44 healthy participants ages 18–30 were recruited from New York, USA and 43 healthy participants from Guangzhou, China. Participants rated their English and Mandarin proficiency (1 = not at all fluent, 5 = completely fluent43), and were retroactively excluded if they were fluent (score of 4–5) in the opposing language or if they were not fluent in the language of their own country (score of 1–2). Three English speakers were excluded for high Mandarin proficiency and one English speaker was excluded for an inability to follow task instructions. Three Mandarin speakers spoke a local dialect fluently and were excluded for low Mandarin proficiency. Forty English speakers (M = 23.5 years, SD = 2.75, n = 22 female) and 40 Mandarin speakers remained (M = 23.0 years, SD = 3.05, n = 20 female). No difference in average age was observed across the cultures (Wilcoxon sum rank test; W = 919.5, two-tailed, P = 0.25, 95% CI for the difference in location [− 1.00, 2.00]).
All participants provided written informed consent and were financially compensated for their participation. English speakers were paid $15 in cash for completing the tasks on the day of the experiment and were paid their choice after a specified delay on a randomly selected trial from the delay discounting task. Mandarin speakers were paid their choice on a randomly selected trial from the discounting task after the delay specified, and were reimbursed for transportation to the lab. The Institutional Review Board at the Icahn School of Medicine at Mount Sinai and the ethical committee of South China Normal University approved this experiment. Compliance with all relevant ethical regulations was ensured throughout the study and methods were carried out in accordance with relevant guidelines and regulations.
Procedures and materials
Participants completed two tasks, a time representation task and a delay discounting task, and five surveys. Surveys were completed in the following order for all participants: event evaluation, demographic, self-continuity, task rationale, and state-trait anxiety inventory (STAI). Experiment order for the time representation and delay discounting tasks was counterbalanced in each study sample. All materials were developed in English and translated to Mandarin by two bilingual researchers at South China Normal University (for translated task instructions, see Supplementary Table S4). English speakers completed the tasks on a Windows Lenovo ThinkPad (15.5in), and Mandarin speakers on either an Acer Aspire e15 (15.6in), Acer V193W (19in), or Acer G195WVAb (19in).
Time representation task
Participants viewed a 700pt. × 700pt. blank white canvas on a computer screen. At the top of the canvas an event was displayed in a 50pt. box. At the center of the canvas was a 60pt. × 120pt. blue avatar. The blue avatar was chosen to avoid visually apparent racial indicators. Overlaid on the stomach of the avatar was a yellow circle with a 10.5pt .radius. Instructions shown in Supplementary Table S4 remained on the screen at all times. Participants were verbally instructed to assume that all events have happened or could happen, meaning they were instructed to assume that all events were realistic. Participants picked up the yellow dot with the mouse and placed it anywhere within the 700pt. × 700pt. canvas using a drag-and-drop approach. Grid coordinates were not apparent to the participant, but were recorded in the background for the final placement of the event.
Participants had unlimited time to answer and each time a new event appeared the yellow dot reappeared on the stomach (center) of the avatar. Participants were not allowed to go back and change the location of previous placements. Events appeared serially and randomly on the screen. The canvas refreshed after each placement to encourage participants to place events based on the represented distance of the event from themselves (the avatar), rather than the relative distance of events to each other. Participants rated 6 events in a practice round before completing the full 72-event task. This training served to initialize all participants to the space using the same set of events. It also sought to familiarize participants with the timespan over which events would be surveyed, such that they would have a mental representation of time on the canvas when beginning the task.
