Bidirectional Prospective Associations between Behavioral Economic Indicators and Drinking Patterns During Alcohol Use Disorder Natural Recovery Attempts

Abstract

Objective.

Behavioral economic (BE) theory posits that harmful alcohol use is a joint product of elevated alcohol demand and preference for immediate over delayed rewards. Despite cross-sectional research support, whether expected bidirectional relations exist between BE indicators and drinking during recovery attempts is unknown. Therefore, this prospective research investigated quarter-by-quarter cross-lagged associations between BE simulation tasks and drinking following a natural recovery attempt. Higher demand and discounting in a given quarter should predict subsequent drinking. Conversely, drinking in a given quarter should predict subsequent higher demand and discounting.

Method.

Community-dwelling problem drinkers were enrolled shortly after stopping heavy drinking without treatment (N = 191). Drinking practices, problems, delay discounting, and alcohol demand (intensity, O_max, P_max, elasticity) were assessed at baseline and 3, 6, 9, and 12-month follow-ups. Longitudinal cross-lagged models related each BE indicator in the previous quarter to drinking status in the next quarter, and vice versa.

Results.

Higher demand intensity (consumption when drinks are free) at quarter 1 distinguished participants who drank heavily in quarter 2 from those who abstained. In turn, heavy drinking participants in quarter 2 had higher intensity at quarter 3 than abstainers and moderate drinkers in quarter 2, and higher intensity at quarter 3 distinguished heavy drinkers in quarter 4 from moderate drinkers (ps <.05). Hypothesized associations for other BE indices were inconsistent or partially supported.

Conclusions.

APT metrics showed some hypothesized prospective associations with drinking during a natural recovery attempt, which supports their ecological validity as relapse risk indicators.

Behavioral economic (BE) models characterize excessive substance use as a pattern of discrete choices involving less valuable, more immediate options (e.g., drinking) over engaging in behavior patterns with delayed positive consequences that are more valuable overall and more conducive to recovery (Bickel et al., 2014; Vuchinich & Heather, 2003). Conversely, recovery is characterized as requiring a temporal shift from a shorter to a longer view of the future and shifting resource allocation away from substance use toward valuable delayed non-drinking rewards, which reinforce and stabilize the shifted recovery behavior patterns (Tucker, 2022).

Behavioral economic theory and research, particularly work guided by the reinforcer pathology model (Bickel & Athamneh, 2020; Bickel et al., 2014), emphasize two factors as influential in affecting changes in patterns of substance use. Alcohol or drugs are consumed because of (1) their high relative reinforcing value compared to other activities and commodities (high drug demand), and (2) a decision-making bias that overvalues smaller immediate rewards relative to larger delayed rewards (high delay discounting). Cross-sectional studies have supported these hypotheses. Substance-related risk groups, including both treatment and non-treatment-seeking samples, show higher demand for substances and greater delay discounting than control groups who are not engaged in risky substance use (Acuff et al., 2020, 2022; Amlung et al., 2017).

Most studies finding such associations used simulation tasks involving choices among hypothetical rewards available at varying prices and/or delays, which tend to show relationships similar to tasks involving the purchase of real commodities (Amlung & MacKillop, 2015). Alcohol demand, for example, is typically assessed using the Alcohol Purchase Task (APT; Murphy & MacKillop, 2006) during which individuals report how many drinks they would purchase during a hypothetical drinking scenario across a series of increasing prices. Choices yield a demand curve from which multiple indices of reinforcer value or demand can be computed, including intensity when substances are free, O_max or the maximum amount spent on substances, P_max or the price at which O_max is reached, and elasticity or the rate of change in substance demand as a function of price (Acuff & Murphy, 2017). Most versions of hypothetical delay discounting tasks assess preferences for immediate versus delayed rewards over a series of smaller, sooner and larger, later rewards (e.g., Kirby & Marakovic, 1996; Richards et al., 1999). Metrics computed based on the resulting discount curve (e.g., the slope of the curve, or k-parameter) reflect individual differences in the devaluation of delayed rewards.

