This PDF is a selection from an out-of-print volume from the National Bureau of Economic Research Volume Title: Economic Aspects of Health Volume Author/Editor: Victor R. Fuchs. editor Volume Publisher University of Chicago Press Volume ISBN: 0-226-26785-7 VolumeUrl:http://www.nberorg/books/fuch82-1 Publication Date: 1982 Chapter Title: Time Preference and Health: An Exploratory Study Chapter Author: Victor R. Fuchs ChapterUrl:http://www.nber.org/chapters/c6546 Chapter pages in book:(p 93-120)
This PDF is a selection from an out-of-print volume from the National Bureau of Economic Research Volume Title: Economic Aspects of Health Volume Author/Editor: Victor R. Fuchs, editor Volume Publisher: University of Chicago Press Volume ISBN: 0-226-26785-7 Volume URL: http://www.nber.org/books/fuch82-1 Publication Date: 1982 Chapter Title: Time Preference and Health: An Exploratory Study Chapter Author: Victor R. Fuchs Chapter URL: http://www.nber.org/chapters/c6546 Chapter pages in book: (p. 93 - 120)
Time Preference and health An Exploratory Study Victor R. fuchs c& This paper reports the results of an exploratory effort in a new area- ne relationship between intertemporal choice, health behavior, and health status. Intertemporal choice(or time preference)is, of course, a subject much discussed by economists and psychologists. (See Maital and Maital 1978. ) There is also a large literature on individual behavior(e. g cigarette smoking, diet, exercise)and health status. This paper, howe,g seems to be the first to attempt to bring these subjects together and to test empirically for possible interrelations In the first section of the paper I briefly review some of the considera tions that suggest that an investigation of time preference might throw light on health behavior and health status. These include empirical stud ies of the relation between schooling and health, epidemiological inves- tigations of the health effects of cigarette smoking, diet, exercise, and the like, and theoretical issues concerning investment in human capital imperfections in capital markets, and optimizing behavior The second section considers the critical problem of the measurement of time preference and reviews some recent efforts by other investigators to measure time preference in contexts other than health. I then describe a pilot questionnaire given to 500 men and women and present the results Victor R. Fuchs is professor of Economics at Stanford University and a research associate at the National Bureau of Economic Research. This research was supported by grants from The robert wood johnson found The Henry J. Kaise Phillip Farrell provided research assistance and made many valuable suggestions. The cipants in the Stanford University Interdisci stimulating comments, and at an early stage I benefited from discussions with Martin Seligman and Richard Thaler. Helpful comments from the participants in the NBEr Conference“Eco Aspects of Health"are also gratefully acknowledged. The con- tributions of many other colleagues are indicated in the references and notes. The research reported here is part of the NBEr,'s research in the Health Economics Program. A pinions expressed are those of the author and not those of the National Bureau Economic
3 Time Preference and Health: An Exploratory Study Victor R. Fuchs This paper reports the results of an exploratory effort in a new areathe relationship between intertemporal choice, health behavior, and health status. Intertemporal choice (or time preference) is, of course, a subject much discussed by economists and psychologists. (See Maital and Maitall978.) There is also a large literature on individual behavior (e.g., cigarette smoking, diet, exercise) and health status. This paper, however, seems to be the first to attempt to bring these subjects together and to test empirically for possible interrelations. In the first section of the paper I briefly review some of the considerations that suggest that an investigation of time preference might throw light on health behavior and health status. These include empirical studies of the relation between schooling and health, epidemiological investigations of the health effects of cigarette smoking, diet, exercise, and the like, and theoretical issues concerning investment in human capital, imperfections in capital markets, and optimizing behavior. The second section considers the critical problem of the measurement of time preference and reviews some recent efforts by other investigators to measure time preference in contexts other than health. I then describe a pilot questionnaire given to 500 men and women and present the results Victor R. Fuchs is professor of Economics at Stanford University and a research associate at the National Bureau of Economic Research. This research was supported by grants from The Robert Wood Johnson Foundation and The Henry J. Kaiser Family Foundation. Phillip Farrell provided research assistance and made many valuable suggestions. The participants in the Stanford University Interdisciplinary Seminar in Decision Analysis offered stimulating comments, and at an early stage I benefited from discussions with Martin Seligman and Richard Thaler. Helpful comments from the participants in the NBER Conference “Economic Aspects of Health” are also gratefully acknowledged. The contributions of many other colleagues are indicated in the references and notes. The research reported here is part of the NBER’s research in the Health Economics Program. Any opinions expressed are those of the author and not those of the National Bureau of Economic Research. 93
