Date of Award
Fall 2022
Document Type
Open Access Dissertation
Department
Epidemiology and Biostatistics
First Advisor
James W. Hardin
Abstract
A discrete choice experiment (DCE) allows researchers to understand how individuals value characteristics of a product or service in hypothetical scenarios and the trade-offs these individuals are willing to make between these characteristics. DCEs quickly gained traction in public health but as more researchers utilized DCEs in broader contexts, several methodological questions have arisen. This dissertation addresses some of these gaps in the literature.
One critique of DCEs is whether individuals would make the same choice in reality as they claimed they would have made in the hypothetical scenario. Perhaps the most efficient way to evaluate the predictive value of DCEs is to use two-stage models such that the predicted probabilities of the associated attributes are used to predict an individual’s characteristics. To lower prediction error and improve accuracy, the derivation of an appropriate sandwich variance estimator is attuned for use in this application.
A practical guide including Stata syntax is constructed for public health researchers on incorporating an opt-out option, comparing effects coding versus indicator coding, assessing between-subject variables and control variables, and running best-worst models for best-worst DCE data. Best-worst models simultaneously incorporate the information from the best and worst indicated choices. There are currently no programs in Stata to run these. Another flaw in Stata’s extensive CM suite for modeling choice data is that it does not assist researchers in setting up their choice data. Thus, several commands have been developed to: (a) manipulate discrete choice data into the format for the CM suite, and (b) estimate best-worst models.
One critique of DCEs is whether individuals would make the same choice in reality as they claimed they would have made in the hypothetical scenario. Perhaps the most efficient way to evaluate the predictive value of DCEs is to use two-stage models such that the predicted probabilities of the associated attributes are used to predict an individual’s characteristics. To lower prediction error and improve accuracy, the derivation of an appropriate sandwich variance estimator is attuned for use in this application.
A practical guide including Stata syntax is constructed for public health researchers on incorporating an opt-out option, comparing effects coding versus indicator coding, assessing between-subject variables and control variables, and running best-worst models for best-worst DCE data. Best-worst models simultaneously incorporate the information from the best and worst indicated choices. There are currently no programs in Stata to run these. Another flaw in Stata’s extensive CM suite for modeling choice data is that it does not assist researchers in setting up their choice data. Thus, several commands have been developed to: (a) manipulate discrete choice data into the format for the CM suite, and (b) estimate best-worst models.
Rights
© 2022, Farahnaz Islam
Recommended Citation
Islam, F.(2022). Extensions of Discrete Choice Experiment Theory for Public Health. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/7032