Adaptation of the Monetary Choice Questionnaire to accommodate extreme monetary discounting in cocaine users

Abstract

Delay discounting, which refers to the phenomenon that rewards decrease in subjective value as the delay associated with their receipt increases, is a paradigm that has been used extensively in substance abuse research to understand impulsive decision making. One common measure to assess delay discounting is the Monetary Choice Questionnaire (MCQ) developed by Kirby, Petry, and Bickel (1999). While the MCQ has great utility because of its simplicity and brief administration time, it is possible that the MCQ produces a ceiling effect in estimating delay discounting parameters in highly impulsive individuals. In the present study, we adapted the MCQ to attempt to address this ceiling effect by extending the original scale with 9 items, and we then compared scores on the original MCQ to the extended MCQ in a sample of active cocaine users. The ceiling effect, while observed in the original MCQ scores for over a quarter of the sample, was largely eliminated with the extended scale. Highly impulsive participants, whose scores on the extended scale exceeded the highest possible score on the original scale, had higher levels of sensation seeking compared to other participants, but not trait impulsivity. The extended MCQ may be useful in populations with high rates of impulsivity, where the original measure may underestimate discounting rates due to a ceiling effect.

INTRODUCTION

Delay discounting describes to the phenomenon that rewards decrease in subjective value as the delay associated with their receipt increases (Mazur, 1987; Rachlin & Green, 1972). Highly impulsive individuals exhibit high rates of delay discounting in laboratory tasks, meaning that they show preference for smaller, immediate rewards over larger, delayed rewards. Research has demonstrated that higher rates of delay discounting are associated with various health risk behaviors, including substance use, problematic gambling, and HIV transmission risk behaviors (Bickel & Marsch, 2001; Chesson et al., 2006; Odum, Madden, Badger, & Bickel, 2000; Reynolds, 2006).

Delay discounting has been utilized extensively as a paradigm for impulsivity in substance users, because increased delay discounting theoretically reflects a fundamental decision making process present in substance using populations (Reynolds, 2006). While decision making is a complex process, part of the phenomenology of substance dependence is that substance-dependent persons often make choices that contradict their stated goals. For example, a person may commit to quitting smoking one day, only to buy a pack of cigarettes the following day.

Delay discounting may be a key factor in how that decision is reached. For those with high rates of delay discounting, more distal rewards, like achieving 30 days of sobriety or having funds to pay bills at the end of the month, are underweighted compared to immediate rewards, such as substance use. Delay discounting has been shown to correlate with other measures of impulsivity, including trait impulsivity and sensation seeking (Caswell, Bond, Duka, & Morgan, 2015; Koff & Lucas, 2011). However other research has shown weak or inconsisent associations (Mitchell, 1999; Vuchinich & Simpson, 1998). Research has demonstrated that higher delay discounting rates are associated with addictive behaviors across many different substances, including nicotine, cocaine, opiates, and alcohol compared to non-drug using controls (MacKillop et al., 2011). Cocaine users in particular have demonstrated greater delay discounting compared to non-drug users (Garcia-Rodriguez, Secades-Villa, Weidberg, & Yoon, 2013; Heil, Johnson, Higgins, & Bickel, 2006; Kirby & Petry, 2004).

One measure that has been successfully used to assess delay discounting in the laboratory is the Monetary Choice Questionnaire (MCQ) (Kirby et al., 1999). The MCQ asks participants to make 27 choices between smaller immediate rewards versus larger delayed rewards (Kirby et al., 1999). By examining the pattern of responses, one can infer a participant’s rate of delay discounting or “k value.” Kirby et al. (1999) found that MCQ scores were correlated with other measures of impulsivity such as the Barratt Impulsiveness Scale, and mean MCQ scores were higher among the substance-users compared to non-drug using controls. Additionally, research has shown that the scores from the MCQ correlate highly with scores from traditional measures of delay discounting, such as adjusting-amount procedures (Epstein et al., 2003). Taken together, this research suggests that the MCQ is a valid means of measuring delay discounting. Given its simplicity and brevity, the MCQ is an ideal measure to administer in time-limited research settings where use of lengthier behavioral tasks to assess delay discounting is not feasible.