ictor R. Fuchs of correlation and regression analyses of their replies. The paper con- cludes with a discussion of questions raised by this exploratory research Background Empirical considerations Cross-sectional studies of the determinants of health status in the United States usually report a strong association between health and years of schooling. This result typically aprality rates)or subjectively ppears regardless of whether health is measured objectively(e.g, mor (e. g, self-evaluation), and is equally robust in studies of differences across groups(e. g, states or cities )or across individuals(e. g, household survey data). Simple correlations between health and years of schooling are usually significant in both the statistical and the practical sense Furthermore, the relation remains strong after controlling for other variables such as income Probably the most thorough investigation of this relationship has been carried out by Michael Grossman in"The Correlation between Health and Schooling"(1975). This study of middle-aged men is particularly notable for two reasons First, a statistically significant effect of schooling on health remains after controlling for a large number of other variables, including family background, health status in high school, income, job satisfaction, and y their early 20s Second, each of the men had at least a high school diploma; the mear level of schooling was over fifteen years. Grossman's finding that the favorable effects of additional schooling persist even at high levels of schooling is in sharp contrast to the relation between income and health which is positive at low levels of income but seems to be much weaker or nonexistent at average or high levels(Auster, Leveson, and Sarachek 1969) While the relationship between schooling and health seems well estab- lished, the mechanisms through which schooling affects health are less clear, Grossman has interpreted the empirical results as support for a make the individual a more efficient producer of health. This efficiency may arise through wiser use of medical care or, what is more likely through differences in cigarette smoking, diet, and other elements of"life yle The view that" the greatest potential for improving the health of the American people.. is to be found in what people do and don,'t do to and for themselves"(Fuchs, 1967) has gained widespread acceptance in re
94 Victor R. Fuchs of correlation and regression analyses of their replies. The paper concludes with a discussion of questions raised by this exploratory research. Background Empirical considerations Cross-sectional studies of the determinants of health status in the United States usually report a strong association between health and years of schooling. This result typically appears regardless of whether health is measured objectively (e.g., mortality rates) or subjectively (e.g., self-evaluation), and is equally robust in studies of differences across groups (e.g., states or cities) or across individuals (e.g., household survey data). Simple correlations between health and years of schooling are usually significant in both the statistical and the practical sense. Furthermore, the relation remains strong after controlling for other variables such as income. Probably the most thorough investigation of this relationship has been carried out by Michael Grossman in “The Correlation .between Health and Schooling” (1975). This study of middle-aged men is particularly notable for two reasons. First, a statistically significant effect of schooling on health remains after controlling for a large number of other variables, including family background, health status in high school, income, job satisfaction, and scores on physical and mental tests taken by the men when they were in their early 20s. Second, each of the men had at least a high school diploma; the mean level of schooling was over fifteen years. Grossman’s finding that the favorable effects of additional schooling persist even at high levels of schooling is in sharp contrast to the relation between income and health, which is positive at low levels of income but seems to be much weaker or nonexistent at average or high levels (Auster, Leveson, and Sarachek 1969). While the relationship between schooling and health seems well established, the mechanisms through which schooling affects health are less clear. Grossman has interpreted the empirical results as support for a household production function model; additional years of schooling make the individual a more efficient producer of health. This efficiency may arise through wiser use of medical care or, what is more likely, through differences in cigarette smoking, diet, and other elements of “life style. ” The view that “the greatest potential for improving the health of the American people. . . is to be found in what people do and don’t do to and for themselves” (Fuchs, 1967) has gained widespread acceptance in re-
Time Preference and Health cent years as the result of numerous studies by epidemiologists and social ntists interested in health These studie ences in health status and in life expectancy associated with such factors as cigarette smoking, diet, and exercise. Not only is a statistical correla tion well established but in many instances there is some understanding of the causal mechanisms as well, e. g, the role of diet and exercise in the prevention of atherosclerosis. What is not understood at all well is the cause of individual variation in health-related behavior From an economic point of view many of these behaviors have common characteristic--they involve trade-offs between current costs and future benefits. The costs may be purely psychic, such as the loss of pleasure from passing up a rich dessert or a cigarette. They may involve time, such as jogging, or they may involve other costs including financial and nonfinancial resources. The expected benefits typically take the form of reductions in the probability of morbidity and mortality from one or more diseases sometime in the future Theoretical considerations e The acceptance of a current cost for a future benefit constitutes an vestment. Becker's development of the theory of investment in human capital(Becker 1964)and Grossman's application of this theory spe- cifically to health(Grossman 1972) provide a convenient framework for thinking about these health behaviors. Suppose individuals differ in their willingness or ability to undertake investments, i. e, they have different time preferences. Such differences might help to explain variations in cigarette smoking, diet, and the like. Furthermore, this approach sug- links with the health-schooling relationship that has bee found by so many investigators There are at least two ways that individual variation in time preference ould explain the correlation between schooling and health. 