The technique of coarsening a variable by reducing it to a dichotomy based on the median is sometimes called the "nefarious median split" or the "invidious median split". This technique throws away information. I suggest you search the archives of this list for "median split".
How many cases do you have?
Are T1 and T2 pre and post measures?
Was treatment randomly assigned? I.e, is this actually a control group or just a contrast group?
Am I correct in assuming that your Hypothesis is that the change in the treatment group is larger than the change in the control/comparison group?
I suggest you start by visualizing your data. Use a series of perspectives.
scatterplots, boxplots, and ladder graphs.
-- a scatterplot of stress T1 (horizontal) by stress T2 (vertical) use different colors/shapes for the two groups. If you are doing this by hand connect the two markers. [Perhaps someone else on the list can suggest how to tie pairs of measures for cases in a scatter plot. ]
-- A scatterplot with Stress on the vertical and T1 vs T2 on the horizontal. Again use different colors/markers for the cases. In the output file try fitting regression and loess lines. Is there striking non-parallelism? Does the loess curve suggest a better fit than the regression line?
-- then EXPLORE creating 6 boxplots: for each group T1, T2 and a variable representing the change.
-- check the archives for "ladder graph" and "Andy". This list most likely has someone who has can help you adapt the example graph to have 2 colors for the lines. If you only have a few dozen cases you can use the example to do this by hand.
There are different ways to analyze change: e.g.,
-- a two way ANOVA with 1 dichotomous IV between the groups and 1 dichotomous IV within the groups. The interaction is what you would be interested in.
-- a REGRESSION with T2 stress as the DV, T1 stress on the first ENTER, and looking at the "variables not in the equation".
Before you start your visualizations, sketch some examples of what the picture would look like IF things were the way you would like them to be.
The impression from the ladder graph is derived from the idea that if things are the way you would like them to be:
The change for the control/comparison group is due to (a) measurement noise and (b) experience with the measure. Whereas the treatment should add change effects in a desired direction.
Ideally the control/comparison group would have fairly horizontal rungs. The treated group would have somewhat consistently slanted rungs.
Do you have random assignment to the treatment and control groups? And was the T1 measure of distress taken at baseline (i.e., before assignment to groups, or at least before the treatment phase)? If so, then one-way ANCOVA would be a sensible method to use. The T2 measure of distress is the dependent variable, the T1 measure the covariate, and group the fixed factor. See the following BMJ Statistics Note by Vickers & Altman for more info: