This dissertation present a Compound Poisson Mixture Regression model of the distribution of transaction frequency and monetary value, and apply it to study donations at a private university in the Midwestern United States. The model captures the joint effect of covariates, recognizing that both response variables emanate from one statistical unit -- a donor. Moreover, the mixture regression framework provides a rigorous and appealing approach to account for heterogeneity and other features in the data. In particular, the framework captures latent, group-level factors through coefficients that vary across the different population segments. The data in the study are from donation records for the 17 year period between 2000-2016, and an alumni survey conducted in the Fall of 2017. The empirical results highlight features of the proposed model, and lead to insights with potential to improve fundraising efforts. Specifically, the results show that the proposed model captures behavioral differences manifested as heterogeneity in either donation amounts, frequencies, or both response variables. Interestingly and in spite of the inclusion of subjective factors assessed through the survey, the results suggest that between-segment differences are not explained by the available data, i.e., the between-segment heterogeneity is unobserved. The results show that covariates, including a number of subjective factors, i.e., connectedness/psychological distance, perceptions of donation impact, and willingness to volunteer, display stratified marginal effects on either transaction amounts, frequencies, or compound effects on both response variables. We discuss how characterization of such effects supports development of targeted fundraising/marketing strategies. In order to deal with heterogeneous issues arising from the Compound Poisson Mixture Regression model, and to provide a practical way to control rating scale bias in a broader field, we present a method to estimate and control for individuals' rating scale biases appering in responses to surveys about their experiences, attitudes, feelings and perceptions. The approach is based on the Rasch model, and is motivated by the increasing use of survey data in marketing research. Without relying on additional objective information for anchoring purposes, the proposed approach utilizes only survey data itself to provide individual-question level bias correction, with impacts of both individual rating scales and specific questions accounted for. We apply the method to study data from an alumni survey at a private university in the Midwestern United States. Specifically, we use the bias-corrected parameters to estimate the relationships between attitudes and donation behavior. The results show that the bias-corrected survey data significantly improves model accuracy. Moreover, we observe that the marginal effects of survey variables from the bias-corrected model turn out to be different with model with original survey data in certain variables, which indicates that rating scale biases may impact insights related to the effects of alumni attitude. While the (practical) effectiveness of the proposed bias correction method is illustrated, we discuss limitations in the Rasch Model-based method. To further generalize accounting for heterogeneity in transportation field, this dissertation presents a segmentation analysis of households in the Chicago Metropolitan Area based on reported travel outcomes. The data are from the travel tracker survey conducted between 2007-2008 by the Chicago Metropolitan Agency for Planning. In our analysis, we assume that unobserved, group-level factors play a pivotal role in determining/explaining the heterogeneity observed across the population in terms of mode choice and distance traveled. As a benchmark, we consider a segmentation model relying exclusively on distance traveled by personally-owned vehicle or taxi, an approach used the literature. The results suggest additional information on trips of other modes is useful and validates our joint segmentation approach. Our analysis of the Chicago data suggests that the population consists of 4 segments of households. Aggregate analysis of the travel outcomes in each ZIP code highlights complicated inter-dependencies among travel behavior, residential location, and public transport coverage. Nevertheless, disaggregate analysis (of the correlations in the cluster membership probabilities) suggests that socioeconomic and demographic factors play stronger role in travel outcomes, than do build environment factors. The discussion concludes the actual relationship between urban form and travel behavior is not as simple as it seems in analysis of their statistical relationship, and relevant policies are also supported by our findings.

Creator

• Bao, JIngyuan

DOI

• https://doi.org/10.21985/n2-vjma-tj47

Subject

• Marketing
• Transportation

Language

• en

Alternate Identifier

• etdadmin_upload_810588
• http://dissertations.umi.com/northwestern:15511

Date created

• 2021-01-01

Resource type

• Dissertation

Rights statement

• In Copyright