2 First, sup- pose that differences in time preference are established early in life, are relatively stable, and do affect subsequent behavior. These differences might be due to differences in the education or income of parents, the stability of the family, the values associated with different religions, or other background characteristics. Given variation in time preference, it would not be surprising to observe that individuals with low rates of time discount would invest in many years of schooling and would also invest in health-enhancing activities. According to this view schooling has no direct effect on health; the observed correlation is due to both schooling and health as depending upon time preference A second possibility(the two explanations are not mutually exclusive) is that schooling actually affects time preference; those with more schoo ing are more willing to invest at a lower rate of return. Thus more schooling could result in better health by increasing investments in
95 Time Preference and Health cent years as the result of numerous studies by epidemiologists and social scientists interested in health.’ These studies report significant differences in health status and in life expectancy associated with such factors as cigarette smoking, diet, and exercise. Not only is a statistical correlation well established, but in many instances there is some understanding of the causal mechanisms as well, e.g., the role of diet and exercise in the prevention of atherosclerosis. What is not understood at all well is the cause of individual variation in health-related behavior. From an economic point of view many of these behaviors have a common characteristic-they involve trade-offs between current costs and future benefits. The costs may be purely psychic, such as the loss of pleasure from passing up a rich dessert or a cigarette. They may involve time, such as jogging, or they may involve other costs including financial and nonfinancial resources. The expected benefits typically take the form of reductions in the probability of morbidity and mortality from one or more diseases sometime in the future. Theoretical considerations The acceptance of a current cost for a future benefit constitutes an investment. Becker’s development of the theory of investment in human capital (Becker 1964) and Grossman’s application of this theory specifically to health (Grossman 1972) provide a convenient framework for thinking about these health behaviors. Suppose individuals differ in their willingness or ability to undertake investments, i.e., they have different time preferences. Such differences might help to explain variations in cigarette smoking, diet, and the like. Furthermore, this approach suggests possible links with the health-schooling relationship that has been found by so many investigators. There are at least two ways that individual variation in time preference could explain the correlation between schooling and health.* First, suppose that differences in time preference are established early in life, are relatively stable, and do affect subsequent behavior.’ These differences might be due to differences in the education or income of parents, the stability of the family, the values associated with different religions, or to other background characteristics. Given variation in time preference, it would not be surprising to observe that individuals with low rates of time discount would invest in many years of schooling and would also invest in health-enhancing activities. According to this view schooling has no direct effect on health; the observed correlation is due to both schooling and health as depending upon time preference. A second possibility (the two explanations are not mutually exclusive) is that schooling actually affects time preference; those with more schooling are more willing to invest at a lower rate of ret~rn.~ Thus more schooling could result in better health by increasing investments in
ictor R. fuchs health. The empirical portion of this paper, based on a single cross section survey, cannot distinguish between these two hypotheses, but we can test for possible relations between schooling and time preference Empirical investigation of time preference through survey questions designed to elicit marginal rates of time discount depends critically on capital markets being"imperfect. "If capital markets were perfect (i. e, if individuals could borrow and lend without limit at a single market rate of interest)marginal rates would be equal for all regardless of time prefer ence. Differences across individuals in time preference might still result in differences in nontradeable health-related activities but these would not be predictable from the replies to interest rate questions. However, if capital markets are not perfect(an assumption of this paper), individuals may well have different rates of interest at the margin, and these may be related to health behavior and health status Let us imagine a two-period world. Suppose utility in each period depends upon consumption of goods(G). Utility in the first period also is a function of some activity C(for simplicity assumed to be free with respect to G)which affects health(and therefore utility) in period two For example C1 might be cigarette smokin U1=U1(G1,C1) U2=U2(G2, H2) where H2=H(C1) a wealth compensated increase in the rate of interest(r)will,ceteris paribus, alter the allocation of wealth between G1 and G2. But if the marginal utility of C depends on the quantity of G,(and the marginal utility of H2 depends on the quantity of G2), the change in r will also affect C(and H2). If G, and C(and G2 and H2) are substitutes, an increase inr will lead