SMEP: Power for Statistical Tests

SMEP: Power for Statistical Tests

Status: partially implemented (part of PR #711)

auxiliary code: Effect Sizes

Use Cases

power and sample size calculation :

having the power equation, we can solve for any of the variables.

"non-central tests": equivalence tests, not clear, ?

example: chisquare gof test, to test that the distance between distribution is larger than a threshold

Possible Problem

For more complex models it is difficult to specify the parameters, effect sizes and assumptions.

Calculation

explicit:

Under normal assumption we have explicit formulas for some tests, like t-tests

Monte Carlo:

The range of alternatives can be huge. What supporting code can we provide to make it easier?

Implementation

requirements :

• easy to expand: I don't expect we will add a lot in one big push
• usage for standalone: e.g. for sample size calculations
• attached to test classes/functions: to get the power of a test case

Cases

(stubs)

t-test:

easy

f-test:

easy

TOST:

requires a "special" integral, see SAS documentation

chisquare (gof)

???

References

• SAS Manual
• R pwr: used as benchmark for tests
• GPower