A well-known result in spatial voting theory is that, for a one-dimensional issue space and under certain mild conditions,
political parties choose platforms coinciding with the median voter's position. This result does not carry over to multi-dimensional
issue spaces however, since then an equilibrium only exists under very restrictive assumptions. This implies that there are
in general always policy positions for the challenger that defeat the incumbent party and the model therefore predicts that
the challenger always wins the election and that policy positions of subsequent governments keep on fluctuating. These results
hinge critically on the assumption that voters as well as political parties have full information of voters' and parties preferences.
This is indeed a very demanding assumption. In recent advances in computational political economy, the assumption that political
candidates exactly know voters' preferences is relaxed. Instead, political candidates are assumed to experiment with different
policy positions in order to find the position where the probability of winning is the highest (see for example Kollman, Miller
and Page, American Political Science Review 1992). In this paper we extend this analysis in two directions. First of all we
introduce interest groups, in which voters are organized that feel strongly about a certain position on a certain issue. These
interest groups influence the election outcome by, on the one hand, providing information about its members preferences to
the political candidates and, on the other hand, advising these members on how to cast their vote in the upcoming election.
Secondly, in addition to adaptive political parties we investigate incomplete information and adaptive behavior on voters'
behalf. Voters do not act independently of each other, but they interact and influence each other in determining their preferred
policy positions. In particular, group formation and herd behavior plays an important role. We explicitly model these characteristics
which leads to a, highly computational, agent-based model of spatial competition. The model features clustering of preferences
and the endogenous formation of social groups. These social groups lead to the creation of interest groups. The size and position
of the interest groups is therefore determined endogenously and depends upon the interaction structure within the population
of voters and on the election outcomes. Hence, the model leads to a two-way interaction between the size and the position
of the interest group and the policy positions of political candidates.