to an increase in C, and a decrease in H2. If the relationship is complementary (which seems less plausible to me), the reverse would be It should be emphasized that(given imperfect capital markets ) differ- ences across individuals in marginal rates of interest can be the result of differences in underlying preference functions(indifference curves)or differences in opportunities to borrow and lend. In general, it will not be possible to distinguish between these sources empirically, although con trolling for family income(as a proxy for"opportunities" )may move the analysis somewhat closer to a focus on preference functions per se Because time preference is probably only one of many factors affecti the demand for cigarettes, jogging, or other health- related behaviors, we can hardly expect perfect correlation among these activities. Differences in time preference across individuals, however, should result in some positive correlations among these behaviors
96 Victor R. Fuchs health. The empirical portion of this paper, based on a single crosssection survey, cannot distinguish between these two hypotheses, but we can test for possible relations between schooling and time preference. Empirical investigation of time preference through survey questions designed to elicit marginal rates of time discount depends critically on capital markets being “imperfect.” If capital markets were perfect (i.e., if individuals could borrow and lend without limit at a single market rate of interest) marginal rates would be equal for all regardless of time preference. Differences across individuals in time preference might still result in differences in nontradeable health-related activities, but these would not be predictable from the replies to interest rate questions. However, if capital markets are not perfect (an assumption of this paper), individuals may well have different rates of interest at the margin, and these may be related to health behavior and health status. Let us imagine a two-period world. Suppose utility in each period depends upon consumption of goods (G). Utility in the first period also is a function of some activity C1 (for simplicity assumed to be free with respect to G) which affects health (and therefore utility) in period two. For example C1 might be cigarette smoking: u1= U1(GI,CJ U2 = U2(G2,H2) where H2 = H(Cl). A wealth compensated increase in the rate of interest (r) will, ceteris paribus, alter the allocation of wealth between GI and G2. But if the marginal utility of C1 depends on the quantity of G1 (and the marginal utility of H2 depends on the quantity of G2), the change in r will also affect C1 (and H2). If GI and C1 (and G2 and H2) are substitutes, an increase in r will lead to an increase in C1 and a decrease in H2. If the relationship is complementary (which seems less plausible to me), the reverse would be true. It should be emphasized that (given imperfect capital markets) differences across individuals in marginal rates of interest can be the result of differences in underlying preference functions (indifference curves) or differences in opportunities to borrow and lend.5 In general, it will not be possible to distinguish between these sources empirically, although controlling for family income (as a proxy for “opportunities”) may move the analysis somewhat closer to a focus on preference functions per se. Because time preference is probably only one of many factors affecting the demand for cigarettes, jogging, or other health-related behaviors, we can hardly expect perfect correlation among these activities. Differences in time preference across individuals, however, should result in some positive correlations among these behaviors
97 Time Preference and Health Measurement of Time Preference In recent years there have been several attempts to measure time preference through household survey techniques. The objectives of the investigators have varied greatly, but the general approach has been similar: the respondent is confronted with a hypothetical situation involv ing different sums of money at different points in time and is asked to express a preference which will implicitly reveal a rate of time discount Thomas and Ward (1979) Psychologists Ewart A. C. Thomas and Wanda E. Ward were in terested in relations between time preference and various psychological measures of temporal orientation and measures of optimism or pessi mism. They were also interested in possible effects of time preference on saving and spending behavior. Their sample consisted of 63 college tudents who were asked 24 open-ended time preferer questions of the If offered $100 now or X dollars in six months, what would be the smallest amount of money(x dollars ) you would accept rather than the Some questions gave the future amount and asked the respondent to choose a current value; others gave both amounts and asked for the time period that would make them commensurate. Still others were formi lated as payments rather than as receipts, and some were expressed in terms of goods rather than dollar amounts Implicit discount rates were found to be negatively correlated with future time orientation and positively correlated with"big spending The group results were considered satisfactory, but the measurement of time preference was"disappointing to the authors because of the"high instability of parameter estimates for individual subjects West(SRI)(1978 Economists involved in the Seattle-Denver income maintenance ex periment were interested in time preference because the bias introduced by the finite length of the experiment(compared to a national program of indefinite life)would vary depending upon the households rate of time discount( Metcalf 1974). The families in the experiment(more than 1, 500 in each city) were asked a large number and variety of time preference questions. Some were open-ended, similar to those of Thomas and Ward Some were"cascades"of the following type Suppose you had a choice between a cash bonus of $100 today and $200 a year from now; which would you choose?
