statsExpressions 2.2.0

• New helper complete_subjects() prepares long-format repeated-measures
  data the same way the paired tests do. It drops rows that cannot be paired,
  validates the subject identifier, and keeps only subjects observed in every
  condition. Packages that plot or summarize repeated-measures data (such as
  {ggstatsplot}) can use it so that they work on the same observations as
  the statistical tests.

• Repeated-measures analyses (paired = TRUE) now validate the subject
  identifier via complete_subjects(), and thus in
  long_to_wide_converter(), two_sample_test(), oneway_anova(), and
  pairwise_comparisons():

  • A subject with more than one row in a level of x now gives an
    informative error. Previously, depending on type, this produced
    list-column warnings, cryptic errors, or results that silently averaged or
    discarded the duplicated observations.
  • Rows with a missing subject.id are dropped. Previously they were treated
    as observations of a single shared subject.
  • Subjects without a row for every level of x are dropped, just like
    subjects with a missing response. Previously they were counted in the
    reported sample size, and the Bayesian two-sample test failed.
  • A row with a missing x value now removes only that row rather than every
    row of the subject.

• oneway_anova() with type = "robust" and paired = TRUE now reports the
  correct Algina-Keselman-Penfield effect size and confidence interval.
  Previously the internal row identifier was treated as an extra condition,
  which inflated the estimate (e.g., 0.66 instead of 0.32 for bugs_long).

• Columns in the input data that share a name with a function argument or an
  internal variable no longer change the results. For example, a column named
  digits or n.text in add_expression_col() input, a grouping variable
  named digits or centrality in centrality_description(), a or b in
  Bayesian pairwise_comparisons(), pair in pairwise_contingency_table(),
  or es.text in tidy_model_expressions() input could previously produce
  wrong expressions or errors.

• Functions that report a sample size no longer warn that 'big.mark' and 'decimal.mark' are both ',' when options(OutDec = ",") is set (#146).

• Corrected the documentation of Bayesian correlation columns, missing-value
  counts in centrality_description(), degrees of freedom, paired sample
  sizes, and adjusting p-values after expressions have been generated.