Paired or unpaired data: please help

#1
Hi all,

I have this question and have been toiling on it for days. Am not sure if I'm tackling it correctly.

A group of patients attending a particular program were assessed at the start of program and then at the end of the program. Some in the group were prescribed medications before end of program. Data set provided is only for those not prescribed medications. I need to assess effect of medication by comparing values at the end between the two groups.

My question is: should I consider the outcome as paired or unpaired data? I assumed unpaired and used the unpaired two sample t test to measure difference but am having doubts if this is correct.

Thanks in advance.

Cliff
 

trinker

ggplot2orBust
#2
cliff75 said:
My question is: should I consider the outcome as paired or unpaired data? I assumed unpaired and used the unpaired two sample t test to measure difference but am having doubts if this is correct.
Why did you assume unpaired?

Some in the group were prescribed medications before end of program.
It also sounds like you have a control group. Is this correct?
 
#3
Thanks for your reply. I assumed unpaired because some were prescribed the treatment, thus it looks like they had a control group. In fact the analysis concerns comparison of total cholesterol in the two groups (prescribed and not prescribed statins). Participants were from the same cohort not randomised into two separate groups.
 

trinker

ggplot2orBust
#4
What you have is a repeated measures with a control group. I think a repeated measures anova could be a good choice (though you could use ancova here as well). The only thing that isn't clear to me is:

cliff75 said:
In fact the analysis concerns comparison of total cholesterol in the two groups (prescribed and not prescribed statins). Participants were from the same cohort not randomised into two separate groups.
How was the determination of prescribed and not prescribed made?
 
#5
This isn't made clear. Though it could be simple random sampling.


I need to (i) assess the effect of statins by comparing total cholesterol at the end of the programme between the two groups
(ii)examine the difference in total cholesterol at the end of the programme between the two groups, adjusting for total cholesterol levels at initial assessment.

I am using STATA to calculate the statistics and am pretty new to all of this.

Thanks once again.
 

trinker

ggplot2orBust
#6
This isn't made clear. Though it could be simple random sampling.
If it is random then my previous suggestions would be options. If the groups receive treatment based on cholesterol level then my suggestions aren't appropriate and I don't really know what it.

(i) assess the effect of statins by comparing total cholesterol at the end of the programme between the two groups
(ii)examine the difference in total cholesterol at the end of the programme between the two groups, adjusting for total cholesterol levels at initial assessment.
Are these your hypotheses? If so they really are the same question if done correctly. It's not fair to just compare end results as in your (i). You have to take into account pre levels as in (ii).

If this is an assignment (I'm getting that feeling it is) then your prof may want you to actually compare the poor choice for answering (i) with the better choice for answering (ii). If this is an assignment and your prof wants what I just stated you have already completed the poorer choice of assessing ending differences between the control and treatment group without taking into account pre treatment values.