97 Time Preference and Health Measurement of Time Preference In recent years there have been several attempts to measure time preference through household survey techniques. The objectives of the investigators have varied greatly, but the general approach has been similar: the respondent is confronted with a hypothetical situation involving different sums of money at different points in time and is asked to express a preference which will implicitly reveal a rate of time discount. Thomas and Ward (1979) Psychologists Ewart A. C. Thomas and Wanda E. Ward were interested in relations between time preference and various psychological measures of temporal orientation6 and measures of optimism or pessimism. They were also interested in possible effects of time preference on saving and spending behavior. Their sample consisted of 63 college students who were asked 24 open-ended time preference questions of the following type: If offered $100 now or X dollars in six months, what would be the smallest amount of money (X dollars) you would accept rather than the immediately available $loo? Some questions gave the future amount and asked the respondent to choose a current value; others gave both amounts and asked for the time period that would make them commensurate. Still others were formulated as payments rather than as receipts, and some were expressed in terms of goods rather than dollar amounts. Implicit diskount rates were found to be negatively correlated with future time orientation and positively correlated with “big spending.” The group results were considered satisfactory, but the measurement of time preference was “disappointing” to the authors because of the “high instability of parameter estimates for individual subjects.” West (SRI) (1978) Economists involved in the Seattle-Denver income maintenance experiment were interested in time preference because the bias introduced by the finite length of the experiment (compared to a national program of indefinite life) would var‘y depending upon the household’s rate of time discount (Metcalf 1974). The families in the experiment (more than 1,500 in each city) were asked a large number and variety of time preference questions. Some were open-ended, similar to those of Thomas and Ward. Some were “cascades” of the following type: Suppose you had a choice between a cash bonus of $100 today and $200 a year from now; which would you choose?
Victor R. Fuchs If the respondent chooses $200, the question is repeated, with $175 substituted for $200, and so on until the respondent chooses $100. Some cascade questions go up instead of down; some involve payments rather than receipts; and some involve different time periods The mean interest rates implicit in the replies of these low income respondents were typically quite high, but the correlation between ques tions was typically low(r= about l or 2). The author, richard w. West, expressed some concern that"the measures are not reliable""(p Maital and Maital (1978) A paper by an economist and a psychologist, Shlomo Maital and Sharona Maital, reviews some of the economic and psychological litera ture on time preference and reports the results of a survey of 515 israeli adults. The Maitals'focus is on the role of time preference in the ir tergenerational transmission of income inequality. They asked one cade question involving choice between a sum of money now and hi sums one year from now. A similar question in which gift certificates for a week's shopping at a supermarket were substituted for money was asked in an attempt to measure the real as opposed to the nominal implicit rate The implicit interest rate was negatively correlated with years of schooling(r=-08)and with a dummy variable which took a value of 1 if the subject and the subjects father were born in Israel (r=-. 12). The nominal rate was negatively correlated with income(=- 14), but the real rate was not. The authors concluded that the ability to defer gratifica tion is part of the process of socialization and that after adolescence the tification is quite stable"(. 192). This may be correct, but it is not clear that the conclusion follows from their results Thaler(1979) In a questionnaire administered to approximately 75 college student Richard Thaler posed a large number of open-ended money choices primarily to learn how the implicit interest rate varies with the amount of money involved, the time period, the starting point of the comparison and whether the choice involves receipt or payment. He found that the implicit rate was lower the larger the amount of money and the longer the time period. Also, choices involving two points both in the future typi cally invoked a smaller implicit interest rate than choices involving the present versus the future. He concluded that there is a "psychic fixed cost"to waiting, as well as a cost that varies with amount and time I included a few questions on health status in the Thaler questionnaire and found a significant negative correlation between health and median
