Election Night Forecasting

Abstract

The context and problems of election night forecasting are described. The linear models proposed attempt to reduce parameterization by splitting constituencies into three types-special, "three-party" and "two-party". The analysis is based on a modified ridge regression approach. The election night February 1974 performanceof the method as implemented for the BBC is described. Some results concerning univariate and multivariate ridge regression are given.

Forecasting on British Election Night 1997

“…Throughout the night, commentary on particular constituencies as they declare, and on the state of the parties, is interlaced with forecasts which are constantly updated; these forecasts form the driving force for broadcast discussion by both psephologists and politicians. The forecasting method used by the BBC on May 1st, 1997, is broadly unchanged since its inception in 1974 (Brown and Payne, 1975). In this paper we describe modi®cations to the basic method that were introduced for the 1997 election; some of the changes were designed to cope with the problems experienced with the election night forecast in the previous general election in 1992, whereas others were necessitated by the political context of the 1997 election itself.…”

Ridge Regression and James-Stein Estimation: Review and Comments

“…Brown and Payne[7], on the other hand, do not see this scaling as a cure-all. In addition,Brown [6] has pointed…”

Estimation of the constant term when using ridge regression

“…This discrepancy between theory and practice has not been commented upon except in a brief remark in [2] and more extensively in [1]. In the next section we provide several arguments as to why estimating b 0 by F is preferable to the estimator of bo given in (4).…”

“…By contrast, in most published applications of OR regression (e.g. [2,4,6,7]) the estimators used are…”