ABSTRACT
Whenever a promising new approach to data analysis becomes available many researchers are keen to try it out. In this chapter, we describe some of our attempts to use the new statistical models termed random coefficient or hierarchical or multilevel structural models (after Mason, Wong, & Entwisle, 1984). Of course, multilevel models are not really new models because they have been available for over two decades (e.g., Aitkin & Longford, 1986; Hartley & Rao, 1967; Harville, 1977; Jennrich & Sampson, 1976; Lindley & Smith, 1972; Novick, Jackson, Thayer, & Cole, 1972). However, several recent reviews show a surge of interest in multilevel models (see Bock, 1989; Bryk & Raudenbush, 1993; Goldstein, 1987; Hoc & Kreft, 1994; Longford, 1993; Muthen & Satorra, 1989; Raudenbush, 1988). These statistical models have become popular for a very good reason—multilevel models seem to provide a reasonable way to deal with the complex statistical and mathematical problems of “nested” or “clustered” data collection designs.
