I have access to two surveys of the population in a specific country - one where the country is divided into 10 regions that are used as sampling strata, and another one where the country is divided into 26 regional strata. Each of these 10 and 26 regional strata are again divided into an urban and a rural stratum. From the urban and rural strata inside each region a number of respondents are drawn, securing an urban/rural-balance corresponding to the urban/rural-balance of each region.
What I want to do is to test for regional effects on my dependent variable. The regions I want to test for are, however, not necessarily corresponding to the strata of the surveys. For the first survey I will have to divide some of the 10 regional strata into two in order to obtain regions that correspond to the regions that I want to test for. For the second survey I need to merge two or three of the 26 regional strata together in order to obtain the wanted regions.
I think that the first option sounds like a bad idea, since I will not know anything about the sample of my new micro regions. But what about the second option? Am i doing anything wrong if I merge two or three regional strata together in order to use them as a dummy macro-regions in my regression analysis? And is the second option, in any case, better than the first?