How to do ANOVA (or similar) analysis when there are multiple groups?

#1
I have 40 patients I have made some measurements in. 20 have a disease (group A), and 20 are reference/control patients (group B)

The measurements I've taken are the conduction velocity of electrical impulses in various regions of the patients' hearts. Let's call them regions 1-5.

I want to measure whether patients in group A have significantly different velocities PER REGION as compared with patients in group B ie. an overall analysis of whether group 1 patients have different velocities to group 2 per region, but also multiple comparisons for each region ie. region 1 velocities are slower in group A than group B, region 2 velocities are not significantly different between groups etc.

I'm using GraphPad Prism and have done some analysis I'm not sure how I arrived at (2 way Anova and row statistics with Multiple Mann Whitney tests as I think some of the results were non-standard distributed, though it was a while ago so I can't remember why I did those specific tests).

Just wonder if they were correct or whether I need to do a different analysis? Have done quite a bit of searching but haven't found a definitive answer as yet.
 
Last edited:

Karabiner

TS Contributor
#2
I do not know what you mean by "per region". Do you mean, if they have different velocities in general,
across 5 regions? This could be done with a multivariate analysis of variance (MANOVA).

By the way, there is no such thing as a nonparametric distribution. There are nonparametric tests.

With kind regards

Karabiner
 
#3
I mean that it's likely that some regions will have different velocities overall in group A vs group B, and I want to find those differences.

Regions 1-4 are regions of interest (where a difference between groups A and B might exist if our hypothesis is correct)
Region 5 is a "control" region where having the disease shouldn't influence velocity so velocities should be similar between A and B

So what I want to know is

Does having the disease (group A) correlate with there being a difference (hypothesised to be a slowing of velocity) per region between A and B ie. are velocities in region 1 slower in A versus B, are velocities in region 2 slower in A vs B etc. with an expectation that velocities in region 5 would be similar between A and B.

*non-standard distribution - have edited OP

Thanks
 

fed2

Active Member
#4
you should probably compute the change score (Region x - Region 5) for each subject. Then compare between groups with prism, what you describe with the row-stats seems right. I would not recommend worrying about multiplicity adjustment regarding the regions at this juncture. In prism, you can select 'do not adjust' in the analysis options dialogue. Looking at the correlation between regions will be useful.

boxplots may help as well.
 

Karabiner

TS Contributor
#5
Ok, in that case I would suggest you do 4 separate t-tests,
for regions 1 to 4.
With region 5 it is a bit difficult. A statistically non-signficant
difference between groups does not necessarily mean
equivalence, it could well be a type II error. But you can
do a t-test in region 5 anyway.

With kind regards

Karabiner
 
#7
Thanks for advice - I'll give this a go and see if I get different results to the original analysis (which for the life of me I can't quite figure out how I did!)
 

fed2

Active Member
#8
heres how you did it:
1) open prism select grouped data, 20 replicates
2)each row is region, 2 columns A, B
3) analyze -> multiple t-tests per row (or something like that)
4) select mann whitney and whatever else you want to select

great job!