Assessing statistical significance of spread.

Good morning! My apologies if my question seems obvious to some. I am by no means a mathematician. Here is a little background to my problem. I am a Clinical Pharmacist running an anticoagulation clinic for several hundred patients. I can categorise my patients into 3 distinct groups, each with a different target 'blood clotting time' e.g. group 1 has a target of '2.0', group 2 has a target of '2.5' and group 3 has a target of '3.0'

My assumption is that those with higher target have a higher spread of results i.e. their level of anticoagulation changes more dramatically (from their target) than those with lower targets.

What I would effectively be doing is looking at the variation of theirs results over time (which I am assuming to be of normal distribution) and comparing the spread of these results between the 3 groups. Can anybody point me in the right direction regarding an appropriate statistical test anything else I might be overlooking? Thanks, Dan


Active Member
Dan, much depends on the data you will end up collecting. Their type and size. Generically speaking, you may end up using a time series model which depends on the group ID. It is hard to say more at this point, unfortunately... Once you have collected the data, you can post them here and we will know more.
Edited: posted before I read previous reply.
Thanks for your reply staassis. My initial thought it to record 'difference between target and observed result' for my groups, as ultimately I am trying to demonstrate larger variance between my 3 groups. Can you forsee any issues with this data collection?

My thoughts so far is that I should calculate target - actual value for all my patients within the 3 groups. As I am going to have positive and negative results, I could not possibly work out a 'mean' and I am trying to demonstrate deviation from the target (positive and negative). I'm assuming the variance to be greater in group 2, and greater again in group 3.
I could calculate mean positive deviation from target and mean negative deviation from target separately and use a Welch's test to demonstrate larger spread between the three groups? Am I barking up the wrong tree? Dan
Last edited:


No cake for spunky
I don't know any time series model that would address this type of question. They tend to assume stationarity which assumes the mean and variation don't change over time (although the data is often not stationary which requires special methods in practice). There are GARCH/ARCH models, but I am not familiar with those.