Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Apr 1.
Published in final edited form as: Exp Clin Psychopharmacol. 2019 May 9;28(2):225–234. doi: 10.1037/pha0000299

Persisting on the Past: Cross-sectional and Prospective Associations Between Sunk Cost Propensity and Cannabis Use

Michael J Sofis 1, Shea M Lemley 1, Alan J Budney 1, Catherine Stanger 1, David P Jarmolowicz 2
PMCID: PMC6842029  NIHMSID: NIHMS1025426  PMID: 31070426

Abstract

Prevalence of cannabis use in the U.S. continues to rise and 30% of cannabis users eventually meet criteria for Cannabis Use Disorder (CUD). One response to this problem is to develop decision-making constructs that indicate vulnerability to a SUD that might not be gleaned from diagnostic criteria. Unfortunately, there is limited evidence that decision-making constructs consistently relate to cannabis use. Interestingly, those who exhibit the sunk cost bias, an over-generalized tendency to persist based on past investment, and those who use cannabis, both tend to focus on the past and perseverate more than their counterparts. Despite this overlap, no studies have assessed whether the sunk cost bias is positively associated with cannabis use. In two experiments with undergraduates, relations between cannabis use and the propensity to engage in the sunk cost bias were examined using negative binomial models. Experiment 1 (n=46) evaluated the association between sunk cost bias propensity (using hypothetical costs and rewards) and frequency of cannabis use over the past 30 days. Greater sunk cost propensity was associated with more frequent cannabis use after controlling for demographics and alcohol use. In Experiment 2 (n=103), more frequent cannabis use during a six-week follow-up period was predicted by greater sunk cost propensity at baseline (using a real cost and reward-based task), independently and after controlling for mental health symptoms, alcohol use, and demographics. These findings provide preliminary evidence that a propensity to exhibit the sunk cost bias may be an important feature associated with cannabis use.

Keywords: Sunk cost bias, cannabis use, persistence, marijuana, sunk cost propensity

An estimated 10 percent of individuals in the U.S. use cannabis each year, with roughly 3 out of 10 users eventually meeting criteria for Cannabis Use Disorder (CUD; Hasin et al., 2015). Cannabis use is associated with an increased risk of altered judgement, psychosis, poor educational and vocational outcomes, anxiety disorders, respiratory problems, neurocognitive impairment, and lower academic achievement (Hall, 2009; Volkow, Baler, Compton, & Weiss, 2014). Studies of the generic DSM and ICD criteria used to diagnose all substance use disorders (SUD) suggest that the clinical manifestation of CUD is more similar than different from other SUDs (Budney, 2005; Hasin et al., 2013). Multiple federal agencies in the last decade in the U.S. have placed a greater emphasis on developing markers of addiction, defined as behavioral or biological measures that indicate vulnerability and presence of an SUD that may not otherwise be captured using current DSM or ICD diagnostic criteria (Kwako, Bickel, & Goldman, 2018). Such markers may be transdiagnostic across SUDs (Bickel, 2015) or may be differentially relevant based on type of SUD (MacKillop, 2013). One such measure is delay discounting (DD; Bickel, Koffarnus, Moody, & Wilson, 2014), i.e., the propensity to devalue future rewards (Moody, Franck, Hatz, & Bickel, 2016), which has shown a robust relationship with the use and misuse of various substances (Amlung, Vedelago, Acker, Balodis, & MacKillop, 2017; Bickel, Jarmolowicz, Mueller, Koffarnus, & Gatchalian, 2012; MacKillop, 2013). For substances such as nicotine, opioids, and alcohol, DD is positively associated with risk for developing a SUD and SUD severity (Bickel, 2015) and has been shown to decrease with effective treatment (Koffarnus, Jarmolowicz, Mueller, & Bickel, 2013). In the context of cannabis use and CUD, however, the DD findings have been more equivocal (Johnson et al., 2010; Kollins, 2003; Strickland, Lile, & Stoops, 2017). Given the potential utility of DD as a transdiagnostic marker of addiction, further exploration of the relationship between cannabis use and DD is warranted. It may also be useful to examine alternative behavioral markers that may contribute to our understanding of cannabis use patterns and the development of CUD.

The sunk cost bias, a form of maladaptive persistence in response to past investment of time, effort, or money (Arkes & Blumer, 1985) may be one useful candidate. Two basic examples of the sunk cost bias are as follows: (1) individuals will eat more at an all-you-can-eat buffet if they paid more to initially access the buffet (Siniver, Mealem, & Yaniv, 2013) and (2) individuals will more be more likely to follow advice that cost money versus free advice (Gino, 2008). Sunk cost paradigms typically manipulate the amount of the initial investment (e.g., cost of a buffet) and assess the degree to which participants’ responding in terminal investment contexts (e.g., amount of caloric intake at buffet) is influenced by the amount of the initial investment.

The sunk cost bias is generally thought to be a trait-like tendency that primarily occurs due to overgeneralization of two normally adaptive rules: namely, “waste not want not” and “persevere” (Arkes & Ayton, 1999; Bruine de Bruin, Parker, & Fischhoff, 2007; Stanovic & West, 2000). These two rules are thought to impact the sunk cost bias through initial and terminal investment contexts, respectively. A tendency to avoid losses (e.g., waste) breaks the axiom that only future benefits and costs, and not past ones, should be considered when choosing whether to pursue a given action (Arkes & Blumer, 1985; Stanovic & West, 1998). For example, paying more at an all-you-can-eat buffet is an irrecoverable cost; a rational choice leaves only considering the costs and benefits in present and potential future contexts when deciding how much to eat. Second, the sunk cost is thought to be an overgeneralized form of perseverance because completing the effect typically entails incurring greater costs than other available options. This can be illustrated by continuing to watch an unpleasant movie (incur additional time costs) simply because of the time invested into watching it (i.e., sunk costs), despite having the option of not watching the movie any longer, which would reduce the total time costs incurred. Such perseveration in the context of sunk costs has been conceptualized as a failure to demonstrate cognitive flexibility, defined as the ability to consider multiple decision options and switch attention between them when appropriate (Emich & Pyone, 2018). An inability to attend to all the possible alternatives when making a terminal investment decision may be more likely when overly focused on a past investment (Emich & Pyone, 2018). Moreover, one way in which the sunk cost bias can be increased is through perseveration, by enhancing the anticipated regret associated with passing on a terminal investment rather than completing it (Wong & Kwong, 2007).

