Skip to contents

infercalcR

Explicit observed statistical procedures, diagnostic calculations and finalized results in R. infercalcR supplies the computational core of the inference suite; inferguidR provides guidance and prospective design planning, and infereportR provides tables, text and plot annotations.

This is development software. Use the checked revision recorded with an analysis. A successful calculation is not universal scientific validation. The package is not a CRAN release.

Documentation, including tutorials and the function reference, is available at https://cttir.github.io/infercalcR/.

Start 

spec <- infercalcR::method_spec("t_means_two_independent_welch")
r <- infercalcR::run_inference(spec,
  x = c(5.2, 6.1, 4.9, 7.3, 5.8, 6.5, 7.1, 4.7),
  y = c(3.8, 5.9, 4.2, 5.1, 6.7, 4.8, 3.9, 5.3, 6.0),
  result_id = "primary-comparison")
infercalcR::result_data(r)
infercalcR::result_metadata(r)
infercalcR::describe_data(data.frame(value = c(1, 2, NA),
  category = factor(c("a", "b", "a"))))

The mean difference is x minus y. Sidedness, confidence level and method options are explicit; data-driven pretests never choose the procedure. Inspect status, reason, estimand, orientation and population before using an estimate. Missing values, unsupported inference, numerical failure and legitimate infinite limits remain distinct.

method_catalogue() lists current procedure identities, available adapters and optional backends. ?observed-methods and the input-signature inventory describe each method’s accepted data/design arguments. The 68 executable analysis templates and their three callable numerical helpers were extracted with their original scope, assumptions and limits. The advisor-only one-sample variance method is explicitly unsupported as an observed adapter.

Results and reporting

as_inference() explicitly converts supported htest, lm/glm, Boschloo and classifier results. It never chooses or fits a replacement analysis. Model conversion finalizes its declared numerical summaries once, before reporting. Unknown sample sizes are not inferred from degrees of freedom.

The common result is an ordinary S3 object containing finalized data, metadata and method-specific payload. Raw and adjusted p-values remain separate. adjust_inference() requires a declared complete family and rejects silent readjustment. Reporting may select rows but cannot change that family. See CONTRACT.md for the shared schema and legacy boundaries.

# Optional reporting consumer; the calculation object needs no raw observations.
infereportR::report_table(r)
infereportR::report_text(r)

Dependencies and installation

R 4.1.0 or later is required. The new result, descriptive and ordinary statistical core uses only R-distributed base packages. The package never installs anything at runtime. It has no dependency on either suite consumer.

Legacy identity-bearing functions retain optional digest/jsonlite for canonical JSON, SHA-256 fingerprints and correctly rounded decimal parsing. They fail explicitly if those packages are absent. boschloo_inference() uses the same numerical engine without requiring historical identities. Plain count metrics also work without contributed packages.

Specialist procedures explicitly use cluster, nlme, MASS, nnet, lme4 or geepack. These are feature-local optional dependencies, with separate executed validation lanes. Installing another package never silently switches the model, covariance method or interval. Existing testthat regression coverage is a development integration lane; new tests are ordinary R scripts.

Install the development version from GitHub, for example with remotes::install_github("CTTIR/infercalcR"), or from a local checkout:

R CMD build infercalcR
R CMD INSTALL infercalcR_0.1.0.tar.gz
R CMD check --no-manual infercalcR_0.1.0.tar.gz

Install this provider before either consumer. The source archive includes NAMESPACE and Rd help; installation needs no documentation generator. Set INFER_SUITE_FULL=true for the required supported-backend tests. Minimal checks may explicitly skip optional integration lanes and cannot substitute for full validation.