98 Victor R. Fuchs If the respondent chooses $200, the question is repeated, with $175 substituted for $200, and so on until the respondent chooses $100. Some cascade questions go up instead of down; some involve payments rather than receipts; and some involve different time periods. The mean interest rates implicit in the replies of these low income respondents were typically quite high, but the correlation between questions was typically low (r= about .1 or .2). The author, Richard W. West, expressed some concern that “the measures are not reliable” (p. 23). Maital and Maital (1978) A paper by an economist and a psychologist, Shlomo Maital and Sharona Maital, reviews some of the economic and psychological literature on time preference and reports the results of a survey of 515 Israeli adults. The Maitals’ focus is on the role of time preference in the intergenerational transmission of income inequality. They asked one cascade question involving choice between a sum of money now and higher sums one year from now. A similar question in which gift certificates for a week’s shopping at a supermarket were substituted for money was asked in an attempt to measure the real as opposed to the nominal implicit rate of interest. The implicit interest rate was negatively correlated with years of schooling (r = - .08) and with a dummy variable which took a value of 1 if the subject and the subject’s father were born in Israel (r = - .12). The nominal rate was negatively correlated with income (r = - .14), but the real rate was not. The authors concluded that the ability to defer gratification is part of the process of socialization and that “after adolescence the propensity to delay gratification is quite stable” (p. 192). This may be correct, but it is not clear that the conclusion follows from their results. Thaler (1979) In a questionnaire administered’ to approximately 75 college students, Richard Thaler posed a large number of open-ended money choices primarily to learn how the implicit interest rate varies with the amount of money involved, the time period, the starting point of the comparison, and whether the choice involves receipt or payment. He found that the implicit rate was lower the larger the amount of money and the longer the time period. Also, choices involving two points both in the future typically invoked a smaller implicit interest rate than choices involving the present versus the future. He concluded that there is a “psychicjked cost” to waiting, as well as a cost that varies with amount and time. I included a few questions on health status in the Thaler questionnaire and found a significant negative correlation between health and median
99 Time Preference and health implicit interest rate across individuals. This result led me to undertake the larger pilot survey described in the next section The Pilot Survey In November 1979, Stephen and ann Cole conducted a survey measur- ing time preference, health status, and health behavior as well as a large number of family background and current socioeconomic variables Telephone interviews approximately twenty minutes in length were con ducted with 508 individuals living in Nassau and Suffolk Counties(on chong Island just east of New York City). Respondents were selected nrough a random sample of telephone numbers; interviews were com pleted with 58% of the eligible respondents. The characteristics of the respondents conformed closely to census data for those two counties, but the possibility of selection bias remains, especially with respect to some of the family background variables The sample was restricted to individuals aged 25-64, and interviewers were instructed to obtain an approximately equal distribution between female and male respondents. The respondents differ from a national sample w espect to religion(55% Catholic and 17% Jewish), race(3% black), and schooling(about one year above the national average). They are also somewhat more affluent and in slightly better health. Allowing for the predominantly suburban middle-class character of the two coun ties, the distributions of replies on the health, health behavior, family background, and socioeconomic variables conform closely to those obtained in national surveys The principal approach to the measurement of time preference was hrough a series of six questions asking the respondent to choose between a sum of money now and a larger sum at a specific point in the future, o e.g,"Would you choose $1, 500 now or $4,000 in five years?"The amount and the time period varied, as did the interest rate implicit in each question. The lowest implicit rate was 10. 1% per annum(continuousl compounded); the highest was 51. 1%. This dichotomous choice type of ecause it was deemed simpler for the respondent than the open-ended or cascade type questions discussed previously. In addition to the implicit interest rate series of pe question with an explicit interest rate(beginning at 6% and rising to 50%)was asked. The survey also included four attitudinal questions e.g., "Do you agree or disagree with this statement: It makes more sense to spend your money now rather than save it for the future "Also, each respondent was asked to choose an expected rate of change of prices for the coming year. The final time preference questions dealt with respondent,'s use of credit during car purchases or through unpaid credit cards
