On The Log-Linear Analysis Of Multiple Response Data

Log-linear procedures have been recommended for analyzing data that are in the form of counts or frequencies (Iacobucci and McGill 1990). In its current form, the recommended procedure does not deal with multiple response data (i.e., more than one thought per response category). The paper proposes analytical modifications that researchers need to consider when fitting models with this type of data. The problems and recommended modifications are illustrated with two distinct data sets. The first involves consumers' attribution responses to negative word-of-mouth communication about a brand whereas the second pertains to consumers' cognitive responses toward an advertisement. Results in each case indicate that researchers may generate erroneous conclusions about the significance of estimated parameters if the recommendations are ignored.