Choosing an engine

Thirteen methods compute their test with infercalcR’s own base-R code. Each of them can instead be computed by the validated original: the function DescrTab2 uses for that test or, for tests DescrTab2 does not offer, the established package.

d <- data.frame(id = 1:6, a = c(0, 1, 0, 1, 1, 0), b = c(1, 1, 0, 0, 1, 1),
  c = c(1, 0, 0, 1, 0, 1))
spec <- infercalcR::method_spec("cochran_q", engine = "reference")
r <- infercalcR::run_inference(spec, data = d)   # computed by DescTools
infercalcR::result_metadata(r)$engine            # engine, package, version, call

engine = "base" is the default, so results never depend on which packages are installed. The originals are optional. When one is missing or older than the version the benchmarks were captured with, engine = "reference" stops with an error; it never falls back to the base engine. Methods whose test is computed by R’s stats package, or whose model is fitted by stats or an optional model backend, have a single engine, which either setting runs. inst/methods/provenance.csv notes where infercalcR computes tests or summaries from such a fitted model itself.

Method Original
boschloo_one_sided, boschloo_two_sided_central, boschloo_two_sided_minlike Exact::exact.test (exact2x2’s Boschloo test, which DescrTab2 uses, drops tied tables)
unconditional_exact_z_pooled Exact::exact.test
cochran_q DescTools::CochranQTest
cochran_armitage_trend DescTools::CochranArmitageTest
jonckheere_terpstra DescTools::JonckheereTerpstraTest, permutation route
stuart_maxwell_marginal_homogeneity DescTools::StuartMaxwellTest
variance_brown_forsythe DescTools::LeveneTest with center = median
mcnemar_exact_conditional, mcnemar_midp exact2x2::exact2x2 with paired = TRUE
brunner_munzel brunnermunzel::brunnermunzel.test
anova_repeated_one_way_gg stats::anova.mlm with test = "Spherical"

Both engines are benchmarked on the originals’ help-page examples and agree within a relative difference of 1e-8, with two exceptions compared under approved rules in inst/methods/engine-deviations.csv: the Monte Carlo p-value of the Jonckheere-Terpstra test, and the pooled-Z p-value, for which infercalcR reports a certified upper bound on the supremum (absolute tolerance 1e-8) and Exact’s grid search can fall slightly below it, further for p-values close to 1. inst/methods/provenance.csv records, for every method, its source, its original and the publication its base implementation follows; ?observed-methods lists those publications.

Evidence and provenance

New tests compare matched stats calls, historical synthetic fixtures, independent small-support enumeration and scientific invariants. Backend parity verifies the adapter, rather than independently certifying the backend algorithm. Historical JSON remains readable through inferguidR with unchanged identities; new results can legitimately record new producing-package provenance. Trusted RDS regression fixtures are not an untrusted interchange format.

Raban Heller. MIT license; see LICENSE.md. Record package versions, repository revisions, method settings and sessionInfo() with reproducible work.

Acknowledgments and sources

infercalcR is part of an inference suite with inferguidR and infereportR. The suite builds on the work of others. We gratefully acknowledge that work here and cite it, so that every method can be traced to its source.

DescrTab2, Institute of Medical Biometry, Heidelberg University

The suite is deeply inspired by DescrTab2 [1], the R package for publication-quality descriptive statistics tables developed at the Institute of Medical Biometry (IMBI), Heidelberg University, by Jan Meis, Lukas Baumann, Maximilian Pilz and Lukas Sauer, with contributions from Lorenz Uhlmann and Csilla van Lunteren. Its documented test-choice tree, its validation statement and its practice of computing each test with an established implementation (R’s stats package, exact2x2, DescTools and nlme) are the model for how the suite chooses, computes and reports statistical tests. DescrTab2 was archived on CRAN on 18 October 2025; its source code remains available on GitHub. We thank its authors and IMBI Heidelberg for making their work openly available.

Methodenberatung, University of Zurich

The methods-consulting pages of the University of Zurich [2] inspired the guidance in this suite: their decision tree and step-by-step decision assistant for choosing a statistical method, and their worked examples and explanations of each method.

