Bibliometrics with the subject matter put back in
The Slop Index
A growing share of what is published as medical and public health research is political advocacy in the costume of science. The conclusion is settled before the study begins, the data are gathered only because a journal will not print an opinion without a table or graph. The abstract exists to lend an argument the authority of a citation. The result is a literature that confirms, at length and with confidence intervals, propositions that were never in doubt by their authors, put forth for the purpose of advancing a political agenda. None of this shows up in the h-index, which counts a commentary cited by other commentaries exactly as it counts a trial that changed practice.
The Slop Index is my attempt to characterize that difference. It scores an author on how much of his or her body of work comprises new, serious research, and how much of it is political commentary. It does not ask only whether a topic is political; many honest scientific topics either are or eventually become political, and policy deserves to be informed by science. My score asks whether a paper on such a topic earned its place the way any other paper must, with a real design, a hard endpoint, and an effect the field found worth citing. The aim is to distinguish actual contributions to knowledge about contested questions from advocacy that has borrowed the vocabulary of research, and to do it with open source methodology.
What is the h-index counting?
The h-index is often used as a measure of a researcher's impact. It counts citations, but it does not ask what was cited or whether it is worth citing. A viewpoint in JAMA and a randomized trial in JAMA count the same. This tool scores every PubMed-indexed paper by an author on three axes, reports each separately, and combines them into one labeled composite, the Total Slop Index. Everything runs in your browser: PubMed and NIH iCite are queried directly, and the scoring code is the open-source Python package executed in-page.
| Year | Role | Tier / design | Outcome | Exposure | Cites | RCR | Politicized topic | Value | Paper |
|---|
Tier A = original data, B = synthesis or modeling, C = commentary, U = no abstract in PubMed (held out). RCR is the NIH Relative Citation Ratio (1.0 = field average). Value is the per-paper substance score before normalization; see Methodology.
Weights: every number in the composite, with sliders to see how the verdict depends on them
The topic-neutral preset is the honest companion to any score you quote: it shows how much of a number is the editorial position and how much is the methodology.
Reading the number
Zero is good and one is bad, and the useful cutoff is 0.57. That number is the score of a bibliography whose average paper has nothing to recommend it and nothing to penalize, so anything above it means the typical paper is net negative under the weights, and anything below it means the typical paper adds something. The calibration table turns that into three practical bands. Below 0.47 is a research record: bench scientists, trialists, and surgeons and statisticians with hard endpoints land here, from about 0.18 (Karikó) to 0.44. 0.47 to 0.65 is mixed: real empirical work with enough highly politicized topics and political agenda, or enough commentary, to pull the average toward zero; health economists and the senator-elect epidemiologist sit here. Above 0.65 is a commentary record: the median paper is negative, most of the output is highly politicized topics and political agenda, a third or more is opinion and pure advocacy. Nobody on the calibration table scored above 0.84 or below 0.18, so a result outside that range deserves a second look at the author match before it is believed.
Two cautions on any single number. A short record (under about 15 papers) swings wildly on one or two papers, so quote the band rather than the decimal. And always quote the topic-neutral score beside it; the gap between the two is the size of the editorial position, and the reader is entitled to know it.
Read before quoting a number
A score is only as good as the author match and the index behind it. PubMed does not index books, op-eds, policy reports, or most economics and social-science journals, and its full-name indexing begins around 2002, so early careers are clipped. The classifier is rule-based and conservative; a paper it cannot place lands in commentary with a low-confidence flag. The interpretation page shows where seventeen named researchers fall, so you can see what 0.3 or 0.7 means before you quote one.