Statistical Power Guidelines


 
Are there standard guidelines used for choice of statistical power?  This really depends on the situation and the business, safety, customer, and other risks associated with Type 2 error.

Having offered this disclaimer, here are some thoughts that might provide a useful starting point for the discussion.


In the absence of an informed decision based on business risk it may be useful to start with the following suggestions:
  • For safety-critical issues, a statistical power no less than 90% might be appropriate. (Type 2 error of no greater than 10%.) Perhaps higher if warranted.
  • For standard business/engineering applications a statistical power no less than 80% is sometimes used. (Type 2 error of no greater than 20%.)
  • For those who suggest that even lower statistical power is OK, this begs the question as to whether or not the testing is a good use of business resources.
Of course if the testing is performed with smaller sample sizes (and lower statistical power) but from the testing a statistically significant difference IS found, then Type 2 error is no longer an issue.



DISCLAIMER:

The choice of statistical power depends on the business situation as noted above and is the decision of those sponsoring the study, who should seek advice from a statistician who works with the statistical methods that will be employed.




Comments