TY - THES A1 - Huber, Sandra T1 - The role of polls for election forecasting in German state elections N2 - German state elections are in focus of this work due to the decreasing importance of the "catch all parties" and rise of the AfD in 2013. As small parties like the AfD first reached the 5% threshold in state parliaments (e.g. in the Saxony state election 2014), state elections can be used as barometer elections for the national ones. Further, state elections fill the gap between the 4-year national election cycle and provide additional information for the national election. The aim of this thesis is to forecast state elections based on polling data from different institutes. Despite occurring errors in polls like measurement or sampling errors - which are also discussed in this work - forecasting is made with aggregate models depending on short term polling data. Irregular polling data have to be customized to generate daily data to apply parametric regression based models. To forecast single vote shares in multi-party elections, the range of methods varies from basic methods like averaging over nonparametric regression based methods to dynamic linear models. KW - forecasting KW - election KW - polls KW - Umfragen KW - Deutschland KW - Landtagswahl KW - Demoskopie KW - Wahlprognose KW - Geschichte 2014 - 2017 Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:739-opus4-5552 ER -