Skip to contents

Changelog

Source: NEWS.md

uSDT 0.2.0

CRAN release: 2026-11-05

  • Scoped random seeds preserve the session’s RNG state and generator settings. Simulated studies use distinct seeds and reproduce across core counts.
  • Bayesian examples document their compiler requirements and use at most two cores. Long bootstrap and simulation examples remain in \donttest{}.

Breaking changes

  • Tasks are now named awareness and influence throughout the API. Use these names in usdt_data_tasks(), task-specific lists, task_levels, labels, dichotomize and sensitivity-term arguments. The old names are no longer accepted.
  • The awareness code is "A", with model terms d_A and c_A. Planning columns use trials_awareness, trials_influence, reliability_awareness and reliability_influence.

Bayesian estimation

  • hsdt(estimation = "bayesian") fits with brms and Stan and returns the brmsfit in $fit. Both rstan and cmdstanr are supported. Compiled models are cached. Defaults are four chains with 1,000 warmup iterations and 5,000 kept draws per chain. Sampling reports divergences, tree depth, R-hat and effective sample sizes.
  • usdt_priors() specifies priors for task means, SDs and correlations. plot() checks their predictive and hypothesis implications.
  • hsdt(latent = "regression") places the influence-location prior on the latent intercept instead of the mean. This changes the induced prior of the mean difference. The intercept prior also affects inference on the correlation.
  • unequal_variances = TRUE estimates signal SDs for Bayesian fits.
  • usdt_tests() and the individual hypothesis functions accept Bayesian hsdt and custom brms fits. They report posterior means, SDs, central credible intervals and two-sided posterior sign p-values.
  • usdt_bf() tests point, directional and interval hypotheses about diff, rho, intercept and slope. Point nulls use Savage-Dickey density ratios. Directions and regions compare posterior with prior odds. Reports give both log Bayes factors. By default, prior/posterior plots accompany them. Use plot = FALSE to suppress drawing and plot(bf) to draw later.
  • Bayes factor plots label the tested hypothesis and quantity. A point-null panel reports BF01. A directional or interval panel reports BF10.
  • plot() and usdt_reliability() support Bayesian fits.
  • The Bayesian tutorial covers priors, posterior summaries, Bayes factors and the latent regression prior.

Equivalence and one-sided tests

  • usdt_equivalence() accepts frequentist hsdt and custom glmer fits. Hypotheses use the syntax of usdt_bf(), with explicit bounds for equivalence (TOST), non-inferiority, one-sided and minimal-effect tests. Reports include the interval used for the decision. Bootstrap replicates are used when available. Unsupported inference is marked not estimable.

Study planning

  • usdt_population() defines population parameters, optionally from a frequentist pilot fit. Multiple values create scenarios. update() changes them. A latent intercept can replace the influence mean.
  • usdt_power() evaluates designs or grids using a sampling-distribution approximation. It supports the hypotheses of usdt_equivalence(). combine = "all" adds joint rejection power. Non-estimable outcomes count as failures to reject, and populations in H0 are flagged.
  • usdt_power(method = "simulation") simulates and fits whole studies, repeating the median split for dichotomized tasks. It reports Monte Carlo error, Wilson intervals and fitting failures. summary() adds bias, standard errors, interval widths and coverage. Results are reproducible across core counts. Unequal-variance populations can be simulated and analysed with the equal-variance model.
  • usdt_plan() selects subjects, trials or both to meet a power or precision target at minimum cost, or maximize power within a budget. Ranges identify free values. Single numbers fix them. Check a chosen plan with usdt_power(plan, method = "simulation").
  • usdt_mdes() finds detectable effects or equivalence regions. vary selects another population parameter to vary.
  • plot() supports populations, power curves, plans and detectable effects. The planning tutorial illustrates the workflow.

Data preparation and plots

  • usdt_aggregate() counts binary trials by grouping columns for custom binomial models with covariates.
  • Plots share panel styling and captions that wrap to the figure width. Plot methods accept caption = FALSE.
  • Regression plots accept nrow = 2 to stack the observed and latent panels.
  • ROC legends show sensitivity and AUC inside the panel.

uSDT 0.1.0

CRAN release: 2026-09-29

First release.

Data preparation and descriptive estimates

  • usdt_data_tasks() and usdt_data_long() prepare paired direct and indirect measures and report the column mappings and preparation diagnostics. Every column, level and format argument takes either one value for both tasks or one per task as list(direct = ..., indirect = ...), so the two tasks may differ in the columns they use, in the values those columns take, and in the format they arrive in.
  • meyen_split() dichotomizes continuous responses using a within-subject median pooled across conditions, following Meyen et al. (2022).
  • sdt_moments() computes subject-level SDT estimates and optional sampling variances without fitting a hierarchical model.

Model fitting and inference

  • hsdt() fits a binomial probit mixed model with correlated random sensitivities and returns the three hypothesis tests.
  • sensitivity_diff(), latent_cor() and latent_regression() calculate the mean sensitivity difference, latent correlation and latent regression. usdt_tests() collects the analytical tests in one table.
  • usdt_boot() adds parametric bootstrap inference to the fitted object’s hypothesis table when enough usable replicates are available.

Visualization and reliability

  • plot() for a fitted hsdt object draws observed and latent regressions, shrinkage, subject intervals and model-implied ROC curves.
  • usdt_reliability() estimates task reliability from the fitted between-subject sensitivity variance and trial-level measurement variance.

Example data

  • vadillo_awareness and vadillo_cuing provide trial-level data from Experiment 2 of Vadillo, Malejka and Shanks (2025).