statsmodels.stats.oneway.confint_effectsize_oneway

statsmodels.stats.oneway.confint_effectsize_oneway(f_stat, df, alpha=0.05, nobs=None)

Confidence interval for effect size in oneway anova for F distribution

This does not yet handle non-negativity constraint on nc. Currently only two-sided alternative is supported.

Parameters:

f_statfloat

F-statistic for which the effect size confidence interval is computed.

dftuple

degrees of freedom df = (df1, df2) where

• df1 : numerator degrees of freedom, number of constraints

• df2 : denominator degrees of freedom, df_resid

alphafloat, default 0.05

Significance level for the confidence interval.

nobsint, default None

Total number of observations. If None, then it is set to df1 + df2 + 1.

Returns:

Holder

Class with effect size and confidence attributes

See also

confint_noncentrality

Notes

The confidence interval for the noncentrality parameter is obtained by inverting the cdf of the noncentral F distribution. Confidence intervals for other effect sizes are computed by endpoint transformation.

R package effectsize does not compute the confidence intervals in the same way. Their confidence intervals can be replicated with

>>> ci_nc = confint_noncentrality(f_stat, df1, df2, alpha=0.1)
>>> ci_es = smo._fstat2effectsize(ci_nc / df1, df1, df2)