Highly impulsive persons may exhibit an extreme preference for immediate rewards. On the MCQ, these individuals select the immediate reward in all 27 items, resulting in a ceiling effect. In these cases, the 27-item MCQ may underestimate the rate of delay discounting. This ceiling effect has been observed in prior research using the MCQ with substance users. For example, in one sample of alcohol and cocaine users, participants always selected the immediate reward in approximately 22% of the MCQs administered (Black & Rosen, 2011). In addition, a ceiling effect has been observed in other measures of delay discounting among highly impulsive samples and has been cited as a contributing factor to spurious correlations (Johnson, Bickel, & Baker, 2007; Petry & Casarella, 1999; Yoon et al., 2007). Thus, this ceiling effect is of concern when working with populations who exhibit high impulsivity, including stimulant users who engage in high rates of risky behaviors.

To address this ceiling effect, we adapted the MCQ to accommodate extreme discounting by adding 9 items to extend the scale. The purpose of this paper is to describe the adaptation process and to compare performance on the original versus adapted versions of the MCQ in a sample of cocaine users.

METHODS

Participants and procedures

MCQ data was collected as part of a larger study examining the neurocognitive effects of HIV and cocaine use (Meade, Towe, Skalski, & Robertson, 2015). The community-based sample included 101 adult cocaine users who met the following inclusion criteria: (1) ≥4 days of cocaine use in the past month or a positive urine drug screen for cocaine, (2) ≥1 year of regular cocaine use, and (3) lifetime cocaine dependence. Alcohol, marijuana, and nicotine use were permitted, and current alcohol and marijuana dependence were permitted if cocaine dependence was the principal diagnosis. For other drugs, individuals were excluded for lifetime abuse or dependence, history of regular use, any use in the past year, and/or a positive drug screen. Additional exclusion criteria were: English non-fluency or illiteracy; <9^th grade education; serious neurological disorders (e.g., seizure disorder, cryptococcal meningitis) or severe head trauma; severe mental illness; pregnancy; physical disabilities impeding participation (e.g., blindness); and impaired mental status.

Participants were recruited from [MASKED FOR REVIEW] between May 2010 and May 2015 via advertisements in local newspapers and websites, flyers and brochures at community-based organizations and clinics, and participant referrals. All potential participants completed a structured telephone screen to assess preliminary eligibility (e.g., HIV infection, drug use history), and interested individuals were then invited for a comprehensive in-person screening.

After providing written informed consent, participants were given a breathalyzer test to ensure sobriety, and they provided a urine sample for drug and pregnancy screening. Self-reported HIV-positive status was verified by medical record review, and HIV-negative status was confirmed by an OraQuick© rapid HIV test. Participants then completed clinical interviews and questionnaires. Eligible participants returned on another day to complete a neurocognitive assessment that included the MCQ, additional clinical interviews and questionnaires, and another urine drug test. All questionnaires were computerized using audio computer-assisted self-interview (ACASI).

Participants were paid $35 for the screening visit, regardless of eligibility, and $65 for the neurocognitive assessment. All procedures were approved by the institutional review boards at [MASKED FOR REVIEW].

Measures

Original Monetary Choice Questionnaire (MCQ-27)

Participants are presented with choices between smaller, immediate rewards and larger, delayed rewards (e.g., “Would you prefer $54 today or $80 in 30 days?) (Kirby et al., 1999). The MCQ-27 includes a fixed set of 27 items with immediate rewards ranging from $11-78 and delayed rewards ranging from $25-85 with a delay of 7-186 days. Delayed rewards are grouped into 3 categories based on size, with 9 items per category: small ($25-35), medium ($50-60), and large ($75-85).

