Wall of noise? How useful are theories of electoral geography built on socio-demographic composition?