Exhibiting the sunk cost bias has largely been linked to undesirable outcomes (Jarmolowicz, Bickel, Sofis, Hatz, & Mueller, 2016; Rego, Arantes, & Magalhães, 2018; Siniver et al., 2013; Stiegler & Ruskin, 2012). Those who exhibit the sunk cost bias are more likely to focus on the past than those who do not, which is presumed to be a key factor underlying the maladaptive nature of the bias (Strough, Schlosnagle, & DiDonato, 2011; van Putten, Zeelenberg, & van Dijk, 2010). Those who commit the sunk cost bias tend to feel greater regret and avoidance during decision making, both of which suggest a tendency to focus on past events and value how such events relate to present circumstances (Bruine de Bruin et al., 2007)

Despite evidence that the sunk cost bias is likely a trait-like phenomenon (Arkes & Ayton, 1999; Bruine de Bruin et al., 2007; Stanovic & West, 2000) the propensity to exhibit the sunk cost bias has also be shown to be malleable (Emich & Pyone, 2018; Hafenbrack, Kinias, & Barsade, 2014; Lee, Keil, & Wong, 2018). A series of studies have shown that targeting cognitive flexibility with behavioral prompts to consider alternative options have resulted in reduced perseveration following sunk costs through enhancement of participants’ ability to consider context-specific alternative options. One such study showed that among students who sunk costs into semester-long meal plans (i.e., initial investment), those who were prompted to consider diverse aspects of meal enjoyment ate fewer calories (terminal investment) than those who were instructed to focus on the cost of their meal (Emich & Pyone, 2018). Another study observed that a mindfulness intervention shifted temporal focus closer to the present and resulted in a reduced propensity to exhibit the sunk cost bias compared to a control group who did not receive mindfulness training (Hafenbrack et al., 2014). These data suggest that the sunk cost bias can be altered and support the position that the phenomenon may reflect an overgeneralization of rules related to avoiding losses and persevering (Arkes & Ayton, 1999; Strough et al., 2011).

Propensity to exhibit the sunk cost bias may be specifically associated with patterns of cannabis use because compared to non-users, heavy cannabis users (near daily use) show greater response perseveration on the Wisconsin Card Sorting Task (WCST), a measure of cognitive flexibility defined as the ability to consider many aspects of a problem and appropriately switch attention among them (Dahlgren, Sagar, Racine, Dreman, & Gruber, 2016; Lane, Cherek, Tcheremissine, Steinberg, & Sharon, 2007). Heavy cannabis users also show less adaptive response allocation (i.e., exhibit a higher number of responses per reward) to changing reinforcement contingencies than those who do not use cannabis (Lane et al., 2007), a common co-occurrence with overly rigid adherence to rules (Rosenfarb, Newland, Brannon, & Howey, 1992). Higher frequency of monthly cannabis use has also been shown to be associated with annual decreases in cognitive flexibility for up to 17 years in a sample with diverse patterns of cannabis use (Okafor et al., 2019), suggesting that cognitive flexibility is negatively associated with more frequent cannabis use. The threshold of use that results in impaired functioning, however, is not yet known. In sum, individuals who use cannabis may be more prone to maladaptively persist in response to past investments because they are also more prone to focus on the past and to perseverate due to deficits in cognitive flexibility than those who do not use cannabis.

The current Experiments tested whether a greater propensity to exhibit the sunk cost bias was positively associated with more frequent cannabis use in the past month. Experiment 1 used a cross-sectional approach to compare the differential associations of SCP and DD to frequency of cannabis use. The sunk cost task for Experiment 1 entailed a hypothetical monetary investment scenario and an adjusting amount task was employed for DD (Du, Green, & Myerson, 2002). Using a prospective design and a larger sample size, Experiment 2 investigated whether baseline levels of the sunk cost bias and DD predicted which participants would use cannabis and the frequency of their cannabis use during a six-week follow-up period. Unlike Experiment 1, the sunk cost task entailed real rewards (extra credit points) and effortful responding (button clicks). Experiment 2 also assessed symptoms of depression, anxiety, and somatization to determine whether the sunk cost bias was associated with cannabis use after controlling for mental health symptoms.

Experiment 1

Method

Participants.

Participants were recruited from undergraduate applied behavioral science classes at a large midwestern university. The only inclusion criterion was enrollment in the class. Fifty-six participants completed all experimental tasks, however, nine participants were excluded because they provided inconsistent DD data in accordance with criteria described by Johnson & Bickel (2008) and one participant was excluded for exclusively choosing to invest for all trials, for a final sample size of n=46. Participants ranged from age 18 to 30 (Mdn = 20.5 years, IQR 19, 22), and 78% were female. Participant’s annual parent income ranged from “unemployed/disabled” to $200,000+/year but was skewed towards the higher end (Mdn = $101,000–200,000). Three participants did not report parent income but were included in analyses. Participants received extra credit for their participation.

Procedures.

All study procedures were approved by the Human Research Protection Program (HRPP), which is the institutional review board at the University of Kansas (Titled “Relations between Risk and Real and Hypothetical Investment Decisions”). Participants first reviewed an information statement and signed the informed consent document. Participants then completed a brief demographics questionnaire, a sunk cost task, a DD task, and an assessment of substance use during a 1-hr session that included other assessments (not reported).

Measures.

Substance Use.

Participants completed a computerized version of the Timeline Follow-back (TLFB) to obtain valid reports of their substance use during the past 30 days (Sobell, Sobell, Litten, & Allen, 1992). Cannabis use days, days of alcohol use, and total number of alcohol drinks consumed in the last 30 days were collected. Cannabis use groups (Use / no-Use) were created by dichotomizing on whether participants had recently used cannabis (i.e., in the last 30 days). Twenty-one participants (46%) reported cannabis use in the last 30 days (range: 1–30 days; M = 10.00, SD = 9.73). This cannabis use group reported a range of 0 to 20 days of alcohol use (M = 8.43, SD = 5.13) and a range of 0 to 153 drinks (M = 39.24, SD = 37.64) during the prior 30 days. The no-use group reported a range of 0 to 12 days of alcohol use (M = 4.24, SD = 3.44) and a range of 0 to 65 alcoholic drinks (M = 19.88, SD = 20.70) in the last 30 days. Participants were also asked whether they were current cigarette smokers. Five in the cannabis use group (23.8%) reported current cigarette use, while none of those in the no-use group reported current cigarette use in the past 30 days. A non-parametric correlation showed that days of cannabis and alcohol use in the last 30 days were positively correlated (rs = .43, p = .003).

Sunk Cost Task.

