To be more precise:

I need a way to show the decay of effect (indep. var = region; dep. var. = preference for specific party) over time.

From 1990 on I expect that group 1 initially has higher preference for Party A than group 0 but that this preference decays over time. The difference in preference is expected to be smaller in 2020.

Hi Felix. In your case you may first try with separate regressions for each of the two groups and then to compare the results of both regressions in order to determine a difference in the statistical significance between the two groups and accordingly to explain the effect. This can be done if your sample is not small. As I can see your sample is about 30 years or more, so it is OK, you can do this.

Also the other approach is to create/ include a dummy variables as you have already explained that how you plan to divide your groups and to test the effect over the time. It is OK as well. You can use grouping variable using more dummies and it can be done in different ways. Please take care that the number of the dummies should be less/smaller for one then the number of the categories you have in order to avoid a trap. Also you should take care in explaining of the coefficients of the partial regression in such dummies. For instance, in your case, the coefficient of the dummy under ceteris paribus condition is the difference in average viewership of TV between those who lived in A and B region. Then, furthermore you could change the value of the slope creating interaction variable (region A*Political party A) to explore the effect of the preference of the specific party. These interaction variables can be used always when there is an assumption that the influence of a independent variable on dependent variable has been changing. I hope it helps. Good luck with your research.

I do not see the hypothesis you want to test - apart from, that party preferences of two regions are approaching each other. But TV before 1990 was only randomly chosen, but - I think - cannot have any relevance for the party preference. Therefore, there seems no reason for regressions, because there is no explaining variable apart from region, which means it is sufficient to show the relations of party preference for both regions over time.