The object returned by every anovakit function

All eight analysis functions return an object of class anovakit_fit. It is a plain list, so $ extraction works as usual, with print, summary and plot methods for convenience.

Components

method

Character. The analysis that was run, as printed.

call

The matched call, so a result can always say how it was made.

model

The fitted model object: an lm from anova_ancova, a glm (or a negbin from MASS::glm.nb()) from anova_bin, anova_count and anova_glm, the multivariate mlm from anova_manova, an afex_aov from anova_rm (fitted on internal names, see its $internal_names), and the htest returned by oneway.test or kruskal.test from anova_welch and anova_kw, which fit no model. The call of a fitted lm or glm reaches the analysed rows from anywhere, so update(), step() and lmtest::lrtest() work on it directly: for instance update(fit$model, . ~ 1).

anova

Data frame. The omnibus test table. Its attributes record the type of sums of squares, the test statistic and (in anova_rm) the sphericity correction; print() shows them.

effect_sizes

Data frame, or NULL. What it holds depends on the method: partial eta squared with partial omega squared (anova_ancova, anova_manova, and anova_glm on a Gaussian family), a standardised mean difference on the average of the two group variances, with intervals (anova_welch), epsilon squared and eta squared for the rank statistic (anova_kw), partial and generalised eta squared (anova_rm), odds ratios (anova_bin) and incidence rate ratios (anova_count), each a level against its reference level whatever the contrasts, and the deviance explained (anova_glm on any other family) with McFadden's pseudo R squared for the binomial, Poisson and negative binomial families only. Both variance measures are partial: the classical omega squared is only a proportion of variance when the effect sums of squares partition the total, which Type II and Type III sums of squares do not.

emmeans

Data frame, or NULL. Estimated marginal means with intervals at conf_level (group summaries for anova_welch and anova_kw). When grouping factors enter a model additively they are reported per factor, with a term column.

emmeans_object

The emmGrid, a named list of them (one per factor) when the means are reported per factor, or NULL. Pass it to emmeans for contrasts the wrapper does not cover. It carries emmeans' own default confidence level rather than conf_level, so give level = when you summarise it. It is NULL for anova_welch and anova_kw, which fit no model emmeans can use, and for anova_manova, where there is one grid per response under $univariate[[response]]$emmeans_object.

posthoc

Data frame, or NULL. Pairwise comparisons, with the unadjusted p-value in p_value, the adjusted one in p_adjusted and the method in adjustment.

assumptions

Named list of assumption checks. Contents vary by method; each element is a test object, a data frame or NULL. It is empty for anova_kw unless diagnostics = TRUE, since the test assumes no distribution.

plots

Named list of ggplot2 objects. Empty when plots = FALSE. Nothing is ever drawn as a side effect.

data_used

Data frame. The rows and columns the model was fitted on: the analysed columns only, after missing and infinite values, zero weights and incomplete subjects are removed. Covariates are mean-centred there when anova_ancova or anova_manova centred them.

n_removed

Integer. Input rows that are not in data_used: dropped for missing or infinite values (a factor level that is itself NA counts as missing) or a prior weight of zero, or, in anova_rm, for an incomplete within-subject design or because repeated subject-by-cell rows were aggregated (with fun_aggregate, the mean by default). nrow(data_used) + n_removed is always the number of rows given.

conf_level

Numeric. The level used for every interval returned.

notes

Character vector. Everything the function decided on your behalf, could not compute, or thinks you should know – including anything car, glm, emmeans, sandwich or afex said while the model was being fitted. Always read this.

Components individual functions add

Each function returns everything above plus whatever its own method produces. names(fit) lists them all. anova_ancova adds $slopes_test, $simple_slopes, $covariate_means, $model_additive and $model_interaction; anova_rm adds $sphericity, $subjects_dropped, the breakdown of $n_removed, $residuals (within-subject residuals in the row order of $data_used) and $internal_names; anova_manova adds $multivariate, $univariate, $canonical, $canonical_term, $slopes_test (with covariates), $covariate_means and $test; anova_count adds $model_type, $dispersion and $model_dispersion; anova_bin adds $model_stats; anova_glm adds $model_stats and $family. Each is documented on the function that produces it.

See also

print.anovakit_fit, summary.anovakit_fit, plot.anovakit_fit