Sample
The Japanese Study on Stratification, Health, Income and Neighbourhood (J-SHINE) is a longitudinal panel study of young and middle-aged (aged 25–50 years old) community residents in the greater Tokyo metropolitan area of Japan [31]. The wave 1 survey was carried out in 2010. The survey participants were re-contacted in 2012 and 2017 (waves 2 and 3). A computer-based self-report questionnaire (wave 1 and 2) and a self-administered questionnaire (wave 3) were used to collect information covering socio-demographics, behavioural tendencies, personal values and developmental history. A total of 4385 respondents participated in the wave 1 survey (response rate: 31.3%). A total of 2787 individuals (63.6% of the wave 1 respondents) responded to the wave 3 survey. In this study, we used data from respondents who completed both wave 1 and wave 3 surveys, because the wave 1 survey collected information of time preference, and the wave 3 survey asked questions of personal values.
Measures
Time preference
Time discounting and hyperbolic time discounting were measured with the wave 1 questionnaire, by using a three-item choice-based measure as described in an earlier study using the same dataset [18], adopting a scale developed by previous studies [13, 15], with applying a common methodology for large-scale studies [32, 33] (see Additional file 1: Appendix 1). Participants were asked to complete three questions in which they made a series of binary choices between receiving a smaller amount of money sooner or a larger amount of money. Q1 asked participants to choose between receiving 10,000 yen in 1 month (option A) or receiving more (an amount which increased from 9500 to more than 14,000 yen via nine steps) in 13 months (option B). Q2 asked respondents to choose between receiving 1 million yen in 1 month (option A) or receiving more (an amount which increased from 0.95 million to more than 1.4 million yen via nice points) in 13 months (option B). Q3 asked participants to choose between receiving 1 million yen in 13 months (option A) or receiving more (increasing from 0.95 million to more than 1.4 million yen via nice points) in 25 months (option B). If a respondent switched from option A to option B, he or she was given an interest rate (equivalent to the amount of money for option B divided by the amount of money for option A minus 1) as the time discount rate. The maximum likelihood estimation of the cardinal proxy values of the discount rate was applied, derived from the categorical responses in the questionnaire [18]. The proxy time discount rate (discrate) was standardised and averaged across the three questions. Respondents who switched their choices more than once, switched their choice in the category when given an interest rate of 5% or 0%, and/or who chose A for all categories (never switching to B) were considered to be irrational and their data were excluded from the analyses.
Hyperbolic time discounting was measured, comparing time discount rates for Q2 and Q3, following a methodology used in previous study [13]. By definition, if a respondent had a lower rate for Q3 (in a longer future) than for Q2 (in a near future), he or she could be classified as having a hyperbolic tendency. However, the decline in time discount rate in the future is typically not prominent beyond a six-month time frame [12]. Those who had a discount rate for Q3 lower than that for Q2 were classified as having a hyperbolic tendency in this study; otherwise participants were classified as having a non-hyperbolic tendency [13]. The hyperbolic tendency was set as missing if time discount rate either for Q2 or Q3 was missing. In a previous study, time discount score was significantly associated with debt holdings and obesity [13], indicating a piece of evidence for validity of the measure. The score of Q2 (i.e., an interest rate at which a respondent switch from the option A to B) was used as a control variable for the analyses on hyperbolic time discounting.
For a sensitivity analysis of the respondents who were excluded from the analyses for time discounting, five groups were prepared on the basis of irrational and missing responses to the scale and compared with the sample used for the main analysis on time preference; (a) an irrational group 1 (IR1) who selected reward in distant future that was lower (− 5% gain) than one in near future; (b) an irrational group 2 (IR2) who selected reward in distant future without no (0%) gain compared to one in near future; (c) an irrational group 3 (IR3) who kept selecting reward in near future despite of maximum gain to be obtained in distant future; (d) a switcher group (SW) who switched near and distant future rewards multiple times; (e) a missing group who had a missing response on any of three questions on time preference.
Personal values at present and at age 15
Personal values (value priorities and degrees of commitment to the values) at present were measured with the wave 3 questionnaire. To measure value priorities, we used a list of 11 values [6, 7, 34]: not bothering others, being evaluated by others, having and keeping a belief, succeeding economically, improving society, exploring what you are interested in, having influence on society, pursuing active challenges, cherishing family and friends, graduating from a famous school and maintaining a stable life (see Additional file 1: Appendix 2). Respondents were asked to report their value priorities at the time of the survey (‘How important do you think the following values are in your present life?’), rating each on a seven-point Likert scale (1 = Not at all important, 7 = Very important).
