So, I need to do the t-test of values in the following situation:

1. There are two groups of patients, each following a different treatment.

2. For each group, there are 20 independent parameters analysed. For example, blood sugar, blood lipids, liver enzymes, and so on. These parameters are unrelated to each other then.

3. For each of the 20 parameters, there is a value at T0, T1, and T2. We are interested in the variation between T1 and T0, and the variation between T2 and T1.

So, this is what I was instructed to do: get the average (T1-T0) and (T2-T1) values for each of the 20 parameters, put the ones for "treatment A" in one column, the ones for "treatment B" in another column, and then "do the t-test". Looks like this:

View attachment 3450

At first I did the t-test like this: I used all the values from C2 to C41 as the first set of values, and all the values from D2 to D41 for the second set of values. I used two tails in the calculation (values can go up or down), and heteroschedastic type (treatments are rather different).

I'm thinking this is wrong though. I'm thinking this would be the correct way instead: do not use the average values. Instead, I should do a separate t-test using [the values of each parameter (one parameter at the time) in each time interval (one time interval at the time) for all the individual patients' values from the first treatment] as the first set of values, and [the same thing for the second treatment]. Then repeat this for the other time interval. Then repeat it all another 38 times for the other parameters with their two time intervals.

So, basically, that each different parameter and each time interval should be treated as a separate set of data on which to perform the t-test.