The sunk cost task comprised a series of trials involving allocating hypothetical monetary funds to complete projects and was adapted from the task described by (Sofis, Jarmolowicz, Hudnall, & Reed, 2015). Each trial contained an initial investment scenario followed by a terminal investment scenario. In the initial investment scenario, participants were forced to make an initial monetary investment and, based on the cost of that investment (e.g., $35), they had to choose to either complete the terminal investment at a particular cost (e.g., $80), or pass on completing that terminal investment cost and begin a new trial. Prior to beginning the task, participants were instructed:

The current task will involve investing money to complete projects. Each project will cost money to begin ($20 on average) and will cost money to complete ($50 on average). These costs, however, will vary from project to project. You must pay the cost to begin each project but can opt out of completing it. Your goal is to complete 30 projects.

Broadly, the goal of the task was to evaluate how participants would allocate their hypothetical budget of $3,500 to complete 30 terminal investments, or “projects” (See Figure 1 for a schematic of the trial sequence). Throughout the task, the completed project count was updated in the upper left corner of the screen and after completing each initial or terminal investments, the remaining funds in the bank were adjusted in the upper right corner of the screen.

Figure 1.

Figure 1.

Open in a new tab

Schematic of the sunk cost procedure for a given trial.

Initial and terminal link investment amounts for each participant were presented quasi-randomly and without replacement. Investment amounts were quasi-randomly presented to mitigate potential order effects of investment amounts. Each investment amount was presented without replacement so that each participant would receive approximately equivalent exposure to each investment amount.

Each trial began with the presentation of an initial forced choice investment of X amount (X = $5 or $35; quasi-randomly presented without replacement) and text stating, “This project costs $X to begin.”). Participants were required to click an “Invest” button in the middle of the screen. After this initial forced investment, the participant’s funds decreased by the investment amount, and the terminal link appeared (i.e., “This project costs $Y to complete.” (Y = $5, $20, $80, or $95; quasi-randomly selected)) with an “Invest” button in the lower right corner and a “Pass” button in the lower left corner of the screen. Choosing to invest in the terminal link incremented the projects completed count by one, subtracted the cost of the project from the participant’s funds remaining, and presented the next force-choice initial link. Choosing not to invest (i.e., pass) did not impact the count of projects completed or the funds remaining but resulted in presentation of the next forced-choice initial link. Initial and terminal link amounts were quasi-randomly selected without replacement within each set of eight initial-terminal link combination (i.e., one round); after each round, the values were shuffled and presented again in a new round. This process continued until participants completed 30 projects.

Sunk cost propensity (SCP) was the index measure of the sunk cost bias for this study (see “Overall Sunk Cost Effect” index in Sofis et al., 2015). SCP was calculated by subtracting the mean investment percentage of all terminal investments following $35 initial investments (i.e., the higher initial investment amount) by the mean investment percentage of all terminal investments following $5 initial investments. Adapted from Sofis et al. (2015), this approach provides a continuous measure of the sunk cost that ranges from negative values (i.e., more likely to complete terminal investment following $5 initial investments) to positive values (i.e., more likely to complete terminal investment following $35 initial investments). The SCP is labeled as a propensity because the measure is an aggregation of all responding made during the task.

DD Task.

Participants then completed a titrating delay discounting task (Du et al., 2002) in which they made hypothetical choices between a $1,000 larger, later reward available after one of 6 ascending delays (1-day, 1 week, 1 month, 6 months, 1 year, 5 years) or a smaller, sooner reward available immediately. At the beginning of each delay, the amount of the smaller, sooner reward was $500, but this amount was titrated based on the participant’s choice on each trial. If the participant chose the larger, later reward, the value of the smaller, sooner reward increased by 50% of the previous titration value (initially $250), but if the participant chose the smaller, sooner reward, the value of the smaller, sooner reward was reduced by 50% of the previous titration value. The sixth titration trial at each delay determined the participant’s indifference point.

An index of DD, area under the curve (AUC), was calculated for each participant using the summed trapezoid method (Myerson, Green, & Warusawitharana, 2001). First, each delay was converted to a proportion of the longest delay, and each indifference point was expressed as a proportion of the total, undiscounted reward amount. The areas of trapezoids formed by pairs of these successive indifference points (y-coordinates) and delays (x-coordinates) were then determined and summed according to the equation:

AUC = ∑((x2−x1)×(y2+y12)).

Higher AUC values correspond to lower rates of DD.

Analysis Plan.

A Spearman’s rank order correlation failed to demonstrate a significant relation between terminal investment percentage and cannabis use days (rs = .11, p = .52), suggesting that any association between frequency of cannabis use and SCP was likely not an artifact of participants’ broader investment strategies in the sunk cost task. Spearman’s rank order correlations amongst alcohol use days, cannabis use days, DD, and age all failed to demonstrate statistical significance except for the relation between cannabis and alcohol use days (rs = .43, p = .003). A Mann-Whitney U test showed that men used cannabis more frequently than women (U = 73.00, p = .003). Age was included as a covariate given previous associations with the sunk cost bias in the literature (Strough, Mehta, McFall, & Schuller, 2008). Smoking status was not tested as a potential covariate because only five participants were current cigarette smokers.

Because the outcome variable, frequency of days of cannabis use had a high proportion of zeros and over dispersions (variance is much high than mean), a negative binomial regression analysis was conducted to predict the frequency of days of cannabis use in the last 30 days based on SCP and DD after controlling for the effects of days of alcohol use, age, and gender. Continuous variables that were included in the negative binomial regression were transformed into z-scores to improve the model fit and to allow DD to become linearly associated with the logit of the dependent measure. These transformations did not significantly alter the overall model chi-squared value, the significance values, or the Wald statistic scores for any measure. A second negative binomial analysis with SCP as the only predictor variable was run to test whether the effect of SCP on cannabis use days required the presence of the covariates. Separate Omnibus tests of each model were statistically significant (p < .001).

Results

For the overall sample, the mean SCP was .0002 (SD = .0573). Table 1 shows the results of a negative binomial model predicting frequency of cannabis use days. After controlling for days of alcohol use, DD, gender, and age, SCP (standardized Z-score) was positively associated with more days of cannabis use in the last 30 days (RR= 2.32, 95% CI = 1.39, 4.14, p =.002). Days of alcohol use in the last 30 days was positively associated with more days of cannabis use during the same period (RR= 1.56, 95% CI = 1.04, 2.44, p =.04). Age negatively corresponded with days of cannabis use (RR= .52, 95% CI = .29, .87, p =.02) and being female corresponded with more days of cannabis use than being male (RR= 10.7, 95% CI = 4.2, 30.8, p <.001). There was not a significant effect of DD (p = .76). In the second negative binomial model that did not include any covariates, SCP remained positively associated with more days of cannabis use in the last 30 days (RR= 2.35, 95% CI = 1.55, 3.80, p > .01).