Simulation tasks provide brief efficient methods to assess BE-substance use associations that theoretical accounts regard as potential mechanisms of addictive behavior change (Bickel & Athamneh, 2020; Bickel et al., 2014). To date, however, the bulk of research has used simulation tasks to investigate whether trait-like levels of demand and discounting predict concurrent or future substance use typically assessed using “dipstick” summary measures of treatment outcomes at 6 month or longer follow-ups. Few studies (e.g., Dennhardt et al., 2015; Murphy et al., 2019) investigated associations prospectively to examine temporal associations over multiple shorter intervals during active addictive behavior change attempts. For example, Murphy et al. (2019) found support for the role of substance-free reinforcement as a mechanism of change in an evaluation of a brief BE alcohol intervention for college students that included multiple assessments over a 16-month follow-up. Reducing the proportion of total reinforcement associated with alcohol use mediated the intervention effect on alcohol use and alcohol problems, suggesting that reward-related mechanisms account for change even if not directly targeted by an intervention.

In general, if reinforcer pathology variables function as mechanisms of addictive behavior change, dynamic bidirectional associations should exist between BE and substance use variables when both are changing over time during behavior change attempts. Higher demand and steeper discounting should predict substance use (Bickel & Athamneh, 2020), and substance use should increase demand and discounting (de Wit & Mitchell, 2010). Much extant research supports the former association (Amlung et al., 2017), and more limited research supports the latter association, with findings being more consistent for demand and mixed for discounting; e.g., alcohol consumption increased demand assessed by the APT in experimental (Motschman, et al., 2022) and naturalistic (Motschman et al., 2021) research, but substance use among human and nonhuman participants variously increased, decreased, or did not affect discounting across a range of drugs and doses (de Wit & Mitchell, 2010).

To our knowledge, to date no single prospective study has evaluated such cross-lagged bidirectional associations between BE variables and drinking status during active behavior change, which is important in evaluating reinforcer pathology model variables as mechanisms of addictive behavior change and the theory more broadly. Therefore, the present study investigated cross-lagged bidirectional associations between BE simulation measures of alcohol demand and delay discounting and drinking patterns throughout the year after initiation of a natural recovery attempt by untreated persons with alcohol use disorder (AUD). The present research is an extension of our previous research using this sample (Tucker et al., 2016a, 2016b), which showed that baseline measures of APT metrics, particularly intensity and elasticity, predicted initial resolution status (abstinent or drinking moderately) but did not predict terminal drinking status summarized over the entire post-resolution year (continuously resolved abstinent [RA] or resolved non-abstinent [RNA], or unstable resolution [UR] involving relapse). Delay discounting was not associated with either initial or terminal drinking status. Earlier research had established these three outcome groups as molar drinking patterns common in natural recovery samples that evidence reliable differential associations with a range of alcohol problem indicators (e.g., Tucker et al., 2009; cf. Tucker, 2022). Hypotheses in the present study were that higher demand and discounting in a given quarter would predict subsequent drinking and, conversely, that drinking in a given quarter would predict subsequent higher demand and discounting, although the latter association was considered exploratory given the mixed empirical findings.

Method

Participants

Community-dwelling problem drinkers were enrolled shortly after stopping heavy drinking without treatment and were followed prospectively for up to a year. Participants were recruited via media advertisements in the Southeastern United States and were screened by phone and then by questionnaire. Eligibility requirements were age ≥ 21 years; ≥ 2 year drinking problem history assessed using the Michigan Alcoholism Screening Test (MAST; Selzer, 1971), Alcohol Dependence Scale (ADS; Skinner & Horn, 1984), and Drinking Problems Scale (DPS; Cahalan, 1970); recent cessation of heavy drinking, defined as maintaining abstinence or lower risk drinking (<4 standard drinks/day for men, <3 drinks/day for women; National Institute on Alcohol Abuse and Alcoholism [NIAAA], 2005), for at least 3 weeks and up to 3 months without alcohol-focused interventions including professional treatment and mutual help groups (M = 9.99 weeks resolved, SD = 4.34); no current alcohol-related problems or alcohol dependence; and no current nonmedical prescription or other drug use. All study procedures received Institutional Review Board approval, and the research was granted a Certificate of Confidentiality.

Table 1 presents the sample characteristics (N = 191) for demographic, drinking problem severity, and BE indicators. Participant drinking problem histories were similar to outpatient treatment samples (Miller & Munoz, 2013), and 97.9% fulfilled alcohol dependence criteria (American Psychiatric Association, 2000), although this was not an eligibility criterion. Sample sex composition approximated the problem drinker population, and race/ethnicity composition approximated the southern U.S. region where the research was conducted. The vast majority of non-White participants were African American.

Table 1.

