Using a theoretical framework in which voters' choices are the result of ut
ility maximisation decisions, and in which the variation of utility functio
ns across individuals is partly deterministic (depending on their socio-eco
nomic characteristics) and partly stochastic, we estimate constituency-leve
l regression equations to explain how the electoral shares of each of the m
ain parties depend on conditions in the constituency. Whilst social charact
eristics appear to have an impact similar to that predicted in the existing
literature, economic conditions appear to have played a very different rol
e in 1997 from in previous elections.