ori elyada
New member
- Joined
- Mar 26, 2024
- Messages
- 1
I'm an engineer but last time I did statistics was over 20 years ago...
I'm writing a test protocol for a device we're developing.
I know from reference protocols that for a pass-or-fail criteria, with 95% confidence and 95% reliability rate I have to test a sample size of 18 and have 18-of-18 PASS based on the laplace method calculation.
These tests are lengthy and each protocol has 60+ different tests to perform, all which have to to PASS for just one sample to be considered PASS.
Could I define a tighter tolerance on PASS tests, tighter than ACTUALLY required, and reduce the sample size (at least if all of the reduced sample size pass)?
Note the test distribution and stddev are unknown.
for example, per a certain test, my product requires amplification between 50-70 but if I redefine PASS as being between 55-65 can I reduce the sample size to less than 18 and by how much?
If so, does tolerance need to be tightened for ALL 60+ tests by the same amount?
Can i prove anything at all with only one or two samples that are very close to the predicted value? My intuition is that if I designed for an amplification of 60.0 and got 60.1 on one single device tested then it's not a fluke and the design is spot-on.
I'm writing a test protocol for a device we're developing.
I know from reference protocols that for a pass-or-fail criteria, with 95% confidence and 95% reliability rate I have to test a sample size of 18 and have 18-of-18 PASS based on the laplace method calculation.
These tests are lengthy and each protocol has 60+ different tests to perform, all which have to to PASS for just one sample to be considered PASS.
Could I define a tighter tolerance on PASS tests, tighter than ACTUALLY required, and reduce the sample size (at least if all of the reduced sample size pass)?
Note the test distribution and stddev are unknown.
for example, per a certain test, my product requires amplification between 50-70 but if I redefine PASS as being between 55-65 can I reduce the sample size to less than 18 and by how much?
If so, does tolerance need to be tightened for ALL 60+ tests by the same amount?
Can i prove anything at all with only one or two samples that are very close to the predicted value? My intuition is that if I designed for an amplification of 60.0 and got 60.1 on one single device tested then it's not a fluke and the design is spot-on.