Catalog spending. This realistic modeling project requires much more time than a typical exercise. Table 28.16 shows catalog-spending data for the first 9 of 200 randomly selected individuals from a very large (over 20,000 households) data base.18 We are interested in developing a model to predict spending ratio. There are no missing values in the data set, but there are some incorrect entries that must be identified and removed before completing the analysis. Income is coded as an ordinal value, ranging from 1 to 12. Age can be regarded as quantitative, and any value less than 18 is invalid. Length of residence (LOR) is a value ranging from zero to someone's age. LOR should not be higher than age. All of the catalog variables are represented by indicator variables; either the consumer bought and the variable is coded as 1 or the consumer didn't buy and the variable is coded as 0. The other variables can be viewed as indexes for measuring assets, liquidity, and spending. Find a multiple regression model for predicting the amount of money that consumers will spend on catalog shopping, as measured by spending ratio. Your goal is to identify the best model you can. Remember to check the conditions for inference as you evaluate your models.