Sample characteristics for demographic, drinking practices and problem severity, and behavioral economic indicators

Demographic characteristics at enrollment
Biological sex (n and %)
Male	146 (76.44)
Female	45 (23.56)
Race/Ethnicity (n and %)
White	116 (60.73)
Black	70 (36.65)
Othera	5 (2.62)
Married (n and %)	69 (36.32)
Employed part/fulltime (n and %)	72 (37.89)
Age in years (Mean and SD)	50.09 (11.94)
Education in years (Mean and SD)	14.11 (2.57)
Pre-resolution drinking practices and problem severity (Mean and SD)
Problem duration in years	17.74 (13.06)
Pre-resolution year drinking practices (TLFB)
Days well-functioningb	147.78 (128.60)
Alcohol consumed per drinking day (ml ethanol)	207.16 (195.25)
Alcohol Dependence Scale (0 – 47)	18.90 (9.60)
Drinking Problems Scale (0 – 40)	17.60 (9.66)
Post-resolution year drinking practices
Drinking status at resolution onset (n and %)
RA	152 (79.58)
RNA	39 (20.42)
Post-resolution days to first drinking day
Initial drinking status	1-year drinking statusc	Mean (SD)
RA	RNA	124.14 (83.72)
	UR	99.16 (94.49)
RNA	RNA	20.92 (30.03)
	UR	20.50 (18.26)
RA-RNA combined	RNA	58.95 (74.19)
	UR	53.71 (73.39)
Post-resolution days to first heavy drinking day
Initial drinking status	1-year drinking statusc	Mean (SD)
RA	UR	141.50 (131.36)
RNA	UR	92.47 (70.36)
RA-RNA combined	UR	116.32 (106.02)
Post-resolution year IVR-assessed behavioral economic indicators (Mean and SD)
	1-year drinking statusc	Quarter 1	Quarter 2	Quarter 3	Quarter 4
Delay discounting (log k)d	RA	−1.96 (1.26)	−1.92 (1.11)	−1.90 (1.16)	−1.76 (1.14)
	RNA	−1.79 (1.05)	−1.82 (0.93)	−1.82 (0.91)	−1.82 (1.03)
	UR	−1.76 (1.30)	−1.83 (1.12)	−1.65 (1.12)	−1.85 (1.39)
Intensitye	RA	6.84 (13.29)	3.44 (4.35)	2.84 (4.50)	1.87 (3.41)
	RNA	4.15 (2.35)	3.32(2.55)	2.53 (1.98)	4.50 (6.77)
	UR	6.56 (6.29)	3.90 (2.69)	6.00 (7.46)	7.97 (11.47)
Participants refusing drinks at all pricesf (n and %)	RA	21 (27.27)	36 (37.50)	43 (49.43)	52 (63.41)
	RNA	1 (5.00)	3 (13.64)	3 (15.79)	2 (9.09)
	UR	8 (19.51)	3 (7.32)	3 (8.57)	4 (11.76)
Omaxe	RA	−0.04 (4.52)	−1.01 (4.76)	−2.35 (4.84)	−3.59 (4.64)
	RNA	2.25 (2.30)	0.63 (3.80)	1.06 (3.74)	1.51 (2.93)
	UR	0.76 (3.97)	1.50 (3.02)	1.65 (2.82)	1.61 (3.45)
Pmaxe	RA	−1.07 (3.86)	−1.85 (4.08)	−2.89 (4.23)	−3.96 (4.10)
	RNA	1.16 (1.98)	−0.01 (3.42)	0.40 (3.30)	0.87 (2.70)
	UR	−0.23 (3.43)	0.61 (2.65)	0.72 (2.50)	0.30 (2.87)
Elasticitye	RA	0.02 (0.02)	0.02 (0.09)	0.03 (0.03)	0.02 (0.02)
	RNA	0.02 (0.01)	0.03 (0.02)	0.03 (0.03)	0.03 (0.04)
	UR	0.03 (0.03)	0.04 (0.04)	0.04 (0.02)	0.03 (0.02)

N = 191. Notes: Possible score ranges for scaled questionnaires are given in parentheses after the variable name. Higher Drinking Problems Scale scores indicate greater alcohol-related problems; higher Alcohol Dependence Scale scores indicate greater alcohol dependence levels. TLFB = Timeline Followback interview.

^aOther race/ethnicity group included one Hispanic, one Native American, and three participants who did not specify their race/ethnicity.

