females/males, U.S./non-U.S. citizens, freshmen/non-freshman, and lower-scoring/higher-scoring math ACT groups.
A recent paper used structural equation modeling to infer effects of gender-dependent student attitudes on several outcomes of introductory physics courses. The model used is precisely Markov equivalent to an equally plausible approximation in which a key gender-dependent coefficient in determining attitudes is indistinguishable from zero. I also argue that the qualitative framing may attribute to students motivations that actually belong to their teachers.
A recent paper by Salehi et al. claims that the differences found between major demographic groups on scores in introductory college physics tests are due to differences in pre-college preparation. No evidence is produced, however, to show that preparation differences are more causally important than any other differences. In one case, the male/female difference, the paper actually provides evidence that preparation gaps are unimportant.
A recent paper in Sci. Adv. by Miller et al. concludes that GREs do not help predict whether physics grad students will get Ph.D.s. The paper makes numerous elementary statistics errors, including introduction of unnecessary collider-like stratification bias, variance inflation by collinearity and range restriction, omission of needed data (some subsequently provided), a peculiar choice of null hypothesis on subgroups, blurring the distinction between failure to reject a null and accepting a null, and an extraordinary procedure for radically inflating confidence intervals in a figure. Release of results of simpler models, e.g. without unnecessary stratification, would fix some key problems. The paper exhibits exactly the sort of research techniques which we should be teaching students to avoid.
I argue that various derivations of the Born probability rule within strictly unitary quantum mechanics implicitly use anthropic constraints on the properties of the probabilities, but without proposing an ensemble from which those constraints limit a selection. An argument by Mallah obtains the Born probabilities explicitly for standard unitary dynamics but via postulating a hypothetical extra noise-like component of the quantum state. I argue that the implicit anthropic constraints required for well-behaved probability rules would select for a state of the sort Mallah proposes.
Barkhausen noise as found in magnets is studied both with and without the presence of long-range (LR) demagnetizing fields using the non-equilibrium, zero-temperature random-field Ising model. Two distinct subloop behaviors arise and are shown to be in qualitative agreement with experiments on thin film magnets and soft ferromagnets. With LR fields present subloops resemble a self-organized critical system, while their absence results in subloops that reflect the critical point seen in the saturation loop as the system disorder is changed. In the former case, power law distributions of noise are found in subloops, while in the latter case history-induced critical scaling is studied in avalanche size distributions, spin-flip correlation functions, and finite-size scaling of the second moments of the size distributions. Results are presented for simulations of over 10^8 spins.