Use this skill for ANY question about multivariable regression modeling or the R rms package (ols, lrm, orm, cph, psm, blrm, rcs, datadist, anova, validate, calibrate, nomogram, contrast, Predict). Also trigger for: restricted cubic splines, degrees of freedom allocation, overfitting, bootstrap validation, penalized estimation, shrinkage, multiple imputation, binary/ordinal logistic regression, proportional odds model, Cox regression, parametric survival models, generalized least squares, semiparametric ordinal longitudinal models, partial effect plots, nomograms, machine learning vs. statistical models, Frank Harrell's modeling philosophy, safe data mining, spending degrees of freedom, effective sample size, chunk tests, phantom degrees of freedom, chi-squared-minus-df signal, or any question about rms, rmsb, or Hmisc packages. Always use this skill — even for basic rms questions.
Use this skill for ANY question about Bayesian analysis or clinical trial design: posterior probability calculations, prior specification, skeptical priors, sequential and adaptive trial designs, Bayesian stopping rules, posterior exceedance probabilities, ROPE analyses, Bayesian operating characteristics, multiplicity in the Bayesian framework, multiple endpoints and totality of evidence, Bayesian sample size and power for ordinal outcomes, borrowing historical data, Bayesian RCT analysis, blrm, rmsb, Stan, brms, MCMC, and frequentist vs. Bayesian comparisons. Also trigger for: why posterior probabilities beat p-values, why alpha-spending is unnecessary under Bayes, prior justification, simulating Bayesian operating characteristics, reporting Bayesian results to regulators, or the BET course (hbiostat.org/bayes/bet). Use this skill alongside the rms skill for Bayesian regression mechanics, and the principles skill for broader philosophical arguments. Always use this skill for Bayesian methodology.
Use this skill for ANY question about biostatistics for biomedical research: types of measurements, study design, RCTs, observational studies, ANCOVA, sample size, power, confidence intervals, hypothesis testing, p-values, comparing proportions, nonparametric tests, correlation, serial/longitudinal data, observer variability, measurement agreement, propensity scores, information loss from dichotomization, biomarker research, sensitivity, specificity, ROC curves, medical diagnosis, high-dimensional data, reproducible research, and alpha vs. decision error probability. Also trigger for: change from baseline vs. ANCOVA, regression to the mean, subgroup analyses, or Frank Harrell's BBR course (hbiostat.org/bbr). BBR takes precedence over rms for design, measurement, and inference philosophy; rms takes precedence for multivariable modeling mechanics. Always use this skill for study design, statistical inference, or biomedical data analysis questions.
Frank Harrell's R Workflow for reproducible data analysis and reporting (hbiostat.org/rflow). Use this skill whenever the user asks about or pastes code involving: analysis file creation, data import, annotating variables, data dictionaries, variable labels/units, Hmisc functions (describe, label, units, upData, csv.get, getHdata, contents, hlab, hlabs, vlab, Cs, summarize, summary.formula, wtd.*, rcorr, varclus, naclus, missChk, multDataOverview), data.table operations (DT[i,j,by], :=, fcase, melt, dcast, setkey, let, rbindlist, uniqueN), qreport functions (hookaddcap, makecnote, maketabs), ggplot2 graphics for exploratory or descriptive analysis, Quarto document setup and rendering, descriptive statistics workflows, missing data patterns, data overview, caching with targets or knitr, parallel computing with future/parallel, simulation in R, or reproducible research reporting. Also trigger when the user asks for help reviewing R code using any of these packages or idioms, or when they ask how to do something
Apply Frank Harrell's statistical principles and philosophy when the user is reasoning about study design, inference, p-values, Bayesian vs. frequentist methods, model building, variable selection, validation, measurement, or reporting results. Use this skill whenever the user asks how to analyze data, interpret statistical output, choose between methods, design an experiment, or critique a published analysis — even if they don't use the words "Bayesian" or "frequentist". The skill encodes a coherent philosophical stance that should inform all statistical advice given in this project.