Graduation Year

2022

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Economics

Major Professor

Murat Munkin, Ph.D.

Committee Member

Benjamin M. Craig, Ph.D.

Committee Member

Padmaja Ayyagari, Ph.D.

Committee Member

William H. Greene, Ph.D.

Committee Member

Xin Jin, Ph.D.

Committee Member

Lu Lu, Ph.D.

Keywords

discrete choice experiments, EQ-5D, quality-adjusted life years, scale adjustments

Abstract

Unobserved heterogeneity is one of the main concerns for applied economists, particularly when modeling preference estimates for health and healthcare in the stated choice experiments. This thesis illustrates the current state-of-the-art in analyzing preference heterogeneity in health-related stated choice experiment studies and four empirical studies on modeling unobserved preference heterogeneity using some recent advancements in latent class models for controlling scale variation. This dissertation contributes to the field of health economics and econometrics. In chapter1, I have described the motivation, thesis objectives, conceptual framework, and contribution of my research to the existing literature.

Chapter 2 is focused on an in-depth literature review on preference heterogeneity analysis in health and healthcare-related DCEs. The goal of the chapter is to understand the current practices and identify the existing knowledge gaps. The paper systematically summarized current practices that account for preference heterogeneity based on the published discrete choice experiments (DCEs) related to healthcare. The systematic review conducted systematic searches on PubMed, OVID, and Web of Science and two recently published reviews to identify articles. The review included health-related DCE articles published between 01 January 2000 to 30 March 2020. All the included articles also presented evidence on preference heterogeneity analysis based on either explained or unexplained factors or both. The results showed that 342 of the 2,202 (16%) articles met the inclusion/exclusion criteria for extraction. The trend showed that analyses of preference heterogeneity increased substantially after 2010 and that such analyses mainly examined heterogeneity due to observable or unobservable factors in individual characteristics. Heterogeneity through observable differences (i.e., explained heterogeneity) is identified among 131 (40%) of the 342 articles and included one or more interactions between an attribute variable and an observable characteristic of the respondent. To capture unobserved heterogeneity (i.e., unexplained heterogeneity), the studies largely estimated either a mixed logit (n=205, 60%) or a latent-class logit (n=112, 32.7%) model. Few studies (n=38, 11%) explored scale heterogeneity or heteroskedasticity. Providing preference heterogeneity evidence in health-related DCEs has been found as an increasingly used practice among researchers. In recent studies, controlling for unexplained preference heterogeneity has been seen as a common practice rather than explained ones (e.g., interactions); yet a lack of providing methodological details has been observed in many studies that might impact the quality of analysis. As heterogeneity can be assessed from different stakeholder perspectives with different methods, researchers should become more technically pronounced to increase confidence in the results and improve the decision makers’ ability to act on the preference evidence.

Chapter 3 is focused on identifying preference heterogeneity among single-employee young adults for health insurance in the US. Analyses of preference evidence frequently confuse heterogeneity in the effects of attribute parameters (i.e., taste coefficients) and the scale parameter (i.e., variance). Standard latent class models often produce unreasonable classes with high variance and disordered coefficients due to confounding estimates of effect and scale heterogeneity. In this study, we estimated a scale-adjusted latent class (SALC) model in which scale classes (heteroskedasticity) were identified using respondents’ randomness in choice behavior on the internet panel (e.g., time to completion and time of day). Hence, the model distinctly explained the taste/preference variation among classes associated with individual socio-economic characters, where scales are adjusted. Once scale heterogeneity was controlled, we found substantial heterogeneity with four taste classes. Two of the taste classes were highly premium sensitive (economy-class), coming mostly from the low-income group, and the class associated with better educational backgrounds preferred to have a better quality of coverage of health insurance plans. The third class was a highly quality-sensitive class with higher socioeconomic status (SES) background and lower self-stated health conditions. The last class was identified as stayers, who were not premium or quality sensitive. This case study demonstrates that one size does not fit all in the analysis of preference heterogeneity. The novel use of behavioral data in the latent class analysis is generalizable to a wide range of health preference studies.