Table 1.

Experiment 1: Negative Binomial Regression Analysis of Frequency of Cannabis Use

Predictor β SE β Wald’s χ2 df P eβ (Rate Ratio)
Intercept .15 .27 .34 1 .56 1.17
SCP .84 .28 9.31 1 <.01 2.32
DD (AUC) −.06 .18 .09 1 .76 .95
Age −.66 .28 5.48 1 .02 .52
Gender 2.4 .51 22.0 1 <.001 10.74
Days of Alcohol Use .44 .22 4.20 1 .04 1.56
Open in a new tab

Note: The p-values and exponentiated betas of significant predictor variables are bolded and italicized. Coefficient estimates (“β”), standard error of coefficient estimates (SE β), the Wald statistic (χ2), degrees of freedom (df), significance value (p), and the exponentiated values of the coefficients (eβ) are presented in the columns from left to right.

Discussion

Data from Experiment 1 showed that greater SCP corresponded with more frequent cannabis use after controlling for alcohol use, DD, age, and gender, suggesting that the association between SCP and cannabis use was at least partially independent of these other variables. The Wald and the exponentiated values of the predictor coefficients from the model were also more than double those of alcohol use days and only 8% of the total variance of the model was lost by removing alcohol use days. This finding provides initial support of a direct association between SCP and frequency of cannabis use days, especially because days of cannabis and alcohol use demonstrated a medium to large positive relation in the current study (rs = .43, p = .003).

Interestingly, this study showed that SCP, and not DD, positively related to more frequent cannabis use. The failure to find a statistically significant relation between DD and cannabis use in the negative binomial model is congruent with previous findings that suggest that DD is inconsistently related to cannabis use frequency (Aston, Metrik, Amlung, Kahler, & MacKillop, 2016; Strickland et al., 2017; VanderBroek, Acker, Palmer, de Wit, & MacKillop, 2016).

This initial examination of the association between the sunk cost bias and cannabis use was limited by several methodological factors. First, the sample size was small, reducing the power of statistical tests. Second, hypothetical rewards and costs (i.e. money) were used in the sunk cost task. It is unknown whether the current findings would translate to paradigms using real costs and rewards. Although rewards associated with terminal investments in real world scenarios have shown robust demonstrations of sunk cost biases (Arkes & Blumer, 1985; Åstebro, Jeffrey, & Adomdza, 2007; Siniver et al., 2013) there is limited evidence directly comparing the sunk cost bias with hypothetical and real rewards. Third, cannabis use and CUD are strongly associated with mental health issues (See Borodovsky & Budney, 2018 for a review) and recent evidence suggests overlap between the sunk cost and multiple components of mental health (Hafenbrack et al., 2014; Jarmolowicz et al., 2016). Experiment 1 did not control for potential effects of mental health symptoms on cannabis use. Fourth, those who reported cannabis use in the current sample tended to display relatively low rates of use. Studies with larger samples with individuals with a wider range of cannabis use could yield different findings.

Experiment 2

Experiment 2 was designed to replicate and extend the findings of Experiment 1 by including a larger sample, controlling for the effects of mental health symptoms, and using real rewards and actual costly responses in the sunk cost task. Additionally, cannabis use was measured prospectively to improve the internal validity of our examination of the relationship between SCP and cannabis use.

Method

Participants.

A sample of 136 undergraduate participants was recruited from applied behavioral science classes to complete two experimental sessions. Inclusion criteria were the same as for Experiment 1. Of the 136 participants, 80% (n=109) completed the follow-up session. Of these 109, four were excluded for providing inconsistent indifference data on the DD task (Johnson & Bickel, 2008) and two were excluded for missing attention check questions designed to ensure that participants were focused and cooperating, which yielded a final sample of n=103 participants. Participants (89% female; 83% Caucasian) ranged from age 18 to 27 (Mdn = 20.0 years, IQR 19, 21). Participant’s annual parent income ranged from “unemployed/disabled” to $200,000+/year, with a median of $76,000–100,000. A series of Chi-square and Mann-Whitney U tests showed no differences in gender or ethnicity (both χ2′s < 2.54, p’s > .10), age, BMI, or parental income (all U’s > 1664.5, p’s > .49) between those who did and did not provide follow-up data.

Procedures.

All study procedures were approved by the Human Research Protection Program (HRPP), which is the institutional review board at the University of Kansas (Titled “Initial Evaluation of Behavioral Economic Indices as Prospective Predictors of Risk”). For the first experimental session, 0.25% extra credit was provided for completing all experimental tasks and 0.01% of extra credit was awarded for each project completed in the sunk cost task (combined max of 1%). Participants received 1.5% extra credit upon completing the second session. The second session (i.e., follow-up) was performed remotely (online), no sooner than eight weeks after the first session, with the exact date and time determined by the individual participant. On the Monday following the eighth week from the initial session, participants received their unique code hint and a link in an email with instructions for accessing the follow-up tasks on Qualtrics™. On average, participants completed the follow-up session 70 days from the initial session (range 62–84 days). During follow-up, participants reported their substance use behavior that occurred during the six weeks immediately preceding the follow-up reporting session.

Measures.

Initial session.

Experimental tasks included the adjusting-amount DD task used in Experiment 1, the Daily Drinking Questionnaire (DDQ; Collins, Parks, & Marlatt, 1985), a human operant version of the sunk cost task reported in Experiment 1 (described below), and the Brief Symptom Inventory-18 (Derogatis, 2000).

DD task.

The DD task was identical to the one described in Experiment 1.

Daily drinking and daily cannabis questionnaire.

Participants completed a modified and computerized version of the DDQ (Collins et al., 1985) for alcohol use. An index of excessive drinking in a typical week was determined by cutoffs of greater than 7 drinks per week for females and greater than 14 drinks per week for males (NIAAA, 2017). The sample was divided into those who used cannabis (38.8%) and those who did not (61.2%). Among those who used cannabis, 68% qualified as drinking alcohol excessively during typical weeks, whereas 32% qualified for excessive alcoholic consumption among those who did not use cannabis.

BSI-18.

The BSI-18 assesses symptoms of anxiety, depression, and somatization in the last 30 days on a 0–4 point scale (0 = “ Not at all”, 4 = “Extremely”; Derogatis, 2000). The factor structure of the BSI-18 has been validated in samples of illicit substance users (Wang et al., 2010). In the current study, the mean scores for the cannabis group were 7.35 (SD = 5.69) for depression, 9.08 (SD = 4.84) for anxiety, and 4.20 (SD = 4.02) for somatization. The mean scores for the no cannabis use group were 5.30 (SD = 5.58) for depression, 7.52 (SD = 5.69) for anxiety, and 3.68 (SD = 4.45) for somatization.

