Introduction

The ability to delay gratification (i.e., forgoing immediate smaller rewards for later larger rewards [Mischel and Metzner 1962]) has important consequences in everyday life. Choosing to maintain a healthy diet in favor of better long-term health or forgoing current purchases to save for retirement requires the ability to delay gratification. The ability to delay gratification may do more than keep one fit and financially sound; it has also been associated with future academic, social, and interpersonal success (e.g., Mischel et al. 1988).

The ability to delay gratification is associated with two related terms that will also be used in this paper. “Delay of gratification” is conversely related to both “impulsivity” and “delay discounting”. For example, someone who is often able to delay gratification is considered less impulsive than someone who rarely delays gratification. Delay discounting refers to the degree to which an individual devalues rewards expected at a future time (Hirsh et al. 2008). For example, compared to a relatively non-impulsive individual, a highly impulsive person would discount future rewards more (i.e., s/he would consider a future reward as less valuable).

Delay of gratification ability can be observed early in life. For example, Mischel and Underwood (1974) gave children a marshmallow and asked them to wait alone in a room until the researcher returned. The children were informed that if they didn’t eat the marshmallow by the time the researcher returned, they would be given a second marshmallow. Even at these early ages, individual differences in delay of gratification ability were identified—some children waited for the researcher to return prior to eating the marshmallow, whereas others ate it immediately. Interestingly, delay of gratification ability predicts future success. Mischel et al. (1988), for example, tested 4- and 5-year-olds using tasks similar to that just described. Ten years later, follow-up results revealed that those participants who delayed longer as children were rated as more attentive, capable of handling stress, and socially and academically competent than their more impulsive peers.

Although the ability to delay gratification is an obviously important individual difference measure, the cognitive and/or emotional factors underlying such differences are poorly understood. One cognitive factor that may predict impulsivity level is subjective time perception, an individual’s sense of how quickly time passes. Wittmann and Paulus (2008) proposed that individuals, when making delay of gratification decisions, consider the perceived differences in time between the earlier, smaller reward and the larger, later reward. They argued that individuals who overestimate the duration of time intervals assign a higher cost to waiting and will thus prefer smaller, immediate rewards to larger, delayed rewards. The subjective experience of time is dependent on one’s “internal clock speed” (ICS). People with fast and slow ICSs perceive more and less time passing, respectively, than is objectively true. For example, during an objective 30-s period, someone with a fast ICS might subjectively report 45 s elapsing whereas someone with a slow ICS might report the passage of only 20 s. Individual differences in ICS may be useful to understanding impulsive behavior. For example, when measuring impulsivity, a frequently used task requires participants to think about the amount of money they would need at a future time (e.g., 2 months from today) in order to forgo a smaller amount sooner (e.g., today). Here, one’s perception of the specified delay may significantly impact how much money would be needed in the future to forgo today’s reward—the longer the perceived wait (i.e., faster ICS), the more money required.

Only three known studies have been performed to investigate the relationship between ICS and impulsivity level among healthy individuals. Van der Broek et al. (1992) found that relatively impulsive people had nonsignficantly slower ICSs when compared to those with better delay of gratification ability (inconsistent with Wittmann and Paulus 2008). Incorporating university students, both Gerbing et al. (1987) as well as Lennings and Burns (1998) failed to demonstrate any significant ICS-impulsivity relationships. Importantly, however, each of these studies used self-report measures of impulsivity, which may not be a strong indicator of actual behavioral (e.g., Carillo-de-la-Pena et al. 1993; Lane et al. 2003). This is a significant limitation which is addressed in the present research.

