Risk taking for potential losses but not gains increases with time of day

Abstract

Humans exhibit distinct risk preferences when facing choices involving potential gains and losses. These preferences are believed to be subject to neuromodulatory influence, particularly from dopamine and serotonin. As neuromodulators manifest circadian rhythms, this suggests decision making under risk might be affected by time of day. Here, in a large subject sample collected using a smartphone application, we found that risky options with potential losses were increasingly chosen over the course of the day. We observed this result in both a within-subjects design (N = 2599) comparing risky options chosen earlier and later in the day in the same individuals, and in a between-subjects design (N = 26,720) showing our effect generalizes across ages and genders. Using computational modelling, we show this diurnal change in risk preference reflects a decrease in sensitivity to increasing losses, but no change was observed in the relative impacts of gains and losses on choice (i.e., loss aversion). Thus, our findings reveal a striking diurnal modulation in human decision making, a pattern with potential importance for real-life decisions that include voting, medical decisions, and financial investments.

Figure 1

Task Design. Gain trials (green outline) had potential gains and no potential losses. In an example gain trial, a participant chose between a risky option (here, 50% probability of 59 points) and a safe option (here, 100% probability of 35 points). Loss trials (red outline) had potential losses and no potential gains. Mixed trials (blue outline) had both potential gains and potential losses with a safe option that was always worth 0 points. Therefore, on any trial participants could opt for a risky choice. The experimental design enables the role of a gain or loss magnitudes in propensity towards risk to be disambiguated. The participants total score, starting at 500 points, was added or taken away from throughout the game as they win and lose points on each risky choice. The green, red, and blue outlines were added to the figure for descriptive purposes, each trial had the same appearence to the participants.

Figure 2

Loss sensitivity decreases with time of day in individuals. (A) We identified individuals (N = 2599) who completed the risky decision task on two different days between 08:00 and 22:00. Risk taking in loss trials increased with time of day within individuals. (B) Loss sensitivity (α_loss), but not gain sensitivity (α_gain) or loss aversion (log(λ)), decreased with time of day in this within-subject sample. Error bars represent bootstrapped 95% confidence intervals [*p < 0.05, **p < 0.01, ***p < 0.005]. (C) Loss sensitivity was lower for later plays in within-subject analysis. Error bars represent bootstrapped 95% confidence intervals. Data binned on differences less (N = 1,643) or more than 5 h (N = 956) are shown for illustration purposes. [*p < 0.05, **p < 0.01, ***p < 0.005].

Figure 3

Risk taking for potential losses increases with time of day. (A) Risk taking in choice trials with potential losses increased with time of day (N = 26,720, first plays only). This was true in loss trials featuring risky options with equal probabilities of zero or a potential loss, and in mixed trials featuring risky options with equal probabilities of potential gains and losses. The frequency of choosing risky options with only potential gains was unaffected by time of day. Each time bin includes data collected in the interval starting at the time indicated (i.e., the 6am bin includes all data collected from 06:00 until 11:59). Error bars represent the SEM. (B) Risk taking in loss trials increased with time of day in both female (N = 13,054) and male participants (N = 13,666). Risk taking in mixed trials increased with time of day in females but not males. Error bars represent bootstrapped 95% confidence intervals [*p < 0.05, **p < 0.01, ***p < 0.005].

Figure 4

Loss sensitivity and not loss aversion decreases with time of day. (A) Loss sensitivity parameters (α_loss) decreased with time of day in both females and males, consistent with increased risk taking in trials with potential losses. Error bars represent the SEM. (B) Increased risk taking for losses could be partially explained by decreased loss aversion. However, loss aversion parameters (log(λ)) did not decrease with time of day in both samples. Error bars represent SEM. (C) Loss sensitivity decreased more than loss aversion in both females and males. Error bars represent bootstrapped 95% confidence intervals [*p < 0.05, **p < 0.01, ***p < 0.005].