As described by Kirby and colleagues (1999), participants’ hyperbolic discount parameter (k value) is determined by fitting data to the following discount function equation: ${\mathrm{V}}_{\text{immediate}}={\mathrm{V}}_{\text{delayed}}/(1+k\mathrm{D})$, in which V is the reward value in dollars and D is delay in days (Mazur, 1984). Values of k range from 0.00016 to 0.25 for the MCQ-27, with higher values indicating a greater preference for smaller, immediate rewards over larger, delayed rewards. Each possible k value increases by an order of approximately 2.5, resulting in a logarithmic scale. K values are estimated by taking the geometric midpoint between the discount rates associated with each item and then examining the participant’s pattern of responses across trials to determine which k value is most consistent with the response pattern. By examining the pattern of responses in this way, one can infer a participant’s point of indifference between delayed and immediate rewards. To determine the most consistent value, the proportion of a participant’s choices that are consistent with each k value is calculated. The k value that yields the highest proportion is the value assigned to the participant. If two or more values have the same proportion, the participant’s assigned k value is the geometric mean of those values. Because raw k values tend to be highly skewed, k values are normalized using the natural logarithm transformation. K values are also ranked from 1 to 10. Rank 1 and rank 10 represent the lowest and highest possible k values, respectively (e.g., when a participant selects all delayed rewards or all immediate rewards across trials), and ranks 2 through 9 are assigned to the ranges of discount rates between items (e.g., values between 0.00016 to 0.00040 were rank 2).

Extended Monetary Choice Questionnaire (MCQ-36)

To expand the MCQ-27, nine items with higher associated k values were added. Three new items were added to each category (small, medium, large), resulting in 12 items per category and a total of 36 items for the MCQ-36. Additional k values for the extended scale were computed by continuing the logarithmic scale of the original MCQ-27, resulting in new k values of 0.625, 1.5625, and 3.90625. The new items included delay periods lasting 1–7 days and utilized the same 3 delayed rewards as the other items in the category. The immediate reward for each additional item was calculated based on the standard discount function equation, resulting in immediate rewards for these new items ranging from $5-17. Dollar amounts were rounded to the nearest whole dollar, and then the exact k for that dollar amount was calculated with the formula. This resulted in k values for the MCQ-36 ranging from 0.00016 to 4.00. K values were also natural log transformed and ranked for the MCQ-36, with ranks ranging from 1 to 13. Table 1 shows the items included in the MCQ-36 and their calculated k values.

Table 1.

Items in the MCQ-36

Reward Size	Amount today	Amount later	Delay in days	Calculated k valuea
Small	34	35	186	0.00016
Small	28	30	179	0.00040
Small	22	25	136	0.00100
Small	25	30	80	0.00250
Small	19	25	53	0.00596
Small	24	35	29	0.01580
Small	14	25	19	0.04135
Small	15	35	13	0.10256
Small	11	30	7	0.24675
Smallb	6	25	5	0.63333
Smallb	5	30	3	1.66667
Smallb	7	35	1	4.00000
Medium	54	55	117	0.00016
Medium	47	50	160	0.00040
Medium	54	60	111	0.00100
Medium	49	60	89	0.00252
Medium	40	55	62	0.00605
Medium	34	50	30	0.01569
Medium	27	50	21	0.04056
Medium	25	60	14	0.10000
Medium	20	55	7	0.25000
Mediumb	12	50	5	0.63333
Mediumb	13	55	2	1.61538
Mediumb	12	60	1	4.00000
Large	78	80	162	0.00016
Large	80	85	157	0.00040
Large	67	75	119	0.00100
Large	69	85	91	0.00255
Large	55	75	61	0.00596
Large	54	80	30	0.01605
Large	41	75	20	0.04146
Large	33	80	14	0.10173
Large	31	85	7	0.24885
Largeb	17	80	6	0.61765
Largeb	13	75	3	1.58974
Largeb	17	85	1	4.00000

^aExact values calculated using the standard discount function equation, ${\mathrm{V}}_{\text{immediate}}={\mathrm{V}}_{\text{delayed}}/(1+k\mathrm{D})$, in which k represents the k value, V is the reward value in dollars and D is delay in days.