Researchers have also studied clinically impulsive populations to better understand the relationship between subjective time perception and impulsivity. For example, Cappella et al. (1977) studied hyperactive children. These children, and those diagnosed with Attention Deficient Hyper Activity Disorder (ADHD), are characterized as having a limited behavioral inhibition, poor attention, hyperactivity and limited ability to delay gratification (Barkley et al. 2001). Cappella et al. (1977) found that, when compared to a control group, hyperactive children overestimated the passage of time, indicative of a fast ICS. Meaux and Chelonis (2003), conversely, found that their sample of ADHD children exhibited more absolute errors on both time estimation and time production tasks (both measures of ICS) relative to healthy controls. These data suggest that accuracy in time perception may also be an important factor when predicting impulsivity. In all, 12 known studies have evaluated the relationship between ICS and impulsivity in ADHD-related populations, with mixed results. That said, regardless of whether researchers believe fast ICS (e.g., Barkley et al. 2001), slow ICS (e.g., Sonuga-Barke et al. 1998), or ICS accuracy (e.g., Toplak et al. 2003) is more strongly associated with impulsivity, most agree that ADHD patients have altered time perception, and this impairment may be related to the impulsive behaviors they exhibit.

Borderline Personality Disorder (BPD) is another clinical population associated with altered subjective time perception and behavioral impulsivity, including drug use/abuse, excessive spending/gambling, sexual promiscuity, and self-injurious behavior. Experimentally, Berlin and Rolls (2004) found that BPD patients were more behaviorally impulsive than controls as measured by the Matching Familiar Figures Test. In addition, Berlin and Rolls (2004) and Berlin et al. (2005) found that BPD patients, relative to controls, consistently underproduced time (they pressed a button for less time than instructed) and trended toward overestimating how much time had passed. These data suggest that BPD patients may have a fast ICS which, in turn, may be associated with greater impulsivity.

In the emotional sphere, an individual’s reactivity toward positive stimuli may also predict behavioral impulsivity. Gray (1990) proposed two systems that regulate behavior toward aversive and appetitive cues. The Behavioral Inhibition System (BIS) is responsible for withdrawal behavior and is sensitive to punishment cues and novelty. The Behavioral Activation System (BAS), conversely, is sensitive to rewards and is responsible for appetitive behavior. Individuals who have greater BAS strength may be better able to wait for larger rewards because they are more motivated by and responsive to rewards than individuals with lesser BAS strength.

In their influential article, Wittmann and Paulus (2008) expressed a need for more research to be conducted to better understand how time perception predicts behavioral impulsivity among healthy individuals, and this study responds to that call. Based on past research, it was hypothesized that greater behavioral impulsivity would be predicted by gender (males > females; Silverman 2003), lower intelligence quotient (IQ; e.g., Shamosh and Gray 2008), higher self-reported impulsivity (e.g., de Wit et al. 2007), and by lower BAS strength. It was further hypothesized that time perception would predict behavioral impulsivity above and beyond the aforementioned predictors. Specifically, in accordance with Wittmann and Paulus (2008), those with faster ICSs were expected to behave more impulsively relative to those with slower ICSs. Based on past clinical work, people with less accurate ICSs were also expected to be more impulsive relative to people with more accurate ICSs.

Methods

Participants were recruited from an undergraduate Introduction to Psychology course at a private, Midwestern university. A total of 79 participants completed the study and received partial course credit for their participation. This study was approved by the university’s Institutional Review Board.

Procedures

Participants were informed about the procedures of the experiment and provided written consent if they agreed to participate. They were then asked to complete a number of self-report questionnaires, perform two tasks designed to quantify ICS, and complete one task designed to measure behavioral impulsivity (all described below).

Questionnaires

  1. 1)

    The BIS/BAS Scales (Carver and White 1994) consist of 20 questions which assess Behavioral Inhibition System (BIS) and Behavioral Activation System (BAS) Strength. The BIS Scale (7 items) measures one’s general propensity to avoid negative, threatening stimuli whereas the BAS Scale (13 items) assesses the degree to which one approaches appetitive/rewarding stimuli. The BAS Scale consists of three subscales. The Drive subscale assesses the degree to which one works in order to receive a reward. The Reward-Responsiveness subscale assesses the degree to which one has a positive reaction to a reward. The Fun-Seeking subscale assesses the degree to which one seeks out new, pleasant situations. The internal consistency of these scales range from 0.66 to 0.76 (Carver and White 1994).