Operant sunk cost task.

The operant sunk cost task challenged participants to complete as many projects as possible in 15 minutes (thus earning more extra credit), using button-clicks inside a moving rectangle as the cost associated with completing initial and terminal investments. Prior to beginning the task, written instructions were presented on the screen:

Click buttons to earn points. Each point earned in the experiment is worth 0.01 percentage points on your final grade. The button in green (left) requires an average of 20 clicks, and the button in blue (right) requires an average of 50 clicks. You must click the button in green, but you can opt out of clicking the button in blue by clicking the floating arrow. You have 15 minutes to earn points. If you need a break, please raise your hand to contact the experimenter.

After the participant clicked “Okay,” the task began. The initial investment context featured a button rotating within a green field on the left side of the participant’s screen (see top panel of Figure 3). Upon completion of the initial investment, this green field and button disappeared, and the terminal investment context appeared on the right side of the screen. The terminal investment context consisted of a button rotating within a larger blue rectangle (see bottom panel of Figure 3). A grey “Pass” arrow moved up and down beneath this terminal link field. After completing any terminal investment, a small box appeared on the screen stating, “You earned 1 point!” The participant had to click “Okay” to collect the point before the task continued. Clicking the “Pass” arrow terminated the current terminal investment scenario without a point delivery. After the participant collected a point or passed during the terminal investment context, the trial ended and the initial investment context reappeared.

Further, initial investment amounts alternated between 5 and 45 clicks (instead of $5 and $35), with the 5 click initial investment context denoted by a light green rectangular background and the 45 click initial investments denoted by a darker shade of green. The terminal investment values alternated between 5, 20, 100, and 125 clicks, with progressively darker shades of blue representing larger investment amounts. The number of clicks remaining within a given investment context was displayed above the initial or terminal investment field. Every initial investment amount was presented with each terminal investment amount. To isolate the chief effect of interest, the SCP measure was calculated in the identical fashion to that used in Experiment 1, and then collapsed into those who did (SCP > 0) and did not (SCP < 0) exhibit SCP.

Participants had 15 minutes to complete as many projects as possible. Given that extra credit was provided after completing a project, the number of completed projects was used as an index of adaptive performance (M = 36.86, SD = 5.73), and as a co-variate in a negative binomial regression predicting the association between SCP and frequency of cannabis use.

Follow-up session.

Substance use.

The timeline followback procedure was identical to the one used in Experiment 1 except that the reporting period was the previous 42 days (six weeks). Days of cannabis use, days of alcohol use, and total number of alcohol drinks consumed were collected. For those who used cannabis, median frequency of days of cannabis use was 8.5, median alcohol days was 12.5, and median alcoholic drinks was 48. For those who did not use cannabis, median days of alcohol use was 4 and the median for alcoholic drinks was 9.

Analysis Plan.

A negative binominal regression analysis was conducted to evaluate whether SCP and DD predicted the frequency of days of cannabis use during the follow-up period after controlling for the effects of projects completed, all three BSI factor scores, excessive alcohol consumption status, gender, and age. Excessive drinking was included because of the association with alcohol days and cannabis use groups observed in Experiment 1. Continuous measures were transformed into z-scores to improve the fit of the model. These transformations did not significantly alter the overall model chi-squared value, significance values, or the Wald statistic scores for any measure. Prior to performing the primary analyses, a non-parametric correlation (Spearman’s rho) failed to show a significant correlation between frequency of terminal investment clicks and frequency of cannabis use days (rs = −.04, p = .70), suggesting that any effect of SCP on frequency of cannabis use days was not due to the influence of terminal investment responding alone. An observed difference in SCP between those who did, and did not, follow-up failed to reach statistical significance (t =1.913, p=.06), but suggested greater levels of SCP among those who did not follow up. A second negative binomial model with SCP as the only predictor variable was run to test whether the effect of SCP on cannabis use days required the presence of the covariates. Separate Omnibus tests of each model were statistically significant (p < .01).

Results

For the overall sample, the mean SCP was −.0119 (SD = .0786). Table 2 shows the results of a negative binomial model predicting frequency of cannabis use days. After controlling for the number of projects completed, three BSI factor scores, excessive alcohol consumption status, DD, gender, and age, SCP (standardized Z-score) was positively associated with frequency of cannabis use days (RR= 1.49, 95% CI = 1.19, 1.87, p =.001). Excessively drinking during a typical week corresponded with more frequent cannabis use (RR= .097, 95% CI =.06, .17, p<.001). Greater BSI depression scores corresponded with more days of cannabis use (RR= 1.69, 95% CI =1.06, 2.71, p = .03), whereas greater BSI somatization scores were associated with fewer days of cannabis use (RR= .56, 95% CI =.36, .85, p = .006). The positive relation between BSI anxiety scores and frequency of cannabis use trended towards significance (p = .08). Age was positively associated with more days of cannabis use (RR= 1.83, 95% CI =1.34, 2.50, p<.001) and being male corresponded with fewer reported days of cannabis use (RR= 2.44, 95% CI =1.06, 5.59, p = .04). No other predictor variables were significant, including DD (p = .55). In the second negative binomial model that did not include any covariates, SCP remained positively associated with more days of cannabis use in the last 30 days (RR= 1.30, 95% CI = 1.09, 1.55, p > .01).

Table 2.

Experiment 2: Negative Binomial Regression Analysis of Frequency of Cannabis Use

Predictor β SE β Wald’s χ2 df P eβ (Rate Ratio)
Intercept 2.19 .160 188.26 1 <.001 8.93
Excessive Drinking −2.33 .274 72.74 1 <.001 .097
Gender .890 .424 4.40 1 .04 2.44
SCP .398 .115 11.91 1 .001 1.49
Projects .088 .127 .484 1 .49 1.09
Age .602 .160 14.17 1 <.001 1.83
BSI-Depression .527 .240 4.80 1 .028 1.69
BSI-Anxiety .453 .256 3.13 1 .077 1.57
BSI-Somatization −.589 .216 7.46 1 .006 .555
Money AUC .088 .146 .362 1 .547 1.09
Open in a new tab

Note: The p-values and exponentiated betas of significant predictor variables are bolded and italicized. Coefficient estimates (“β”), standard error of coefficient estimates (SE β), the Wald statistic (χ2), degrees of freedom (df), significance value (p), and the exponentiated values of the coefficients (eβ) are presented in the columns from left to right.

