TernTables 1.7.2

New exported functions

• ternStyle(): Applies full TernTables ‘Word’ formatting to any user-supplied tibble — font, header shading, borders, caption, footnote, and citation footer — without running it through the ternG()/ternD() pipeline. Supports subheader_rows, bold_rows, italic_rows, bold_cols, italic_cols, bold_sig (cell-level p-value bolding for non-standard column names), and header_format_follow. Returns a tibble with a ternB_meta attribute for direct use in ternB().

• classify_normality(): Exposes the internal ROBUST normality routing algorithm as an exported function. Returns a tidy tibble with per-variable × per-group statistics (n, skewness, excess kurtosis, Shapiro-Wilk p), the gate that triggered the routing decision, a plain-language gate_reason, and the final parametric/non-parametric routing outcome. Intended for manuscript auditing and responding to reviewer questions about normality assessment.

New parameters

ternG()

• force_normal — Bypass all normality assessment for listed variables; always route to mean ± SD and Welch tests. Per-variable counterpart to force_ordinal. force_ordinal takes priority when a variable appears in both.
• force_continuous — Bypass automatic binary 0/1 detection for listed variables so they are analysed as continuous rather than converted to Y/N categorical.
• show_p — When FALSE, suppresses the P value column and all associated columns (OR, test, normality), producing a descriptive-only grouped table.
• percentage_compute — "column" (default) or "row". Controls the denominator for categorical percentages. When "row", the Total column is automatically suppressed.
• categorical_posthoc — For three-or-more-group comparisons, computes Haberman’s adjusted standardized residuals following a significant omnibus test (p < 0.05). Cells exceeding ±1.96 are annotated with *. The ±1.96 threshold derives from the omnibus chi-squared distribution; no additional multiple-comparisons correction is required.
• p_adjust — When TRUE, applies Benjamini-Hochberg (BH) false discovery rate correction to all omnibus P values.
• p_adjust_display — "fdr_only" (default) replaces the P column with BH-corrected values; "both" retains raw values alongside the corrected column.
• show_missing — When TRUE, appends a Missing: n (%) sub-row beneath each variable showing per-group missing counts. Footnote added automatically.
• show_missingness — FALSE (default), "total", or "group". Appends dedicated missingness percentage column(s): one overall column ("total") or one interleaved column per group ("group"). Counts both NA and string representations of missing data.
• missing_indicators — Custom character vector that replaces the built-in string-NA list when computing missingness columns (requires show_missingness to be active). Matching is case-insensitive.
• zero_to_dash — When TRUE, replaces "0 (0%)" cells with "-" in categorical output.
• round_decimal — Integer; overrides the default 1 decimal place for all continuous summary values. Ignored when round_intg = TRUE.
• font_family — Font used throughout all ‘Word’ output. Defaults to getOption("TernTables.font_family", "Arial"); set package-wide with options(TernTables.font_family = "Times New Roman").
• citation — When TRUE (default), embeds a full package citation as a page footer in every exported .docx and at the end of the methods document.
• open_doc — When TRUE (default), opens the written .docx in the system default application after saving. Set FALSE to suppress (useful in scripted or server-side workflows).
• plain_header — Named character vector (same interface as category_start). Inserts a label-only row with underline formatting and no bold, merge, or border treatments. Also accepted by word_export().
• abbreviation_footnote, variable_footnote, index_style — Structured footnote system. abbreviation_footnote prints first. variable_footnote (named character vector) auto-assigns *, †, ‡ … as superscripts to named variables in column 1 and appends definitions below the table. index_style = "alphabet" uses Unicode superscript letters instead. Pipe- separated keys ("Var A|Var B") assign one shared symbol and footnote to multiple variables.

ternD(): force_normal, force_continuous, plain_header, show_missing, show_missingness, missing_indicators, zero_to_dash, round_decimal, font_family, citation, open_doc — all behave identically to the ternG() descriptions above. show_missingness = "group" raises an informative error directing the user to ternG().

ternP()

• mode — "auto" (default) or "manual". In "manual" mode, the PHI column-name hard-stop is skipped and a prominent warning is emitted; all other cleaning steps still run. Intended for datasets that trigger false-positive PHI flags (e.g. a column named "Address" that contains a non-PHI clinical field). Use drop_cols to explicitly remove identifier columns in this mode.
• extra_na — Character vector of additional strings to treat as NA, appended to the built-in list. Example: extra_na = c("9999", "Not Done", "PENDING").
• drop_cols — Character vector of column names to drop before cleaning begins. Names not found in the data are silently ignored.

