The tabxplor options, and their defaults

tabxplor reads its display, colour, statistics and export defaults from options(), all prefixed tabxplor.. Set any of them for a session with options(), e.g. options(tabxplor.stars = TRUE), or once at the top of a script or .Rmd. The defaults are established when the package loads (.onLoad()); most also have a per-call argument on the relevant function, which always wins over the option.

Display and printing

tabxplor.print: "console" (default): how a table auto-prints. "html" renders the tab_html() table (in the Viewer pane in RStudio/Positron, and as a real html table in rmarkdown/Quarto documents) — recommended when you work in an IDE with a Viewer. "kable" is an accepted synonym of "html" (the pre-2.0.0 name).
tabxplor.stars: FALSE (default): whether cells show significance stars, and at which cut-offs. FALSE (no stars), TRUE (the default ladder c("*" = 0.10, "**" = 0.05, "***" = 0.01)), or a named numeric giving your own – names are the glyphs, values the p-value cut-offs, e.g. options(tabxplor.stars = c("*" = 0.05, "**" = 0.01)). Off for tab(), on for tab_reg(). The LADDER is a render-time reading of each cell's stored p-value, so it is this option alone — change it and every table already built shows the new glyphs. Per-call stars =.
tabxplor.ratio_print: "inverse" (default): prints a multiplicative value below its reference as the inverse — an odds ratio of 0.37 as 1/2.67, a mean ratio of 0.42 as /2.4 — so "2.7 times less" reads as strongly as "2.7 times more", and the same in a bracket. "raw" prints the plain number (0.37), the convention of most journals.
tabxplor.n: "range" (default): how many people a table says are behind its numbers. "range" puts the unweighted base beside the Total cell of a crosstab (100% (9 838)) and in the n column of a regression table, printed as min-max when the blocks rest on different populations — several column variables losing different NAs, or several models. "min" prints the smallest base only, "no" shows no count at all. It replaces the add_n argument, deprecated in 2.0.0. Per-call n =.
tabxplor.color_whole_cell: FALSE (default): EXPERT. A cell that prints SEVERAL fields reads as one number with an aside — 1/1.63*** (31%) — so the cell's rendering grades the number and the aside is set slightly back from the table's own text, following the theme. That covers the colour and, under theme = "print", the typography (bold, italic, underline) alike. Set to TRUE to extend the primary's own rendering over the whole cell instead (the pre-2.0.0 look). There is nothing to choose beyond that: which grey an aside takes belongs to the theme's palette, not to a per-cell option — see set_color_palette(). Console, html and Markdown; Excel renders a cell as a whole either way.
tabxplor.var_names: "both" (default): which variable names the exporters annotate: "both", "rows", "cols", "none". Per-call var_names =.
tabxplor.var_labels: FALSE (default): in exports (markdown / html / Excel / plot), show a variable's label (the haven/labelled label attribute, if it has one) instead of its name. Display only – the table structure keeps canonical names, so name-based select() and references still work; the console always shows names.
tabxplor.cleannames: FALSE (default): clean up variable/level names in output. Also strips a "1-"-style prefix from labelled value labels turned into factor levels. Per-call cleannames =.
tabxplor.total_names: c(row = "Total", col = "Total", tab = "Ensemble", other = "Others") (default): the four synthetic labels a table carries: row and col name the total row and the total column, tab the total table (the one made when there are tab_vars), and other the level other_if_less_than lumps small levels into. A partial vector is allowed – options(tabxplor.total_names = c(tab = "Ensemble", other = "Autres")) leaves the first two alone. It replaces the total_names / totaltab_name / other_level arguments, deprecated in 2.0.0.
tabxplor.shape_auto_max: 12L (default): where shape = "auto" draws the line for a numeric row or tab variable: a column with at most this many distinct whole values is a counted number or a short scale, and keeps one level per value; anything else is continuous and is cut into "sd_bands". Raise it for a long scale, lower it to band more eagerly — or name the variable in shape and decide yourself.

