Читать книгу Multilevel Modeling in Plain Language - Karen Robson - Страница 14

Maybe at this point you are wondering, ‘If groups are so important, maybe I should just focus on group-level effects.’ You may think that a possible solution to these problems is just to conduct analyses at the group level. In other words, to avoid the problem of giving group characteristics to individuals, just aggregate the data set so that we focus on groups, rather than individuals. In addition to having far less detailed models and violating some theoretical arguments (i.e. supposing your hypotheses are actually about individual- and group-level processes), a common problem with this approach is usually the sample size. If we have data on 13,000 students in eight different regions, aggregating the data to the group level would leave us with just eight cases (i.e. one row of data representing one region). To conduct any sort of meaningful multivariate analysis, a much larger sample size is required. As we mentioned earlier in this chapter, focusing on group estimates may lead to an error in logic known as the ecological fallacy where group characteristics are used to generalize to individuals. Thus, there are several reasons not to rely on group estimates.