pecanr 0.4.0

New features

• eta2p() gains a slope argument: supply a hypothetical fixed-effect coefficient (e.g. slope = 0.03) and the effect size is computed for that slope, holding the predictor variance and the error structure fixed from the model. Useful for sensitivity and power analyses (e.g. alongside simr).
• The returned eta2p_lmm object now also reports the numerator ingredients: coefficient, var_x (the predictor variance), var_x_unique, and tolerance.
• Added concise print() and comprehensive summary() methods for eta2p_lmm. Printing an object now shows just the estimate; summary() shows the coefficient, var(X), and the error-variance components.

Removed

• eta2p_omnibus() has been removed pending a revised numerator definition (the current variance-of-the-summed-contribution numerator overstates the effect when a covariate is correlated with the factor).

pecanr 0.3.1

Bug fixes

• Fixed a bug in which random-effect terms stored under numeric-suffixed group names were silently dropped from the error variance. lme4 stores multiple random-effect terms for the same grouping factor – as produced by the double-bar syntax (x || g) or by explicit (1 | g) + (0 + x | g) terms – under group names "g", "g.1", "g.2", and so on. The error-variance calculators matched only the base name, so every component under a suffixed name (typically the random slope) was omitted from the denominator, inflating the effect size. This affected all three design calculators (crossed, nested, and mixed) and both general and operative effect sizes.
• Fixed a related bug for nested-syntax grouping terms such as (1 | photo:model). On current lme4 (observed on 1.1-35), as.data.frame(VarCorr()) reports the combined group name dot-separated ("photo.model"), so the previous colon-based fallback match never fired and the component was silently dropped from the error variance in nested and mixed designs – including the mixed-design example in the package documentation. Both separators are now matched.
• Effect sizes from earlier versions may change for affected models. If a model used the double-bar syntax, explicit uncorrelated random-effect terms, or interaction-grouping terms like (1 | a:b), the previously reported eta2p values were too large (the error denominator was missing components). Values for models specified with single-bar terms per grouping factor (e.g. (x | g)) and simple (1 | g) intercepts are unchanged. Single-bar and double-bar specifications of the same model now yield closely agreeing effect sizes, as expected.

Testing

• New regression test suite (tests/testthat/test-double-bar.R) verifying that eta2p()’s error variance matches a direct hand computation from VarCorr() (to numerical precision) across crossed, nested, and mixed designs for double-bar terms, explicit (1|g) + (0+x|g) terms, multi-slope double-bar terms including interaction slopes, and (1 | a:b) grouping. Also guards against false matches (a grouping variable literally named "g.extra" is not absorbed into "g").

pecanr 0.3.0

New features

• New function eta2p_omnibus() computes a single factor-level partial eta-squared for a multi-level factor or multi-df interaction, corresponding to the omnibus (multi-df) test of that effect rather than to individual contrasts. The variance attributed to the effect is the variance of the summed fitted contribution of all of the effect’s design-matrix columns, which correctly includes the covariances among them; the error denominator matches eta2p(). Use this when pairing an effect size with an omnibus F or chi-square test.
• eta2p() and batch_eta2p() now accept factor predictors directly. The focal effect may be given either as a model coefficient name (e.g. "Group75yr") or as the underlying variable name (e.g. "Group"); a variable that maps to a single coefficient is resolved automatically, and a multi-level factor raises an informative error pointing to eta2p_omnibus() or batch_eta2p(). Predictor variances and factor random slopes are read from the model design matrix and model frame, so manual recoding of factors to numeric is no longer required.
• eta2p() gains an optional confidence interval via parametric bootstrap (ci = TRUE), with ci_level, n_boot, and seed arguments. Because the effect size is a ratio of REML variance components with no closed-form sampling distribution, the interval is obtained by simulating from the fitted model (lme4::bootMer), refitting, and recomputing the effect size on each replicate. This is opt-in, as each replicate refits the model.
• eta2p() and batch_eta2p() gain a partial_predictors argument controlling the numerator for single (non-interaction) predictors. When TRUE (the default), the variance attributed to a predictor is its unique (semipartial) variance – equivalently its total variance times its tolerance, Var(X) * tol(X) – so it reflects only the predictor’s unique contribution and declines with redundancy under correlation. Set partial_predictors = FALSE for the previous total-variance (raw) numerator. For centered, orthogonal designs the two are identical, so this affects only models with correlated predictors.

Changes

• For single (non-interaction) predictors, the default numerator now uses each predictor’s unique (semipartial) variance rather than its total variance (see partial_predictors above). For centered, orthogonal designs this is identical to previous behavior; for models with correlated predictors, the default effect sizes will be smaller than in earlier versions, reflecting only each predictor’s unique contribution. Set partial_predictors = FALSE to reproduce the previous total-variance behavior.
• Interaction effect sizes are now computed from the mean-centered constituent predictors: each component of an interaction is centered before forming the product whose variance enters the numerator (the product itself is not centered). This makes the interaction effect size invariant to the location (means) of its constituent predictors, which the variance of the raw product is not. Effect sizes for main effects are unaffected, as is the value for interactions whose components were already centered (e.g. contrast-coded factors). Note that for models with uncentered continuous interaction components, interaction effect sizes may differ from those produced by earlier versions.
• The default for verbose in eta2p() is now FALSE (previously TRUE), matching batch_eta2p(). Pass verbose = TRUE to restore printed output.

Bug fixes

• Operative effect sizes (operative = TRUE) now work correctly for factor predictors. Within/between classification previously matched the focal effect’s name directly against the data columns, which failed for a factor whose coefficient name (e.g. "Age75") differs from its variable name (e.g. "Age"); the grouping factor could then be misclassified and the wrong intercept variances included in or excluded from the operative denominator. Coefficient names are now resolved back to their underlying variable before classification.
• Fixed a bug in which the random slope of an interaction term (e.g. a random slope on x1:x2) was silently omitted from the error denominator when its random-effect name did not match the fixed-effect product column. The contribution of interaction random slopes is now resolved from the model frame and included. # pecanr 0.2.0

New features

• eta2p() and batch_eta2p() now support design = "mixed" for models with both crossed and nested random effects simultaneously. A canonical example is participants viewing multiple photos of each model: photos are nested within models, but both levels are crossed with participants. Supply both cross_vars and nest_vars to use this design.

Bug fixes

• Fixed a bug in operative effect size calculation for crossed designs. detect_within_between() previously used hardcoded $subj and $item keys to classify grouping factors as within or between, which caused intercept variances to be silently omitted from the operative denominator. Keys are now indexed by actual variable name, so the correct components are always included.

Breaking changes

• batch_eta2p() output columns for within/between status are now named within_<varname> (e.g., within_participant, within_item) rather than the hardcoded within_subj and within_item. Code that references these columns by name will need to be updated.
• Operative effect sizes with 3 or more crossed factors now correctly gate each factor’s intercept variance on its within/between status. Previously, third and higher factors were always included in the operative denominator regardless of whether the effect was within or between those factors.

pecanr 0.1.2

• Initial CRAN release.
• eta2p() computes partial eta-squared for a single fixed effect in a fitted lmer model, supporting crossed and nested random effects structures.
• batch_eta2p() computes partial eta-squared for all fixed effects in a model.
• Crossed designs support any number of grouping factors via cross_vars.
• Nested designs support automatic effect-level detection.
• Operative effect sizes available via operative = TRUE.
• Random slope variances are translated to the outcome scale using sigma^2_slope x sigma^2_X, correctly accounting for predictor scaling and interaction terms.