  2. 2)

    The Barratt Impulsiveness Scale, Version 11 (BIS-11, Patton et al. 1995) consists of 30 questions which participants answer on a four point likert-type scale ranging from “Rarely/Never” to “Always/Almost Always.” The internal consistency of the BIS-11 ranges from 0.79 to 0.83 (Patton et al., 1995).

  3. 3)

    The Mill Hill Vocabulary Scales (MHVS; Raven et al. 1998 ) Senior Form 2: Sets A and B each have 34 words presented along with a set of six other words—one target synonym and five distractors. For each word, the participant was asked to select the synonym. Consistent with prior research (e.g., Rhode and Thompson 2007), MHVS scores were used as a measure of IQ. Test–retest reliability of the MHVS exceeds 0.90 (Raven et al. 1998).

ICS measures

Upon completion of the questionnaires, participants were asked to perform two tasks that measure ICS. The first task asked participants to estimate how long an ‘X’ was presented on a computer screen. Once the ‘X’ disappeared from the screen, participants wrote their estimated duration on a piece of paper. Unbeknownst to participants, the ‘X’ was presented for 75 s. This task was repeated three times, and is a measure that has been frequently used by researchers examining ICS (e.g., Berlin and Rolls 2004; Berlin et al. 2005). Longer time estimations indicate faster ICS relative to shorter time estimations. The second task designed to measure ICS was a time production task (Dougherty, Mathias, et al. 2003). Participants were asked to hold a computer key down for 60 s. Again, participants completed this task three times. Smaller time production durations indicated faster ICS relative to longer time production durations. On both ICS tasks, participants were not allowed to use any external time measurement devices (e.g., watches, etc.). Participants were not given any instructions regarding the strategies they might use during either of these tasks (e.g., counting, etc.).

Behavioral impulsivity measure

The final task participants completed is called the “Two-Choice Impulsivity Paradigm” (Dougherty, Marsh, et al. 2003), a measure of behavioral impulsivity. This task required participants to select one of two icons on a computer screen. One icon was associated with a real-time 5-s wait to receive 5 cents, and the other was associated with a real-time 15-s wait to receive 15 cents. Participants were not aware of the exact length of time they were required to wait, but they did know that one icon was worth 5 cents and the other was worth 15 cents. After receiving instructions on how to perform the task, participants completed 10 “training” choices followed by 50 “real” choices. In the training session, participants learned that one icon was always associated with a 5-cent reward and a shorter wait time, and the second icon was always associated with a 15-cent reward and a longer wait. Upon completion of the 50 real choices, participants were immediately paid the amount of money they earned from the 50 real choices. Thus, participants were paid an amount ranging from $2.50 (50 impulsive choices) to $7.50 (50 non-impulsive choices). Participants were free to leave the study immediately after the Two-Choice Task.

Results

Data were collected from 79 participants (28 females and 51 males) with a mean age of 19.44 (SD = 1.35, range 18–25). However, because their responses on the MHVS would be invalid, data from those participants whose native language was not English (N = 16) were excluded. In addition, data from one subject were excluded in analyses because of computer error (not all data were collected). As a result, analyses were conducted using data from 62 participants (39 men; 23 women).

Calculation of variables

A list of the variables measured and analyzed for each participant can be found in Table 1. Average time estimation (Te) was calculated for each participant by averaging the three time estimation trials recorded for that particular participant. Average time production (Tp) was calculated the same way for each participant.