Discussion

When controlling for age, gender, DD, projects completed, and BSI factors, exhibiting a positive SCP at baseline was prospectively associated with more frequent cannabis use during the follow-up period (six-weeks). Like in Experiment 1, DD did was not significantly related to frequency of cannabis use. Importantly, the prediction of cannabis use frequency at follow-up by SCP held regardless of whether covariates were included in the model and despite the one-month gap between the baseline session and the beginning of the follow-up period. These findings suggest there may be an underlying direct association between cannabis use and SCP.

Of note, the positive association between SCP and cannabis use was observed when using real rewards (i.e., extra credit points) and costly responding (i.e., button clicks). This positive finding provides additional evidence that the processes underlying SCP are related to the frequency of cannabis use. This replication of the findings from Experiment 1 using real rewards and costs rather than hypothetical ones suggests that the association between SCP and cannabis use may generalize across consequence type.

Future studies might prioritize replicating the current study with samples comprised of a wider age range and broader sociodemographic profiles. Future studies also need to include a broader range of cannabis use patterns. Among individuals ages 18–25, the past 30-day prevalence of cannabis use was 21% (CBHSQ, 2018), whereas in the present experiment, 38.8% of participants reported using cannabis in the past 42 days. Few participants, however, reported using at daily or near-daily rates, and no information was collected about how often or how much cannabis participants used each day.

General Discussion

The current experiments are, to our knowledge, the first to assess the sunk cost bias in the context of cannabis use. Findings from cross-sectional and prospective methods showed that greater propensities to exhibit the sunk cost bias corresponded to more frequent cannabis use. These data also demonstrated that more frequent cannabis use related to distinctive trait-like patterns of persistence driven by past investments (i.e., SCP) that generalized across reward context, cost context, and time.

The persistence observed in the context of these studies adds to the literature in this area in at least three ways. First, persistence shown in the current Experiments is empirically shown to be driven largely by the amount of past investments which stands in contrast to much of the research on behavioral persistence (Hajek, Belcher, & Stapleton, 1987; Lejuez, Kahler, & Brown, 2003) or perseveration (Lane et al., 2007). Specifically, there was not a significant correlation between percentage of terminal investments in Experiment 1, or frequency of terminal clicks in Experiment 2, and cannabis use days, suggesting that the effect of SCP on frequency of cannabis use was likely not due to the sole influence of terminal investment responding. This is consistent with previous findings showing that individuals who demonstrate the sunk cost bias focus more on the past and make more decisions based on regret and avoidance than those who do not (Bruine de Bruin et al., 2007; Strough et al., 2011; van Putten et al., 2010). Second, the current demonstration of persistence (i.e., SCP) generalized across two unique behavioral tasks that varied the nature of cost and reward parameters (i.e., hypothetical vs. real). By contrast with most measures of persistence, the current SC paradigm resulted in similar positive associations with frequency of cannabis use across two unique modalities (e.g., hypothetical trial-based vs. operant trial-based task). Third, a limited number of measures of persistence have shown prospective prediction of a relevant health outcome and fewer still have predicted a substance use outcome.

Exhibiting a positive SCP corresponded to an empirical pattern of completing the terminal investment more frequently following larger (rather than smaller) initial investment amounts. The finding that frequency of cannabis use was positively related to SCP is broadly consistent with existing evidence demonstrating a positive relation between higher frequency cannabis use and cognitive inflexibility and perseveration. However, cognitive flexibility and response perseveration were not assessed in the current exploratory experiments, which prevents a direct evaluation of whether either underlied the sunk cost bias.

There were no significant findings pertaining to DD and cannabis use in either Experiment, which is consistent with previous studies that failed to show a significant relation between DD and frequency of cannabis use, particularly because our sample was not comprised of heavy users or those seeking treatment for CUD (Aston et al., 2016; Heinz, Peters, Boden, & Bonn-Miller, 2013; Johnson et al., 2010; Stanger et al., 2012; Strickland et al., 2017). One potential synthesis of this literature is that DD may more closely relate to measures that characterize impairment from heavy or problematic cannabis use, rather than frequency of cannabis use per se. Future research might look to recruit larger samples that display broader use patterns ranging from no use to daily use, and assess additional cannabis-related variables such as age of first use, cannabis-related problems, and method of administration to explore this hypothesis. If such a hypothesis is confirmed, it may be fruitful to assess whether SCP relates to frequency of cannabis use, cannabis-related problems, and DD in a sample of those with CUD.

As a combined effect of initial and terminal investment behavior, sunk cost patterns represent brief sequences of behavior which contrast with approaches that assess decision-making with independent trials (e.g., DD). Also unique to this approach, sequences of decisions can be modeled through time and be used to differentiate how, and to what degree, sub-samples differ. The sequence of initial and terminal investment behavior modeled by the present sunk cost tasks also provides a context conducive to adding terminal investment manipulations such as delay to, or probability of, rewards delivered contingent on completing terminal investments (Zeng, Zhang, Chen, Yu, & Gong, 2013). However, any within-subject manipulations of reward parameters and investment amounts will need to be controlled so that the impact of initial and terminal investment amounts can be experimentally isolated to demonstrate the sunk cost bias. Interestingly, it is unknown whether participants with higher SCPs viewed more costly terminal investments as more valuable than lower cost investments due to the general correspondence between higher-quality goods and services and greater cost. Future research should attempt to address this potential issue.

Our findings suggest that SCP is a behavioral phenomenon that warrants additional exploration in relation to its potential contribution to the initiation and maintenance of cannabis use and CUD. Moreover, SCP may be a decision-making construct associated with cannabis use that could be targeted for change in the context of interventions for those seeking to reduce or quit cannabis use. That said, it is premature to make this assertion based on these initial studies that included only undergraduate students with relatively low frequency cannabis use compared to those diagnosed with CUD. Future studies would benefit by testing this relationship with broader ranges of cannabis use and more heterogeneous demographics.

These initial studies of the association between SCP and cannabis use provide preliminary evidence that an increased propensity to demonstrate the sunk cost bias may be associated with cannabis use. Given the limited evidence of decision-making constructs that consistently relate to cannabis use, the current exploratory experiments suggest that SCP may be a useful alternative behavioral phenomenon for better understanding cannabis use and CUD.

Figure 2.

Figure 2.

Open in a new tab

Depiction of on-screen activity of initial investment (top panel) and terminal investment (bottom panel) contexts during the sunk cost task used in Experiment 2.

Public Significance Statement:

Those who demonstrate over-generalized persistence in endeavors due to past investments of time, effort, or money (i.e., the sunk cost bias) are more likely to focus on the past and perseverate than their counterparts. The present experiments showed that a tendency to exhibit the sunk cost bias was associated with more frequent cannabis use.

