DMAR (version 1.0.0)

ss_power_composite_anova: Sample Size or Composite Power for a One-Way or Factorial ANOVA

Description

Determine the necessary per-cell sample size to achieve a desired level of composite statistical power in a balanced analysis of variance with \(a\) groups (a one-way design) or a factorial arrangement of factors, or, given a per-cell sample size, return the realized composite power. Composite power is the probability that every effect named in effects is statistically significant in the same study, the quantity a design must be planned against when its conclusion requires more than one result to hold at once. Each effect is a main effect or an interaction, tested by its own F test, and any subset of them can make up the composite.

Usage

ss_power_composite_anova(
  factor_levels,
  effects,
  means = NULL,
  sigma = NULL,
  desired_power = 0.85,
  alpha_level = 0.05,
  n_per_cell = NULL
)

Value

A data.frame with term and value columns: the recommended (or supplied) n_per_cell and total N, the

composite_power, the residual_df, then for each effect its marginal power_<label>, purported f_<label>, numerator

df_<label>, and noncentral_parm_<label>, followed by

cells and alpha_level. The result carries the

dmar_ss_power class for tidy and

glance, and a dmar_composite_power_factorial

class so plot() draws the figure.

Arguments

factor_levels

Integer vector of the number of levels of each factor, one entry per factor (each at least 2). A single value a is a one-way design with \(a\) groups; c(2, 3) is a 2 by 3 factorial.

effects

A non-empty list naming the effects in the composite. Each element is a list with factors (a vector of factor indices into factor_levels: one index for a main effect, several for an interaction) and, unless means is supplied, exactly one of f (Cohen's f) or partial_eta_squared. An optional label names the effect in the output and the figure; the default label is the factor indices joined by x. The purported effect sizes are population values the researcher posits, never sample estimates.

means

Optional array of population cell means whose dimensions are factor_levels (a matrix for two factors), or a numeric vector of length prod(factor_levels) in array order. When supplied, each named effect's Cohen's f is computed from the means and sigma, and plot() draws the mean pattern.

sigma

The common within-cell population standard deviation of the outcome, required with means and used only there.

desired_power

Desired composite statistical power (default 0.85). Used only when n_per_cell is NULL.

alpha_level

Type I error rate for each individual F test (default 0.05), the per-test rate, not a rate for the composite event.

n_per_cell

Per-cell sample size (balanced); if supplied, the realized composite power is returned rather than a sample size planned.

Author

Ken Kelley [email protected]

Details

The population effects can be stated two ways: as effect sizes, a Cohen's f or partial eta squared per effect, or as a full array of population cell means together with a common within-cell standard deviation, from which each named effect's f is read off the analysis of variance decomposition of the means. Supplying means lets the plot() method draw the mean pattern itself, so the size of the population effects is on the page.

If the design includes one or more covariates, use ss_power_composite_ancova.

References

Maxwell, S. E. (2004). The persistence of underpowered studies in psychological research: Causes, consequences, and remedies. Psychological Methods, 9, 147--163.

Maxwell, S. E., Delaney, H. D., & Kelley, K. (2027). Designing experiments and analyzing data: A model comparison perspective (4th ed.). Routledge. (See Chapter 7 on factorial designs and Chapter 3 on statistical power.)

See Also

ss_power_composite_ancova for the design with one or more covariates; ss_power_factorial_anova and ss_power_one_way_anova for a single effect

Other sample size for power: power_fisher_exact(), ss_aipe_mixed_effects(), ss_power_R2(), ss_power_R2_sensitivity(), ss_power_c(), ss_power_c_ancova(), ss_power_composite_ancova(), ss_power_composite_ancova_2group(), ss_power_composite_factorial_ancova(), ss_power_composite_factorial_ancova_het(), ss_power_composite_factorial_anova(), ss_power_composite_sem(), ss_power_contrast(), ss_power_equivalence_c(), ss_power_factorial_ancova(), ss_power_factorial_anova(), ss_power_indirect_effect(), ss_power_mixed_effects(), ss_power_one_way_anova(), ss_power_pcm(), ss_power_r(), ss_power_rc(), ss_power_reg_coef(), ss_power_reg_coef_sensitivity(), ss_power_rm_anova(), ss_power_sc(), ss_power_sem(), ss_power_smd(), ss_power_split_plot_anova()

Other composite power: ss_power_composite_ancova(), ss_power_composite_ancova_2group(), ss_power_composite_factorial_ancova(), ss_power_composite_factorial_ancova_het(), ss_power_composite_factorial_anova(), ss_power_composite_sem()

Examples

# A 2 by 3 factorial ANOVA whose conclusion needs both main effects, so the
# design is planned against their composite.
ss_power_composite_anova(
  factor_levels = c(2, 3),
  effects = list(list(factors = 1, f = 0.25),
                 list(factors = 2, f = 0.20)),
  desired_power = 0.80)

# Realized composite power at 40 per cell for a main effect and the
# interaction of a 2 by 2 design, sizes given as partial eta squared.
ss_power_composite_anova(
  factor_levels = c(2, 2),
  effects = list(list(factors = 1,       partial_eta_squared = 0.06),
                 list(factors = c(1, 2), partial_eta_squared = 0.04)),
  n_per_cell = 40)

# The effects can instead be a full pattern of population cell means with a
# common within-cell SD; plot() then draws the mean pattern itself.
cell_means <- matrix(c(10, 12, 11,
                       13, 12, 16), nrow = 2, byrow = TRUE)
fit <- ss_power_composite_anova(
  factor_levels = c(2, 3), means = cell_means, sigma = 4,
  effects = list(list(factors = 1, label = "A"),
                 list(factors = 2, label = "B")),
  n_per_cell = 30)
fit
plot(fit)