Table 1 Descriptive statistics for measured variables
Full size table

Consistent with the recommendations of Glicksohm and Hadad (in press), “ratio scores” for both Tp and Te variables were calculated. These ratio variables—TpRatio and TeRatio—were calculated as Tp/60 and Te/75, respectively. ICS was calculated determining the degree to which each participant’s TpRatio and TeRatio differed from objective time. Specifically, ICS was calculated as (TeRatio—TpRatio)/2. Please note that ICS values higher and lower than 1 indicate internal clock speeds faster and slower than objective time, respectively. Moreover, ICS error was also calculated from these ratio variables using the following equation: ICS error = (abs(TeRatio − 1) + abs(TpRatio − 1))/2. Please note that higher values indicate greater ICS error.

Because the correlation between Set A and Set B of the MHVS was significantly high (r = 0.69, p < 0.001), scores on the two Sets were averaged to create an indicator of IQ.

Outliers

All variables were assessed for outliers. An outlier was defined as more than three and a half standard deviations above or below the mean. Only one outlier was found, and it was on the average time estimation variable. Specifically, one participant had an extremely high time estimation average, and this data point was replaced with the value equal to three and a half standard deviations above the mean.

Type I error correction

In order to correct for the increased probability of type I errors among multiple correlations, q values were calculated using Storey’s Positive False Discovery Rate Procedure (pFDR; 2002). Compared to a Bonferroni correction, the pFDR approach is not as restrictive and results in greater power. A q value is analogous to the p value in that it can be used as a measure of statistical significance but with a few important differences. Whereas p values are calculated for each statistic independently, q values account for all statistics concurrently. Also, a p value indicates the probability of making a type I error among all tests, such that if one considers a p value of <0.05 to be statistically significant, one is asserting that fewer than 5% of the tests conducted will result in type I errors. A q value, on the other hand, indicates the rate of type I errors among those tests deemed statistically significant. For example, if one decides that a q value of <0.05 is statistically significant, one is saying that fewer than 5% of the tests will be type I errors. To more clearly elucidate the differences, imagine testing 100 correlations with an alpha-level set to 0.05. One can then estimate that there will be 5 type I errors. If one sets the pFDR at 0.05 and finds 40 correlations to be statistically significant, one can estimate that fewer than one (0.2) of these correlations are type I errors. For a more detailed description of the pFDR approach and q values, please see Storey (2002).

Reliability of scales

In the present data set, the internal consistencies were as follows: BIS-11 (0.849); BIS subscale of the BIS/BAS (0.680); Drive BAS subscale (0.811); Reward-Responsiveness BAS subscale (0.828); Fun-Seeking BAS subscale (0.773); and the three time estimation and three time production values (0.925 and 0.943, respectively).

Correlation results

Please see Table 2 for a correlation matrix of all variables. Of note, both TpRatio and TeRatio were significantly (and inversely) correlated with one another, p < 0.001, indicating that the time production and estimation tasks were tapping the same construct—presumably ICS. In addition, behavioral impulsivity (quantified as the number of impulsive decisions made on the Two Choice task) was significantly correlated with ICS (r = − 0.311, p = 0.014, q = 0.026) but, surprisingly, in the opposite direction as predicted. As expected, behavioral impulsivity significantly correlated with ICS error (r = 0.428, p = 0.001, q = 0.002). Contrary to predictions, behavioral impulsivity did not significantly correlate with either intelligence (average MHVS) or self-reported impulsivity.

Table 2 Correlation matrix of measured variables
Full size table

Also as predicted, a significant correlation was found between behavioral impulsivity and BAS. Specifically, those participants who scored higher on the Drive and Reward-Responsiveness subscales were less behaviorally impulsive than participants who scored lower on these scales (r = − 0.422, p < 0.05, q = 0.002 and r = − 0.388, p < 0.05, q = 0.005, respectively).

Gender differences

Gender differences in behavioral impulsivity were assessed using an independent t test. The results trended towards significance, t(60) = −1.99, p = 0.051. However, the direction of the relationship ran counter to expectations. Specifically, women made more impulsive decisions (M = 12.51, SD = 13.38) than men (M = 6.54, SD = 9.98) on the Two-Choice Task. Gender did not significantly predict ICS, but it did predict ICS Error, t(60) = −3.67, p < 0.001. Specifically, women (M = 34.05% error, SD = 24.65%) evidenced greater error on the ICS tasks compared to men (M = 16.76%, SD = 12.53%).

