An Investigation of the Contribution of Targeted Marketing Data to the Prediction of Attitudes
An Investigation of the Contribution of Targeted Marketing Data to the Prediction of Attitudes
Principal Investigator: Patricia L. Mokhtarian, Susan G and Christopher D Pappas Professor, School of Civil and Environmental Engineering
Co-principal Investigators: Giovanni Circella, Research Engineer, School of Civil and Environmental Engineering | Kari Watkins, Frederick Law Olmsted Associate Professor, School of Civil and Environmental Engineering
Project Duration: 48 months
Project Budget (Federal UTC Funds): $264,180
Project Budget (Cost-share): $132,094
Institution: Georgia Institute of Technology
Abstract
This project involves the use of machine learning methods to impute attitudes into the Georgia subsample of the 2016-17 National Household Travel Survey, training the algorithms on the responses to a 2017 attitudinal survey administered to a separate statewide sample in Georgia. The “common variables” needed to train the learning function will include socio-economic/demographic and other variables found in both samples, but will be augmented by (1) land use-related variables (obtained from multiple external sources) associated with respondents’ residential neighborhoods, and (2) (for the first time) lifestyle-oriented targeted marketing variables associated with the household/respondent that are purchased from a commercial provider. The project will evaluate the effectiveness of targeted marketing variables for this purpose. The objectives of this project are (1) to impute attitudes into the Georgia subsample of the 2016-17 NHTS, training the imputation functions using attitudinally-rich data collected in Fall 2017 from a sample that is (reasonably) representative of the urban and small-town population of the state of Georgia; and (2) to augment the set of “common variables” available for training the imputation process with information from targeted marketing databases. Achievement of both objectives will involve testing the efficacy of the imputed attitudes for predicting travel-related choices of interest, using a variety of comparisons.