FAQ/SUMM: Power and sample size computation - what is available.

[josef-pkt]

(I lost an overview)

by method:

• power based on effectsize
• power based on explicit distribution (inherent heteroscedasticity)

by response variable:

• mean, continuous
• proportion, value, riskdiff, riskratio, oddsratio
• rates, value, diff, ratio
  • poisson
  • negbin NB2
• variance ?
• ...
• ...

by sampling:

• 1 sample
• 2 indep
• 2 paired/correlated
• k sample, anova-style

effectsize based works generically given effect size definition for specific problem that has a standard distribution, normal, t, F, chisquare

list of available effect size functions

• proportion_effectsize 2indep, equality test, based on arcsine transformation
• oneway.effectsize_oneway k-sample, oneway comparison of means
• ???
• ???

explicit, specific power computation

AFAICS, only a few in stats.proportions
no Power classes for rootfinding for those #8159

• power_binom_tost, power_ztost_prop: 1sample
• power_proportions_2indep: 2indep, equality test, var depends on alternative
• oneway power_equivalence_oneway
• ...
• ...

doc issue:
collect recipes, helper functions for effectsize based sample size and power computation

WIP/ENH:
more explicit power functions and classes
my current context: 2indep for proportion, poisson and negbin, for tests with a margin, nonzero-null-value, and for equivalence tests.
#8138