Acknowledgments

This research was supported by NIDA grant T32DA037202 which had no other role other than financial support.

The authors would like to thank William Fleming and Tadd Schneider for their help in data collection for the current manuscript.

Footnotes

No data, interpretations, or any other information related to this manuscript have been posted on any website, Listserv, or social media site. No information pertaining to the is manuscript has been shared or presented at a conference or meeting of any kind.

All authors contributed in a significant way to the manuscript and have read and approved the final manuscript.

This manuscript has not been posted as a preprint.

No authors have any conflicts of interest associated with this research.

References

  1. Amlung M, Vedelago L, Acker J, Balodis I, & MacKillop J (2017). Steep delay discounting and addictive behavior: a meta-analysis of continuous associations. Addiction, 112(1), 51–62. doi: 10.1111/add.13535 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Arkes HR, & Ayton P (1999). The sunk cost and concorde effects: are humans less rational than lower animals? Psychological Bulletin, 125(5), 591–600. [Google Scholar]
  3. Arkes HR, & Blumer C (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124–140. [Google Scholar]
  4. Åstebro T, Jeffrey SA, & Adomdza GK (2007). Inventor perseverance after being told to quit: the role of cognitive biases. Journal of Behavioral Decision Making, 20(3), 253–272. [Google Scholar]
  5. Aston ER, Metrik J, Amlung M, Kahler CW, & MacKillop J (2016). Interrelationships between marijuana demand and discounting of delayed rewards: convergence in behavioral economic methods. Drug & Alcohol Dependence, 169, 141–147. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Bickel WK (2015). Discounting of delayed rewards as an endophenotype. Biological Psychiatry, 77(10), 846–847. [DOI] [PubMed] [Google Scholar]
  7. Bickel WK, Jarmolowicz DP, Mueller ET, Koffarnus MN, & Gatchalian KM (2012). Excessive discounting of delayed reinforcers as a trans-disease process contributing to addiction and other disease-related vulnerabilities: emerging evidence. Pharmacology & Therapeutics, 134(3), 287–297. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bickel WK, Koffarnus MN, Moody L, & Wilson GA (2014). The behavioral-and neuro-economic process of temporal discounting: A candidate behavioral marker of addiction. Neuropharmacology, 76, 518–527. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Borodovsky JT, & Budney AJ (2018). Cannabis regulatory science: risk–benefit considerations for mental disorders. International Review of Psychiatry, 1–20. doi: 10.1080/09540261.2018.1454406 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Bruine de Bruin W, Parker AM, & Fischhoff B (2007). Individual differences in adult decision-making competence. Journal of Personality and Social Psychology, 92(5), 938–956. [DOI] [PubMed] [Google Scholar]
  11. Budney AJ (2005). Are specific dependence criteria necessary for different substances: How can research on cannabis inform this issue? Addiction, 101(s1), 125–133. doi: 10.1111/j.1360-0443.2006.01582.x [DOI] [PubMed] [Google Scholar]
  12. CBHSQ. (2018). 2017 National Survey on Drug Use and Health: Detailed Tables. Rockville, MD [Google Scholar]
  13. Collins RL, Parks GA, & Marlatt GA (1985). Social determinants of alcohol consumption: the effects of social interaction and model status on the self-administration of alcohol. J Consult Clin Psychol, 53(2), 189–200. [DOI] [PubMed] [Google Scholar]
  14. Dahlgren MK, Sagar KA, Racine MT, Dreman MW, & Gruber SA (2016). Marijuana use predicts cognitive performance on tasks of executive function. Journal of Studies on Alcohol and Drugs, 77(2), 298–308. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Derogatis LR (2000). The Brief Symptom Inventory-18 (BSI-18): administration, scoring and procedures manual. Minneapolis, MN: National Computer Systems Inc. [Google Scholar]
  16. Du W, Green L, & Myerson J (2002). Cross-cultural comparisons of discounting delayed and probabilistic rewards. The Psychological Record, 52(4), 479–492. [Google Scholar]
  17. Emich KJ, & Pyone JS (2018). Let it go: positive affect attenuates sunk cost bias by enhancing cognitive flexibility. Journal of Consumer Psychology, 28(4), 578–596. [Google Scholar]
  18. Gino F (2008). Do we listen to advice just because we paid for it? the impact of advice cost on its use. Organizational Behavior and Human Decision Processes, 107(2), 234–245. [Google Scholar]
  19. Hafenbrack AC, Kinias Z, & Barsade SG (2014). Debiasing the mind through meditation: mindfulness and the sunk-cost bias. Psychological Science, 25(2), 369–376. [DOI] [PubMed] [Google Scholar]
  20. Hajek P, Belcher M, & Stapleton J (1987). Breath-holding endurance as a predictor of success in smoking cessation. Addictive Behaviors, 12(3), 285–288. [DOI] [PubMed] [Google Scholar]
  21. Hall W (2009). The adverse health effects of cannabis use: what are they, and what are their implications for policy? Int J Drug Policy, 20(6), 458–466. doi: 10.1016/j.drugpo.2009.02.013 [DOI] [PubMed] [Google Scholar]
  22. Hasin D, Saha T, Kerridge B, Goldstein R, Chou P, Zhang H, … Grant B (2015). Prevalence of Marijuana Use Disorders in the United States Between 2001–2002 and 2012–2013. JAMA Psychiatry, 72(12), 1235–1242. doi: 10.1001/jamapsychiatry.2015.1858 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hasin DS, O’Brien CP, Auriacombe M, Borges G, Bucholz K, Budney A, … Grant BF (2013). DSM-5 criteria for substance use disorders: recommendations and rationale. Am J Psychiatry, 170(8), 834–851. doi: 10.1176/appi.ajp.2013.12060782 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Heinz AJ, Peters EN, Boden MT, & Bonn-Miller MO (2013). A comprehensive examination of delay discounting in a clinical sample of Cannabis-dependent military veterans making a self-guided quit attempt. Exp Clin Psychopharmacol, 21(1), 55–65. doi: 10.1037/a0031192 [DOI] [PubMed] [Google Scholar]
