Over-simple Stats?

#1
Hi all,

This should hopefully be a really simple question but I think I am confusing myself over a throwaway comment that someone made....

Say I have 20 mice undergoing two different treatments, so 10 mice in each group.

From each mouse I take two muscle biopsies, same muscle different sides (therefore 20 biopsies per treatment group).
From each biopsy I take three histological sections (therefore 60 sections per treatment group)
On each slide i count the number of fibres (ranging from 50-200 fibres per slide).

The sample size (per group) would still only be 10, as it was 10 mice that were originally treated in each group (Sample size of 20 overall)?

Can I then simply average the count per sample (so three counts per biopsy, two biopsies per sample) and compare the two groups of 10 to see whether there is a statistically significant difference beween the two groups? Or is that over-simplifying it?

Any help would be really, really appreciated.

Thanks,
 

seanw

New Member
#2
It depends. If you are just interested in looking at the significance of the difference in count per sample, it seems to me that something like a chi-squared test would be fine. If you want to take the nesting into account and get statistics at each level (not just sample-wide but also biopsy and/or section-wide) then you would need appropriately more complex methods, like hierarchical linear modeling.
 

rogojel

TS Contributor
#3
Hi ,
as seanw said the way you analyse the data could be different, depending on how you set up the experiment and what you are interested in. If the mice in the two groups are randomly selected so that there is no reason to expect any difference between the two groups apart from the treatment and you are not interested in a difference between the two sides then I think you could just build two groups of 60 data points each and use some standard analysis - possibly even as simple as a two sample t test.

If you have reasons to expect thatbthe mice differ and /or you are interested in the effect of the sides on the count then a nested Anova could be used or more complex models.
 
#4
Thank you both very much for the replies, really appreciate it.

The animals in each group were randomly selected, I have no reason to think that there should be any difference between them other than the treatment and i'm not interested in the difference between sides, so I think I will just treat it as simply as possible, as you both suggested - I just didn't want to end up artificially inflating my sample count, pseudo-replication I think it is called.

Thanks again,
 

rogojel

TS Contributor
#5
hi,
probably the safe way would be to build averages for each mouse and use them in the final test. So, I think I was wrong in suggesting to use 60 data points per treatment, it should be only 10, but each one an average of 6 measurements. Or actually using a nested analysis.

Sorry for the confusion.
rogojel