# tools comparing for the same drug produced from two different processes.. - may/may not be assumed to be normally distributed (Oct/21/2009 )

hi..

i m new to using biostats..

i wish to compare two products..

product 1 produced from process 1

product 2 (expected to be very close to product 1) produced from process 2

both product 1 and product 2 will have their own internal variability (in terms of multiple batches), multiple impurities.

since we are talking of comparison in terms of impurities and their levels, i donot know if i can assume that these are normally distributed.

i have 5-8 observations for each product..

i m trying to use the t-test..

let me know what else can be used in such a case, to find out if these product impurity levels are overall same or different.

thanks.

hi there... i m not a statistician but i can suggest what you can do if the impurities are normally distributed...

find the variances of the two products A(s1) and B(s2)

find out the F value (s1/s2)

now using the p value approach (table) and the degrees of freedom in your case is 7 (8-1), see if the observed F value is less than or more than 5% F value (F0.05) and decide the significance of the difference in the impurities percentage.

Pradeep Iyer on Oct 21 2009, 01:22 AM said:

find the variances of the two products A(s1) and B(s2)

find out the F value (s1/s2)

now using the p value approach (table) and the degrees of freedom in your case is 7 (8-1), see if the observed F value is less than or more than 5% F value (F0.05) and decide the significance of the difference in the impurities percentage.

ya.. that would work!

one question.. for impurities.. can the assumption of normal distribution be taken?!

thanks..

normally anything which lies in the close proximity of the mean (no. of observations) follows the normal curve...

For impurities, it depends on how you define the mean of the impurity. If like you say you have mean of 8 samples, that is not enough to predict a normal behaviour as it will need atleast 30 data points to follow normal distribution!!!

So i think with 8 samples it is not easy to go ahead with a normal curve and the t test or other non parametric tests (if applicable) can be explored!!

Or alternatively if for any of the process sample you have more than 30 data points and it follows the normal curve and if the other process 8 samples are very very precise and accurate and the variation is less, you might with great risk assume it to follow normal distribution!!

BEst luck!!!

Do post if any updates or conclusions are drawn!!!