Colours and theme

tabxplor.color_breaks: the colour-break scales (a named list of pct_diff, pct_ratio, odds_ratio, mean_diff, mean_ratio, contrib, zscore, adj_ratio, adj_diff, adj_diff_std). Set with set_color_breaks(). Per-call color_breaks =.
tabxplor.color_style_theme (alias tabxplor.console_theme): the console palette theme, "light" or "dark"; set by set_color_palette() (which auto-detects the editor theme on load). NOT the export theme (tabxplor.theme / tabxplor.export_theme).
tabxplor.console_bold: whether to embolden the reference / total (and coloured) cells in the console, TRUE or FALSE. Auto-detected at load: TRUE in Positron and VS Code (which render ANSI bold at a fixed glyph width), FALSE in RStudio and unknown consoles (there bold is drawn wider and would break column alignment). Override it for your own front-end / font.
tabxplor.theme (alias tabxplor.export_theme): "light" (default): the export theme: "light", "dark", "auto" (follow the reader), or a black-and-white publication palette – "print_ready" picks one per table (marks for a cross-table, the emphasis ladder for a regression), or name it yourself: "print_marks", "print_emphasis", "print_minimalistic" ("bw" is a synonym of the last). See tab_css() for what each says. "auto" needs a stylesheet, so only tab_html(), tab_md() and tab_css() honour it; static backends resolve it to "light". A publication palette reaches every backend, Excel included. Per-call theme =.
tabxplor.print_rules: TRUE (default): every stylesheet tab_css() emits also carries a black-and-white publication palette inside an @media print block, so a table rendered in colour prints (or saves to PDF) publication-ready with no further action. Set FALSE if your printer is a colour one and the colours are the point, or name a palette ("print_emphasis") to print in that one instead of the default "print_minimalistic". "print_marks" and "print_ready" cannot be used here: their marks are cell text, and a print rule can restyle a page but not add characters to it. Per-call print_rules =.
tabxplor.background: "page" (default): what a rendered table paints behind itself. "page" leaves it transparent, so the table sits on the page's own ground, whatever that is; "theme" paints the theme's own background, a card of its own; or name any CSS colour. Change it only where the page is not yours to follow — a dark table dropped into a light document, an html email. The interactive Viewer page paints itself either way, and a publication palette is always a sheet of white paper.

Statistics and confidence intervals

tabxplor.anova: "welch" (default): which one-way ANOVA F is shown for mean columns: "welch" (robust) or "classic" (pooled variance). Both are always stored in the test attribute. Per-call anova =.
tabxplor.design_effect: FALSE (default): a weighted tab() estimates the population but bases every interval and test on the raw number of respondents, so they carry no design effect — and the table's footer says so. Set TRUE and the same intervals account for the unequal weighting, exactly: a weight column IS a survey design (the flat one, ids = ~1), whose variance has a closed form in the per-cell sum(w^2) the aggregate already computes, so the base becomes n_eff = p(1-p) / Var_design(p) in every weighted descriptive confidence interval — factor proportions and means (cell, difference, ratio and the color = "odds_ratio" significance) — and the whole-table tests (test = TRUE) become survey::svychisq / a svyglm Wald F on that flat design. It reproduces survey to the last digit, Kish's (sum w)^2 / sum(w^2) being that same formula with each cell's own sum(w^2) discarded. Being exact rather than a bound, it can make an interval narrower as well as wider. It is blind to clustering and to calibration, which the weights do not record — and those are not symmetric: missing the calibration and the strata costs a few percent, in the safe direction, while missing the clusters of a face-to-face household survey can leave an interval several times too short (see the Weights section of vignette("tabxplor")). It needs the microdata weights, so tab_counts() on pre-aggregated counts cannot apply it (such a table states the raw basis in its footer rather than claiming a correction it does not have). Scope: tab() and its leaves only. tab_reg() never reads it — its crude empirical = companions are always on the weighted basis, beside a model column (survey::svyglm) that always was. For the full design effect — strata, clusters, fpc, calibration — pass a survey::svydesign as data; the option is then not consulted at all. Per-call design_effect =.
tabxplor.conf_level: 0.95 (default): confidence level for the intervals and significance tests. Since 2.0.0 each column records the level it was built at, so the colour thresholds follow the argument and this option is the fallback for a column that never recorded one (a hand-built fmt(), or a table from an older session). Per-call conf_level =.
tabxplor.legend_style: "prose" (default): the colour-legend style in exports (tab_md(), tab_html(), tab_xl()): "prose" (full sentences) or "terse" (the compact one-line form the console uses). The console itself is always terse.
tabxplor.test_lines: "summary" (default): how many crosstab test rows the exporters (tab_md(), tab_html(), tab_xl()) append: "summary" (p-value + effect size), "all" (+ the raw statistic), "stat" (p-value + statistic), or "pvalue" (the single p-value row). The p-value row name states the test used ("pvalue (Chi2, Welch F; survey-design)") and the effect-size row its measure ("Cramer's V, eta2"). N is never added – it is already shown by the n column. The console block always shows N + p-value + effect size.
tabxplor.shape_table (alias tabxplor.spark): "all" (default): in a tab_reg() table, each continuous predictor's OBSERVED SHAPE — the outcome binned against the predictor, on the model's scale, with no model in it — drawn as a small curve in a shape table below the footer, beside the range it is a picture of (13-57% (OR 8.7)). It is the eye-half of the Linearity footer row, and the free one: no fit is involved. With tab_vars, one curve per group; with several outcomes, one per outcome. "all" draws it in every medium; "console" only where you are working, so exported tables stay unchanged; "no" never. TRUE / FALSE are accepted for the first and the last. The curve is drawn TO SCALE on the predictor as the model sees it, so a shape transform visibly straightens it when it is the right cure, and every predictor's curve is the same width. Its vertical window is floored by the data's own sampling noise, so a curve smaller than that is greyed and marked ns — read it as a flat line whatever its shape. In HTML the glyphs become an inline SVG; a plot never draws them (no graphics-device font has them). An ordinal or multinomial outcome has one curve per cut or per category and this draws only the first: reg_check_plots() shows them all.