^bItem added to the MCQ-36 that was not included in the MCQ-27.

The MCQ-36 was computerized using ePrime (Psychology Software Tools, Inc., http://www.pstnet.com). Choices were presented in random order, and participants indicated their response using a computer mouse. Participants completed the full MCQ-36, and then scores were generated for the original MCQ-27 and the extended MCQ-36. Participants were told that all rewards were hypothetical, and participants did not receive any money based on their performance on the task.

Barratt Impulsiveness Scale–Version 11 (BIS-11)

At the neurocognitive assessment, participants completed the BIS-11 (Patton, Stanford, & Barratt, 1995), which is one of the oldest and most widely used self-report measures of trait impulsivity (Stanford et al., 2009). The BIS-11 includes 30 items that describe common impulsive or non-impulsive behaviors and preferences. Participants rate each items using a 4-point scale (rarely/never, occasionally, often, almost always/always), and higher scores indicate greater impulsiveness. All 30 items are summed to yield a total score, ranging from 30 to 120, and there are 6 subscales based on first-order factors determined by a principal component analysis (Patton et al., 1995). The 6 subscales are Attention, Motor Impulsiveness, Self-control, Cognitive Complexity, Perseverance, and Cognitive Instability.

Sensation Seeking Scale–Version V (SSS-V)

Participants also completed the SSS-V (Zuckerman, 1994; Zuckerman, Eysenck, & Eysenck, 1978). In this 40-item, forced-choice questionnaire, participants choose which of 2 statements best applies to them. This inventory was developed to measure individual differences in stimulation and arousal needs and is thought to correlate with impulsivity. Reliability and construct validity for this instrument has been well-established (Zuckerman, 1994; Zuckerman et al., 1978). Each statement is scored as either 0 or 1, and then items are summed to create a total score, ranging from 0 to 40. In addition to a total score, the SSS-V produces 4 subscales that represent different dimensions of sensation seeking: Thrill and Adventure Seeking, Experience Seeking, Disinhibition, and Boredom Susceptibility.

Other measures

Demographic characteristics and detailed assessments of substance use were completed at the screening visit. Interviews included Module E of the Structured Clinical Interview for DSM-IV-TR to assess for lifetime cocaine dependence and other substance use disorders (First, Spitzer, Gibbon, & Williams, 1996) and the Addiction Severity Index-Lite to assess lifetime substance use and associated impairments (McLellan et al., 1992). Frequency of substance use in the past 30 days was assessed using Timeline follow-back methodology (Robinson, Sobell, Sobell, & Leo, 2014; Sobell & Sobell, 1996). A urine toxicology screen for cocaine, cannabis, amphetamine, methamphetamine, oxycodone, methadone, other opioids (including heroin), benzodiazepines, and barbiturates was used to corroborate self-report. Premorbid verbal IQ was estimated using the Wechsler Test of Adult Reading (WTAR), in which participants read 50 words aloud that have atypical grapheme to phoneme translations (Wechsler, 2001). Finally, participants completed a computerized survey that assessed demographics.

Participants completed the Brief Symptom Inventory 18 (BSI-18) at the neurocognitive assessment. The BSI-18 is an 18-item questionnaire which assesses psychological distress. Participants rate their level of distress associated with 18 symptoms over the past week using a 5-point Likert scale (0 Not at all to 4 Extremely) (Derogatis, 1993). An overall score is created by calculating the mean of all 18 items. The HIV Risk Behavior Scale (HRBS), a brief interview, was also completed at the neurocognitive assessment. The HRBS, a well-validated measure, assesses frequency of drug risk behaviors (6 items) and sex risk behaviors (6 items) in the past 30 days (Darke, Hall, Heather, Ward, & Wodak, 1991). For the HRBS, items are summed to create a total score.