^bDays well-functioning = abstinent days plus drinking days < 4 drinks for women and < 5 drinks for men.

^cTerminal drinking status based on drinking practices and consequences over the entire post-resolution year. RA = Resolved Abstinent; RNA = Resolved Non-abstinent; UR = Unstable Resolution.

^dLog-transformed delay discounting assessed quarterly during the post-resolution year using Monetary Choice Questionnaire (MCQ; Kirby & Marakovic, 1996) via Interactive Voice Response (IVR) system.

^eIntensity, O_max, P_max, and elasticity of alcohol demand assessed quarterly via IVR during the post-resolution year using the Alcohol Purchase Task (APT). Means and standard deviations presented by the participant’s terminal drinking status.

^fFrequency and percentage of participants who refused drinks at all prices including $0. Participants who refused drinks when free (i.e., intensity demand = 0) were included in the calculation of means and standard deviations of intensity demand. O_max and P_max were log-transformed.

Resolution onset was the most recent date that participants began abstaining (resolved abstinent [RA]) or drinking moderately (resolved non-abstinent [RNA]) for 3 weeks or more. This 3-week minimum assured obtaining a sample of participants who had made a serious quit attempt but were early in recovery when outcomes are not fixed, thereby facilitating essential variation in quarterly and 1-year drinking status (RA, RNA, or UR [unstable resolution]) to evaluate the hypotheses. At enrollment, 79.6% of participants were RA, and 20.4% were RNA. Based on drinking over the entire follow-up interval (terminal drinking status), among participants with post-resolution drinking data (n = 176; 92.1% of the enrolled sample), 58.5% were RA, 13.6% were RNA, and 27.8% were UR, defined as 1 or more relapses involving daily drinking in excess of NIAAA (2022) heavy drinking thresholds (4+/5+ drinks for women/men). From initial resolution throughout the post-resolution year, 37.5% of participants changed their drinking status at least once; of these, 34.8% changed it more than once. Initially resolved participants (RA or RNA) who began drinking again and drinking heavily, started, on average, during quarters 1 and 2, respectively.

Transparency and Openness

The following sections describe data collection procedures, measures, and how analysis sample sizes were derived, including data exclusions.

Procedures and Measures

Data were collected during two 1.5- to 3.0-hour in-person interviews (baseline and 12-months) and 3 briefer telephone-based assessments at 3, 6, and 9-month follow-ups. The in-person interviews were conducted in locations convenient for participants. After participant sobriety was verified by breathalyzer (Lifeloc FC20, Wheat Ridge, CO), informed consent was obtained, including a request to interview a collateral informant (e.g., spouse) by phone to verify participant reports of eligibility criteria and post-resolution drinking (completed for 82.3% of participants). At enrollment and 12 months later, an expanded Timeline Followback (TLFB) interview (Sobell & Sobell, 1992; Vuchinich et al., 1988) assessed drinking practices, income, and expenditures covering the pre- and post-resolution years, respectively. BE simulation tasks were administered by laptop computer or via an interactive voice response (IVR) using a cell phone. The follow-up phone assessments covered the interval since the previous assessment, followed by IVR administration of the BE measures. Participants received $75 for completing each of the two in-person assessments and $25 for each of the three phone assessments, plus a $50 bonus if they completed all five assessments. Procedures and measures that generated study predictor and dependent variables are described next in order of administration.

Drinking Practices and Assignment of Drinking Status

Using standard procedures involving recall aids, the baseline TLFB assessed daily beer, wine, and liquor consumption in standard drinks for the 12 months prior to resolution onset, defined as the first day of heavy drinking cessation for a minimum of 3 weeks. The follow-up TLFB covered the time since resolution onset through the day preceding the 12-month interview. The 3, 6, and 9-month phone assessments included a TLFB interview covering the interval since the last assessment. Daily reports of drinks consumed were converted to ml of 190-proof ethanol, and the number of days “well-functioning” was computed as the sum of abstinent days and days below sex-specific heavy/binge drinking thresholds during the past year (< 4 drinks for women, < 5 drinks for men). The TLFB is a well-established method to assess daily drinking, and summary variables have good reliability and accuracy unaffected by the length of the recall interval (e.g., Sobell & Sobell, 1992) or by assessment modalities (interview, online, phone) (Falk et al., 2010; Witkiewitz et al., 2015). Moreover, comparisons of time-matched TLFB and IVR reports of drinking including among natural recovery samples show very good concordance (e.g., Simpson et al., 2011; Tucker et al., 2007). Therefore, the final TLFB follow-up assessment, which provided continuous daily drinking reports for the entire post-resolution year for all participants regardless of their adherence to the exact quarterly follow-up schedule, was used to establish participants’ drinking status during each quarter.