In the fourth chapter, I controlled for heteroskedasticity and heterogeneity (taste and scale) simultaneously and estimated Dutch EQ-5D-5L values using conditional, heteroskedastic, and scale-adjusted latent class (SALC) logit models by maximum likelihood. After controlling for heteroskedasticity, the PC and BWS values were highly correlated (Pearson's correlation: 0.9167, CI: 0.9109-0.9222) and largely agreed (Lin's concordance: 0.7658, CI: 0.7542-0.7769) on a pits scale. In terms of preference heterogeneity, some respondents (mostly young men) failed to account for any of the EQ-5D-5L attributes (i.e., garbage class), and others had a lower scale (59%; p-value: 0.123). Overall, the SALC model produced a consistent Dutch EQ-5D-5L value set on a pits scale, like the original study (Pearson's correlation:0.7295; Lin's concordance: 0.6904). This paper shows the merits of simultaneously controlling for heteroskedasticity and heterogeneity in health valuation. In this case, the SALC model was dispensed with a garbage class automatically and adjusted the scale for those who failed the PC dominant task. Future analysis may include more behavioral variables to better control heteroskedasticity and heterogeneity in health valuation.

Using discrete choice responses from a Peruvian valuation study, chapter 5 estimated EQ-5D-5L values on a quality-adjusted life-year (QALY) scale accounting for latent heterogeneity in scale and taste, as well as controlling heteroskedasticity at task level variation. In this chapter, I conducted a series of latent class analyses, each including the 20 main effects of the EQ-5D-5L and a power function that relaxes the constant proportionality assumption (i.e., discounting). Taste class membership was conditional on respondent-specific characteristics as well as their experience with the time trade-off tasks. Scale class membership was conditional on behavioral characteristics such as survey duration and self-stated difficulty level in understanding tasks. Each analysis allowed the scale factor to vary by task type and time (i.e., heteroskedasticity). The results indicate three taste classes: a quality-of-life oriented class (33.35%) that placed the highest value on levels of severity, a length-of-life oriented class (26.72%) that placed the highest value on life span, and a middle class (39.71%) with health attribute effects lower than the quality class and life span effect lower than the length-of-life oriented class. The EQ-5D-5L values ranged from -2.11 to 0.86 (quality-of-life oriented class), from -0.38 to 1.02 (middle class), and from 0.36 to 1.01 (length-of-life oriented class). The likelihood of being a member of the quality-of-life class was highly dependent on whether the respondent completed the time trade-off tasks (p-value <0.001). The results also show two-scale classes as well as heteroskedasticity within each scale class.

To understand heterogeneity in health valuation within the national population of US, the research in chapter 6 is aimed to conduct a scale-adjusted latent class model on DCE paired comparison data. The valuation study may identify latent groups that place different absolute and relative importance (i.e., scale and taste parameters) on the attributes of health profiles. We estimated EQ-5D-5L values on a quality-adjusted life-year (QALY) scale accounting for latent heterogeneity in scale and taste, as well as controlling heteroskedasticity at task level variation. We conducted a series of latent class analyses, each including the 20 main effects of the EQ-5D-5L and a power function that relaxes the constant proportionality assumption (i.e., discounting. Taste class membership was conditional on respondent-specific characteristics as well as their experience with the time trade-off tasks. Scale class membership was conditional on behavioral characteristics such as survey duration and self-stated difficulty level in understanding tasks. Each analysis allowed the scale factor to vary by task type and time (i.e., heteroskedasticity). The results indicate three taste classes: a quantity-of-life oriented class (36.28%) that placed the highest value on the lifespan attribute and least value on the quality of life attributes, a slight change in the quality of life sensitive class (34.29%) who placed the lowest importance on the lifespan attribute and were highly sensitive to a slight change from no problem to any problem in any health dimension, and a third class (29.49%) who were also quality of life-oriented nonetheless put more weight when health condition changed from moderate to severe problems. The likelihood of being a member of the quantity of life-oriented classes was associated with comparative lower educated people (p<0.05). The results also show two-scale classes as well as heteroskedasticity within each scale class.

Download

Included in

Economics Commons

Share

COinS