99 Time Preference and Health implicit interest rate across individuals. This result led me to undertake the larger pilot survey described in the next section. The Pilot Survey In November 1979, Stephen and Ann Cole conducted a survey measuring time preference, health status, and health behavior as well as a large number of family background and current socioeconomic variables.s Telephone interviews approximately twenty minutes in length were conducted with 508 individuals living in Nassau and Suffolk Counties (on Long Island just east of New York City). Respondents were selected through a random sample of telephone number^;^ interviews were completed with 58% of the eligible respondents. The characteristics of the respondents conformed closely to census data for those two counties, but the possibility of selection bias remains, especially with respect to some of the family background variables. The sample was restricted to individuals aged 25-64, and interviewers were instructed to obtain an approximately equal distribution between female and male respondents. The respondents differ from a national sample with respect to religion (55% Catholic and 17% Jewish), race (3% black), and schooling (about one year above the national average). They are also somewhat more affluent and in slightly better health. Allowing for the predominantly suburban middle-class character of the two counties, the distributions of replies on the health, health behavior, family background, and socioeconomic variables conform closely to those obtained in national surveys. The principal approach to the measurement of time preference was through a series of six questions asking the respondent to choose between a sum of money now and a larger sum at a specific point in the future,” e.g., “Would you choose $1,500 now or $4,000 in five years?” The amount and the time period varied, as did the interest rate implicit in each question. The lowest implicit rate was 10.1% per annum (continuously compounded); the highest was 51.1%. This dichotomous choice type of question was used because it was deemed simpler for the respondent than the open-ended or cascade type questions discussed previously.” In addition to the implicit interest rate series of questions, a cascade type question with an explicit interest rate (beginning at 6% and rising to 50%) was asked. The survey also included four attitudinal questions, e.g., “DO you agree or disagree with this statement: It makes more sense to spend your money now rather than save it for the future.” Also, each respondent was asked to choose an expected rate of change of prices for the coming year. The final time preference questions dealt with the respondent’s use of credit during car purchases or through unpaid balances on bank credit cards
Victor R. Fuchs One of the purposes of the pilot survey was to determine whet respondents would, in a brief telephone interview, give sensible answ to hypothetical money choice questions when the interest rates implicit the questions are far from transparent. The data presented in Table 3.1 suggest that many respondents do give sensible replies; some do not. The six implicit interest rate questions ask the respondent to choose between taking a smaller prize now or waiting for a larger prize. a priori we expect he fraction of respondents taking the prize now to diminish as the implicit interest rate rises. Table 3. 1 shows that this did occur. For the sample as a whole, 76% chose now for the question with an implicit interest rate of 10.1%o per annum; only 33% did so when the implicit interest rate was 51.1% Not only do the group results conform to a priori expectations, but almost two-thirds of the respondents gave replies which were internally consistent for each individual. A set of replies was defined as consistent if e respondent never answered"now"to a question with an implicit interest rate that was higher than the rate in another question to which the answer was" wait. 712 The last three columns of Table 3. 1 show results for the sample divided into three groups: those with consistent answers those whose answers would be consistent if one reply were reversed (about one-fourth of the sample), and those respondents whose replies require two or three reversals in order to achieve consistency(about 10% of the sample). 3 The relation between the fraction taking the prize now and the implicit interest rate is much weaker for those respondents with inconsistent answers and much stronger for those with consistent answers. Most of the results reported here are based on analyses limited to those respondents with consistent replies Table 3. 2 presents the results of regressions in which each questio each individual is treated as an observation. when the regressions are run OLS, the dependent variable is dichotomous, taking a value of 1 if the Table 3.1 Mean Probability of Taking Prize Now Number of inconsistent answers respondents 2 or 3 number ( per annum (N=329)(N=124)(N=51) 028033
100 Victor R. Fuchs Empirical Results One of the purposes of the pilot survey was to determine whether respondents would, in a brief telephone interview, give sensible answers to hypothetical money choice questions when the interest rates implicit in the questions are far from transparent. The data presented in Table 3.1 suggest that many respondents do give sensible replies; some do not. The six implicit interest rate questions ask the respondent to choose between taking a smaller prize now or waiting for a larger prize. A priori we expect the fraction of respondents taking the prize now to diminish as the implicit interest rate rises. Table 3.1 shows that this did occur. For the sample as a whole, 76% chose “now” for the question with an implicit interest rate of 10.1% per annum; only 33% did so when the implicit interest rate was 51.1%. Not only do the group results conform to a priori expectations, but almost two-thirds of the respondents gave replies which were internally consistent for each individual. A set of replies was defined as consistent if the respondent never answered “now” to a question with an implicit interest rate that was higher than the rate in another question to which the answer was ‘‘wait.’r12 The last three columns of Table 3.1 show