Participants were assigned baseline, quarterly, and terminal drinking statuses based on data from TLFB reports coupled with DPS and ADS assessments of alcohol-related consequences if any drinking was reported at any follow-up. Quarterly and terminal drinking statuses were computed over 3-month intervals and the entire post-resolution year, respectively. Possible statuses were: (1) resolved abstinent (RA) for intervals involving continuous abstinence; (2) resolved non-abstinent (RNA) for intervals involving drinking below sex-specific heavy/binge drinking thresholds and no reported alcohol-related problems; or (3) unresolved (UR) for intervals involving one or more relapse episodes exceeding heavy/binge drinking thresholds. For participants who could not be reached at 12-months, their terminal drinking status was based on partial data assessed by phone at 9-months (n = 18), 6-months (n = 15), or 3-months (n = 1 who relapsed during the first 3 months), whichever was most recent; 15 participants had insufficient follow-up data to assign a terminal drinking status.

Monetary Choice Questionnaire

Delay discounting was assessed using the 21-item Monetary Choice Questionnaire (MCQ; Kirby & Marakovic, 1996). Each item prompted a choice between an immediate amount of money or a larger, delayed amount of money (e.g., $110 today or $300 in 7 days?). Participants were instructed: “Although this is an imaginary choice, try to choose as if these were real dollars available to you at different time delays.” The monetary amounts and delays varied across items. An overall discounting rate (k) was computed from the pattern of choices made across items, which reflects the slope of the hyperbolic discount function characteristic of reward devaluation over time (Mazur, 1987). Higher k-parameters indicate more immediate reward preferences.

Alcohol Purchase Task

Demand for alcohol was assessed using a standard APT (Murphy & MacKillop, 2006) that assessed how many standard drinks one would consume across a range of drink prices ($0-$20). Participants received the following written instructions:

The next set of questions asks how many drinks you would consume if they cost different amounts of money. Imagine that you are drinking in a typical situation when you drink. Assume you did not drink any alcohol before making these decisions. Also assume that you will drink every drink that you request. You cannot stockpile or bring drinks home with you. The available drinks are in a standard size, either a 12 oz. domestic beer, a 5 oz. wine, a 1.5 oz. shot of hard liquor, or mixed drinking containing 1 shot of liquor. Press the numbers on the phone keypad to indicate the number of drinks you would have at each price and enter zero for no drinks. Please respond honestly, as if you were actually in each situation.

Respondents then reported how many drinks they would consume across increasing drink prices ($.00, $0.25, $0.50, $1.00, $1.50, $2.00, $2.50, $3.00, $4.00, $4.50, $5.00, $6.00, $7.00, $8.00, $9.00, $10.00, $12.00, $15.00, $20.00). APT metrics of alcohol value were computed for analysis, including three observed (intensity, Omax, Pmax) and one derived (elasticity) (MacKillop et al., 2009; Murphy & MacKillop, 2006). Elasticity was calculated using the beezdemand package in R (Kaplan, 2018) by fitting the Hursh and Silberberg (2008) demand model to systematic APT data.

Data Analysis

The analysis sample was determined based on sufficient drinking data and systematic APT response patterns in line with the recommendations of Stein et al. (2015). Because the MCQ does not involve establishing equivalence points, discounting data cleaning procedures like Johnson and Bickel’s (2008) do not apply; further, the outlier threshold does not apply since a k value >1 is impossible on the MCQ. Because k-parameters are skewed, they were natural log-transformed for analyses. Of the 191 participants, 15 participants were excluded who did not have sufficient drinking data to determine their drinking status at 12 month follow-up to eliminate potential prediction inaccuracy. Elasticity was calculated by excluding participants who had non-systematic responses (between 12% to 20% depending on the assessment timepoint) or insufficient data points (between 16% and 28% depending on the assessment timepoint) to estimate elasticity, which was defined as the following: individuals providing less than five non-zero consumption values across prices, evidencing more than one increase in consumption across increasing prices, a lack of sensitivity to price (i.e., all consumption the same non-zero value), and/or had more than one missing consumption value across all prices. Nonsystematic APT data were determined on a quarterly basis. For non-purchasers who refused drinks at all prices, O_max and P_max were coded as 0. Due to excess skewness and kurtosis, O_max and P_max were also natural log-transformed for analyses after adding a small constant (i.e., .001) to handle the value of 0.