The preprocessing summary now includes a Missingness Summary section listing every column with at least one missing value (count and percentage), computed on the fully cleaned data. The section is suppressed when no missing values exist.

word_export(), ternB(), ternStyle(): font_family, round_decimal, citation (as described for ternG()). word_export() and ternStyle() additionally accept bold_sig (named list for cell-level p-value bolding of custom tibbles with non-standard p-value column names).

write_cleaning_doc(): open_doc and citation added for consistency with all other Word-output functions.

write_methods_doc() redesigned: generates a single dynamic paragraph tailored to the actual run (descriptive, two-group, or three-or-more-group) rather than a fixed three-section boilerplate. boilerplate = TRUE writes a comprehensive reference document covering all five standard configurations. ternB(methods_doc = TRUE) generates one labeled methods section per table, deduplicating tables with identical configurations.

Bug fixes

• fisher.test() segfault (ternG): Tables larger than 2×2 now route directly to Monte Carlo simulation (simulate.p.value = TRUE, B = 10,000). Previously the C-level exact algorithm could segfault on tables with many levels, killing the R session — a condition that tryCatch cannot intercept.
• Degenerate single-level categorical crash (ternG): Variables reduced to one non-NA level after string-NA removal now display NA (insufficient variation) for the p-value instead of crashing the entire ternG() call.
• methods_filename = NULL crash (ternG, ternD, write_methods_doc): Passing NULL explicitly now resolves to the default "TernTables_methods.docx" rather than throwing a dirname(NULL) error.
• categorical_posthoc crashes (ternG): Fixed three independent issues — (1) stdres returning a vector instead of a matrix caused an if() condition to error; (2) NA factor level names bypassed %in% guards causing subscript-out-of-bounds; (3) cat_posthoc_fisher_display was referenced before being extracted from .ternG_env.
• Blank-string factor levels (ternG): Factor levels named "" (common in UNOS/SRTR haven_labelled data) produced NA (NA%) because as.data.frame.matrix() silently renames "" to "X" via make.names(). Fixed by saving column names before conversion and restoring them explicitly; all 2D cell lookups now use integer indexing via match().
• Blank page between ternB() tables: Stripped the default blank paragraph from each temp document, eliminating spurious blank pages in combined output.
• Citation footer bleed in ternB(): Page footer from one table’s temp document no longer carries over into subsequent tables in the combined file.
• CLD dependency resolution: rstatix and multcompView moved to Imports (were Suggests), ensuring compact letter display superscripts always work without a separate install step.
• Wide-table page overflow (word_export): fit_to_width(6.5) is now applied after autofit() only when the table exceeds Letter page width (6.5 in), preventing column truncation in ‘Word’ and PDF exports.
• Name-cleaning false positives: Fixed .apply_cleaning_rules() patterns that incorrectly transformed variable names containing medical abbreviations and unit suffixes (e.g. "Gy" in radiation dose variables).
• bold_sig column-name mismatch (word_export, ternStyle): word_export() renames columns internally (e.g. "P" → "P value", spaces → \n when line_break_header = TRUE). The bold_sig lookup was comparing caller-supplied names against already-renamed columns, so HR bolding was silently skipped. Fixed by building an original-to-renamed name map before the lookup.

Internal changes

• CRAN compliance — <<- eliminated: All accumulator patterns using <<- inside nested closures in ternG.R, ternD.R, and ternP.R replaced with new.env(parent = emptyenv())-based counters. No user-visible change.
• set.seed() replaced with withr::with_seed(): RNG state is now scoped locally within the Monte Carlo Fisher’s exact fallback, restoring the caller’s state after the call. withr added to Imports.
• ROBUST Gate 2 — kurtosis added: Excess kurtosis (|excess kurtosis| > 7) now triggers non-parametric routing alongside absolute skewness > 2. Normality decision logic extracted to utils_normality.R.
• CLD letter ordering: Removed center-based letter re-mapping; multcompLetters() default alphabetical ordering is now used directly, aligning with standard CLD conventions.
• PHI detection tightened: patient_id, subject_id, participant_id, and clinical-event date patterns removed from the flag list; only personal-identity patterns (DOB, DOD) remain.
• Shared missing-value helpers: .tern_missing_strings() and .is_missing_value() extracted to utils_preprocess.R as a single source of truth used by both ternP() and the show_missingness columns in ternG()/ternD().
• Independence-of-observations scope: Explicit notes added to the package help page, ternG() description, vignette, and README stating that all tests assume independent observations.

