Download Free Essays On Discrete Choice Models Book in PDF and EPUB Free Download. You can read online Essays On Discrete Choice Models and write the review.

This dissertation focuses on the identification and estimation of discrete choice models. In practice, if the error term is independent of the covariates and follows some known distribu- tion, the discrete choice model is usually estimated using some parametric estimator, such as Probit and Logit. However, when the distribution of the error is unknown, misspecification would in general cause the estimators inconsistent even if the independence between the covariates and the error still holds. The two chapters relax the assumptions on the error distribution in the discrete choice models and propose semiparametric estimators.
This dissertation consists of three individual studies that are concerned with discrete choice models with latent variables. In all three studies, a special attention is paid to specification and estimation of latent constructs. The first study aims to extends the methodological framework for latent class model by incorporating latent variables in the class assignment rule. The second study focuses on goodness of fit of discrete choice models with latent variables. The last study is concerned with discrete choice models with latent variables applied in the context of hurricane evacuation. The work presented here aims to explore the complexity and challenges of modeling latent constructs that are often overlooked.
Unobserved heterogeneity is comprehensively acknowledged as an important feature to be considered in discrete choice modeling. Over the last decade, there were abundant studies showing the great outperformance of capturing unobserved heterogeneity of Mixed-Mixed Logit(MM-MNL) models. However, most empirical researches still use mixed logit(MIXL) models or latent class(LC) models which introduced strong assumptions on distributions of marginal utility. In this dissertation, a Mixed-Mixed Logit model(MM-MNL) that assumes a non-parametric mixing distribution for marginal utility is discussed. Consequently, three empirical studies solving different transportation problems are introduced.
This dissertation consists of three essays divided into chapters. In chapter 1, I analyze the identification of a simultaneous binary response model without nonadditive unobservable random terms, and suggest an estimation method. In particular, the derivatives of structural equations are identified and estimated. The identification relies on a special regressor, which enters the underlying structural equation linearly. All other exogenous variables held constant, variation on this special regressor generates variation on the structural equation which determines the latent endogenous variable in a known way, so we can recover the conditional distribution of the structural equations. The estimator can be constructed using a least-squares method, after replacing the elements of a matrix with kernel density and density derivative estimates. The estimator is shown to be consistent and asymptotically normal. In chapter 2, I examine the determinants smartphone adoption among the elderly in South Korea. The advent of smartphones has caused a dramatic change in access to information and media, leading to a super-connected world of real-time services. Meanwhile, the constant dissemination of new technologies makes the digital divide multi-layered. In particular, older persons fall far behind the overall population in the access and use of new devices. To understand the technological environment following the introduction of smartphones and other smart mobile devices, I examine individual, household, and regional factors that can influence the preferences of the elderly with regard to obtaining a smartphone. I find that smartphone ownership among the elderly is mainly determined by personal rather than family characteristics. Also, I find that the area where a person lives has a significant effect on the probability of their owning a smartphone. In chapter 3, I analyze the evolution of preferences for brands in digital camera market. A consumer considers the value of a brand, as well as product characteristics when deciding which product to buy. One way to capture this effect is to use brand-specific dummy variables. However, including brand-specific dummy variables does not fully account for the variation of the unit sales of compact digital cameras, since the preference for digital camera brands evolves over time. Assuming that the brand preference is affected by the advertising expenditure of each brand and the reputation among consumers, I suggest a method to capture the time-varying brand preference under the specification of BLP model.