I have following variables:
-"adtype" (1 is print ad with testimonials, 2 is print ad without testimonials.)
-"AT1" which is "ad attitude immediately after watching ad"
-"AT2" which is "ad attitude after a week"

I would like to compere the ad attitudes. I would like to figure out whether the adtype has a impact of the change of ad attitude, if yes, how strong.

I used dependent T-Test (AT1 and AT2) and found out that there is a significant difference, but I want to figure out the impact of adtype, so I used "split file by adtype", again I have significant difference between AT1 and AT2:

adtype ad with T:
AT1, means 5.9
AT2, means 5.0
adtype ad without T:
AT1, means 5.8
AT2, means 4.6

But still, I cannot see how the adtype influsses my results. What should I do then?

I consider to compute a new variable AT_change=AT1 - AT2, then a T Test with AT_change with adtype. I expect to see that in group ad with T, the change is smaller than group ad without T. Does this make sense?

You could do a repeated measures ("mixed") analysis of variance, with a within-subject factor (time of measurement) and a group factor. The interaction between these factors will tell you whether groups changed differently over time.

Hi Karabiner, thanks you very much for your reply. I thought at the beginning since I have only two groups (adtypes) so it is not suitable to run anova.

But I assume there would be interactions so an anova should be fine. Now I find out that time and adtype has a interaction for "product attitude", interaction p=.045, plot looks exactly like this:

(can't upload my pic therefore I find a similar pic online)

Here comes the new problem: for my "brand attitude", the two lines cross just like the plot of "product attitude", but the p-value of interaction is 0.571, how does it happen?

Crossing lines do not necessarily mean that there is a statistically significant interaction.
The interaction has to be large, compared with the "noise" (variances) in the data.
Moreover, if your sample sizes are small, effects in the sample have to be large
to become statistically significant.
Better create a plot which also shows you the dispersion of the data, e.g.
4 box-and-whisker plots. Or at least add indicators for the standard deviations
to your line plot.

I'm still working on this issue and now I have a problem interpreting the outputs, would you mind to have a look at it and help me figure out where is my problem?

ad attitude (T-Test)
Adtype 1: 6.9
Adtype 2: 6.5

ad attitude (ANOVA, within factor: time, goup factor: Adtype)
AT1 (right after watching): 6.9
AT2 (right after watching): 6.6
AT1 (one week later): 6.2
AT2 (one week later): 5.9

main effect of time, no interaction between time and adtype.

Problem no.1: AT right after watching.
I'm aware that T-Test and ANOVA use different methods to calculate, but which result should I report in this case?

I try to figure out
1. whether the adtype makes a different in influssing ad attitude, therefore I use T-Test and compare the means. So i guess I should report the result of T-Test, but they are different than the results of anova.
2. whether the ad attitude changes after a week in different goups, therefore repeat measure ANOVA. there is no interaction between time and group, main effect of time proved. Doesn't mean AT1 AND AT2 changed after one week, or the AT generally changed after one week? (because no interaction between time and group). I'm quite confused now.