# Don't know how to statistically justify sample size

#### kevstrerini

##### New Member
I'm a newby to the forum, thanks in advance for your help.

I'm planning to test some parts but because of all the replicates I can only afford to test 6 parts. The success criteria is after accelerated aging and chemical treatment the destructive pull force of the materials must not be less than 25% of their original baseline pull force. I haven't received the parts yet and do not have a clue of the pull force or standard deviation so doing a power and sample size in minitab leaves me with too many variables. And even then I don't think this seems like a 2 sample t. Any suggestions? Thanks,
Kevsterini

#### GretaGarbo

##### Human
Just guess and make up 6 data points. Ask your colleagues to do the same. Compute the standard deviation, take the average of the sd:s and multiply that with 3. (People tend to underestimated the "spread".) Now you have some guess work.

Investigate the published literature. Is there anybody that has done measurements on similar materials? (It doesn't matter what the mean is. It is a rough guess of the spread you want.)

Now, consider if there is anything that would persuade you to more than n=6? If not, then the sample size maximum is n=6.

Is there anything that could persuade you to do less than n=6? Do you trust anything with less than n=6? Then the sample size is fixed at n=6.

From your guess work, is it meaningful at all to run the experiment? If so, run the experiment at n=6.

#### noetsi

##### No cake for spunky
Sample size is tied to uncertainty of whether you have measured the true population. You can list a margin of error around your results for a given certainty level (most commonly 95 percent) and your reader can decide if the results you find are reasonable.

#### GretaGarbo

##### Human
Sample size is tied to uncertainty of whether you have measured the true population.
No, sample size is about the uncertainty in an estimated parameter. That you have measured the true population (i.e. the correct one) is taken for granted.

#### noetsi

##### No cake for spunky
As I understand it the uncertainty is whether the estimated parameter is the same as the population estimate.

#### GretaGarbo

##### Human
As I understand it the uncertainty is whether the estimated parameter is the same as the population estimate.
The estimated parameter will never be equal to the population parameter - at least not for continuous variables like with the normal distribution. But the variation in the estimated parameter can estimated with the standard error.

Last edited by a moderator: