Confounding sources of variance in the macro-analysis of electoral data

. Four sources of variance are examined in the construction of ecological models of electoral behaviour: aggregation bias, contextual effects, temporal and spatial diffusion. The proper specification of both micro‐ and macro‐models of political behaviour requires these four sources of variance to be taken into account. Ecological analysis may make valid inferences to individual behaviour once the models are properly specified. Electoral data for Britain between 1955 and October 1974 are used to test a series of structural equation models specifying developmental relationships between class and electoral behaviour and taking into account the four potential sources of bias.