Quality assurance procedures

All MCQ data were evaluated to ensure that participants responded in a consistent manner and appeared to understand the task. Within each category of delayed rewards (small, medium, large), consistency was evaluated by examining the highest proportion of consistent choices. Cases where ties resulted (e.g., 2 or more k values have equally high proportions) were considered indicative of potentially inconsistent responding because a tie indicates that multiple k values are equally likely to represent a participant’s true point of indifference. While ties are generally handled by taking the geometric mean of the k values, this method has potential for bias when the pattern of responding is inconsistent. Therefore, we established more rigorous standardized criteria to evaluate when ties result in k value estimates that are potentially not valid. The k value within a category was excluded if either of the following criteria were met: (1) there was a tie for the highest proportion between 2 k values and those k values were separated by more than 1 other k value, or (2) there was a tie between 3 or more k values for the highest proportion. This first criterion was selected because the geometric mean of two tied k values which are separated by only 1 other k value likely does closely approximate a participant’s actual point of indifference, but the geometric mean of 2 highly discrepant k values (i.e., separated by more than 1 other k value) may not provide as accurate an approximation. For example, responses in the large category for one participant resulted in a 2-way tie between k values of rank 6 and rank 10, and a geometric mean between these 2 values would result in a k value of rank 8. If one of these responses was an error or mistake by the participant (due to lack of comprehension or response imputation error), a rank of 8 would not be as closely representative of the participant’s actual point of indifference. The second criterion was selected because 3- and 4-way ties generally result in very low proportions. For example, responses in the small reward category for one case resulted in a 4-way tie, with each k value having a proportion of 0.58. When no exclusion rule was met, the k values within each category were considered valid.

Cases were also evaluated for overall consistency. Cases were marked as overall inconsistent when either: (1) k value proportions were less than or equal to 0.80 in at least 2 categories, (2) proportions in all 3 categories resulted in ties, and (3) k values across 2 or more categories were excluded during scoring. These criteria were selected because they each indicate that inconsistent responding occurred across 2 or more reward categories. When a case was marked as overall inconsistent, all k values for that case were excluded.

Using these procedures, 2 cases (2%) were marked as highly inconsistent overall and excluded from analysis, resulting in 99 cases with valid MCQ data. In the individual reward categories for the remaining 99 cases, no cases were marked as invalid in small, 1 case was marked as invalid in medium, and 2 cases were marked as invalid in large. For those cases, that category value was considered invalid, but the other two categories were marked as valid and therefore the case was retained for analysis.

Data analysis plan

Quantitative analysis was conducted using SPSS 21.0.0 (SPSS Inc., Chicago, IL). Descriptive statistics were used to characterize the sample and determine the geometric mean for the raw k value, mean ln k value, mean rank value, and the total number of cases with the maximum score for both the MCQ-27 and MCQ-36. Participants whose k value on the MCQ-36 exceeded the maximum possible k value of 0.25 on the MCQ-27 were classified as extreme responders. Independent sample t-tests and chi-squared tests were conducted to identify differences in demographic and other relevant variables across extreme responders and non-extreme responders. Variables identified as significantly different between groups and relevant demographic characteristics (age, race, gender, and years of education) were included as covariates in the remaining analyses. Univariate analyses of co-variance (ANCOVAs) were performed to examine differences between extreme versus non-extreme responders on the BIS, SSS, and other relevant variables.

RESULTS

Participant characteristics

The majority of participants were male (66%) and African American (92%). Participants were 46.29 years old on average (SD = 8.46) and 49% of the sample was HIV-positive. Approximately 24% of the sample identified as gay, lesbian, or bisexual. Participants reported 12.24 years of education on average (SD = 2.40) and 84% reported being unemployed. Participants had used cocaine on average 10.50 of the past 30 days (SD = 7.64) and reported a mean of 17.80 years of regular cocaine use (SD = 7.48). The vast majority (88%) reported smoking as their usual route of cocaine administration, and 85% of the sample tested positive for cocaine in a urine drug screen completed on the day the MCQ was administered.

