Skip to contents

The noncentrality interval for the proportion of variance in outcome explained by group, as effectsize::eta_squared(partial = FALSE, alternative = "two.sided") computes it for a one-way ANOVA: the 2-sided interval for the noncentrality \(\lambda\) of the F statistic, each bound turned into \(\lambda / (\lambda + df_{error})\). The interval reaches 0 when the test is far from significant. Until 0.4.3 this was a percentile bootstrap, whose resampled eta squared is 0 or above, so 40 samples with equal group means all had an interval above 0 (FR-24).

Usage

sframe_eta_sq_ci(outcome, group, R = 2000, conf.level = 0.95, seed = NULL)

Arguments

outcome

A numeric vector.

group

A grouping vector of the same length.

R

Kept so existing calls still run. Unused since 0.4.3.

conf.level

Confidence level. Defaults to 0.95.

seed

Kept so existing calls still run. Unused since 0.4.3, as the interval uses no resampling.

Value

A named numeric vector: estimate, lower, upper, with the attributes level and method ("noncentral F"). The bounds are NA with fewer than 3 complete observations, fewer than 2 groups, or no variation within the groups.

Examples

sframe_eta_sq_ci(mtcars$mpg, mtcars$cyl)