What's New in jjstatsplot 1.0.52

ClinicoPath

2026-08-14

This release adds one analysis, changes the behaviour of several options, and tightens what the module tells you about your own data. The behaviour changes matter more than the new features: a few of them alter numbers or labels you may already be quoting, so read the “Changed behaviour” section before upgrading a running project.

New analysis: Dot Chart

jjdotchart() — Dot Chart (Summary vs Reference Value) — collapses each group to a single summary point and tests those points against a reference value.

jjdotchart(
  data           = mydata,
  dep            = "biomarker_level",   # continuous outcome
  group          = "hospital_site",     # one point per level
  testvalue      = 12,                  # the reference to test against
  typestatistics = "parametric"
)

The thing to understand before using it: n is the number of groups, not the number of patients. Ten sites with 500 patients each is n = 10 for this test, with 9 degrees of freedom. The analysis states this in its own panel, and the summary table reports the per-group n so the aggregation stays auditable.

Useful options: grvar (split into panels), centralityplotting / centralitytype (add a labelled centrality line, which is not the reference line), conflevel, k (decimal places), showSummaryTable.

Changed behaviour

Line Chart — reference lines have their own switch

refline = 0 previously meant “no line”, which made the most common clinical reference impossible to draw: zero is where change-from-baseline, a difference and a log fold-change all sit. There is now a showRefline switch.

linechart(
  data        = mydata,
  xvar        = "visit_week",
  yvar        = "change_from_baseline",
  showRefline = TRUE,     # <- now required
  refline     = 0,        # <- and zero finally works
  reflineLabel = "No change"
)

Scripts that relied on a non-zero refline alone need showRefline = TRUE adding. Nothing else about the line changed.

Line Chart — the confidence band is a model fit

With confidence = TRUE the shaded band is the 95% interval of a fitted model, not the uncertainty of the plotted points: a straight-line fit by default, or a LOESS fit when smooth = TRUE. The drawn line connects the observed values, so band and line describe different things unless you enable smooth. The analysis now says so in its panel.

Bar Charts and Pie Charts — no more phantom Fisher test

On a 2×2 table with low expected counts, these analyses used to suggest switching typestatistics to "nonparametric" “to obtain Fisher’s exact test”. That switch never produced a Fisher test — the plotting package computes the same uncorrected Pearson chi-square for every frequentist option — and the copy-ready Methods sentence went as far as naming a test that had not been run.

Now:

• the chart subtitle shows Fisher’s exact test itself (with the odds ratio and its confidence interval) whenever the table is 2×2 and an expected count falls below 5;
• larger sparse tables keep the chi-square subtitle and the panel hands you the exact p-value to quote instead;
• the Methods sentence names the test that actually ran.

If you previously quoted a p-value from a sparse 2×2, re-check it — the exact test can fall on the other side of 0.05 from the chi-square it replaced.

Segmented Total Bar Charts — chi-square is now opt-in

The chi-square here is computed on the summed Value Variable, which is a contingency table only when that variable counts cases. Integrality cannot establish that: a whole-number measurement passes too, and the statistic then scales with the unit you measured in. Statistical tests therefore now require y_is_count = TRUE, an explicit statement that the variable counts cases.

jjsegmentedtotalbar(
  data                   = mydata,
  x_var                  = "treatment_arm",
  fill_var               = "response_category",
  y_var                  = "n_patients",
  y_is_count             = TRUE,        # <- required for the test
  show_statistical_tests = TRUE
)

The summary table also separates Rows Analysed from Summed Value; the single “Total Observations” column used to report the sum, which read as a patient count.

Automatic Plot Selection — sampling is disclosed and configurable

sampleLarge draws a random subset of large datasets for plotting speed, and every statistic is then computed on that subset. The panel now says so explicitly rather than only reporting a smaller row count, which read like missing-data exclusion.

The threshold and the retained size are no longer hard-coded:

statsplot2(
  data            = mydata,
  dep             = "tumor_response",
  group           = "treatment",
  sampleLarge     = TRUE,
  sampleThreshold = 50000,   # only sample above this many rows
  sampleSize      = 20000,   # keep this many
  seed            = 42       # reproducible draw
)

Discarding rows costs power, so prefer turning sampling off before reporting a result. The reported observation count now also counts usable observations rather than rows, so it no longer over-states N when values are missing.

Analyses without a dedicated vignette yet

These ship in this release but do not yet have a walk-through of their own. The option lists below are complete and current; fuller vignettes are planned.

• hullplot() — Hull Plot. Draws concave hulls around clusters. Key options: x_var, y_var, group_var, color_var, size_var, hull_concavity, hull_alpha, hull_expand, confidence_ellipses, outlier_detection, show_statistics.
• jjsegmentedtotalbar() — Segmented Total Bar Charts (100% stacked). See the section above for the options that changed.
• statsplot2() — Automatic Plot Selection. Chooses the plot and test from the variable types; see the sampling section above.
• raincloud() — covered in 08-advancedraincloud alongside advancedraincloud().

Not in this release

Some vignettes here describe analyses that are still in development and are not part of jjstatsplot 1.0.52 — bbcplots, advancedbarplot, economistplots, jsjplot, jjtreemap and basegraphics. Each carries a notice at the top. Only analyses that are in the shipped module are documented as available.

Several vignettes were removed or relocated in this release:

Was	Now
jjridgestats	Removed — superseded by jjridges(), which ships. The two are not drop-in replacements (different option names entirely).
jjriverplot	Moved to ClinicoPathDescriptives, where the analysis lives as riverplot().
advancedtree	Moved to meddecide, where the analysis lives as treeadvanced().
jjsankeyfier, jjstreamgraph	Removed — no analysis of either name exists in any module.