statuser 0.4.1 (2026-10-31)

Improvements

• plot_freq(): character and factor inputs are supported (previously bare character vectors failed with an obscure xlim error). Thanks to Daniel Bankin for the report. New x.order controls category order on the x-axis: factor levels (default for factors), alphabetical, frequency ("freq" / "freq.asc"), -1 to reverse the default, or a custom name vector.
• interprobe(): draw = "none" skips plotting (and save.as) while still returning simple slopes, Johnson–Neyman results, and related output.
• interprobe(): when histogram = TRUE, the sample-size distribution is drawn as its own panel under each curve plot (shared x-axis) instead of being carved out of the GAM y-axis, so frequency labels no longer collide with axis ticks.
• Documentation expanded for interprobe() (inputs, model paths, how x/z are classified, spotlights, plot options, and return value); \usage lines wrapped to satisfy CRAN Rd line-width checks. Added the taco spiciness example via the bundled tacos_data dataset and the interprobe vignette.
• Package-wide pass over user-facing messages: spelling/grammar fixes and clearer wording in interprobe(), stimulus.plot(), and related helpers.

Bug fixes

• t.test2(): t.test2(outcome, group) now runs a two-sample test when group is a character, factor, or logical vector with two levels (for example t.test2(df$y, df$group)). That call used to be passed to t.test() as a second sample, which failed or returned an empty result. Thanks to Rafael Mancuso (https://github.com/raffaelemancuso) for the report (https://github.com/urisohn/statuser/issues/7). Formals stay function(x, ...) so the S3 method for t remains consistent with the generic.
• lm2() print tests: exact-fit models (undefined SE/t/p under current estimatr) no longer assert significance stars or >999.
• interprobe(): discrete focal predictors with a continuous moderator on the x-axis now build frequency bins along the moderator (and return bin_from / bin_to only when breaks exist), fixing empty-break assignment errors.
• interprobe(): Johnson–Neyman panel now shows the spotlight legend for discrete focal predictors (previously only categorical and continuous cases got a JN legend).

statuser 0.4.0 (2026-10-09)

New functions

• Added round2(): like round(), but always returns a character vector with the requested number of decimals. By default (leading.zero = FALSE), drops leading zeros when every finite value has |x| < 1 (e.g. 0.34 → .34); mixed vectors keep leading zeros for all elements.

Improvements

• interprobe(): Johnson–Neyman regions of significance now report what share of the sample falls in each range, e.g. [2.50 to 4.80] (23.4% of the sample).
• interprobe(): moderator.on.x.axis now defaults to "auto" (moderator on the x-axis for binary focal predictors; focal predictor on the x-axis when it has more than 3 values).

Bug fixes

• print.lm2(): undefined SE / t / p (exact fits under estimatr 2.0.1+, or dropped coefficients) now print as -- instead of erroring in column alignment. Thanks to Alex Coppock for the report.
• stimulus.plot(): null.method = "demean" now uses the two condition-mean columns from get.means.condition() (previously selected the stimulus-label column and one mean, which errored with character stimulus IDs). Thanks to ceo@unicupcompany.com for discovering this.
• stimulus.plot(): with flip.conditions = TRUE, mean columns are named from the factor levels actually used in the t-tests (previously names always followed sorted condition order while contents followed the reversed levels).
• stimulus.plot(): null.method = "demean" heterogeneity p-values now use the same flipped observed effects as the null resamples (previously compared unflipped observed effects to flipped null draws).
• interprobe(): continuous Johnson–Neyman grids rename x/z to the real variable names before adding covariates at their means (avoids duplicate-name newdata columns).
• interprobe(): bare (unquoted) names for x, z, and y now work with data= and model= even when those names exist only as columns (not in the calling environment). Missing columns report e.g. 'value' is not a variable in data.frame 'df'.
• interprobe(): cols of length 2 is allowed when the focal predictor has two levels (previously required length 3 even for binary x).

statuser 0.3.1 (2026-06-24)

Improvements

• lm2(): improved display rounding for printed regression tables. Near-zero numerical noise (e.g. from near-perfect fits) displays as .000 instead of long scientific-notation strings; values with |x| < 1 always omit the leading zero (including zero). Extreme t statistics display as >999. New optional round argument on lm2() and print.lm2(): NULL uses magnitude-based rules, round = -1 uses R default format(), and non-negative values fix decimal places.
• plot_cdf(): xlim passed via ... is now fully honored — ECDF curves are drawn over the requested range, not only the data range.
• table2(): chi2= is accepted as an alias for chi= (if both are passed, they must agree).