  25. Jarmolowicz DP, Bickel WK, Sofis MJ, Hatz LE, & Mueller ET (2016). Sunk costs, psychological symptomology, and help seeking. SpringerPlus, 5(1), 1699. doi: 10.1186/s40064-016-3402-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Johnson MW, & Bickel WK (2008). An algorithm for identifying nonsystematic delay-discounting data. Experimental Clinical Psychopharmacology, 16(3), 264–274. doi: 10.1037/1064-1297.16.3.264 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Johnson MW, Bickel WK, Baker F, Moore BA, Badger GJ, & Budney AJ (2010). Delay discounting in current and former marijuana-dependent individuals. Experimental and Clinical Psychopharmacology, 18(1), 99–107. doi: 10.1037/a0018333 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Koffarnus MN, Jarmolowicz DP, Mueller ET, & Bickel WK (2013). Changing delay discounting in the light of the competing neurobehavioral decision systems theory: a review. J Exp Anal Behav, 99(1), 32–57. doi: 10.1002/jeab.2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Kollins SH (2003). Delay discounting is associated with substance use in college students. Addictive Behaviors, 28(6), 1167–1173. [DOI] [PubMed] [Google Scholar]
  30. Kwako LE, Bickel WK, & Goldman D (2018). Addiction biomarkers: dimensional approaches to understanding addiction. Trends in Molecular Medicine, 24(2), 121–128. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Lane SD, Cherek DR, Tcheremissine OV, Steinberg JL, & Sharon JL (2007). Response perseveration and adaptation in heavy marijuana-smoking adolescents. Addict Behav, 32(5), 977–990. [DOI] [PubMed] [Google Scholar]
  32. Lee J, Keil M, & Wong K (2018). Does a Tired Mind Help Avoid a Decision Bias? The Effect of Ego Depletion on Escalation of Commitment. Applied Psychology, 67(1), 171–185. [Google Scholar]
  33. Lejuez C, Kahler C, & Brown R (2003). A modified computer version of the Paced Auditory Serial Addition Task (PASAT) as a laboratory-based stressor. The Behavior Therapist, 26(4), 290–293. [Google Scholar]
  34. MacKillop J (2013). Integrating behavioral cconomics and behavioral genetics: delayed reward discounting as an endophenotype for addictive disorders. Journal of the Experimental Analysis of Behavior, 99(1), 14–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Moody L, Franck C, Hatz L, & Bickel WK (2016). Impulsivity and polysubstance use: a systematic comparison of delay discounting in mono, dual, and tri-substance use. Experimental and Clinical Psychopharmacology, 24(1), 30–37. doi: 10.1037/pha0000059 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Myerson J, Green L, & Warusawitharana M (2001). Area under the curve as a measure of discounting. Journal of the Experimental Analysis of Behavior, 76(2), 235–243. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. NIAAA. (2017). Women and Alcohol. National Institute of Health; Retrieved from https://pubs.niaaa.nih.gov/publications/womensfact/womensfact.htm [Google Scholar]
  38. Okafor C, Plankey M, Li M, Chen X, Surkan P, Shoptaw S, … Cook R (2019). Association of Marijuana Use with Changes in Cognitive Processing Speed and Flexibility for 17 Years in HIV-Seropositive and HIV-Seronegative Men AU - Okafor, Chukwuemeka N. Substance Use & Misuse, 1–13. doi: 10.1080/10826084.2018.1495736 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Rego S, Arantes J, & Magalhães P (2018). Is there a Sunk Cost Effect in Committed Relationships? Current Psychology, 37(3), 508–519. [Google Scholar]
  40. Rosenfarb IS, Newland CM, Brannon SE, & Howey DS (1992). Effects of self‐ generated rules on the development of schedule‐controlled behavior. Journal of the Experimental Analysis of Behavior, 58(1), 107–121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Siniver E, Mealem Y, & Yaniv G (2013). Overeating in all-you-can-eat buffet: paying before versus paying after. Applied Economics, 45(35), 4940–4948. [Google Scholar]
  42. Sobell LC, Sobell MB, Litten R, & Allen J (1992). In Timeline follow-back: A technique for assessing self-reported alcohol consumption (pp. 41–72). Totowa, NJ: Humana Press. [Google Scholar]
  43. Sofis MJ, Jarmolowicz DP, Hudnall JL, & Reed DD (2015). On sunk costs and escalation. The Psychological Record, 65(3), 487–494. [Google Scholar]
  44. Stanger C, Ryan S, Fu H, Landes R, Jones B, Bickel W, & Budney A (2012). Delay discounting predicts treatment outcome among adolescent substance abusers. Experimental and Clinical Psychopharmacology, 20, 205–212. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Stanovic KE, & West RF (1998). Individual differences in framing and conjunction effects. Thinking and Reasoning, 4(4), 289–317. [Google Scholar]
  46. Stanovic KE, & West RF (2000). Individual differences in reasoning: implications for the rationality debate? Behavioral and Brain Sciences, 23(5), 645–665. [DOI] [PubMed] [Google Scholar]
  47. Stiegler MP, & Ruskin KJ (2012). Decision-making and safety in anesthesiology. Current Opinion in Anaesthesiology, 25(6), 724–729. [DOI] [PubMed] [Google Scholar]
  48. Strickland JC, Lile JA, & Stoops WW (2017). Unique prediction of cannabis use severity and behaviors by delay discounting and behavioral economic demand. Behav Processes, 140, 33–40. doi: 10.1016/j.beproc.2017.03.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Strough J, Schlosnagle L, & DiDonato L (2011). Understanding decisions about sunk costs from older and younger adults’ perspectives. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 66B(6), 681–686. [DOI] [PubMed] [Google Scholar]
  50. van Putten M, Zeelenberg M, & van Dijk E (2010). Who throws good money after bad? Action vs. state orientation moderates the sunk cost fallacy. Judgment and Decision Making, 5(1), 33–36. [Google Scholar]
  51. VanderBroek L, Acker J, Palmer AA, de Wit H, & MacKillop J (2016). Interrelationships among parental family history of substance misuse, delay discounting, and personal substance use. Psychopharmacology, 233(1), 39–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Volkow N, Baler R, Compton W, & Weiss S (2014). Adverse Health Effects of Marijuana Use. The New England Journal of Medicine, 370(23), 2219–2227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Wang J, Kelly BC, Booth BM, Falck RS, Leukefeld C, & Carlson RG (2010). Examining factorial structure and measurement invariance of the Brief Symptom Inventory (BSI)-18 among drug users. Addict Behav, 35(1), 23–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Wong K, & Kwong J (2007). The role of anticipated regret in escalation of commitment. Journal of Applied Psychology, 92(2), 545–554. [DOI] [PubMed] [Google Scholar]
  55. Zeng J, Zhang Q, Chen C, Yu R, & Gong Q (2013). An fMRI study on sunk cost effect. Brain Res Mol Brain Res, 1519, 63–70. [DOI] [PubMed] [Google Scholar]

RESOURCES