Impartial redistricting: a Markov chain approach to the "Gerrymandering problem"

Abstract: After every U.S. national census, a state legislature is required to redraw the boundaries of congressional districts in order to account for changes in population. At the moment this is done in a highly partisan way, with districting done in order to maximize the benefits to the party in power. This is a threat to U.S's democracy. There have been proposals to take the re-districting out of the hands of political parties and give to an "independent" commission. Independence is hard to come by and in this thesis we want to explore the possibility of computer generated districts that as far as possible to avoid partisan "gerrymandering". The idea we have is to treat every possible redistricting as a state in a Markov Chain: every state is obtained by its former state in random way. With some technical conditions, we will get a near uniform member of the states after running sufficiently long time (the mixing time). Then we can say the uniform member is an impartial distribution. Based on the geographical and statistical data of Pennsylvania, I have achieved the Markov Chain algorithm with several constraints, done optimization experiments and a web interface is going to be made to show the results.