Bug fixes

• plot_freq(), plot_density(), plot_cdf(), plot_means(), lm2(), and related helpers: automatic titles and axis labels now strip data-frame prefixes from $ expressions while keeping the rest of the expression.
• plot_means(): user-supplied ylim is respected — the axis is no longer pulled down to 0 for n= labels; the lower limit may extend slightly below your requested minimum only to make room for sample-size labels.
• table2(): chi-square test again prints in APA format (χ²(df) = ..., p = ...) when printing results with chi=TRUE.

statuser 0.3.0 (2026-05-20)

New functions

• Added stimulus.plot() for stimulus plots in matched/treated-stimulus designs (means and effects plots with null resampling).
• Added stimulus.beeswarm() for compared-stimulus designs.
• Added clear_stimulus_cache() to reset resampling cache used by stimulus plotting functions.

Improvements

• t.test2(): now reports Cohen’s d for two-sample independent tests (returned as d and printed after sample sizes).
• t.test2(): missing-value note no longer double-counts rows where the grouping variable is NA.
• lm2(): explanatory notes are no longer printed with the regression table; they are stored when the model is printed and shown via the new lm2_notes() function. The print output ends with a reminder to run lm2_notes().

Deprecated

• lm2() and print.lm2(): the notes argument is deprecated (ignored). Passing notes = TRUE or notes = FALSE emits a message directing users to lm2_notes().

statuser 0.2.1 (2026-04-25)

changes to pass CRAN inspection

• plot_means(): removed the save.as argument (no longer saves plots to PNG/SVG).
• t.test2(): documented S3 registration as t.test2 for base::t to avoid recurring CRAN S3 consistency notes while keeping direct t.test2(...) usage unchanged.
• desc_var(): removed dependency on labelled; now relies on new function var_labels

New function

• Added var_labels() to get/set these labels for vectors and data frames. This avoids depending on labelled

statuser 0.2.0 (never on CRAN)

New functions

• Added plot_means() for barplots of group means with confidence intervals, optional comparisons and interaction-style contrasts, up to three grouping variables in the formula, optional clustered standard errors, and saving to PNG or SVG.
• Added interprobe() for probing interactions: simple slopes and Johnson–Neyman style marginal-effect curves for x × z, using GAM or linear models.

Improvements

• plot_freq(): improved behavior when there are many distinct values; grouped plots support up to four levels of the grouping variable.
• lm2(): mean column shows percentages for factor levels and factor×factor interactions as intended; baseline factor level row shows the percentage of observations like other levels.
• desc_var(): when the response variable is missing, only the intended error is shown (no spurious “restarting interrupted promise evaluation” warning).
• scatter.gam(): supports direct coordinates and formula syntax (including df$y ~ df$x); axis labels use short variable names for $ formulas; ylim and other ... arguments no longer conflict with internal plot arguments; small vertical buffer on the lower frequency panel.
• Messages that used red4 now use red2 for readability in dark terminals (including messages from plot_freq(), plot_density(), plot_cdf(), clear(), twolines(), and exit()).
• t.test2() performance improvements.

Bug fixes

• desc_var() evaluates calls such as df$x correctly when data is not supplied, instead of mistakenly using a same-named object from the search path.
• text2() applies vector arguments passed in ... (e.g. col) per label, fixing wrong colors when labels differ (notably for plot_means() value labels).
• plot_cdf() restores only selected graphics parameters on exit (mar, mgp) so layouts using par(mfrow = ...) are not reset. Similar graphics-state cleanup was applied in other plotting functions where appropriate.

statuser 0.1.9 (2026-02-26)

• plot_freq(), plot_density(), and plot_cdf(): optional comparison of two vectors (in addition to formula syntax); legend placed higher on the plot.
• table2(): clearer dimension labels for expressions such as df$var > 10; very long labels are truncated.