Which statistical analysis?

lah

New Member
#1
I am comparing two sets of data. I want to know if the time taken for a sample to clot taken using technique A are different to the gold standard technique B. Which is the best test to use, t-test? paired or unpaired and why. The total number of samples tested is between 20-30. (The original sample was tested using technique A, and the original sample was tested using technique B).

Technique A (seconds) = 10, 12, 14, 20, 30, 25
Technique B (seconds) = 13, 10, 16, 22, 34, 30
 

hlsmith

Omega Contributor
#3
You don't explicitly state whether the above values are paired? So one technique had a clotting time of 10 and the other technique had a clotting time of 13 on the exact same sample unit, correct? How can you apply two different things to the same unit? For clarification you can give me a drug then another drug, but I may not be naïve to the process the second time around or I may not be the same person since time has elapsed.

I little more detail is needed, thanks.
 

lah

New Member
#4
You don't explicitly state whether the above values are paired? So one technique had a clotting time of 10 and the other technique had a clotting time of 13 on the exact same sample unit, correct? How can you apply two different things to the same unit? For clarification you can give me a drug then another drug, but I may not be naïve to the process the second time around or I may not be the same person since time has elapsed.

I little more detail is needed, thanks.
Yes, for example from a tube I took 1ml of sample and tested the time to clot using technique A. I took another 1ml of sample and tested the time to clot using technique B.
 

hlsmith

Omega Contributor
#5
Time values can be at risk for skewness. You can likely just test the differences between the two values in a one-sample ttest compared against zero or use Wilcoxon sign rank test if data are skewed.
 

Karabiner

TS Contributor
#6
Why should a statistcal test be arried out? It would not answer the question of how reliable is the second instrument, as far as I can see. A t-test would tell us whether the mean score of one instrument is lower than the mean of the other (or, actually, here it probably wouldn't tell us, because of low power). But if one instrument has low reliability, its mean could be near the gold standard, while the single measurements would vastly under- or overestimate the true values. Or, was the original question whether the other instrument systemantically over- or underestimates the true values?

With kind regards

Karabiner