Events were selected using a pre-task survey in an independent sample of 30 English speakers in the USA (M = 21.0 years, SD = 2.75, n = 17 female) and 30 Mandarin speakers in China ages 18–30 (M = 20.6 years, SD = 2.58, n = 15 female). No difference in average age was observed across the cultures (Wilcoxon sum rank test, W = 485.5, two-tailed, P = 0.60, 95% CI for the difference in location [− 1.00, 2.00]). Individuals were sent an online survey with 184 internationally known events and were asked to rate the events on arousal [0 = not exciting-7 = very exciting], familiarity [0 = not familiar-7 = very familiar], valence [0 = negative-7 = positive], using a sliding scale and date (when did/will this happen?) [further past, closer past, closer future, further future, and never] via a multiple-choice response. Within each culture, mean arousal, valence, and familiarity ratings were calculated for each event by averaging scores across the 30 participants. Results were then filtered to remove events that participants were not familiar with (avg. familiarity score of 1.5 and below) or events that were considered too unrealistic (10 or more participants selected “never” for the date rating). This filtering occurred in each culture individually and results were merged to examine which events remained in both cultures.
Thirty-six past and 36 future events were selected and non-parametric two-tailed Wilcoxon sum rank tests verified that there were no differences in arousal, familiarity, or valence for past events (Supplementary Fig. S6, arousal: W = 697.5, P = 0.58, familiarity: W = 802.5, P = 0.08, valence: W = 636.5, P = 0.90) or for future events between the cultures (Supplementary Fig. S6, arousal: W = 713.5, P = 0.46, familiarity: W = 487, P = 0.07, valence: W = 507, P = 0.11). See Supplementary Table S5 for a complete list of training events and their Mandarin translations and Supplementary Table S6 for a complete list of task events and their Mandarin translations. In the current sample, English speakers were significantly more aroused by and familiar with the selected events (Supplementary Fig. S6, Wilcoxon sum rank test, two-tailed, arousal: W = 1,095.5, P = 0.005, familiarity: W = 1,154.5, P < 0.001). No difference was observed in valence ratings (Wilcoxon sum rank test, W = 802.5, two-tailed, P = 0.98). English speakers were significantly more educated than Mandarin speakers (Pearson’s chi-squared test, χ2(2) = 26.23, P < 0.001), however, education level was not related to the familiarity for the events in either culture (English: Spearman’s rho = 0.08, S = 9,820.3, two-tailed, P = 0.63, Mandarin: Spearman’s rho = 0.27, S = 7,764, two-tailed, P = 0.09).
Delay discounting task
Participants viewed 51 choices and selected whether they would like a smaller amount of money now or a larger amount of money after a variable delay (task used in33). The English task displayed amounts in USD ($) and the Mandarin version in CNY (¥). For English speakers, the now options ranged from $10 to $34 and the delay amounts were either $25, $30, or $35. For Mandarin speakers, the now options ranged from ¥33.50 to ¥113.90 and the delay amounts were either ¥83.75, ¥100.50, or ¥117.25. To convert the currencies, we examined the purchasing power in each country by comparing the amount of time an individual would have to work in New York and Beijing to earn a Big Mac. According to the UBS in 2018, an individual would have to work 51.0 m in Beijing and 15.2 m in New York, leading to a conversion of 1 USD = 3.35 CNY (https://www.ubs.com/minisites/prices-earnings/en/cities/beijing/). The delays varied from 1 to 180 days and each participant was given an unlimited amount of time to respond. After making a response, a white check mark appeared on the screen for 500 ms to indicate which answer was just chosen, followed by a 1000 ms inter-trial interval, and the subsequent question. Now and later options appeared randomly on either the right or left side of the screen.
To encourage realistic decisions, participants were informed that they would be paid their response on a randomly selected trial before completing the task. To explain this concept, English speakers were verbally given the example (Mandarin speakers the translated equivalent) that if they selected $25 in 30 days as one of their answers, this would be placed into a lottery with all of their other responses. One response would be randomly selected and if this was chosen, they would be mailed a check for $25 after 30 days. After this example the experimenters again emphasized that the payment and delay would be based on a randomly selected trial from their responses. The task was programmed using E-Prime 2.0 (https://pstnet.com).