Longitudinal analyses were then conducted using Mplus v.8 (Muthén & Muthén, 1998–2017) following the approach used in Tucker et al. (2021). To examine post-resolution quarter-to-quarter bidirectional relations between BE indicators and drinking status, cross-lagged models were estimated separately for each BE indicator by relating the indicator assessed in a given quarter (time t) to drinking status in the subsequent quarter (time t + 1), and vice versa. As drinking status was a categorical variable with three groups (RA, RNA, UR), it could not be treated simultaneously as a predictor and an outcome in the cross-lagged models with all four quarters of data. Therefore, drinking status was coded as two dummy variables for inclusion as predictors of BE indicators, while specifying it as a nominal categorical outcome predicted by BE indicators. Models were estimated for two time points at a time (i.e., quarters 1 and 2, 2 and 3, and 3 and 4, separately). For example, to include drinking status at quarter 1 as a predictor of the BE indicator at quarter 2, two dummy variables were created with the UR group as the referent (i.e., comparing RA to UR and RNA to UR), and drinking status at quarter 2 was specified as a nominal categorical outcome with the UR group as the referent, allowing for multinomial logistic regression differentiating RA and RNA from UR by the BE indicator at quarter 1. Models estimated for two time points were then put together in Figure 1 to present the results comprehensively over four quarters.

Figure 1.

Cross-lagged models for quarter-to-quarter associations between drinking status and behavioral economic indicators with significant associations during the post-resolution year. Significant paths are shown in bold (*p < .05, ** p < .01, *** p < .001). Reference group is UR: RA (1) vs. UR (0); RNA (1) vs. UR (0).

Participants who had partial missing data (e.g., missing data at one of two quarters in the cross-lagged models) were included in the analyses, and missing data were adjusted for using the full information maximum likelihood method implemented in Mplus. To avoid potential biases due to non-normality of variables, MLR, a robust estimator, was used. Conventional fit statistics were not available due to the multinomial logistic regression part in the models predicting drinking status with BE indicators. Because of the extent of missing data on elasticity (due largely to drink refusal across all prices and some non-systematic response patterns), the analysis sample sizes for models for elasticity were not suitable for a structural equation modeling framework. Thus, models for elasticity were estimated in SAS 9.4, by using ordinary least squares (OLS) regression for predicting the continuous BE indicators with drinking status and multinomial logistic regression for predicting drinking status with BE indicators.

Analysis sample sizes varied depending on BE indicators and time points due to different number of cases missing on all outcome variables. Sample sizes for delay discounting were 168, 162, and 159 for models involving quarters 1 and 2, quarters 2 and 3, and quarters 3 and 4, respectively. The respective APT sample sizes were 169, 162, and 160 for intensity and 168, 160, and 159 for O_max and P_max. Sample sizes for the elasticity models were further reduced due to the exclusion of participants for whom elasticity demand could not be calculated. For OLS regressions predicting elasticity with drinking status, the sample sizes were 45, 40, and 35 for models involving quarters 1 and 2, quarters 2 and 3, and quarters 3 and 4, respectively. For multinomial logistic regressions predicting drinking status with elasticity, the sample sizes were 83, 67, and 54, respectively. All models consistently included baseline covariates selected based on the recovery literature and our prior studies (e.g., Tucker et al., 2009, 2016a, 2016b) to control for potential confounders. Covariates were demographic variables (sex, age, white/nonwhite race, married/unmarried status, education), pre-resolution drinking-related characteristics (problem duration, ADS scores, pre-resolution year days well functioning), and initial resolution status (RA or RNA). Materials and analysis code for this study are available to qualified investigators by emailing the corresponding author.

Results

Descriptive information about post-resolution year drinking patterns and behavioral economic indicators are presented in Table 1 quarter-by-quarter for each terminal drinking status group. Of note, a sizeable minority of participants refused drinks at all prices (22% to 42% depending on the quarter), and drink refusal was more common in the RA than RNA or UR groups (for quarters 1 to 4, respectively, χ²(2) = 4.80, 15.69, 21.85, and 37.97; p = .091 for quarter 1; ps <.001 for quarters 2 to 4).