TernTables 1.6.4

• CRAN resubmission addressing two reviewer comments on v1.6.3.
• Eliminated all uses of <<- in R/ternG.R, R/ternD.R, and R/ternP.R. Accumulator patterns replaced with new.env(parent = emptyenv())-based counters; the dplyr::across() lambda in ternP.R replaced with an explicit for loop. No change to user-visible behavior.
• Replaced bare set.seed() with withr::with_seed() in the Monte Carlo Fisher’s exact fallback, scoping the seed locally and restoring the caller’s RNG state. withr added to Imports.
• Removed Maria V. Aslam from Authors@R at her request.

TernTables 1.6.3

• CRAN resubmission addressing reviewer comments on v1.3.1.
• All \dontrun{} blocks replaced with \donttest{} across all exported functions. Examples that write files to disk use \donttest{} because the ‘Word’ export operations may exceed 5 seconds on slower machines; no missing software or API keys are required.
• Software and package names now quoted in the Description field of DESCRIPTION per CRAN policy: ‘Word’, ‘Excel’, ‘tibble’, ‘officer’, ‘flextable’, ‘writexl’, ‘rstatix’.
• Package-level documentation (?TernTables) updated: normality routing description corrected to four-gate algorithm; .onLoad now explicitly initialises TernTables.line_break_header option.
• val_format() now uses the Unicode plus-minus symbol (±) for consistency with ternG() and ternD() output.
• utils_naming.R: single-word variable names (e.g. age, sex, race) now correctly flow through the abbreviation map and capitalisation rules in .apply_cleaning_rules().

TernTables 1.6.2

• Added odds ratio (OR) support for two-level categorical variables in ternG() (e.g. Male/Female shown with 1.00 (ref.) and computed OR with 95% CI). Previously only Y/N and 0/1 binary variables received ORs.
• Added factor_order = "mixed" option to ternG() and ternD().
• Post-hoc compact letter display (CLD) superscripts implemented in utils_posthoc.R with center-based letter ordering (highest center = “a”).
• Fallback to Monte Carlo simulation for Fisher’s exact test when the exact algorithm cannot complete due to workspace limits (B = 10,000; seed fixed via getOption("TernTables.seed")).
• Liberalized PHI column name detection in utils_preprocess.R to reduce false positives on common research variable names.

TernTables 1.6.1

• Integrated tern-tables.com web app into all public-facing documentation: README, vignette, and package help page (?TernTables).
• Added URL: and BugReports: fields to DESCRIPTION.

TernTables 1.6.0

• Added ternP(): preprocessing function for raw CSV/XLSX data. Handles string NA conversion, whitespace trimming, empty column and blank row removal, and case normalisation. Hard-stops on PHI column name patterns and unnamed columns with data.
• Added write_cleaning_doc(): writes a ‘Word’ audit document recording every transformation applied by ternP().
• Added bundled example data tern_colon_messy.csv to inst/extdata/csv/ for use in ternP() examples.
• line_break_header parameter refined: improved behaviour for edge-case column widths in ternG(), ternD(), and word_export().

TernTables 1.5.0

• Added ROBUST normality routing (consider_normality = "ROBUST", now the default). Four-gate algorithm applied per group: (1) n < 3 fail-safe to non-parametric; (2) |skewness| > 2 to non-parametric; (3) all groups n ≥ 30 to parametric via CLT; (4) Shapiro-Wilk p > 0.05 to parametric. Implemented in both ternG() and ternD().

TernTables 1.4.0

• Added ternB(): combines multiple TernTables result tibbles into a single formatted ‘Word’ document.
• Added table_caption parameter to ternG(), ternD(), and word_export(): places a bold caption above the table in the ‘Word’ output.
• Added table_footnote parameter to ternG(), ternD(), and word_export(): adds a merged footer row below the table.
• Added Welch ANOVA for 3+ group continuous comparisons (parametric path); previously only Kruskal-Wallis was available for 3+ groups.
• Added line_break_header parameter to ternG(), ternD(), and word_export(): wraps long column headers onto two lines in ‘Word’ output.
• write_methods_doc() now generates a methods paragraph tailored to ternB() multi-table output.
• Fixed bug where P values were not displaying in three-group comparisons.
• Fixed dynamic bolding/non-bolding of table captions.

TernTables 1.3.1

• Initial CRAN submission.
• ternG(): grouped comparison table for 2- and 3-level group variables, with optional odds ratios (OR_col), normality testing, and post-hoc test framework.
• ternD(): descriptive summary table with no group comparisons.
• word_export(): exports any TernTables tibble to a formatted ‘Word’ document.
• write_methods_doc(): generates a boilerplate statistical methods paragraph.
• val_format() and val_p_format(): formatting utilities for publication-ready numeric and P value display.