UCLA

With the increased efforts of firms to collect data at a more detailed level and the recent advancement of data storage technology, empirical researchers have encountered new challenges in extracting values from large-scale data. Answering to the calls in the era of “big data”, the present dissertation introduces two applications on how to deal with large-scale data problems in marketing: The first essay shows an application on how to extract a specific information from multiple datasets in an online advertising context, and the second essay introduces a new estimation method which enables to measure heterogeneity from large-scale data.

The first essay proposes a structural modeling approach to enhance the precision in measuring an online advertising effect on purchase probability by jointly estimating individual level purchase and store browsing behavior. Many researchers have pointed out that measurement of the ad effect is difficult because the effect size is small and the unexplained variation in purchase is large. Our method extracts information on the difference between latent purchase utility of consumers who have been exposed to ads and who have not been exposed to ad from the difference of observed browsing behavior between those two groups of consumers.

The second essay introduces a novel approach which significantly improves the efficiency of the estimation process for random coefficients models by employing the Stochastic Gradient Descent algorithm. Simulation studies under random coefficients logit model settings demonstrate that our SGD estimator outperforms the BFGS approach, which is one of the most commonly used approaches, in both efficiency and accuracy.