Application of Machine Learning Algorithms on Consumer Choices
In recent years, machine learning algorithms have been widely used in marketing to address the challenges posed by increasingly complex marketing contexts and large-scale unstructured data. Marketing researchers also have emphasized the need to prioritize accurate estimates of causal effects using machine learning algorithms. This dissertation aims to extend the application of machine learning algorithms to consumer choices, incorporating the two themes outlined above. In the first essay, we employ an advanced machine learning algorithm in the causal inference-generalized synthetic control (GSC) method to investigate the impact of household store choices on household sustainability. Specifically, we examine the effect of shopping at club stores on a household's food carbon footprint. Using a process-based Life Cycle Assessment (LCA) model and Nielsen Consumer Panel data, we calculate the household-level food carbon footprint between 2007 and 2017. Using GSC and Diff-in-Diff approaches, we find a consistent average treatment effect indicating that households generate 7-9% more per capita carbon emissions after purchasing groceries at club stores. We also observe a heterogeneous treatment effect that shows a larger increase for low-income and small households. Furthermore, we find that larger package sizes in club stores have a more significant impact on increasing the food carbon footprint. Our study is the first in the marketing field to introduce a carbon emissions calculation model and explore the relationship between retail formats and household sustainability. These findings have significant implications for policymakers and managers. In the second essay, we apply text mining algorithms to understand users' choices in the crowdfunding market using unstructured data. Specifically, we investigate how changes in platform specialization in terms of platform size and backers' composition (i.e., donors vs. buyers) influence platform participant behaviors and campaign outcomes in the event of a crowdfunding platform split-up. Our results show a higher probability of reaching funding goals for campaigns on a reward-based platform (main platform) after the launch of a donation-based platform. This is due to fewer campaigns being launched on the main platform after the split, and creators providing more visual campaign information (i.e., images and videos) to mitigate information asymmetry that is more of a concern after the platform split-up, as the increased proportion of buyers, who are more sensitive to such visual information than donors. Our findings support the notion that potential backers' motivations for supporting a campaign drive creators' information disclosure strategies. The study provides significant managerial implications for platforms and participants.
Zhu, Purdue University.
Marketing|Computer science|Artificial intelligence|Home economics
Off-Campus Purdue Users:
To access this dissertation, please log in to our