Regression results

It was theorized that ICS would predict behavioral impulsivity above and beyond other significant predictors. To test this, a hierarchical linear regression was performed with those variables that were significantly correlated with behavioral impulsivity. BAS-Drive and BAS-Reward-Responsiveness were entered in Step 1 and ICS was entered in Step 2. The overall model was significant (R 2 = 0.294, F[3,58] = 8.05, p < 0.001) but BAS-Reward-Responsiveness only trended to predict behavioral impulsivity (β = − 0.240, p = 0.072) when controlling for BAS-Drive (β = − 0.268, p < 0.05). As predicted, after controlling for BAS, ICS (β = − 0.283, p < 0.05) remained a significant predictor and accounted for an additional 7.9% of the variance in behavioral impulsivity.

The influence of ICS accuracy on behavioral impulsivity was also analyzed using hierarchical linear regression. BAS-Drive and BAS-Reward-Responsiveness were once again entered in the first step and ICS Error was entered in the second step. The overall model was statistically significant (R 2 = 0.299, F[3,58] = 8.24, p < 0.001) but BAS-Drive (β = − 0.176, p = 0.208) and BAS-Reward-Responsiveness (β = − 0.232, p = 0.08) were not significant when controlling for ICS Error (β = 0.314, p < 0.05), which accounted for 8.4% of additional variance in behavioral impulsivity.

A final linear model was constructed predicting behavioral impulsivity from BAS-Drive, BAS-Reward-Responsiveness, ICS, and ICS Error. Again, the overall model was significant (R 2 = 0.377, F[4,57] = 8.62, p < 0.001) but BAS-Drive was not significant (β = − 0.145, p = 0.278) and BAS-Reward-Responsiveness trended toward significance (β = − 0.244, p = 0.053). ICS (β = − 0.281, p < 0.01) and ICS Error (β = 0.312, p < 0.01) were both significant predictors of behavioral impulsivity.

Discussion

There were two main findings from the present research. First, ICS significantly predicted behavioral impulsivity above and beyond known factors, though in the opposite direction as expected. Specifically, the slower one’s ICS, the more impulsive he or she behaved. Second, ICS error also predicted behavioral impulsivity above and beyond other significant predictors. As expected, the more error in the perception of objective time, the more impulsively he or she behaved. This suggests that people who act in an impulsive manner are relatively unable to track time, an intuitively important factor in determining whether a future reward is “worth the wait” or not.

The correlation matrix revealed a few unexpected results. First, contrary to previous findings (Silverman 2003), women tended to make more impulsive decisions than men. Second, although inversely related as predicted, intelligence did not significantly correlate with behavioral impulsivity. We note that the undergraduate population recruited for this study does not adequately represent a full range of intelligence scores and, thus, a restricted range in IQ may have attenuated the observed correlation. Furthermore, and also inconsistent with some past research (e.g., Berlin and Rolls 2004; de Wit et al. 2007; Wittmann et al. 2007; but see Carillo-de-la-Pena et al. 1993; Lane et al. 2003), self-reported impulsivity and behavioral impulsivity did not significantly correlate. One possible reason for this is the “real life” nature of the behavioral impulsivity paradigm used in the present research. Unlike the vast majority of research which uses hypothetical questions to assess impulsivity (e.g., “Would you rather receive $XX today or $YY 2 months from now?”), the Two-Choice Task requires participants to make decisions with pragmatic consequences, both in terms of actual money earned and time waited. It is possible that past correlations between self-reported and “behavioral” impulsivity were potentiated due to common method variance; answering introspective (self-reported impulsivity) and delay discounting (behavioral impulsivity) questions that are both void of “real life” consequences. How might self-report impulsivity measures be altered to improve their ability to predict “real-life” decisions involving delay of gratification? Given the predictive value of emotional self-report measures in the present research (i.e., BAS), it is possible that self-report impulsivity measures may be improved by incorporating emotional aspects to the questions used. The BIS-11, for example, focuses almost exclusively on the respondent’s cognitive and behavioral processes, to the exclusion of their emotional processes. Pertinent questions may tap such discrete emotions as excitement (to receiving a reward), hope (for attaining a reward), pride (for having persevered to attain a larger, later reward), and remorse (for impulsively opting for an earlier, smaller reward).

