Sample Size Calculation

[#1]

Hi All,

I am planning an experiment to understand the lifetime of a product. Samples will be cut-out from the product. A batch of product is 15000 and I will cut smaller pieces suitable for bending test. Samples will be stored in different chemicals and there will at least 5 time-periods. I take out after each t1, t2, ... samples out of the solution an test. By this my goal is to fit the degradation and define the limits. My problem is to define sample size for each solution + storage time combination. This will be a destructive testing, so I cannot use the same sample for another measurement. Power will be at least 90 and confidence interval 90.

[#2]

An approach is described in Statistical Methods for Reliability Data 1st. ed. by Meeker and Escobar. Please check out section 22.5 Planning Accelerated Degradation Tests.

To do it, you need:

1. Prior knowledge about the model and parameters, with which you can simulate data.
2. The range of settings, which you can experiment with, e.g. total sample size, sample time (value of t1, t2), number of testing units at individual sample times.

The detailed procedure is described in subsection 22.5.3 Simulation Procedure. The idea is to repeatedly simulate data at different combinations of settings, fit models, get results of interest. And finally, evaluate the results and find the setting that gives you the desired properties.

The section discusses accelerated degradation, but it is also applicable to non-accelerated situation (sounds like yours is so), your simulation just does not involve an acceleration factor.

The section discusses repeated measures degradation, but it is also applicable to destructive degradation. Your simulation simulate individual points, not individual paths. And you use a destructive degradation model to estimate your quantity of interest.