results for the sample divided into three groups: those with consistent answers, those whose answers would be consistent if one reply were reversed (about one-fourth of the sample), and those respondents whose replies require two or three reversals in order to achieve consistency (about 10% of the sample).I3 The relation between the fraction taking the prize now and the implicit interest rate is much weaker for those respondents with inconsistent answers and much stronger for those with consistent answers. Most of the results reported here are based on analyses limited to those respondents with consistent replies. Table 3.2 presents the results of regressions in which each question to each individual is treated as an observation. When the regressions are run OLS, the dependent variable is dichotomous, taking a value of 1 if the Table 3.1 Mean Probability of Taking Prize Now Implicit compound All Number of inconsistent answers respondents 1 2 or 3 Question interest rate number (% per annum) (N=504) (N=329) (N= 124) (N=51) 30 10.1 .76 .78 .75 .61 32 15.7 .61 .66 .56 .34 28 19.6 .58 .59 .60 .41 29 30.5 .52 .52 .48 .61 33 40.2 .34 .35 .28 .41 31 51.1 .33 .25 .37 .71
101 Time Preference and Health Table 3.2 Regressions of Probability of Taking Prize Now on Interest Rate Variables Number of inconsistent answers respondent 2 or 3 Intercep implicit interest Question simple .0032* implicit interest (.0006) (.0011) -0020]【-.0007](-0037][-0053] Respondent explicit 0054* Interest rate (.0008) (.0009) (.0019 (.0025) (%o per annum Regressions based on person-question observations. The OLs regression co shown first with their standard errors in parentheses below. The marginal effects probability)from the logistic regressions are in brackets <.05 reply is"now"and 0 if it is"wait. The right side variables are the compound interest rate implicit in each question, the simple implici interest rate, and the individuals explicit interest rate given in reply to the cascade question mentioned in the previous section. We see that the probability that a given individual will reply"now"to a given question falls sharply as the interest rate implicit in the question rises, and rises rapidly as the individual,'s explicit interest rate rises. These results hold for the entire sample and are particularly strong for those respondents classified as consistent, but do not hold for the other respondents. Logis tic regressions estimated by a maximum likelihood procedure give similar results when evaluated at the mean probability of taking"now. " (See marginal effects in brackets. The contrast between the compound interest rate and the simple interest rate coefficients, depending upon the consistency class, suggests one possible reason why some respondents give inconsistent replies. 14 The two interest rates are, of course, highly correlated, but not perfectly o. Those giving consistent replies seem to have been influenced by the implicit compound rate, while those with the most inconsistent replies seem to have been influenced primarily by the simple rate. We also see that there is a close connection between the explicit rate and the probabil
101 Time Preference and Health Table 3.2 Regressions of Probability of Taking Prize Now on Interest Rate Variables Number of inconsistent answers All respondents 0 1 2 or 3 N R2 Intercept Question compound implicit interest rate (% per annum) Question simple implicit interest rate (% per annum) Respondent explicit interest rate (% per annum) 2952 ,106 ,733 ( ,022) (.0012) - .0073** [ - .0071] - .0017** (.OO06) [ - .0020] (.000S) .0054** [ .OO64] 1956 ,158 ,783 ( ,026) - .0111** (.0014) [ - .0126] - ,0008 (.0007) [ - .0007] .0068** (.0009) [ .OO90] 719 277 ,082 ,026 .733 ,414 ( ,046) (.074) - .0037 .0106* ( ,0024) (.0040) [ - ,00341 [.0135] (.0011) (.OO19) [ - ,00531 .0020 ,0010 (.OO19) (.0025) - .0032** - .0042* [ - ,00371 [ ,00191 [.0010] Notes: Regressions based on person-question observations. The OLS regression coefficients are shown first with their standard errors in parentheses below. The marginal effects (at mean probability) from the logistic regressions are in brackets. *p < .05 **p < .01 reply is “now” and 0 if it is “wait.” The right side variables are the compound interest rate implicit in each question, the simple implicit interest rate, and the individual’s explicit interest rate given in reply to the cascade question mentioned in the previous section. We see that the probability that a given individual will reply “now” to a given question falls sharply as the interest rate implicit in the question rises, and rises rapidly as the individual’s explicit interest rate rises. These results hold for the entire sample and are particularly strong for those respondents classified as consistent, but do not hold for the other respondents. Logistic regressions estimated by a maximum likelihood procedure give similar results when evaluated at the mean probability of taking “now.” (See marginal effects in brackets.) The contrast between the compound interest rate and the simple interest rate coefficients, depending upon the consistency class, suggests one possible reason why some respondents give inconsistent re~1ies.l~ The two interest rates are, of course, highly correlated, but not perfectly so. Those giving consistent replies seem to have been influenced by the implicit compound rate, while those with the most inconsistent replies seem to have been influenced primarily by the simple rate. We also see that there is a close connection between the explicit rate and the probabil-
