Election night forecasting : Australian House of Representatives elections

Abstract

This thesis develops a statistical methodology, based on early counts, for prediction of final outcomes in Australian House of Representatives elections. The basis of the method is that early counts, after being reduced to a two party preferred vote for each Division, can be corrected for bias which typically arises in early returns because of the nonrepresentative nature of the sample. An analysis based on two party swing is used to predict results in Divisions where no counts have been recorded. The model provides estimates of variance from which statements on the probabilistic uncertainty of the seat being won by either party can be made. Basically the method requires the assumptions that patterns of bias will remain consistent from one election to the next and that average swing is a useful predictor where no information is available. That is, there are no violent fluctuations from State to State or between Divisions within States. It is also assumed that electoral redistributions will not greatly affect the nature of the bias in each of the new Divisions. The method is evaluated using data from the 1974, 1975 and 1977 elections for NSW Divisions only. Predictions are made for the 1975 data based on an analysis of the 1974 data. An analysis of the pooled 1974 and 1975 data provides input for predictions of the 1977 result. The results of these analyses suggest that the model performs satisfactorily and it certainly offers much more than the methods currently used. Show more