Statistical Design

Here are a few notes on Statistical Design from class today...

As you design your study it is important to realize that statistical significance is different than practical significance. Your results can be statistically significant, but not practically important.

Determining the Sample size
The following four pieces of information are needed to determine the sample size:

1. Acceptable type on error (alpha) - standard value is .05
   
2. Acceptable type two error (beta) - standard value is .2 (power of .8)
3. Standard deviation - pooled estimate of the standard deviation (sq. rt. of the error mean squared error), includes block effects and covariates
4. Meaningful difference you want to detect

There are various "rules of thumb" that can help you figure this out based on having the above information. For instance, the following is a basic formula that could be used:

n=16/(delta/sigma)^2

where delta is the change in means between the two groups you might be comparing.

Here is another rough regression rule of thumb:

n=15*p

Where p is the number of parameters being used.

Finally, here are some steps to cover when designing your statistical experiments.

Steps in Design

• Question of interest
• Score of inference
• Understand the system
• Response
• Other factors
• Planning
• Statistical anaysis
• Sample size

The often overlooked considerations are very important as new studies are designed.