Participants completed an event evaluation survey where they evaluated the 72 time representation events on arousal, familiarity, valence, and date. The scales mirrored those of the pre-task survey, however, since participants were told to assume all events were realistic, the date question no longer included a never option. The event evaluation survey was administered to English speakers via SurveyMonkey (https://www.surveymonkey.com) and Mandarin speakers via Wenjuanwang (https://www.wenjuan.com). Participants completed a demographic survey where they self-reported their age, sex, race, language fluency, education, occupation, income, socioeconomic status, financial security, and religious or spiritual views. On the task rationale survey, participants explained their decisions in the discounting task, their strategy for the time representation task, and sketched their time representation schemas (data not included). On the self-continuity survey, participants reported how similar they felt to their past self, 10 years ago, and future self, 10 years from now (survey used in54, see Supplementary Fig. S7 for results). Lastly, participants completed the standard STAI, which consisted of 40 questions examining their state and trait stress (see Supplementary Fig. S7 for results). We did not observe any significant correlations between state stress, trait stress, future self-continuity, and past self-continuity and psychological distance or discount rate in either English or Mandarin speakers (Supplementary Table S7). These surveys were administered to English speakers using REDCap (https://www.project-redcap.org) and Mandarin speakers using Wenjuanxing (https://www.wjx.cn).
All analyses were conducted in R v. 3.6.0, except for the discount rate estimation, which was run in MATLAB R2015a (The MathWorks, Inc.). Statistics were computed using non-parametric Wilcoxon sum rank, Wilcoxon signed rank, and Kruskal Wallis tests, parametric Two Sample t-tests, Pearson correlations, and Spearman correlations as indicated in the text. All tests were two-tailed. A Cohen’s d effect size was used when comparing linearity across cultures and was calculated by dividing the mean difference by the pooled standard deviation. A Cohen’s q effect size was calculated for the comparison of correlations across English and Mandarin speakers55. An eta-squared (η2) effect size was calculated for the linearity ANOVA using the eta_sq function from the sjstats package in R56.
We quantified linearity by fitting a linear model, lm(Y coordinate ~ X coordinate) to each participant’s event placements and extracting the residual standard deviation associated with the model. We computed event distances by calculating the length of the vector connecting each event placement to the center of the canvas (stomach of the avatar) and PD by averaging the vector lengths across all event placements for each participant. We then dissected this PD value into the average PD for past-rated events and future-rated events in each participant. In an exploratory analysis, we examined whether the difference in PD between “further” and “closer” past/future events was associated with discount rate, however, we did not find evidence for a relationship (Supplementary Table S8).
Discount rates were estimated using a logistic regression model in MATLAB, where participants’ choices were fit using maximum likelihood estimation34. The subjective values of options were estimated using a hyperbolic discounting model: SV = A/(1 + kD). A represented the amount of the option, SV the subjective value of the option, D the delay, and k represented an individual discount parameter. Analyses were conducted using scripts provided by the Kable Laboratory.
The linearity multiple linear regression model was built using the lm() function in R and served to ensure that the effect remained when controlling for demographic factors. The regressions estimated the Type III Sum of Squares. Regressors and their scales for the linearity model included: mean event arousal, familiarity, and valence ratings (1–7), age (18–30), sex (male, female), education (high school, college, post graduate), and culture (English, Mandarin). Linear mixed models were built to examine the relationship between event distances and discount rates. These models were built using the lme4 and lmerTest packages in R and estimated significance using the Type III Sum of Squares. Fixed effects and their scales included: culturally adjusted income bracket (1–7, decline to respond, I don’t know), socioeconomic status (1–10), financial security (1–5, decline to respond), age (18–30), sex (male, female), education (high school, college, post graduate), culture (English, Mandarin), timeframe (past, future), and log k. Event name (72 total events) and participant (74 total participants) were included as random effects. In the models above, event ratings, age, SES, psychological distance, and log k were coded as continuous variables. Sex, education, culture, income bracket, financial security, and timeframe were coded as factor variables.