Priority-Setting and Personality: Effects of Dispositional Optimism on Preferences for Allocating Healthcare Resources

Abstract

In a publicly financed health system, it is important that priority-setting reflects social values. Many studies investigate public preferences through surveys taken from samples, but to be representative, these samples must reflect value judgments of all relevant population subgroups. In this study, we explore whether, next to better-understood sources of heterogeneity such as age, education or gender, also differences in personality play a role in how people want to set limits to health care. We investigate the influence of dispositional optimism: whether someone anticipates a good or bad future. This is an important personality dimension that has been shown to widely reverberate into people’s lives and that can also be expected to influence people’s views on health care. To test our hypothesis, we asked a representative sample of the Belgian population (N = 750) to complete both the revised life orientation test and a discrete choice experiment about allocating healthcare resources, and we investigated the relationships between both measurements. We found that more pessimistic individuals were less supportive of using patients’ age as a selection criterion and more hesitant to invest in prevention. Since individual dispositions are usually not part of the criteria for selecting representative samples, our findings point at a potential non-response bias in studies that elicit social values.

Appendices

Appendix A: Example of a Choice Set

Medical interventions A and B are exactly equally expensive, and they apply to a similar number of patients. If you were forced to make a choice, which of both interventions should be reimbursed by the government? To make it easier for you, we have highlighted in yellow the characteristics that differ between both interventions. There are no right or wrong answers; we are interested in your opinion.

	A	B
What type of intervention is it?	Curative (meant to cure patients who are ill)	Preventive (meant to prevent healthy persons from becoming ill)
How big is the probability of success of the intervention?	2 in 3 is successful	Always successful
How often do adverse effects occur?	Often	Often
How severe is the illness for which the intervention is developed?	Not lethal, but everyone who gets the disease will experience a severe and lasting reduction in quality of life	Lethal, everyone who gets the disease will die from it
Does the patient cause the disease through his or her own lifestyle?	Not at all	Not at all
How long does it take before the patient becomes ill/shows signs/symptoms of illness?	Within a year	Within a year
At what age does the patient become ill?	0–10 years	40–50 years
Your preference	◘	◘

Appendix B: MNL Model

Formally, the MNL model employs random utility theory which describes the utility that a respondent attaches to intervention j (j = 1, 2) in choice set s (s = 1, …, 14) as the sum of a systematic and a stochastic component:

$$ U_{js} = {\mathbf{x^{\prime}}}_{js} {\varvec{\upbeta}} + \varepsilon_{js} . $$

In the systematic component, ${\mathbf{x^{\prime}}}_{js} {\varvec{\upbeta}},{\mathbf{x}}_{js}$ is a vector containing the attribute levels of intervention j in choice set s. Additionally, in our analysis, this vector includes the interactions between the attribute levels and the LOT-R score or any other respondent variable under investigation. The vector β is the vector of parameter values indicating the importance respondents attach to the different attribute levels and interactions. The stochastic component $\varepsilon_{js}$ is the error term capturing the unobserved sources of utility. Under the assumption that the error terms are independently and identically Gumbel distributed, the MNL probability that a respondent chooses intervention j in choice set s is

$$ p_{js} = \frac{{\exp \left( {{\mathbf{x^{\prime}}}_{js} {\varvec{\upbeta}}} \right)}}{{\exp \left( {{\mathbf{x^{\prime}}}_{1s} {\varvec{\upbeta}}} \right) + \exp \left( {{\mathbf{x^{\prime}}}_{2s} {\varvec{\upbeta}}} \right)}}. $$

To estimate the parameter vector β, we used a maximum likelihood estimation approach, which maximizes the probability of obtaining the responses from the selected data sample. A positive estimate has a positive effect on the total utility, whereas a negative estimate has a negative effect. We computed the overall significance of the attributes and interactions by means of likelihood ratio (LR) tests. Such tests evaluate the difference in goodness of fit between nested models. More specifically, they compare the goodness of fit of an unrestricted or full model to a restricted model in which one or more parameters have been set to zero.