Can I apply a Cochran-Armitage test (chi-squared linear trend test) to compare two categories of data?

Question

I have a problem where I am trying to determine whether there is statistical significance from two groups of data.

I have two columns which show the frequency of visits to their personal trainer based on receiving notifications about their fitness.

	Score	BEFORE	AFTER
severe	3	28	78
elevated	2	262	346
moderate	1	1344	1523
low requirement	0	2311	2400

There are different notification severity categories labelled as "severe", "elevated", "moderate" and "low requirement". The "before" period is the baseline with "suppressed" notifications, which means that we the investigators are the only ones that know of the notification status during this time. There is an "after" period where now people are receiving these notifications.

I want to see whether the notifications had an impact on the frequency of visits to their personal trainers.

I've come across the Cochran-Armitage test for trend (chi-squared test for linear trend) and am wondering whether this would be appropriate?

Answer 1

Yes, this be an appropriate situation for the Cochran-Armitage test, assuming the before and after treatments are independent. That is, that the same participants aren't included in each.

Answer 2

I don't see the need to use a chi-square test (or the one for trend,aka CA test), because you have a categorical variable and a numerical variable as I see (number of visits or number of people, not sure!), and since it's a before and after protocol, I think a paired t-test is more suitable for this (if the normality is not violated of course, otherwise use wilcoxon test)