Agreeing to disagree with generalised decision functions

We develop a framework that allows us to emulate standard results from the “agreeing to disagree" literature with generalised decision functions (e.g. Bacharach (1985)) in a manner the avoids known incoherences pointed out by Moses and Nachum (1990). We analyse the implications of the Sure-Thing Principle, a central assumption. The upshot is that the way in which states are described matters, and that the results fail if decisions are allowed to depend on interactive information. Furthermore, using very weak additional assumptions, we extend all previous results to models with a non-partitional information structure in a coherent manner. Finally, we provide agreement theorems in which the decision functions are not required to satisfy the Sure-Thing Principle.