Estimating the effect of individual time preferences on the use of disease screening.

• Author: Bradford, W. David
• Position: Survey

1. Introduction

   Time preferences are considered a fundamental characteristic of economic behavior. Standard utility theory, set in a dynamic model, has strong predictions about the effect of different rates of discounting on an individual's behavior. In general, we expect that higher rates of discounting for an individual will lead her to more strongly shift consumption of economic goods to the present and economic bads to the future, relative to a person with lower rates of preference for the present. Preventative health care can be categorized into two types: primary prevention and secondary prevention and screening. Primary prevention often requires patients to engage in activities they do not enjoy today (for example, reducing the intake of high-fat and high-sodium foods, exercising, losing weight, consuming pharmaceutical products, etc.) to prevent the onset of disease. Patients who discount the future more heavily should be less likely to demand primary preventative health care than patients with low rates of time discounting. However, secondary prevention involves screening and medical care intended to detect disease that may already be present and to prevent its advancement. Thus, while people who have high rates of discounting would still prefer to shift unpleasant health care into the future, their past neglect of primary prevention may raise the likelihood of disease such that the increased clinical need outweighs the economic tendency toward procrastination. Thus, more complex interactions between time preferences and the use of secondary prevention and screening are possible. Despite the potential importance of this time discounting effect on the demand for preventative medicine, the issue has not been heavily studied to date.

   We investigate the direct impact of higher discount rates for an individual patient on her utilization of secondary prevention health screens using a compensating variations method. We evaluate one standard screening tool for men (prostate exams), two screening tools for women (PAP smears and mammography), and three general screens (dental exams, blood pressure tests, and cholesterol tests). To do this, we conducted a nationally representative survey of 2000 individuals over age 40. In addition to a set of standard demographic and economic questions and respondents' recent utilization of health care screening tests, individual rates of time preference were elicited by asking respondents to imagine they had won a lottery that will pay them $10,000 one year from that day, or some higher value six years from that day. (Respondents were also told the interest rate that a savings account would pay to generate the offered higher future payment.) They were then asked whether they would prefer the one-year delayed payout or the six-year delayed payout. Follow-up questions were asked to permit tighter bounds on the range of discount rates. All payments (and so, interest rates) were randomly assigned uniquely to each respondent.

   With the data in hand, we model the joint likelihood that a respondent's latent discount rate lies within the interval indicated by their responses to the survey questions and that the respondent uses each of the six screening services; this model is estimated using a two-step maximum likelihood (LIML) method. We find that respondents have a discount rate of approximately 25.1% per year, on average, and that this discount rate increases with age. We find that discount rates have a generally negative relationship to the likelihood of screening, though a positive relationship is found for one disease screen--that for prostate cancer.

   The results from these models should be of interest to economists in general, as well as health policy makers. For economists, this will be one of the few attempts to integrate a direct estimate of individual agents' actual discount rates with their demand for a time-dependant service. Consequently, the results will inform an important, but understudied, intersection between economic theory and empirical estimation. For policy makers, the information gain with respect to the demand for preventative services should be similarly informative. Clinicians are often frustrated by the difficulty in convincing patients to consume preventative health care. This reluctance is typically taken as an indication that patients are poorly informed, and so education programs are proposed as a solution. These results suggest, however, that at least some patients are in part making rational decisions based upon their discounting of the future.

   The article proceeds by reviewing the literature on the estimation of individual rates of time preference and on models that predict the demand for preventative health care. Section 3 presents the details of our empirical models. Section 4 presents the results, and section 5 concludes with a discussion of the implications of this work and suggestions for future research.

2. Discount Rates and the Demand for Preventative Medicine

   Michael Grossman (1972) introduced the concept of health as a component of human capital, which depreciates and in which investments can be made. Since that seminal contribution, a number of economists have investigated many dynamic aspects of health production and health care demand (Wagstaff 1986; van Doorslaer 1987; Wagstaff 1993; Grossman and Kaestner 1997; Zweifel and Breyer 1997). Theoretically, there have been a number of contributions that have explicitly modeled the role of time preferences on general human capital investments, of which health care is one. The Becker and Murphy (1988) model of rational addiction is perhaps the most successful of these. In that model, agents have foresight, and make human capital (and other consumption) decisions based upon the current utility and future utility generated. They find that higher rates of time preference tend to lead to lower current consumption of goods but will increase current consumption of addictive products. As Grossman (2000) notes, the Becker-Murphy model predicts a discount rate effect only under certain circumstances (the result is generally ambiguous in sign, and uncertain in magnitude). Ehrlich and Chuma (1990) explore the general implications of the Grossman (1972) model more completely and do pay particular attention to the impact of time preference. They find that increasing the rate of discounting the future tends to reduce investments in health capital--though this result holds only on average. The empirical research we present will test these "average" predictions from the Grossman (1972) and Ehrlich and Chuma (1990) models.

   While the theoretical guidance is relatively clear with respect to the impact of time preference in health care demand, direct empirical tests of these predictions are notably absent from the literature. A number of authors have tested the effect indirectly, by demonstrating a schooling--health investment relationship that is consistent with an inverse relationship between discounting and human capital investment (Farrell and Fuchs 1982; Berger and Leigh 1989). However, these are only indirect tests and subject to multiple interpretations. Consequently, in the words of Grossman (2000, p. 401): "definitive evidence with regard to the time preference hypothesis is still lacking." While definitive evidence may be long in coming, we will at least present direct evidence in this work.

   One paper that does examine time and risk preferences impacts on the use of medical screening exams is Picone, Sloan, and Taylor (2004)--though again the test with respect to time preferences is necessarily indirect. The authors propose a simple two-period model of expected utility maximization over the decision to undergo cancer screening. They predict that higher rates of discounting would lead to a reduction in the demand for screening. However, in their model the likelihood of disease does not depend upon the health state; there is no long-term clinical benefit from early detection (either in terms of likelihood of successful treatment or mortality), and they do not explicitly model time preferences (rather they make inferences about time preferences by assuming specific functional forms for utility). From a data perspective as well, the authors lack a direct measure of time preferences. Using the Health and Retirement Survey, they are only able to categorize individuals into short time-horizon or long time-horizon groups, which may or may not be directly correlated with having high or low time preferences over the long run. Despite these limitations, which the authors discuss, they find that women with longer life expectancies and self-identifying as having a long time horizon are more likely to undergo cancer screening. We will improve upon this work in two areas. Theoretically, we will expand their model to permit more direct assessment of the role of time preferences, which generates a somewhat more complex set of results. Empirically, we will have a direct estimate of each person's underlying discount rate, which we can then use as an explanatory variable in models that track their actual use of several types of screening (not just mammography and PAP smears among women). Unlike Picone, Sloan, and Taylor, though, we will not have a measure of respondents' risk preferences.

   One reason that direct evidence on the relationship between time preference and health care demand is scarce is the difficulty in measuring time preference. Until fairly recently, the art of estimating individual discount rates has been poorly explored. However, there is currently a strong, and growing, literature on which to draw. (1) There are three primary methodologies for assessing individual discount rates. The first is to use natural experiments in which individuals must choose between alternatives with differential time dimensions, such that a discount rate can be inferred. An example of this literature is Warner and...