Contextual effects and other influences: using multilevel modelling to study the extent and causes of spatial variations in post-war Australian federal voting
Associate Professor David Charnock Faculty of Media, Society and Culture, Curtin University
Refereed paper accepted for presentation at the Australasian Political Studies Association Conference University of Adelaide 29 September - 1 October 2004
David Charnock: Contextual effects and other influences on post-war Australian federal voting
Introduction1 The statistical techniques of multilevel modelling have been developed since the early 1980s as researchers in various fields have recognised the importance of accounting for the effects of data structures that are composed of different levels of units of analysis. These levels are usually hierarchically structured and it is this structure that most multilevel methods attempt to model2. In the area of political science, an example is individual voters, located within constituencies, located in turn within regions such as states. An example from educational research is individual school students, grouped into classes, which in turn occur within schools, and one from demography is individual families or households, located within villages, located in turn within some sort of larger geographic regions such as provinces. If data are available for each of the levels in these structures, then the effects at various levels can be studied simultaneously using multilevel models.
While the use of multilevel models for data structures like these is sometimes motivated by technical statistical considerations3, the models are naturally of most interest when there are substantive reasons for believing that individual behaviour or attitudes might be linked to the group context. In the voting example mentioned above, an incumbency advantage at the constituency level might be such an effect, as would a contextual effect of the social class composition of an area.

下一页