Most appropriate test for working out significance? Thank you in advance

#1
Hi all,

I am new to statistics and seem to be getting myself more confused the more I read.

I would like to work out the p value (level of significance 0.05).

I have a group of patients all having an MRI scan. As an example, let us say 50 have a suspicious appearance, and 25 have a non-suspicious appearance.
Upon biopsy, 20 of those with a suspicious appearance have cancer. 5 of those with a non-suspicious appearance have cancer on their biopsy.

I was initially led to believe using the chi-squared test for nominal categorical data may be my best bet. However, I am now reading that it may be worth using the Fisher's exact test as one of my cells has an expected frequency of 5 or less.

Extremely grateful for any help you may be able to give. Hope the above is clear.
 

Karabiner

TS Contributor
#2
According to the data you present, no cell has an expected (!) frequency of 5 or less.
So if that's the data you want to analyse, you can use Chi². But the Fisher test can be used
anyway.

The main problem here is a possibly uninteresting research question. You would
find an assocition between MRI result and biopsy result which is statistically significant
(i.e. it is not exactely 0.00000000... in the population), which will come as no surprise,
I suppose. But the test will not tell you how strong the association is (just that it's
supposedly not 0.00000...).


With kind regards

Karabiner
 
Last edited:
#3
What is the question?
There are two distributions; mri folks and not suspicious folks--a subset of mri folks---all of whom got a biopsy.
The data looks like this to me:
mribiocancer.jpg
The only question I can see is: If everybody gets a biopsy, why does anyone get an mri?
(Chi sqrd test looks at 2 sets of numbers and finds the probability that they're from the same distribution. I'd not suspect that any of these numbers fits the "same distribution?" question.)