Comparing two methods of preparation

#1
Hi all,
I am now processing data for my thesis and wonder what statistical method would be appropriate to use to compare two of the methods I used.

In the first method, I prepare an ion exchange column by following a given procedure. In the second, I change one parameter in the preparation procedure. In each case, I prepare a total of three parallel ion exchange columns (series).

When a column is prepared, I use it to measure the half-life of a radionuclide (5 times per column) What I want to do in my data analysis is to compare the resulting half-lives from Method 1 and Method 2.

My data set looks something like this:
Method 1:
Series 1: 70,1 - 69,2 - 72,0 - 67,5 - 72,1
Series 2: 66,8 - 70,9 - 71,3 - 73,1 - 70,0
Series 3: xxxxx
Same for method 2

From what I’ve been reading, I believe that paired T-test is not suitable here. Is there any other common statistical analysis I can perform?

i greatly appreciate your help!
 

katxt

Well-Known Member
#2
I presume that 70,1 means 70.1 in my part of the world.
While I don't fully understand what is happening, it looks to me like it is probably a nested situation.
If that is true, you have effectively just three data points for each method - each point being the average of the 5 estimates in the series.
If the data looks like your example, a 2 sample t test between the methods using those two sets of 3 points would be appropriate. However, the difference would have to be quite large for a test this small to show a significant difference.
 
#3
Is it a better idea to instead use a two-sided t-test and compare each average, in each method, to the theoretical value? This is in order to see which method gives less amount of “significant differences” (say method 1 has t-obs > t in 2/3 cases, while method 2 has same in 3/3 cases). Or would I need many more parallel series in each method to be able to conclude that?
 

katxt

Well-Known Member
#4
Is it a better idea to instead use a two-sided t-test and compare each average, in each method, to the theoretical value?
It's not clear to me just what that means, but I think that you will still have only 3 data points per method.
Or would I need many more parallel series in each method to be able to conclude that?
Probably. It's a pity that your thesis supervisor didn't ask for a power analysis before you started so you would have some idea how many series in each method you would need to detect a desired difference. You may have decided at that early stage that the project needed data beyond your budget and that another investigation had better chance of success.
On the other hand, the difference may be so big that it can catch it with only 3 series per method. I hope so. kat