A potentially important finding is the negative relationship between BAS strength and behavioral impulsivity, two constructs which (to our knowledge) have never before been studied simultaneously. The notion that persons with higher BAS-Drive and BAS-Reward-Responsiveness strengths are less impulsive is intuitively pleasing. Because people with higher BAS-Drive and BAS-Reward-Responsiveness scores seek and respond to rewards more than people lower on these scales, higher BAS individuals may be willing to forego immediate gratification in order to receive larger rewards. In the present study, it appears that the benefit of receiving a greater monetary reward outweighed the longer wait time for those individuals with greater BAS-Drive and BAS-Reward-Responsiveness strength.

Why does faster ICS predict decreased behavioral impulsivity? Although Wittmann and Paulus (2008) provided a sound theoretical argument supporting the opposite relationship, there is another possible explanation for the negative relationship observed in the present study. Although highly speculative in nature, a person with a relatively slow ICS—who may, for example, feel as if only 4 min passes over an actual, 5 min span—may be frequently surprised by the high rate of speed with which objective time passes. Because time passes faster than s/he appreciates, a slower-ICS individual may frequently feel rushed when performing everyday tasks. Therefore, while performing a “real life” impulsivity task (e.g., the Two Choice task), participants with slower ICSs may respond impulsively to complete the task quickly.

The present research has limitations. First and foremost, no instructions were given on either the time production or time estimation tasks, nor were follow-up questionnaires administered to determine the strategies used on these tasks. That said, based on past research which has made such queries after similar ICS tasks (e.g., Fraisse 1963; Glicksohm and Hadad, in press), the vast majority (if not all) of our participants likely employed a counting strategy. Counting leads to different results than pure interval timing (see Rattat and Droit-Volet, in press). Although many have used paradigms similar to those contained herein, other researchers have used strategies to mitigate the use of counting on ICS tasks. These strategies include (1) asking participants not to count (e.g., Hicks et al. 1976) and asking them to perform a secondary task such as (2) reading randomly-presented digits (e.g., Wearden et al. 1997) or (3) repeatedly saying something (“blablabla”) to disrupt counting (e.g., Delgado and Droit-Volet 2007). Each of these strategies have relative disadvantages, such as requiring participants to be truthful (1), disproportionately influencing the time perception of individuals with lower working memory capacities (2), and producing noise in the time perception task (e.g., Rattat and Droit-Volet, in press). As such, for the present study, we decided to increase the reliability of time perception data by allowing people to count on several trials (3 each) of two different ICS tasks (time estimation and time production). The present study should be replicated using ICS measures that mitigate counting. Second, only one behavioral impulsivity measure was used, and the validity of the assessment may be improved by incorporating additional behavioral impulsivity measures. Third, although the Two Choice Task has the benefit of being “real life”, it is a hybrid task in that people make 50 dichotomous choices (impulsive or delay gratification) that scale up to a single continuous measure (level of impulsivity). Behavioral impulsivity tasks that are independently continuous or dichotomous may relate differently to such constructs as ICS and self-reported impulsivity. Therefore, in future research, both continuous and dichotomous tasks might be incorporated.

Although this study does have its limitations, it has helped advance an area of study that has received relatively scant attention. Whereas previous research has focused on the role of ICS in clinical populations with impulsive behaviors, the current research shed light onto this relationship in a healthy population. The current research has helped identify fertile areas for future investigation such as a possible distinction between “future-oriented” and “in-the-moment” impulsivity.