MCQ-27 and MCQ-36 descriptive characteristics

Table 2 presents geometric means of raw k values, mean ln k values, and mean rank values for the MCQ-27 and the MCQ-36 overall and within each reward category. With the extended scale, the MCQ-36 produced higher k values than the MCQ-27 overall and across each category. Figure 1 shows the number of participants with each rank value in the MCQ-36. A total of 26 participants (26%) were assigned a k value of > 0.25 in either their overall mean or within at least one individual reward size category on the MCQ-36. Seventeen participants (17%) were assigned a k value of > 0.25 for their overall mean k value. Eighteen participants (18%) received a k value of > 0.25 in the small category, 15 (15%) in medium, and 16 (16%) in large. A total of 4 participants (4%) were assigned the maximum possible k value of 4.00 in either their overall mean or within at least one individual reward size category. Two participants (2%) were assigned the maximum k value of 4.00 for their overall mean, with 2 participants (2%) receiving the maximum value in the small category, 2 (2%) in the medium category, and 4 (4%) in the large category.

Table 2.

Average k values, ln k values, and ranks for the MCQ-27 and MCQ-36

Reward Size	K value Geo. Mean (95% CI)a		ln k value M (SD)		Ranked k value M (SD)
	MCQ-27	MCQ-36	MCQ-27	MCQ-36	MCQ-27	MCQ-36
Small N = 99	0.06 (0.05–0.08)	0.08 (0.05–0.10)	−2.76 (1.34)	−2.57 (1.62)	8.08 (1.58)	8.21 (1.81)
Medium N = 98	0.05 (0.04–0.06)	0.06 (0.04–0.08)	−3.00 (1.35)	−2.84 (1.61)	7.82 (1.58)	7.93 (1.80)
Large N = 97	0.04 (0.03–0.05)	0.05 (0.03–0.07)	−3.20 (1.38)	−2.97 (1.80)	7.59 (1.63)	7.78 (2.02)
Overall mean N = 99	0.05 (0.04–0.06)	0.06 (0.04–0.08)	−2.98 (1.20)	−2.78 (1.51)	7.84 (1.43)	7.99 (1.69)

^aGeo. Mean = Geometric Mean of raw k values, 95% CI = 95% confidence intervals for each geometric mean.

Figure 1.

Histograms of ranked k-values across reward categories and by overall mean for the MCQ-36. Shaded bars represent ranks from the original MCQ-27 items. Additional ranks from the extended k value scale included in the MCQ-36 are unshaded.

Comparison of extreme responders to non-extreme responders

We compared the 26 extreme responders to the 73 non-extreme responders on other participant and substance use characteristics (e.g., occupational status, days of cocaine use) to identify other potential control variables. A significantly higher proportion of non-extreme responders (89%) were unemployed compared to extreme responders (69%) for occupational status (χ²(1, N = 99) = 5.55, p = 0.018). There was also a significant difference between groups on sexual orientation (χ²(1, N = 99) = 3.88, p = 0.049), such that a larger proportion (39%) of extreme responders identified as gay, lesbian, or bisexual compared to non-extreme responders (19%). There were no other significant differences between groups. Therefore, in addition to age, race, gender, and years of education, occupational status and sexual orientation were included as covariates in the subsequent ANCOVAs.

Table 3 presents the adjusted means and F-values for each ANCOVA. Several group differences were apparent on the SSS-V, with extreme responders having higher scores overall and on the experience seeking and disinhibition subscales. On the BIS-11, groups did not significantly differ across the total or subscale scores. Groups also did not differ on premorbid verbal IQ, HRBS score, or BSI total score. To reduce the risk of a Type I error due to multiple comparisons, we applied a familywise Bonferroni adjustment to the significance level for these ANCOVAs so that significance levels were set at p < 0.003. No results were significant using this adjusted significance level. Traditional p values (i.e., p < 0.05, 0.01, and 0.001) are noted in Table 3.

Table 3.