G*Power

The power and sample-size procedures of inferguidR follow the test inventory of G*Power 3.1 [3, 4], and the distribution calculator of infercalcR follows the G*Power 3.1.9.7 calculator. Their results are compared with values recorded from G*Power 3.1.9.7. G*Power is neither distributed with nor called by these packages.

Software

infercalcR is written in R [5] and computes with R’s stats package. Specialist procedures use the optional backends cluster [6], nlme [7, 8], MASS and nnet [9], lme4 [10] and geepack [11]; digest [12] and jsonlite [13] create canonical identities. Boschloo’s test follows Boschloo (1970) [14]. The reference engines call the optional originals DescTools [15], exact2x2 [16], Exact [17] and brunnermunzel [18]. Use citation("infercalcR") for this package and citation() or citation("<package>") for R and the packages you used.

References

  1. Meis J, Baumann L, Pilz M, Sauer L (2022). DescrTab2: Publication Quality Descriptive Statistics Tables. R package version 2.1.16. https://doi.org/10.32614/CRAN.package.DescrTab2. Source code, including the development version 2.1.26: https://github.com/imbi-heidelberg/DescrTab2.
  2. Universität Zürich, Methodenberatung. Datenanalyse mit SPSS: decision tree (Entscheidbaum), decision assistant (Entscheidassistent) and method pages with examples and explanations. https://www.methodenberatung.uzh.ch/de/datenanalyse_spss.html. Accessed 3 October 2026.
  3. Faul F, Erdfelder E, Lang A-G, Buchner A (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175-191. https://doi.org/10.3758/BF03193146.
  4. Faul F, Erdfelder E, Buchner A, Lang A-G (2009). Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41(4), 1149-1160. https://doi.org/10.3758/BRM.41.4.1149.
  5. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.
  6. Maechler M, Rousseeuw P, Struyf A, Hubert M, Hornik K. cluster: Cluster Analysis Basics and Extensions. R package. https://doi.org/10.32614/CRAN.package.cluster.
  7. Pinheiro JC, Bates DM (2000). Mixed-Effects Models in S and S-PLUS. Springer, New York. https://doi.org/10.1007/b98882.
  8. Pinheiro J, Bates D, R Core Team. nlme: Linear and Nonlinear Mixed Effects Models. R package. https://doi.org/10.32614/CRAN.package.nlme.
  9. Venables WN, Ripley BD (2002). Modern Applied Statistics with S. Fourth edition. Springer, New York. https://doi.org/10.1007/978-0-387-21706-2.
  10. Bates D, Mächler M, Bolker B, Walker S (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1-48. https://doi.org/10.18637/jss.v067.i01.
  11. Halekoh U, Højsgaard S, Yan J (2006). The R package geepack for generalized estimating equations. Journal of Statistical Software, 15(2), 1-11. https://doi.org/10.18637/jss.v015.i02.
  12. Eddelbuettel D. digest: Create Compact Hash Digests of R Objects. R package. https://doi.org/10.32614/CRAN.package.digest.
  13. Ooms J (2014). The jsonlite package: A practical and consistent mapping between JSON data and R objects. arXiv:1403.2805. https://arxiv.org/abs/1403.2805.
  14. Boschloo RD (1970). Raised conditional level of significance for the 2 x 2-table when testing the equality of two probabilities. Statistica Neerlandica, 24(1), 1-9. https://doi.org/10.1111/j.1467-9574.1970.tb00104.x.
  15. Signorell A. DescTools: Tools for Descriptive Statistics. R package. https://doi.org/10.32614/CRAN.package.DescTools.
  16. Fay MP (2010). Two-sided exact tests and matching confidence intervals for discrete data. The R Journal, 2(1), 53-58. https://doi.org/10.32614/RJ-2010-008.
  17. Calhoun P. Exact: Unconditional Exact Test. R package. https://doi.org/10.32614/CRAN.package.Exact.
  18. Ara T. brunnermunzel: (Permuted) Brunner-Munzel Test. R package. https://doi.org/10.32614/CRAN.package.brunnermunzel.