HTML / `tab_html()` export

tabxplor.tab_kable_css (alias tabxplor.kable_css): TRUE (default): inline the stylesheet with each tab_html() / tab_md() table (self-contained). Set FALSE in a many-table document that emits tab_css() once at the top. Per-call css =.
tabxplor.tab_kable_tooltips: TRUE (default): show the per-cell hover tooltips (counts, confidence intervals, differences...) in html tables. Set FALSE once per document when every table auto-prints and tooltips are unwanted. Per-call tooltips =.
tabxplor.kable_popover: FALSE (default): use click popovers instead of hover tooltips. Per-call popover =.
tabxplor.tab_kable_num_font: the HTML/markdown number-font CSS stack. Monospace by default so figures stay column-aligned (set a proportional stack to revert).
tabxplor.output_kable: FALSE (default): make tab() render its result with tab_html() before returning it — a convenience for .Rmd/.qmd documents. Since 2.0.0 it only renders: it no longer changes the shape of the built object (that is output_list).

Excel / `tab_xl()` export

tabxplor.xl_font_text: "DejaVu Sans Condensed" (default): text (labels/headers) font. Per-call font_text =.
tabxplor.xl_font_num: "DejaVu Sans" (default): number font without stars. xlsx records ONE name (no fallback list), so set a font installed where the workbook is opened. Per-call font_num =.
tabxplor.xl_font_num_stars: "Cascadia Mono" (default): number font with stars (monospace, so stars align). Per-call font_num_stars =.
tabxplor.xl_ratio_cells: "fold" (default): what a ratio / odds-ratio cell HOLDS in the workbook: "fold" (the default) the signed fold, so Excel prints what the console prints and the cell stays a number; "raw" the untransformed ratio; "text" the exact display string. Per-call ratio_cells =.

Plot, paths and language

tabxplor.export_dir: NULL (default): default directory for exported files (NULL = the working / typed path).
tabxplor.lang: "auto" (default): the colour-legend language: "auto" (follows the R/OS locale), "en" or "fr". Per-call lang =.

Parallel build

tabxplor.parallel: FALSE (default): build the independent units of one call on parallel CPU cores (needs the mirai package): the per-row_var tables of a tab(), the models / tab_vars groups / outcomes of a tab_reg(). The result is byte-identical to the serial one. FALSE (default) never dispatches. "auto" (or TRUE) takes half the cores this session may actually use, at least 2 and at most 4 — so 2 on a dual-core laptop, 2 on a 4-core machine, 4 on 8 cores or more. An integer takes that many verbatim. The count respects options(mc.cores), a container's CPU quota and an HPC allocation (it reads them through the parallelly package when installed), and never exceeds 2 under R CMD check. It stays OPT-IN because starting the pool BLOCKS for about a second, so the first parallel table of a session is always slower than the serial one; from the third it is ahead. It pays off for MANY evenly sized units against a small or medium data frame — 24 tables run about 2.8x faster on 4 workers — and is a loss for few units or multi-million-row data, where shipping the population to each worker eats the gain. Set it once at the top of a script: options(tabxplor.parallel = "auto"). A model comparison (stats = "compare_*") is always serial and says so when asked: it is a test BETWEEN the fits, so they are built together. For one call only, wrap it in withr::with_options(list(tabxplor.parallel = "auto"), ...). The pool persists for the session; release it with tab_parallel_stop().
tabxplor.parallel_min: 2L (default): the smallest UNIT count worth dispatching – row_vars for tab(), models for tab_reg() (fewer runs serially, since the setup would outweigh the gain).