Hey scientitions,
I'm wondering if any of you fine anons have any constructive comments/thoughts about the following.
I have 10 measurements of the useful life of a device. So I don't have a lot of data. I was wondering if any of you agree with using a conservative confidence interval using student t and chi square to estimate mean and standard deviation of my device's life . Sorry about my english - not a native speaker.
thoughts? thanks again anons.
gonna bump once or twice waiting for a based anon to chime in.
no one?
You could bootstrap it
don't use parametric statistics (Student-T or chi-squared) search on "small sample analysis"
>>7665378
thanks anon
http://www.measuringu.com/blog/small-n.php
these guys suggest a student t confidence interval for the mean, which is what I was leaning towards. My real uncertainty is how to estimate my standard deviation.
OP bumping one last time
Try 15n at least for normale distribution
>>7666219
Ideally, yes. But I'm stuck with n=10 as I'm dealing with someone else's data
Since your data measures the lifetime of some process, an exponential distribution makes far more sense than a gaussian.
You can construct the exponential confodence interval by computing L=1/sample mean, then taking the lower and upper bounds as
(-ln(0.95)/L, -ln(0.05)/L)
respectively.
>>7666295
Thanks Anon. I think I will try this. I've also plotted a frequency plot of my data and it doesn't look normally distributed. THere's a clump of data at lower times and some outliers at larger times. I'm looking at Weibull and Raleigh distributions, as my data is actually in cycle count at failure, rather than time (though I could convert that to time).
thanks again.