Figure 1 presents the hypothesis-relevant quarter-to-quarter cross-lagged associations between BE indicators and drinking status and shows the significant paths in bold with associated p-values (for intensity, O_max, P_max, and log k). As hypothesized, some associations between intensity of alcohol demand and drinking status were found to be bidirectional. Participants who had higher intensity at quarter 1 were more likely to be UR at quarter 2 than RA (p = .011). In turn, UR participants at quarter 2 had higher intensity at quarter 3 compared to RA (p = .004) and RNA (p = .017) participants at quarter 2. Further, participants who had higher intensity at quarter 3 were more likely to be UR at quarter 4 than RNA (p = .044).

Some significant associations were found for other BE indicators in the hypothesized direction, but none were bidirectional in nature. Drinking status predicted O_max and P_max during subsequent quarters, but O_max and P_max did not predict subsequent drinking status. Compared to RA participants, UR participants at quarter 2 showed higher O_max at quarter 3 (p <.001). For P_max models, UR compared to RA participants showed higher P_max at subsequent quarters across quarters 1 to 3 (ps <.05). In addition, RNA participants at quarter 3 showed higher P_max at quarter 4 compared to UR participants (p = .021). Elasticity of alcohol demand was not associated with drinking status at any point during the post-resolution year.

For delay discounting (log k), the only significant and unexpected association was that participants with greater discounting at quarter 2 were more likely to be RNA than UR at quarter 3 (p = .010). None of the other associations were significant.

Discussion

Some support was found for hypothesized bidirectional associations between post-resolution drinking status and demand intensity assessed by the APT, which reflects alcohol consumption levels when drinks are free and unconstrained by monetary price. Intensity predicted shifts in molar drinking patterns during the middle of the year following a natural recovery attempt when drinking patterns were changing in the direction of resuming or increasing drinking (per Table 1). Participants who had relapsed mid-year had higher intensity during the next quarter compared to those who were abstinent or drinking moderately. Higher intensity also distinguished those who relapsed from those who were abstinent or drank moderately. The overall pattern of intensity results provided prospective support for select hypothesized bidirectional associations between BE indicators as predicted by the reinforcer pathology model (Bickel et al., 2014).

A related interesting APT finding was that a sizeable minority of participants (> 20% of the sample) refused to purchase drinks, even when free, and drink refusal was significantly more likely among RA participants. Although previous research with active drinking risk groups have not reported APT drink refusals at these levels, Meshesha et al. (2020) found similar elevated APT drink refusal levels (32.5%) at baseline in a study of women seeking outpatient treatment for AUD, and intensity was significantly associated with higher drinks per drinking day. As the authors noted, refusing hypothetical drinks, even when free, aligned with participants’ goal to reduce or stop drinking. Thus, drink refusal on the APT may occur among a sizeable minority of persons with AUD who are in active recovery given that this pattern has been observed in both a clinical and natural recovery sample. Taken together, these results suggest that intensity as a measure offers unique advantages in clinical and field assessments. A brief simple question—”How much would you drink if drinks are free?”—appears to provide a meaningful estimation of the likelihood of continued abstinence or resumption of drinking, including heavy drinking, among persons with AUD who are trying to recover.

Other BE indicators showed some limited significant associations in the hypothesized direction in the RA vs. UR comparisons, but none were bidirectional in nature. UR drinking status at one quarter predicted higher O_max and P_max during subsequent quarters, but O_max and P_max did not predict subsequent drinking status. One significant association involving the RNA group was unexpectedly found late in the post-resolution year; i.e., RNA participants at quarter 3 showed higher P_max at quarter 4 compared to UR participants, suggesting that by the end of the year they were willing to pay relatively higher prices for the limited number of drinks they purchased. Elasticity of alcohol demand was not associated with drinking status at any point during the post-resolution year. These non-significant elasticity results are likely due to the reduced sample sizes necessitated by the exclusion of participants for whom valid elasticity estimates could not be calculated.

No support was found for hypothesized bidirectional associations between delay discounting and drinking status. A potential reason for these negative findings is that the present sample was generally middle-aged, and discounting rates tend to be steeper at younger ages and typically stabilize in established adulthood (Read & Read, 2004), thus reducing variability necessary to detect associations with shifts in drinking. Unexpectedly, however, higher delay discounting mid-year was associated with subsequent moderation drinking compared to heavy drinking. No directional associations opposite to predictions were found for the log k comparisons between the UR and RA status groups.