Comparison of extreme responders to non-extreme responders using ANCOVA

Measure	Adjusted means		F value
	Non-extreme responders N=73	Extreme responders N=26
Barratt Impulsiveness Scale
BIS-11 Total Score	65.18	65.52	F(1,91) = 0.02
Self-control	14.69	15.60	F(1,91) = 1.40
Motor Impulsiveness	14.51	13.71	F(1,91) = 1.40
Attention	9.93	10.28	F(1,91) = 0.29
Cognitive Instability	4.66	5.04	F(1,91) = 1.04
Cognitive Complexity	13.16	13.12	F(1,91) = 0.01
Perseverance	8.23	7.78	F(1,91) = 0.87
Sensation Seeking Scale
SSS-V Total Score	14.22	17.04	F(1,91) = 5.14*
Thrill and Adventure Seeking	3.23	3.52	F(1,91) = 0.24
Experience Seeking	4.62	5.56	F(1,91) = 4.21*
Disinhibition	4.14	5.48	F(1,91) = 8.27**
Boredom Susceptibility	2.23	2.47	F(1,91) = 0.37
Estimated premorbid verbal IQ	86.83	82.71	F(1,90) = 2.04
HRBS Sex Composite	3.37	4.34	F(1,91) = 1.19
BSI Total Score	0.49	0.53	F(1,91) = 0.11

Notes. The covariates included were age, race, gender, years of education, sexual orientation, and occupational status. Means presented represent the adjusted means after controlling for all covariates in the ANCOVA.

BIS-11 point scales: Total: 30–120, Self-control: 6–24; Motor Impulsiveness: 7–28; Attention: 5–20; Cognitive Instability: 3–12; Cognitive Complexity: 5–20; Perseverance: 4–16.

SSS-V point scales: Total: 0–40; all subscales: 0–10.

^*p < .05,

^**p < .01,

^***p < .001

DISCUSSION

In this sample of active cocaine users, the modified MCQ-36 appeared to be successful in reducing a ceiling effect on k scores compared to the MCQ-27. There was a clear ceiling effect using the original MCQ-27 items, with over a quarter of participants producing k values that exceeded the maximum possible k value. This ceiling effect was reduced substantially with the MCQ-36 to only 4 participants receiving the maximum possible k value. Based on these results, use of the extended MCQ-36 may be warranted in populations with high rates of impulsivity, such as out-of-treatment cocaine users, in order to get a more accurate estimate of discounting parameters.

There were some notable differences between extreme responders and non-extreme responders on the SSS, though these were not statistically significant after correcting for multiple comparisons using the Bonferroni adjustment. Overall, extreme responders had higher levels of sensation seeking compared to non-extreme responders, and showed particularly high levels of sensation seeking in Disinhibition and Experience Seeking. Disinhibition reflects both social and sexual disinhibition that is expressed in substance-use behaviors and variety in sexual partners (Zuckerman et al., 1978). Experience seeking represents the pursuit of new sensations and experiences through the mind and senses, and activities such as travel and a nonconforming life-style (Zuckerman et al., 1978).

There were few differences between extreme responders and non-extreme responders on cocaine use, trait impulsivity, and many other participant characteristics. The lack of differences on cocaine use variables is very likely the result of strict eligibility criteria for the sample related to frequency and severity of cocaine use. Use of these strict criteria resulted in a relatively homogeneous sample in terms of substance use characteristics.

The findings from this study have important implications for measuring delay discounting in highly impulsive populations. In our results, scores based on the MCQ-27 yielded k values that underestimated delay discounting due to a ceiling effect for a large proportion of our participants. Other studies using the 27-item MCQ with substance users reported mean k values similar to those in the current study sample (Gonzalez et al., 2012; Kirby & Petry, 2004). Therefore, it is possible that others have found a similar ceiling effect, although it has not been reported. The MCQ-36 represents a way to assess a wider range of delay discounting in highly impulsive individuals that captures variability that would otherwise be missed by the shorter MCQ-27. Furthermore, while yielding a potentially more accurate estimate of delay discounting, the addition of 9 items in the MCQ-36 did not substantially increase the length of time of administration. Future research might compare this extended MCQ against other validated measures of delay discounting. Additionally, future research could examine whether scores on the MCQ-36 are more predictive of risk behaviors.