The study compares two approaches to predicting the distribution of parliamentary seats among parties under the d’Hondt system of seat allocation. The first approach proposed by Bochsler (2010) allows computation of seat distribution based on expected vote share (opinion poll data) at the district level. Flis et al. (2018) offer an alternative formula using countrywide results. The former approach makes it possible to apply district-specific weights, while the latter’s strength lies in its simplicity. In case of substantive differences in distribution of support for parties between districts and under the assumption that district-level voting patterns remain stable over time, the first approach should lead to more accurate prediction. This expectation is tested using election results of four major Polish parties from the period between 2001 and 2015. The empirical analysis suggests that the procedure relying on historical weights yield consistently better results only in case of the centrist-liberal Civic Platform (PO). Also with respect to the PO one may observe consistent overestimation of its expected share of parliamentary seats while using the method not controlling for regional differences. The seat share of the greatest competitor of the Civic Platform, the Law and Justice (PiS) party, on the contrary, can be reliably estimated based on its countrywide vote share. The study explains how these findings reflect differences in the spatial distribution of respective electorates of the two parties.