Commitment to value was measured via the Personal Values Questionnaire II (PVQ-II) [35]. The original version consists of nine items, but one item was dropped for the Japanese PVQ-II [36, 37]. Each item was rated on a five-point Likert scale to calculate the total score, with a higher score indicating greater commitment to a value. The internal consistency and concurrent and structural validity have been tested for the Japanese version [36, 37].
Personal values (both areas of priority vales and commitment to value) at age 15 were also measured based on a retrospective recall in the wave 3 survey. We set a time point for the recall of personal value at age of 15, when both time preference [38] and personal values are considered stable in adolescence [39]. Respondents were asked to recall their value priorities at age 15 (‘When you were 15–16 years old, how important did you think the following values were in your life?’) using the same list of 11 value areas, rating each on a seven-point Likert scale (1 = Not at all important, 7 = Very important). We modified the PVQ-II questions to the past tense and asked participants to rate their commitment to a value that was most important for them in their age of 15 [34]. Cronbach’s alpha coefficients for the PVQ-II were 0.711 and 0.741 for age 15 and adulthood, respectively.
Several unpublished studies of Japanese adults reported evidence for reliability and validity of the measures of areas of priority values. In a test-retest reliability study of areas of priority values at age 15, the scores reported twice with a two-week interval correlated moderately (ranging from 0.533 to 0.705 in intraclass correlations, ICCs), with one exception (0.372 in ICC for cherishing familiar people), while test-retest reliability statistic was not available for areas of priority values at present. For validity, the agreement (ICCs) between corresponding areas of priority values at age 15 and at present were greater for those who reported that their personal values were same since age 15 than for those who reported that these were different. The results indicate that areas of priority values at age 15 and at present can be measured separately by using the present questionnaire. On the other hand, for the PVQ-II, a pilot study of 500 adults in Japan reported acceptable levels of internal consistency and factor-based validity for the age-15 version and for the present version [34]. Cronbach’s alpha coefficients for the age-15 version and the present version of the PVQ-II were 0.680 and 0.732, respectively, in the present sample. The agreement (ICCs) between PVQ-II scores at age 15 and at present were greater for those who reported that their personal values were same since age 15 (0.596) than for those who reported that these were different (0.385), again indicating validity of asking these questions separately for age 15 and present.
Covariates
Sex, age, educational attainment, job status, and marital status were measured in the questionnaire as part of the wave 1 survey. We did not use household income as a covariate, because approximately 20% of respondents had missing values [18]. As this factor may be associated with personal values at age 15, two indicators of childhood socioeconomic status were measured in the questionnaire during the wave 1 survey. Parental education was defined as the highest educational attainment among respondents’ parents; this was dichotomised into university graduate or higher and less than university education. Economic hardship at age 15 was assessed by a question asking about the household economic situation when a respondent was 15 years old (with a 5-point scale from difficult to affluent), and dichotomised into yes (experiencing hardship) or no.
Statistical analysis
Average scores of commitment to value (PVQ-II) and priority values as well as of time discount rate and hyperbolic time discount rate, were calculated. A linear regression analysis was conducted for the commitment to value score and a score for each priority value on time discount rate (crude), additionally adjusting for sex and age groups (sex- and age-adjusted), and further adjusting for educational attainment, job status, marital status, parental education, and economic hardship at age 15 (fully adjusted). A similar linear regression analysis was conducted of the commitment to value score and a score for each priority value on the dichotomous score on hyperbolic time discount rate (crude), additionally adjusting for the response to Q2 as a proxy of time discount rate, sex, and age groups (sex, age-, and base rate-adjusted), and further adjusting for educational attainment, job status, marital status, parental education, and economic hardship at age 15 (fully adjusted).
For the sensitivity analyses on the irrational and missing groups, scores of priority values and commitment to value were compared among the six groups (i.e., the sample for the main analysis, IR1–3, SW, and missing groups) for crude (one-way analysis of variance), sex- and age-adjusted and fully-adjusted models (analysis of variance).
The p-value for significance was set at < 0.05. However, considering multiple significant testing for the analysis for 11 priority values, we reset p for significance at < 0.0045 (=0.05/11), applying Bonferroni’s correction method. All statistical analyses were conducted by suing IBM SPSS version 26.