The mixed models introduced in the examples presented so far are reviewed in Table 6.1. These examples illustrate several features of the range of mixed models that can be specified. Each model term may be a variate, like ‘latitude’, in which each observation can have any numerical value, or a factor, like ‘town’, in which each observation must come from a specified set of levels. Each part of the model (the fixed-effect model and the random-effect model) may have more than one term, as in the random-effect model ‘block + line’. Factors may be crossed, as in the model ‘brand*assessor’, which is equivalent to
Alternatively, they may be nested, as in the model ‘day/presentation/serving’, which is equivalent to
Crossing and nesting of factors is specified using the notation of Wilkinson and Rogers (1973).
Table 6.1 Mixed models introduced in earlier chapters.
|Example||Response variable||Fixed-effect model||Random-effect model||Design illustrated||Where discussed|
|Effects on house prices in England||logprice||latitude||town||Regression analysis on grouped data||Chapter 1; Sections 3.18 and 4.2|
|Sensory evaluation of ravioli||saltiness||brand*assessor||day/presentation/serving ...|