Hello, first of all just a comment regarding the "median-split" approach: Using the median to create a categorical variable from a continuous one is poor statistical practise and connected to several problems, as summarized e.g. in the paper of Macmallu, Zhang and Rucker (2002). Regarding the problem with the sign: Are all other variables the same in the two regression model? Could you provide some of the data or outputs? In priciple you are right, the sign should be the same, as long as the "0" is the baseline category in the factorial regression model...