Some Notes about the Sample Size Testing Provided in this Workbook
The following worksheets provide an automatic and simplifed way to estimate appropriate sample sizes in a wide variety of
sampling circumstances. These involve comparing a test population to a known population, or comparing two different test
populations to each other. Values are provided for the following cases:
1. Means
2. Standard Deviations
3. Proportions (or yields)
4. Rates
Generally, to provide a reasonable estimate, four values must be input (in the yellow fields in the worksheet):
1. Alpha (a): The probability of rejecting the null hypothesis when it is true (e.g. saying there is a difference when there is not).
2. Beta (b): The probability of accepting the null hypothesis when it is not true (e.g. saying there is no difference when
in fact there is a difference of the magnitude that you wanted to find).
3. The Critical Difference: The difference from the base value (1 sample) or the difference between (2 samples) that the
experimenter determines critical to detect (with 1 - b confidence). For means and proportions, the critical difference is
a straight difference in values (although for means tests, what is really important is the difference relative to the
expected standard deviation). For standard deviations and rates, the critical difference is usually a ratio corresponding
to a fractional increase or decrease.
4. A Base Value: For one sample tests, some information needs to be supplied about the known population. For two
sample tests, usually an estimate of the approximate value of some population parameter is required to make a
reasonable sample size estimate. In this latter case, if the samples provide information contrary to these initial
parameter estimates, it is likely that the power of the test (b error) will be compromised.
The nature of the testing must be included. If the value is being tested either for being greater than or less than the known or
expected value (but not both), then the test is called "1 sided." If the value is tested only for inequality, the test is "2 sided."
It is expected that the results of the sampling will be tested using the standard methods. Only for the "1 Rate" section has
a criterion value been included.
Notes About the Sample Size Estimation Methods
Different methods have been used for estimating sample sizes. When exact calculations are not available, generally an
approximation of the base distribution to a normal distribution is made. For two-sided tests, the sample size is calculated so
that the worst case situation will be tested at the stated a and b errors. In such cases, if the true result is on the less
conservative side, the b error will be smaller in the conclusion. The following list explains the methods used:
1. Means
1 Sample: For known std dev, exact. For unknown std dev, nearly exact for good std dev estimate
2 Sample: For known std devs, exact. For unknown std dev, nearly exact for good std devs estimate
2. Standard Deviations
1 Sample: Exact (uses the Chi-Square/df distribution)
2 Sample: Exact (uses the F distribution)
3. Proportions (or yields)
1 Sample: Uses normal approximation after "arcsine of the square root" transformation
2 Sample: Uses normal approximation after "arcsine of the square root" transformation
4. Rates
1 Sample: Exact (uses the Chi-Square distribution, which exactly corresponds to Poisson for integer values)
2 Sample: Uses normal approximation after "square root" transformation to approximate Poisson