Thus, the two directional associations that were counter to hypotheses both involved the RNA group, which is more common in natural recovery than treatment-seeking samples, and both occurred after RNA participants had maintained moderate drinking for many months. In our related research using additional samples that focused on shifts in monetary expenditure patterns during the post-resolution year (Tucker et al., 2021), the RNA group also showed distinctive shifts in spending compared to both the RA and UR outcome groups during the second half of the post-resolution year. These shifts were characterized by higher expenditures on big ticket items (i.e., housing, durable goods, and related insurance) and lower expenditures on savings and financial planning. These dynamic changes unique to the RNA group were interpreted as their shifting spending in ways that yielded higher overall alcohol-free rewards that reinforced recovery behaviors while they enjoyed some limited drinking without problems. In addition to suggesting that moderation entails different behavioral regulation processes than abstinent and relapse outcomes, these monetary allocation shifts point to the importance of considering alternative substance-free reinforcement in BE accounts of substance use disorders and recovery.

The role of alternative reinforcement has long been heavily emphasized in molar BE accounts of substance use (e.g., Rachlin, 1997; Vuchinich, 1995; cf. Tucker et al., this issue), although it is de-emphasized in the reinforcer pathology model that stresses the role of drug demand and delay discounting, in some accounts to the point of reification as critical causal mechanisms. This is increasingly considered a serious limitation of the reinforcer pathology model given available evidence (e.g., Bailey et al., 2021), and the less than robust support found in this research for bidirectional BE-drinking associations predicted by the model adds to this concern and directs attention to considering the role of alternative reinforcers (Tucker et al., in press). In the present study, among APT metrics only intensity evidenced some of the predicted bidirectional associations, which is generally consistent with prior research selectively supporting the utility of intensity to predict drinking and drug-related outcomes in clinical and experimental research (e.g., Acuff et al., 2020; Kiselica et al., 2016; Murphy et al., 2009; Strickland et al., 2019).

Limitations of the research qualify the present findings. First, in addition to the reduced sample sizes for elasticity analyses, the number of participants with moderation drinking statuses at enrollment and during the post-resolution year was modest compared to participants who abstained or relapsed. This further qualifies the non-significant associations observed for several BE variables and merits investigation using larger samples of lower risk drinkers in addition to participants who abstain or relapse. Also as noted, the associations of interest warrant study using participants with a broader age range, particularly younger risky drinkers who tend to have higher discount rates and are in a developmental period when “maturing out” of heavy drinking is normative. Finally, the present discounting task was limited to monetary choices at different delays. Given evidence that discounting patterns differ for different commodities (Odum & Rainaud, 2003) and that substance use varies with the availability of substance-free alternatives (Acuff et al., 2022; Murphy et al., 2019), investigating associations of shifts in drinking patterns during recovery that vary with discounting and the availability of commodities other than money should be a future research priority.

With these qualifications, the present study suggests that demand intensity has some prospective utility to predict shifts in quarterly drinking patterns during a natural recovery attempt, and vice-versa. This adds to evidence of its ecological validity and predictive utility for drinking-related outcomes in different drinking risk populations. To our knowledge, this is the first prospective study to assess bidirectional BE-drinking associations at regular intervals during the first year of a recovery attempt. Overall, the less than robust bidirectional associations for demand and discounting indicators do not provide much support for the reinforcer pathology model. In line with Rachlin’s (e.g., 1992, 1997) emphasis on relating contextual environmental features to molar patterns of behavior, greater emphasis on the role of alternative reinforcers and substitute/complement relationships in understanding and changing addictive behavior is recommended (Tucker et al., in press).

Public Health Significance.

Most persons with alcohol-related problems do not seek help, many recover on their own, and studying natural recoveries has potential for increasing the appeal, effectiveness, and population impact of services for alcohol-related problems. This prospective study of natural recovery provided some support that alcohol demand metrics using the Alcohol Purchase Task tracked shifts in drinking patterns during the year following initial cessation of heavy drinking without treatment. Alcohol demand metrics—particularly intensity or consumption when drinks are free—may be useful in clinical and field settings to assess the likelihood of maintaining recovery or resuming heavy drinking among persons attempting to recover.